In Silico Promoter Analysis can Predict Genes of Functional Relevance in Cell Proliferation: Validation in a Colon Cancer Model

Abstract

Specific combinations of transcription-factor binding sites in the promoter regions of genes regulate gene expression, and thus key functional processes in cells. Analysis of such promoter regions in specific functional contexts can be used to delineate novel disease-associated genes based on shared phenotypic properties. The aim of this study was to utilize promoter analysis to predict cell proliferation-associated genes and to test this method in colon cancer cell lines. We used freely-available bioinformatic techniques to identify cell-proliferation-associated genes expressed in colon cancer, extract a shared promoter module, and identify novel genes that also contain this module in the human genome. An EGRF/ETSF promoter module was identified as prevalent in proliferation-associated genes from a colon cancer cDNA library. We detected 30 other genes, from the known promoters of the human genome, which contained this proliferation-associated module. This group included known proliferation-associated genes, such as HERG1 and MCM7, and a number of genes not previously implicated in cell proliferation in cancer, such as TSPAN3, Necdin and APLP2. Suppression of TSPAN3 and APLP2 by siRNA was performed and confirmed by RT-PCR. Inhibition of these genes significantly inhibited cell proliferation in colon cancer cell lines. This study demonstrates that promoter analysis can be used to identify novel cancer-associated genes based on shared functional processes.

Background

The methods of analysis of the colon cancer transcriptome described thus far produce large quantities of data in their output (Alon et al. 1999; Saha et al. 2001). Given the often arbitrary nature of the statistical thresholds for determining disease association, the functional relevance of many “over-expressed” genes is often unclear (Kothapalli et al. 2002; Troyanskaya, 2005). The absence of hypothesis in many microarray papers has yielded as many questions as answers (Shih et al. 2005).

One approach to this “data overload” is to focus on specific biological processes rather than individual genes that are altered in malignant cells. Such processes are driven by transcription factors that are common to genes which share similar functional contexts e.g. proliferation, invasion (Qiu, 2003; Werner, 2001). The promoter regions of these genes contain patterns of transcription factor binding sites (promoter modules) that form the basis for such co-regulation. These modules contain at least two transcription factor binding sites separated by a defined distance (Fessele et al. 2002). By identifying the promoter modules prevalent in genes that are known to share a common biological function, one can use these as a starting point to detect previously unknown genes that are involved in this process (Werner, 2001). The presence of these modules in a genes’ promoter region can positively or negatively influence functional processes. In this manner, a network of co-regulated genes can be determined that are implicated in specific processes (Liu et al. 2003). This approach has been used successfully in detecting interferon-responsive genes in inflammation, and novel cell-junction associated proteins (Cohen et al. 2006; Klingenhoff et al. 1999).

The purpose of this study was to use bioinformatic techniques to determine promoter modules common to those genes in the colon cancer transcriptome that are involved in cell proliferation. In this paper we utilize an integrated bioinformatics pathway to identify novel genes associated with cell proliferation in colon cancer, and validate this approach in an in vitro model.

Methods

Bioinformatic techniques

An outline of the bioinformatics pipeline is illustrated in Figure 1. A transcriptional profile of colorectal cancer was produced by comparing cDNA libraries obtained from normal colon and colon carcinoma with Digital Differential Display (DDD), as previously described (Moss et al. 2006). Briefly, the relative abundance of ESTs in colon cancer libraries was compared to normal tissue libraries, and those genes significantly over-expressed in colon cancer were extracted. The output was ontologically classified using Onto-Express to select those transcripts associated with cell proliferation (Khatri et al. 2002). The accession numbers of these transcripts were uploaded to Gene2Promoter (Genomatix Software GmbH), a software program that allowed identification of promoter regions based on the individual transcripts in a gene expression profile (Werner, 2001). The promoter sequences from Gene2Promoter were submitted to Frame-Worker, (FrameWorker, 2006) and once a model common to the input promoters was identified, its presence was screened for in known promoters of the human genome using Model Inspector (Model Inspector 2006). Briefly, all matches for individual elements of the module which score above a pre-set threshold are located in the promoter database. These individual elements are combined to match the organization (element order and distances) of the input module, to evaluate the fit of the model. Finally, Bibliosphere was utilized to examine the characteristics of selected genes based on the published literature (Scherf et al. 2005).

