Networking of differentially expressed genes in human cancer cells resistant to methotrexate

Background The need for an integrated view of data obtained from high-throughput technologies gave rise to network analyses. These are especially useful to rationalize how external perturbations propagate through the expression of genes. To address this issue in the case of drug resistance, we constructed biological association networks of genes differentially expressed in cell lines resistant to methotrexate (MTX). Methods Seven cell lines representative of different types of cancer, including colon cancer (HT29 and Caco2), breast cancer (MCF-7 and MDA-MB-468), pancreatic cancer (MIA PaCa-2), erythroblastic leukemia (K562) and osteosarcoma (Saos-2), were used. The differential expression pattern between sensitive and MTX-resistant cells was determined by whole human genome microarrays and analyzed with the GeneSpring GX software package. Genes deregulated in common between the different cancer cell lines served to generate biological association networks using the Pathway Architect software. Results Dikkopf homolog-1 (DKK1) is a highly interconnected node in the network generated with genes in common between the two colon cancer cell lines, and functional validations of this target using small interfering RNAs (siRNAs) showed a chemosensitization toward MTX. Members of the UDP-glucuronosyltransferase 1A (UGT1A) family formed a network of genes differentially expressed in the two breast cancer cell lines. siRNA treatment against UGT1A also showed an increase in MTX sensitivity. Eukaryotic translation elongation factor 1 alpha 1 (EEF1A1) was overexpressed among the pancreatic cancer, leukemia and osteosarcoma cell lines, and siRNA treatment against EEF1A1 produced a chemosensitization toward MTX. Conclusions Biological association networks identified DKK1, UGT1As and EEF1A1 as important gene nodes in MTX-resistance. Treatments using siRNA technology against these three genes showed chemosensitization toward MTX.

M Me et th ho od ds s: : Seven cell lines representative of different types of cancer, including colon cancer (HT29 and Caco2), breast cancer (MCF-7 and MDA-MB-468), pancreatic cancer (MIA PaCa-2), erythroblastic leukemia (K562) and osteosarcoma (Saos-2), were used. The differential expression pattern between sensitive and MTX-resistant cells was determined by whole human genome microarrays and analyzed with the GeneSpring GX software package. Genes deregulated in common between the different cancer cell lines served to generate biological association networks using the Pathway Architect software. R Re es su ul lt ts s: : Dikkopf homolog-1 (DKK1) is a highly interconnected node in the network generated with genes in common between the two colon cancer cell lines, and functional validations of this target using small interfering RNAs (siRNAs) showed a chemosensitization toward MTX. Members of the UDP-glucuronosyltransferase 1A (UGT1A) family formed a network of genes differentially expressed in the two breast cancer cell lines. siRNA treatment against UGT1A also showed an increase in MTX sensitivity. Eukaryotic translation elongation factor 1 alpha 1 (EEF1A1) was overexpressed among the pancreatic cancer, leukemia and osteosarcoma cell lines, and siRNA treatment against EEF1A1 produced a chemosensitization toward MTX. C Co on nc cl lu us si io on ns s: : Biological association networks identified DKK1, UGT1As and EEF1A1 as important gene nodes in MTX-resistance. Treatments using siRNA technology against these three genes showed chemosensitization toward MTX.

B Ba ac ck kg gr ro ou un nd d
The large amount of information obtained with highthroughput technologies like expression microarrays needs to be processed in order to be comprehensible to molecular biologists. In this regard, many computational methods have been developed to facilitate expression data analysis. Gene clustering, gene ontology and pathway analyses are commonly used [1,2]. Pathways are manually generated diagrams that represent knowledge on molecular interactions and reactions [3] and they can be used to visualize the involvement of the differentially expressed genes in specific molecular, cellular or biological processes. However, the complexity of higher organisms cannot be explained solely as a collection of separate parts [4]; in organisms, pathways never exist in isolation, they are part of larger networks, which are more informative and real [5]. Gene networks are capable of describing a large number of interactions in a concise way, and provide a view of the physiological state of an organism at the mRNA level. Biochemical networks can be constructed at several levels and can represent different types of interactions. Literature mining allows the extraction of meaningful biological information from publications to generate networks [6]. Taking into account the progress in gene expression profiling, elucidating gene networks is an appropriate and timely step on the way to uncovering the complete biochemical networks of cells [5].