Figure 1.

Summary of bioinformatics methods used. References for each method contained in text.

Gene expression

Public gene expression repositories derived from microarray data from normal colon, colonic cancers and colon cancer cell lines, were interrogated for genes of interest. The normal colon microarray profile originated from pooled samples from normal colonic tissue (Gene Expression Omnibus tissue GSM44680) hybridized to the Affymetrix GeneChip Human Genome U133 Array (Ge et al. 2005). The results are expressed in log2 of user-provided counts for comparison to other normal tissues. Colon cancer tissue expression profile was obtained from the transcriptome of 10 colorectal adenocarcinomas hybridized to the U95a Affymetrix GeneChip and compared to other human cancers (Su et al. 2001). Finally, the microarray data from a primary colon cancer (SW480) and a metastatic colon cancer cell line (SW620) hybridized to the Affymetrix GeneChip Human Genome U133 Array was surveyed (Provenzani et al. 2006). The results are expressed in log2 of user-provided counts for comparison between the cell lines.

Cell lines

The Caco₂ human colonocyte cell line was purchased form ATCC (LGC Promochem, U.K.) and the T84 cells were a kind gift from Dr. Cormac Taylor, UCD. Cell lines were cultured in minimum essential medium (Caco₂) or mixture of Dulbecco’s modified Eagle’s medium and Ham’s F12 medium under standard conditions (T84).

siRNA transfection

Prior to transfection 1×10⁵ cells were seeded in 500 μl of medium in each well of a 24 well plate and cultured until 50–80% confluent (∼24 hours). For transfection, 0.5 μg of custom-designed siRNA (Dharmacon, IL, U.S.A.) was diluted in 100 μl medium and 1.5 μl RNAifect transfection reagent added (Qiagen, U.K.) at a 1:3 ratio and added to each well as per protocol. Three controls were used for each experiment; a positive control of laminin siRNA for mRNA quantification, a positive control of fluorescent-labeled siRNA for microscopy, and negative controls of medium only, transfection reagent only and scrambled siRNA only. The transfected cells were incubated for 24 hours under normal conditions.

RT-PCR

RNA extraction was subsequently performed from cells using the RNeasy kit (Qiagen, U.K.), and reverse transcribed using SuperScript II (Promega, U.K.). Quantitative PCR was performed using an ABIPrism Taqman PCR machine. Expression levels of individual genes were normalized to 18s RNA.

Cell proliferation assay

In order to determine the effect of siRNA on cell proliferation rates, transfected CaCO₂ cells were seeded into 96-well plates at a concentration of 1×10⁴ cells in 100 μl per well and allowed to adhere overnight. The MTS cell proliferation assay (Promega, U.K.) was used to assess proliferation rates at 48 hours, based on absorbance at 490 nm in an ELISA plate reader. Proliferation ratios were based on comparison of mean absorbance values for transfected and untransfected wells using one-way ANOVA.

Statistical analysis

Statistical analysis of laboratory results was performed using StatView software (SAS Institute, Cary, NC). Normalised gene expression was analysed using ANOVA, after testing for equality of variance. A p < 0.05 was considered significant. The differential expression profiles, promoter analysis and module detection all contain integral statistical thresholds for results as described in the results section.

Results

An EGRF/ETSF transcription factor module is prevalent in cell proliferation-associated genes over-expressed in colorectal cancer

Digital Differential Display comparison of normal colon to colorectal cancer cDNA libraries identified 163 transcripts differentially expressed in colon cancer, of which 16 were classified as involved in cellular proliferation (supplementary 1)(Moss et al. 2006). These 16 genes were the source material for promoter screening. The loci of these 16 genes were entered into Gene2Promoter, which detected 30 unique promoters assigned to 30 transcripts in the mapped regions; all transcripts with at least one exon identical to one of the mapped exons and their promoters were listed (Table 1). Fifteen of these promoters had been experimentally verified, and the other 15 were computational predictions based in sequence location and content.