In this work, we use biological association networks (BANs) as a tool to define possible targets for gene therapy in combination with methotrexate (MTX). This approach could serve to minimize the development of MTX resistance acquired by cancer cells, which remains a primary cause of therapy failure in cancer treatment [7]. A role in MTX resistance was established for the three node genes selected, namely those encoding Dikkopf homolg 1 (DKK1), UDP-glucuronosyltransferases (UGTs; UGT1As) and Eukaryotic translation elongation factor 1A1 (EEF1A1). Human cell lines were routinely grown in Ham's F12 medium supplemented with 7% fetal bovine serum (both from Gibco/Invitrogen, Grand Island, NY, USA) at 37°C in a 5% CO 2 humidified atmosphere. Resistant cells were routinely grown in selective DHFR medium lacking glycine, hypoxanthine and thymidine (-GHT medium; Gibco), the final products of dihydrofolate reductase (DHFR) activity. This medium was supplemented with 7% dialyzed fetal bovine serum (Gibco).
M Mi ic cr ro oa ar rr ra ay ys s Gene expression was analyzed by hybridization to the GeneChip ® Human Genome U133 PLUS 2.0 from Affymetrix, containing over 54,000 transcripts and variants. Total RNA for oligo arrays was prepared from triplicate samples of every sensitive and resistant cell line using the RNAeasy Mini kit (Qiagen, Germantown, Maryland, USA) following the recommendations of the manufacturer. Labeling, hybridization and detection were carried out following the manufacturer's specifications.
M Mi ic cr ro oa ar rr ra ay y d da at ta a a an na al ly ys se es s Gene expression analyses were performed using three samples of both sensitive and resistant cells for each of the seven cell lines studied. These analyses were carried out with the GeneSpring GX software v 7.3.1 (Agilent Technologies, Santa Clara, CA, USA), using the latest gene annotations available (March 2009). This software package allows multi-filter comparisons using data from different experiments to perform the normalization, generation of restriction (filtered) lists and functional classifications of the differentially expressed genes. Normalization was applied in two steps: 'per chip normalization', by which each measurement was divided by the 50th percentile of all measurements in its array; and 'per gene normalization', by which all the samples were normalized against the median of the control samples (sensitive cells). The expression of each gene was reported as the ratio of the value obtained for each condition relative to the control condition after normalization of the data. Then, data were filtered using the control strength, a control value calculated using the Cross-Gene Error model on replicates [9] and based on average base/proportional value. Measurements with higher control strength are relatively more precise than measurements with lower control strength. Genes that did not reach this value were discarded. Additional filtering was performed to determine differentially expressed genes. A first filter was performed by selecting the genes that displayed a Pvalue corrected by false discovery rate (Benjamini and Hochberg false discovery rate) of less than 0.05. The output of this analysis was then filtered by fold expression. Thus, lists of genes differentially expressed by at least twofold were generated for each of the seven resistant cell lines.
C Co om mm mo on n g ge en ne es s b be et tw we ee en n c ce el ll l l li in ne es s The lists of genes differentially expressed by at least twofold with a P-value <0.05 including multiple testing correction for each cell line were divided into two groups: overexpressed and underexpressed genes. Comparisons of lists of overexpressed genes were performed using Venn diagrams in GeneSpring GX. Lists of underexpressed genes were also compared using the same approach. All lists were compared in pairs and lists of genes in common between each pair were generated.
G Ge en ne er ra at ti io on n o of f b bi io ol lo og gi ic ca al l a as ss so oc ci ia at ti io on n n ne et tw wo or rk ks s BANs were constructed with the aid of Pathway Architect software v3.0 (Stratagene-Agilent). Briefly, this software package generates interaction networks starting with the genes in a given list (entities) taking into account the information present in a database of known molecular interactions. The lists correspond to the collection of differentially expressed genes under specific conditions. The database of molecular interactions is composed of more than 1.6 million interactions divided into different classes (binding, regulation, promoter binding, transport, metabolism, protein metabolism and expression). The interactions are extracted from the literature using a Natural Language Processing (NLP) tool run on Medline abstracts (NLP references), plus those obtained from external curated databases like BIND (Biomolecular Interaction Network Database) [10] and MINT (Molecular INTeraction) [11]. Interactions in the interaction database are scored according to five different categories: maximum, high, medium, low and minimal. Curated interactions (BIND and MINT sources) are given the maximum quality score, as are any interactions that have at least three NLP references. Pathway Architect gathers all that information to construct novel views as to how the entities in a list could be interacting with each other, even including entities not present in the original list (neighbors resulting from the expanded interaction). Customized analyses were performed to select relevant interaction networks with an associated high confidence index since such networks are likely to mirror biological significance. One-step expansion (using the expand network command) of the original set of entities with maximum score interaction were then analyzed by setting an advanced filter that included the categories of binding, expression, metabolism, promoter binding, protein modification and regulation. This procedure gives a final representation formed of a collection of nodes with different degrees of interrelationship. Some gene products from the original lists were not significantly connected with other members or neighbors and, therefore, were removed from the final view. Finally, members of the network were matched with expression levels.
T Tr ra an ns sf fe ec ct ti io on n o of f s sm ma al ll l i in nt te er rf fe er ri in ng g R RN NA As s a ag ga ai in ns st t s se el le ec ct te ed d g ge en ne es s HT29 cells (30,000) were plated in 1 ml of -GHT medium and transfection was performed 18 hours later. For each well, Lipofectamine™ 2000 (Invitrogen, Carlsbad, CA, USA) in 100 µl of serum free -GHT medium was mixed in Eppendorf tubes with 100 nM of small interfering RNA (siRNA) in 100 µl of serum free -GHT medium. The mixture was incubated at room temperature for 20 minutes before addition to the cells. MTX (2 x 10 -8 M) was added 48 hours after siRNA treatment and 3-(4,5-Dimethylthiazol-2-yl)-2,5diphenyltetrazolium bromide (MTT) assays [12] were performed 3 days after MTX addition. Treatment of MDA-MB-468 and MIA PaCa-2 cells was performed following the same protocol but using Metafectene™ (Biontex, Martinsried, Germany). A non-related siRNA was used as negative control; it was transfected in parallel with the other siRNAs and used to normalize the results.
The siRNAs were designed using the software iRNAi v2.1. (The Netherlands Cancer Institute, Amsterdam, The Netherlands) Among the possible alternatives, sequences rich in A/T at the 3' end of the target were chosen. Then, BLAST resources in NCBI were used to assess the degree of specificity of the sequence recognition for these siRNAs. Only the siRNAs that reported the target gene as the only mRNA hit, or some family members in the case of siUGT1A, were selected. The sequences for the sense strand of all siRNAs used are available in Table 1.
H He ea at t m ma ap p g ge en ne er ra at ti io on n A global comparison of all cell lines was performed using GeneSpring GX v 7.3.1. The triplicate samples for each condition in each of the seven cell lines (42 samples) were imported into one single experiment. Normalization was performed in two steps: 'per chip normalization' (as described above) and 'per gene normalization', by which the samples were normalized against the median of all samples. Lists of genes displaying a false discovery rate-corrected Pvalue <0.05 were generated for each cell line. As a filter, these values had to appear in at least two out of the seven cell lines. A hierarchical clustering method in GeneSpring was used to group genes on the basis of similar expression patterns over all samples. The distance matrix used was Pearson correlation, and average linkage was used as clustering algorithm. The same clustering method was used to group the cell lines on the basis of similar patterns of gene expression.

R Re ea al l--t ti im me e R RT T--P PC CR R
Gene mRNA levels were determined by real-time RT-PCR. Total RNA was extracted from cells using Ultraspec™ RNA reagent (Biotecx, Houston, TX, USA) following the recommendations of the manufacturer. For determining genenode mRNA levels upon siRNA treatment, cells were treated as described above and total RNA was prepared 48 hours after transfection using the same reagent. In either case, complementary DNA was synthesized in a total volume of 20 µl by mixing 500 ng of total RNA and 125 ng of random hexamers ( One hundred nanograms of DNA and the assays-on-demand Hs00758822_s1 for DHFR and Hs99999901_s1 for 18S were used for real-time PCR amplification.