Table 1.

Genes associated with cell proliferation in colon cancer that had promoter regions identified (verified = published experimental verification, predicted = transcript with 5’ end confirmed by Gene2Promoter 2004).

mRNA	Locus	Transcript/TSS	Quality Level
NM_002394	SLC3A2 (Loc 6520)	AK090758_1	Verified
		AK094620_1	Verified
		NM_002394	Predicted
NM_005916	MCM7	AK055379_1	Verified
		AK096959_1	Verified
		NM_005916	Predicted
NM_014865	CNAP1	AK022511_1	Verified
		AK125155_1	Verified
		AK128354_1	Verified
NM_001034	RRM2	AK092671_1	Verified
		AK123010_1	Verified
		NM_001034	Predicted
NM_002707	PPM1G	AK127593_1	Verified
		NM_002707	Predicted
		NM_177983	Predicted
NM_000077	CDKN2A	NM_058195	Predicted
NM_016343	CENPF	NM_016343	Predicted
NM_002447	MST1R	NM_002447	Predicted
NM_001255	CDC20	NM_001255	Predicted
NM_004526	MCM2	AK128291_1	Verified
NM_005186	CAPN1	AK025380_1	Verified
		AK097277_1	Verified
		NM_005186	Predicted
NM_004494	HDGF	AK096411_1	Verified
		NM_004494	Predicted
NM_003334	UBE1	AK097343_1	Verified
		NM_003334	Predicted
NM_002335	LRP5	NM_002335	Predicted
NM_002032	FTH1	NM_002032	Predicted
NM_005030	PLK	NM_005030	Predicted

The identified promoter sequences were investigated using FrameWorker software, which detects patterns in transcription factor binding sites (TFBS). We searched for modules containing at least 2 elements (TFBS), at a distance of 5–50 nucleotides apart, and adjusted the quorum constraint (prevalence threshold) until the program identified a common module. No individual module was common to all the input promoter sequences. However, one complex module, containing members of the EGRF and ETSF transcription factor binding site families, was present in 18/30 (60%) of the input promoters (Fig. 2). The specificity score of this model had a p-value of 0.0059 e.g. the probability that an equal or greater number of sequences with a model match would be obtained in a randomly drawn sample of the same size as the input sequence set. The relative occurrence of individual model matches in a background promoter sequence set of 5000 human promoters scanned with this module was 0.27 and 0.50 for EGRF and ETSF respectively.

Figure 2.

EGRF/ETSF module is common to proliferation genes expressed in colon cancer. EGRF element (purple), ETSF element (yellow) and combined module (grey) location in promoter region of representative sample of input loci relative to transcription start site (TSS, red arrow). Graphical output generated by FrameWorker software.

This EGRF/ETSF module contains members of the Early Growth Response Factor family and the ETS factor family at a distance of 6–44 base pairs between elements. The matrices (transcription factors) of the EGRF family were EGR1, EGR2, EGR3, EGR4 and Wilms tumour suppressor. The re-value, an expectation value of the number of matches per 1000 base pairs of random DNA sequence for each individual matrix, ranged from 0.03–0.35 for the EGRF elements. ETS1, ETS2, ELK1 and NRF2 were the components of the ETSF element, with re-values of <0.01–2.05. The free version of the software does not detail the exact sequences of the modules, as it is their relative location, rather than sequence, that determines a module’s functional activity.

The EGRF/ETSF module identifies novel proliferation-associated genes

The known promoters of the human genome were screened for the EGRF/ETSF module using Model Inspector, based not on sequence alignment, but detection of individual elements and their position relative to each other. At the time of the experiment, the database contained 46,119 promoters with known transcripts. A total of 102 matches for the selected proliferation-associated module were detected in 30 genes (Table 2). All matches contained a model score of ≥85% specificity. The chromosomal locations of these genes were widely dispersed throughout the genome, excluding the possibility of co-regulation due to overlapping sequences (data not shown).

Table 2.

Identified genes that contain the EGRF/ETSF promoter module in their promoter regions.