P Pr re ep pa ar ra at ti io on n o of f t to ot ta al l e ex xt tr ra ac ct ts s f fo or r w we es st te er rn n b bl lo ot tt ti in ng g Total extracts from cells, either sensitive or MTX-resistant, were used to assay DHFR protein levels. Cells were washed twice with ice-cold phosphate-buffered saline and scraped in 200 ml lysis buffer (50 mM Hepes, 500 mM NaCl, 1.5 M MgCl 2 , 1 mM EGTA, 10% glycerol (v/v), 1% Triton X-100 and protease inhibitor cocktail). Cells were incubated in ice for 1 hour with vortexing every 15 minutes and then centrifuged at 14,000 rpm at 4°C for 10 minutes. Five microliters of the extract were used to determine protein concentration by the Bradford assay (Bio-Rad, Munich, Germany). The extracts were frozen in liquid N 2 and stored at -80°C. Fifty micrograms of both sensitive and resistant cell total extracts were resolved by 15% SDS-PAGE. Transference to PVDF membranes (Immobilon P, Millipore, Bedford, MA, USA) using a semidry electroblotter was followed by incubation with an antibody against DHFR (Davids Biotechnologie, Regensburg, Germany), and detection was accomplished using secondary horseradish peroxidase-conjugated antibody and enhanced chemiluminescence, as recommended by the manufacturer (Amersham/GE Healthcare, Buckinghamshire, UK). To normalize the results, blots were re-probed with an antibody against actin (Sigma, St. Louis, MO, USA).
F Fi ig gu ur re e 1 1 Heat map of differentially expressed genes. Lists of differentially expressed genes with a t-test P-value <0.05 including multiple testing correction were generated for each cell line. A hierarchical clustering method in GeneSpring GX v 7.3.1 was used to construct both the gene tree and the sample tree, as described in Methods. Data are shown in a matrix format: each row represents a single gene, and each column represents a cell line. Red indicates overexpressed genes (expression levels over the median) and green indicates underexpressed genes (expression levels under the median; see legend). The pattern and length of the branches in the dendrograms reflect the relatedness of the samples or the genes.
T Tr ra an ns sf fe ec ct ti io on ns s, , c co o--t tr ra an ns sf fe ec ct ti io on ns s a an nd d l lu uc ci if fe er ra as se e a as ss sa ay ys s HT29 cells, either sensitive or MTX-resistant, were seeded into six-well plates the day before transfection at a density of 2 x 10 5 cells/well in Ham's F12 medium containing 7% fetal bovine serum. Transfection was performed using FUGENE™ HD (Roche). For each well, 6 µl of FUGENE™ HD in 100 µl of serum-free medium was incubated at room temperature for 5 minutes. The mixture was added to 1 µg of TOPFLASH (Millipore) and incubated at room temperature for 20 minutes before addition to the cells. In co-transfections, 1 µg of TOPFLASH was mixed together with 2 µg of pBATEM2-CDH before the addition of FUGENE™ HD in serum-free medium. The total amount of DNA was kept constant at 3 µg, adding empty vector when necessary. Luciferase activity was assayed 30 hours after transfection.
In all cases, cell extracts were prepared by lysing the cells with 200 µl of freshly diluted 1x Reporter Lysis Buffer (Promega). The lysate was centrifugated at 13,000 g for 2 minutes to pellet the cell debris and the supernatants were transferred to a fresh tube. A 15-µl aliquot of the extract was added to 15 µl of the luciferase assay substrate (Promega) and the luminiscence of the samples was read immediately on a Gloomax 20/20 luminometer (Promega); light production (relative light units) was measured for 10 s. Each transfection was performed in triplicate. Protein concentration was determined by the Bradford assay and used to normalize the results.

S St ta at ti is st ti ic ca al l a an na al ly ys se es s
Data are presented as mean ± standard error (SE). Statistical analyses were performed using the unpaired t-test option in GraphPad InStat version 3.1a for Macintosh (GraphPad Software). P-values <0.05 were considered to be statistically significant.
R Re es su ul lt ts s G Ge en ne es s d de er re eg gu ul la at te ed d i in n m me et th ho ot tr re ex xa at te e--r re es si is st ta an nt t c ca an nc ce er r c ce el ll l l li in ne es s In a previous study, we analyzed the differential gene expression between sensitive and MTX-resistant cells derived from the human colon cancer cell line HT29 We compared the gene expression patterns of all the studied cell lines together. Lists of genes displaying a false discovery rate-corrected P-value <0.05 were generated for each cell line. Then, hierarchical clustering in GeneSpring GX was used to construct a heat map displaying both a gene tree and a sample tree (Figure 1), as described in Methods. Two facts could be extracted from this representation. First, there is a high correlation between cell lines sharing the same tissue origin. The two colon cancer cell lines studied (HT29 and Caco-2) are more highly correlated in gene expression with each other than with all the other cell lines. The breast cancer cell lines studied (MCF-7 and MDA-MB-468) also showed similar gene expression, although the degree of correlation is slightly lower than that for the colon cancer cell lines. The other three cell lines studied (MIA PaCa-2, K562 and Saos-2) displayed different gene expression from the colon or the breast cancer cell lines, and thus cluster apart from them. Second, gene expression of the resistant cells is more closely correlated with that of their sensitive cell counterparts than with any other sample or cell line. T Ta ab bl le e 3 3 G Ge en ne es s d di if ff fe er re en nt ti ia al ll ly y e ex xp pr re es ss se ed d i in n c co om mm mo on n a am mo on ng g c co ol lo on n c ca an nc ce er r c ce el ll l l li in ne es s r re es si is st ta an nt t t to o M MT TX X D Di ih hy yd dr ro of fo ol la at te e r re ed du uc ct ta as se e s st ta at tu us s i in n a al ll l c ce el ll l l li in ne es s s st tu ud di ie ed d As DHFR is the target for MTX, and was upregulated in MTX-resistant cells, we validated its overexpression in all cell lines studied. Real-time RT-PCR was used to quantify the mRNA levels, and DHFR protein levels were determined by western blotting in both sensitive and MTX-resistant cell lines (Table 2). Copy number determination revealed dhfr amplification only in HT29, Caco-2, MCF-7 and MIA PACA-2 resistant cells (Table 2).  (Tables 3, 4 and 5, respectively). This approach allowed us to identify differentially expressed genes with a common trend in expression among the cell lines compared. BANs were constructed using the Pathway Architect software as described in Methods starting with the lists of genes differentially expressed in common between both colon cancer cell lines, both breast cancer cell lines and among the pancreas cancer, leukemia and osteosarcoma cell lines (Figure 2a,b,c, respectively). In the BANs generated, DKK1 is a highly interconnected node in the colon cancer cell lines, UGT1A family members formed a network of genes differentially expressed in breast cancer, and EEF1A1 was commonly overexpressed in pancreatic cancer, leukemia and osteosarcoma. A BAN including all the genes of the three lists of differentially expressed genes was constructed ( Figure 3). DKK1, UGT1A and EEF1A1 all seemed to be important nodes of this newly constructed network, and thus were selected for further study. Given that the node genes DKK1, UGT1As and EEF1A1 were overexpressed in cells resistant to MTX (Table 6), we investigated the effect of decreasing their mRNA levels by means of siRNAs on the sensitivity to this chemotherapeutic agent. We also performed treatments with siDHFR in order to assess the role of DHFR in MTX resistance. HT29 and MDA-MB-468 cell lines were used as models of colon and breast cancer, respectively, and MIA PaCa-2 cells were selected as the model for the other three cell lines. Previously, it was confirmed that the mRNA levels of the three genes were decreased 48 hours after siRNA treatment (Figure 4a-c). Cells were pre-incubated with individual siRNAs for 48 hours before the addition of methotrexate. The presence of 100 nM of either siDKK1 or siDHFR caused increases in MTX cytotoxicity in HT29 cells of 50% and 65%, respectively, com-  Given that the UGT1A family is involved in the metabolism of other drugs, we also performed combination treatments with siUGT1A and SN38, the active metabolite of the anticancer drug irinotecan. Transfection of the siRNA was performed as described above, and 1 nM SN38 was added 48 hours after siRNA treatment. These incubations led to a significant (P < 0.01) increase of 46% in SN38 sensitivity.