Accession	Genea	Model score	Effect on proliferationb	Expression in colon cancerc	Public microarray datad	References
AB000381	GML	89%	negative	Yes - in cell lines	No	Oncogene 1996; 13 (9) 1965–7. Int J Clin Oncol. 2001 Apr; 6(2):90–6
AB001517	TMEM1	90%	?	?	No
AB001523	PWP2	90%	?	?	No
AB003173	WRN	90%	positive	Yes – in unmethylated tumours	Yes	DNA Repair 2004; 3(5): 475–482 Proc Natl Acad Sci USA. 2006 Jun 6; 103(23):8822–7
AB003469	MCM5	90%	?	Yes	Yes	Clin Cancer Res. 1999 Aug; 5(8):2121–32
AB004270	MCM7	90%	positive	?	Yes	Oncogene 2006; 25(7): 1090–9
AB005647	NPR2	90%	?	?	No
AB006075	HMG-CoA synthase	89%	?	Yes	No	Mol Carcinog. 2001 Nov; 32(3):154–66.
AB006684	AIRE	91%	?	?	No
AB007828	NDN	88%	negative	?	No	Gene 1998; 213(1–2): 65–72
AB008496	COL4A3	90%	negative	?	No	J Biol Chem 2000; 275 (28):21340–8
AB008502	TLX2	90%	?	?	No
AB008681	ACVR2B	92%	positive	Yes – in cell lines	No	Dev Biol 2004; 266(2): 334–45 Gut. 2001 Sep; 49(3):409–17
AB008822	TNFRSF1 1B	92%	negative	?	No	J Clin Invest 2001; 107(10):1235–4
AB009071	KCNH2	88%	positive	Yes	Yes	J Biol Chem 2003; 278(5):2947–55 Cancer Res. 2004 Jan 15; 64(2):606–11
AB009667	Klotho	95%	?	?	No
AB009777	NID2	85%	?	?	No
AB012286	ITGB4	90&	positive	?	Yes	Cancer Res 2005; 65(23):10674–9
AB012668	hFUCT-7	91%	?	?	No
AB015751	APLP2	96%	?	?	No
AB016243	SLC9A3R2	91%	?	?	No
AB016656	LIMK2b	91%	?	?	No
AB016767	TERT	95%	?	?	No
AB017018	HNRPDL	94%	?	?	Yes
AB017547	SPR	93%	?	?	Yes
AB017567	LIPT1	95%	?	?	No
AB017602	PDE9A	90%	?	?	Yes
AB018192	PHC1	95%	?	?	No
AB018401	DHH	90%	positive	?	No	Development 2004 Oct; 131(20):5009–19
AK001326	TSPAN3	90%	positive	?	Yes	J Cell Biol 153:295–305

^a.HUGO accepted gene name

^b.Published experimental evidence of effect on cell proliferation

^c.Experimental evidence of increased protein or mRNA expression in colon cancer

^d.Upregulation of gene in public microarray database of expression relative to normal (Diehn et al. 2003)

The products of the 30 genes were entered into Bibliosphere (Bibliosphere, 2006) to determine their functional context based on the scientific literature e.g. published experimental evidence of a role in affecting cellular proliferation (Table 2). Eleven of these genes (37%) have been implicated in cell proliferation in the literature, including KCNH2 and MCM7 (Lastraioli et al. 2004; Yoshida et al. 2003) (Table 2). Six of the genes have been described in the literature as expressed in colonic neoplasia based on experimental data, and nine are up-regulated in colon cancer gene expression profiles in public databases (Table 2) (Diehn et al. 2003). As a control functional context, a common disease process, inflammation, was explored in the 30 identified genes using Bibliosphere; only one (TLX2) has been associated with inflammation (data not shown).