A non-related siRNA against the luciferase gene was used as a negative control in all experiments. Transfection of this siRNA was performed in parallel with the other siRNAs, and was used to normalize the results. As DKK1 is known to be transcriptionally regulated by the Wnt pathway, we investigated the degree of activation of this signaling pathway in MTX-resistant HT29 cells compared with their sensitive counterpart. Cells were transiently transfected with the reporter plasmid TOPFLASH, bearing three T-cell factor (TCF) binding sites. A transcriptional activation of 26-fold resulted from the transfection of TOPFLASH in the resistant cells, while no significant activation was observed upon transfection in the sensitive cells ( Figure 5). Additionally, co-transfections of TOPFLASH with an expression plasmid for E-cadherin (pBATEM2-CDH) were performed in both cell lines. As shown in Figure  5, overexpression of E-cadherin in the resistant cells led to a marked decrease in TOPFLASH activity, down to basal activity. No significant changes in transcriptional activity were observed when these co-transfections were performed in the sensitive cells.  Genes differentially expressed by at least twofold with a P-value <0.05 including multiple testing correction were compared between MCF7 and MDA-MB-468 cells as described in Table 3.  Genes differentially expressed by at least twofold with a P-value <0.05 including multiple testing correction were compared between MIA PaCa-2, K562 and Saos-2 cells as described in Table 3. Genome Medicine 2009, 1 1: :83 F Fi ig gu ur re e 2 2 BANs of differentially expressed genes in common between cell lines. The lists of common genes between both colon cancer cell lines, between both breast cancer cell lines, and among the other three cell lines studied (representative of pancreatic cancer, leukemia and osteosarcoma) were used to construct BANs with the Pathway Architect software. Expanded networks were constructed for each list -( (a a) ) colon cancer, ( (b b) ) breast cancer and ( (c c) ) the other three cell lines -by setting an advanced filter that included the categories of binding, expression, metabolism, promoter binding, protein modification and regulation (see legend). Only proteins are represented. Overlapping of the expression levels was also performed (red for overexpressed genes and green for underexpressed genes; translucent shading represents genes that were not in the list and were added by the program from the interactions database). The BANs show some node genes that were studied further (those with arrows pointing to them).

D Di is sc cu us ss si io on n
The main objective of this work was to explore whether node genes could be identified from BANs constructed starting from genes differentially expressed in MTX-resistant cells from different human cancer cell lines, representative of five tissues. Those putative node genes may then be used as targets to increase the sensitivity toward MTX.
We started by determining and comparing the patterns of differential gene expression associated with MTX resistance in seven cell lines. The only differentially expressed gene in common among all the cell lines studied was dhfr. Its overexpression, at both the mRNA and protein levels, was confirmed in the MTX-resistant cells studied. The mRNA upregulation can be explained either by gene amplification of the dhfr locus, a well recognized mechanism for MTX resistance [15][16][17], or by an increase of dhfr transcription rate. In accordance with this, HT29, CaCo-2, MCF-7 and MIA PACA-2 resistant cells display an increased dhfr copy number. Indeed, amplification of the 5q14 locus, including dhfr and flanking genes, has been described in HT29 MTXresistant cells [13]. On the other hand, the other cell lines studied bear no changes in dhfr copy number. Thus, drug resistance in MDA-MB-468, K562 and Saos-2 cells may be caused by any of the other known mechanisms for MTX resistance [18][19][20][21][22]. Additionally, one has to take into account that MTX causes the differential expression of many genes that may be direct or indirect regulators of cell proliferation, survival or apoptosis, and that this expression pattern can contribute to modulation of the resistance phenotype. As described in this work, the overexpression of DKK1, UGT1As or EEF1A1 could represent a mechanism, parallel to DHFR overexpression, that plays a role in MTX resistance; the possible contribution of the overexpression of each of these genes is addressed below. Other work has determined genes that correlate with the capability of parental cells to resist treatment using concentrations assessed to be clinically achievable in tumor tissue [23]. Although this approach is different to the goal of our work, to identify genes differentially expressed in cells with acquired resistance to high concentrations of MTX, we compared the results obtained by Gyorffy et al. [23] with ours and found only six genes in common, namely CD99, CKMT1, DHRS2, IGFBP7, MAP7 and MYO1E.