Suppression of TSPAN3 and APLP2 inhibits cell proliferation in colorectal cell lines

The experiments above identified genes containing a promoter module that is frequently present in genes associated with cell proliferation in colon cancer. In order to determine the functional significance of the presence of this module in these genes, the role of their knock-down by siRNA on cell proliferation was determined. We screened the identified genes using Bibliosphere for 1) reports of expression in colon cancer, 2) reports of a role in cell proliferation (Table 2). As our interest was in novel proliferation-associated genes which may be relevant to colon cancer, we selected three genes not previously reported as altered in colon cancer; TSPAN3, NDN and APLP2. They have been described as having positive, negative and unknown roles in cell proliferation (Taniura et al. 1999; Tiwari-Woodruff et al. 2001). The TSPAN3 gene contains the EGRF/ETSF module at position 491-418 on the negative strand at 15q24.3. The APLP2 gene contained the EGRF/ETSF module at position 2492–2532 on the positive strand at 11q24. The NDN gene contained the EGRF/ETSF module at position 1268-1179 on the negative strand at 15q11.2-q12.

Gene expression of each gene in colon cancer was first measured from 3 diverse microarray databases; one from 36 types of normal tissue, one from 174 epithelial tumors that included 10 colorectal adenocarcinomas, and a third from primary and metastatic colorectal cell lines (Ge, Yamamoto, Tsutsumi, Midorikawa, Ihara, Wang and Aburatani, 2005; Provenzani, Fronza, Loreni, Pascale, Amadio and Quattrone, 2006; Su, Welsh, Sapinoso, Kern, Dimitrov, Lapp, Schultz, Powell, Moskaluk, Frierson, Jr. and Hampton, 2001). Both TSPAN3 and APLP2 were expressed in normal colon, and colon cancer cell lines, although only TSPAN3 was relatively over-expressed in colonic adenocarcinoma tissue relative to other tumours (Figure 3). NDN was not expressed in normal colon, adenocarcinoma or colon cancer cell lines.

Figure 3.

TSPAN3 and APLP2 are expressed in normal and neoplastic colon. Expression of TSPAN3, APLP2 and NDN in microarray profiles of: (a) normal colon (Log2 of user-provided count of gene expression on oligonucleotide microarray (Affymetrix U133) using pooled RNA) (b) primary colon cancer cell line (black bars) and metastatic colon cancer cell line (grey bars) (Log2 of user-provided count of gene expression on oligonucleotide microarray (Affymetrix U133) using pooled RNA) (c) 174 human epithelial tumors (Co; colon samples, red, increased gene expression; green, decreased expression; black, median level of gene expression. The color intensity is proportional to the hybridization intensity of a gene from its median level across all samples.

Cells were transfected with siRNA designed to provide at least 70% silencing of expression, and mRNA levels and cell proliferation quantified (Reynolds et al. 2004). TSPAN3 expression in T84 colon cancer cell line was confirmed by RT-PCR (Fig. 4a). siRNA caused a 62% inhibition of TSPAN3 expression at 24 hours (Fig. 4a, p < 0.05). Confirmation of cellular uptake was observed using the labeled fluorescent siRNA (Fig. 4b). This led to a 40% reduction in cellular proliferation at 48 hours in T84 cells (Fig. 4c, p < 0.05). Neither scrambled siRNA or transfection agent alone affected cell proliferation. APLP2 expression in colon cancer cells was confirmed by RT-PCR (Fig. 5a). siRNA caused a 45% inhibition of APLP2 expression at 24 hours (Fig. 5a, p < 0.05). Confirmation of cellular uptake was observed using the labeled fluorescent siRNA (Fig. 5b). This inhibition led to a 40% reduction in cellular proliferation at 48 hours in CaCo2 cells (Fig. 5c, p < 0.05). Neither scrambled siRNA or transfection agent alone affected cell proliferation.

Figure 4.

Inhibition of TSPAN3 expression by siRNA inhibits colon cell line proliferation. (a) RNA extracted from transfected and untransfected T84 cells after 24 hours was reverse transcribed to cDNA and probed for TSPAN3 using Taqman PCR (expressed in arbitrary units normalised to 18s RNA). (b) T84 cells in a 96-well plate were transfected with a control fluorescent-labeled siRNA to confirm transfection efficiency. (c) Proliferation of transfected T84 cells in a 96-well plate was assessed after 24 hours using the MTS Cell Proliferation Assay. Control = media only, vehicle = media and transfection agent (Lipofectamine), scrambled siRNA = transfection agent and scrambled siRNA, and TSPAN3 siRNA = custom-designed TSPAN3 siRNA (10 nm)

Figure 5.