Hierarchical clustering of all samples indicated that the MTX-resistant cells and their sensitive counterparts were highly correlated with regards to gene expression with each other. This is in accordance with reports showing that two breast tumor samples from the same patient before and after treatment with doxorubicin pair together in a hierarchical clustering [24]. Moreover, the same authors proposed that the molecular program of a primary tumor could generally be retained in its metastases. Similar results were obtained with leukemic cells from patients [25].
Interestingly, as shown in Figure 1, the gene expression patterns for the cell lines from the same tissue origin were very similar. Sets of coordinately expressed genes provide gene expression signatures that can indicate where to find targets suitable for gene therapy. Thus, we generated a list of genes differentially expressed in common between the two colon cancer cell lines. This list included, among others, genes encoding DHFR, the target for MTX, three members of the AKR family and ENO2, which we have previously studied as modulators of MTX resistance [8,13], and DKK1.
BAN construction using the genes differentially expressed in common between both colon cancer cell lines identified DKK1 as a highly interconnected node of the network, which could as such be a candidate druggable gene. DKK1 is a secreted protein involved in embryonic development [26] and is classically considered to function as an inhibitor of the canonical Wnt signaling pathway [27] (see [28] for a review). However, it does not take an active part in the Wnt/βcatenin pathway in colon cancer cells, as mutation of adenomatous polyposis coli (one of the components of Wnt pathway) occurs in most human colon cancers [29,30], thus disconnecting the effector part of the signaling cascade from the Wnt receptors, where DKK1 exerts its inhibitory effect [31,32]. This situation led us to hypothesize that DKK1 could have other cellular functions aside from its role in the Wnt pathway. Indeed, a role for DKK1 overexpression in cancer [26], including hepatobastomas [33] and breast cancer bone metastasis [34], aggressive tumors, epithelial-mesenchymal transition [35] and proliferation [36], has been previously suggested, although its precise mechanism of action has not yet been elucidated. In the case of HT29 MTX-resistant cells, the role of DKK1 is unclear, although it seems to be related to the resistant phenotype, since treatment with the siRNA against DKK1 mRNA showed a chemosensitization toward MTX. In keeping with this, Katula and collaborators [37] showed that folate deficiency led to the downregulation of DKK1, and that MTX inhibited DKK1 transcription. Thus, DKK1 overexpression in HT29 MTX-resistant cells could constitute a mechanism to overcome the transcriptional repression exerted by MTX.
It is worth noting that we had previously proposed the activation of the Wnt/ß-catenin pathway to be an important step in MTX resistance in HT29 colon cancer cells [13]. In this cell line, E-cadherin is chromosomically lost and underexpressed, thus allowing β-catenin to play its function in gene transcription. In order to shed some light on the possible role of the Wnt pathway in the overexpression of DKK1 in HT29 resistant cells, we performed transient transfection experiments with a luciferase reporter of β-cateninmediated transcriptional activation. These experiments showed that the Wnt pathway was more active in the HT29 resistant cells than in HT29 sensitive cells, and that reexpression of E-cadherin in the resistant cells resulted in lower β-catenin-mediated transcriptional activation, probably due to recruitment of β-catenin to the adherent junctions. Interestingly, DKK1 is transcriptionally regulated by Wnt/βcatenin signaling. Thus, constitutive activation of this signaling pathway through β-catenin, downstream of adenomatous polyposis coli, could represent a mechanism for DKK1 overexpression in HT29 MTX-resistant cells.

(f)
UGT1As were the only genes differentially expressed in common between both breast cancer cell lines analyzed. UGTs comprise a family of membrane glycoproteins that come from one single gene located on chromosome 2q37, rendering nine functional UGT1A proteins by alternative splicing of 13 different first exons with the common exons 2 to 5 [38]. UGTs are involved in phase II metabolism of a wide range of metabolites, both endogenous and exogenous [39]. Glucuronidation is an important metabolic process, as it carries out the biotransformation of lipophilic substrates into hydrophilic glucuronides, which can be more easily removed from the body. Among the different products that can undergo glucuronidation, we find analgesics, sex hormones, flavonoids, rifampicin, bilirubin and tobacco-specific carcinogens [40,41]. Some anticancer drugs, such as topotecan, irinotecan, SN-38 (the active metabolite of irinotecan), doxorubicin and 4hydroxytamoxifen, have also been described to be substrates of UGT1A activity [42][43][44][45][46][47]. Metabolism by UGT1A family members has been described to induce resistance toward daunorubicin (in both cell lines and rat hyperplastic liver nodules) [ [66], doxorubicin [67] and MTX resistance [68], maybe due to its ability to inhibit apoptosis [69]. It has been proposed that EEF1A overexpression promotes cell growth and replication by contributing to an overall increase in protein translation. Antisense-mediated abrogation of EEF1A1 expression inhibits tumorigenesis and anchorage-independent cell replication in prostate tumor cells [70]. Our functional analyses using siRNA technology against EEF1A1 are in keeping with these results, and show a chemosensitization of MIA PaCa-2 cells, thus stating a role for EEF1A1 in MTX resistance in this cell line.
In summary, our results provide evidence that node genes can be identified by constructing BANs with lists of genes differentially expressed in common between cell lines resistant to MTX. RNA interference technology has enabled us to demonstrate a role for DKK1, UGT1As and EEF1A1 in MTX resistance.

C Co on nc cl lu us si io on ns s
BANs were constructed using genes differentially expressed in common between cells resistant to MTX from seven human cancer cell lines representative of five tissues. We have been able to identify important node genes in the BANs, namely DKK1 in colon cancer cells, UGT1As in breast cancer cells and EEF1A1 in pancreatic cancer, leukemia and osteosarcoma cells. These three genes were functionally validated using siRNAs against their respective mRNAs, which resulted in increased sensitivity to MTX. . Thirty hours after transfection, luciferase activity (relative light units (RLU)) was assayed. The protein content was used to normalize the luciferase activity for each sample and is expressed relative to that of pGL3 basic vector (mean ± standard error of the mean for triplicate wells). *P < 0.05.   Priest DG, Noordhuis P, Schornagel JH, Pinedo HM, Peters GJ, Assaraf YG: A A s st tr ru uc ct tu ur ra al ll ly y a al lt te er re ed d h hu um ma an n r re ed du uc ce ed d f fo ol la at te e c ca ar rr ri ie er r w wi it th h i in nc cr re ea as se ed d f fo ol li ic c a ac ci id d t tr ra an ns sp po or rt t m me ed di ia at te es s a a n no ov ve el l m me ec ch ha an ni is sm m o of f a an nt ti if fo o-l la at te e r re es si is st ta an nc ce e. . J Biol Chem 1998, 2 27 73 3: :30189-30198. 21. Roy K, Tolner B, Chiao JH, Sirotnak FM: A A s si in ng gl le e a am mi in no o a ac ci id d d di if ff fe er r-e en nc ce e w wi it th hi in n t th he e f fo ol la at te e t tr ra an ns sp po or rt te er r e en nc co od de ed d b by y t th he e m mu ur ri in ne e R RF FC C--1 1 g ge en ne e s se el le ec ct ti iv ve el ly y a al lt te er rs s i it ts s i in nt te er ra ac ct ti io on n w wi it th h f fo ol la at te e a an na al lo og gu ue es s. . I Im mp pl li ic ca a-t ti io on ns s f fo or r i in nt tr ri in ns si ic c a an nt ti if fo ol la at te e r re es si is st ta an nc ce e a an nd d d di ir re ec ct ti io on na al l o or ri ie en nt ta at ti io on n o of f t th he e t tr ra an ns sp po or rt te er r w wi it th hi in n t th he e p pl la as sm ma a m me em mb br ra an ne e o of f t tu um mo or r c ce el ll ls s. . J Biol Chem 1998, 2 27 73 3: :2526-2531. 22. Zhao R, Sharina IG, Goldman ID: P Pa at tt te er rn n o of f m mu ut ta at ti io on ns s t th ha at t r re es su ul lt ts s i in n l lo os ss s o of f r re ed du uc ce ed d f fo ol la at te e c ca ar rr ri ie er r f fu un nc ct ti io on n u un nd de er r a an nt ti if fo ol la at te e s se el le ec ct ti iv ve e p pr re es ss su ur re e a au ug gm me en nt te ed d b by y c ch he em mi ic ca al l m mu ut ta ag ge en ne es si is s. . Mol Pharmacol 1999, 5 56 6: :68-76.