Inhibition of APLP2 expression by siRNA inhibits colon cell line proliferation. (a) RNA extracted from transfected and untransfected T84 cells after 24 hours was reverse transcribed to cDNA and probed for APLP2 using Taqman PCR (expressed in arbitrary units normalised to 18s RNA) (b) T84 cells in a 96-well plate were transfected with a control fluorescent-labeled siRNA to confirm transfection efficiency (c) Proliferation of transfected T84 cells in a 96-well plate was assessed after 24 hours using the MTS Cell Proliferation Assay. Control = media only, vehicle = media and transfection agent (Lipofectamine), scrambled siRNA = transfection agent and scrambled siRNA, and APLP2 siRNA = custom-designed APLP2 siRNA (10 nm)

Discussion

This study has demonstrated the use of promoter modules as bioinformatic “bait” to delineate key regulatory networks in colon cancer, and to identify novel biological players in cell proliferation. It is based on the premise that genes expressed in similar disease states share a common “footprint” of transcriptional regulatory processes for specific functional activities. The relative order and spacing of these transcription factor (TF) binding sites (TFBSs) within a module are often highly conserved through evolution, highlighting their importance in regulation. This conservation allows us to use computational searching to pinpoint these clusters of known TF binding sites rather than specific nucleotide sequences (Berman et al. 2002). Although the shared process selected for this study, proliferation, is not unique to cancer cells, it is a dominant process that partially defines this disease state. The presence of the EGRF/ETSF module in the promoter region of genes associated with cell proliferation suggests a role for this module in the regulation of cellular proliferative activity (Lantingavan, I et al. 2005). Although this influence could be positive or negative, its frequency in the promoter regions of over-expressed genes in colon cancer cDNA libraries, compared to random promoter sets, suggests a pro-proliferative effect in this setting. The experimental data clearly suggests a role for the studied genes in cell proliferation, although whether this is actually dependent on the identified promoter module, would require further studies.

This strategy predicted a role for TSPAN3 in cell proliferation in colorectal cancer that has not previously been described. TSPAN3 is a member of the tetraspanin family of cell surface receptors that have been implicated in the cell proliferation process in oligodendrocytes (Tiwari-Woodruff, Buznikov, Vu, Micevych, Chen, Kornblum and Bronstein, 2001). Our data demonstrating its expression in normal and neoplastic colon, and the negative effects of its inhibition on cell proliferation in colon cancer cell lines, confirming the predicted role based on promoter analysis. Further work will be required to determine whether this effect is unique to colon cancer cells, and which component of proliferation is involved. APLP2 is an amyloid-like protein precursor that plays a role in G-coupled signaling. It may be required for epithelial cell growth in wounds (Siemes et al. 2006). This study suggests a role for APLP2 in colon cancer cell proliferation, which may be due to its key function in genomic segregation (von der et al. 1994).

Although this study validates this approach in identifying co-regulated genes, the activation of the promoters involved has not been experimentally tested. As this work focuses on functionally relevant associations between genes and disease, we sought to examine the functional end-point primarily. The confirmation of alterations in expression and proliferation validates the computational predictions. We have not focused on the descriptive aspects of the module discussed e.g. sequence, as it is the strategic organization of the EGRF/ETSF matrix in the promoters of interest, rather than sequence composition, that confers its functional properties (Dohr et al. 2005). Our intention was proof-of-concept evidence that could validate this bioinformatic approach.

In conclusion, this study demonstrates that an integrated in silico promoter analysis approach can be used to delineate novel cancer-associated genes. We have described a previously unreported role for TSPAN3 and APLP2, in cell proliferation in colon cancer based on a common promoter module. Further study of this module may provide increased understanding of this regulatory network.

List of abbreviations:

siRNA

small interfering RNA,

EGRF

early growth response family,

ETSF

E26 transformation-specific family.