A genomic strategy for the functional validation of colorectal cancer genes identifies potential therapeutic targets

Summary/Abstract

Genes that are highly overexpressed in tumor cells can be required for tumor cell survival, and have the potential to be selective therapeutic targets. In an attempt to identify such targets, we combined a functional genomics and a systems biology approach to assess the consequences of RNAi-mediated silencing of overexpressed genes that were selected from 140 gene expression profiles from colorectal cancers (CRC) and matched normal mucosa. In order to identify credible models for in-depth functional analysis, we first confirmed the overexpression of these genes in 25 different CRC cell lines. We then identified five candidate genes that profoundly reduced the viability of CRC cell lines when silenced with either siRNAs or shRNAs, i.e., HMGA1, TACSTD2, RRM2, RPS2, and NOL5A. These genes were further studied by systematic analysis of comprehensive gene expression profiles generated following siRNA-mediated silencing. Exploration of these RNAi-specific gene expression signatures allowed the identification of the functional space in which the five genes operate, and showed enrichment for cancer specific signaling pathways, some known to be involved in CRC. By comparing the expression of the RNAi signature genes with their respective expression levels in an independent set of primary rectal carcinomas we could recapitulate these defined RNAi signatures, therefore establishing the biologically relevance of our observations. This strategy identified the signaling pathways that are affected by the prominent oncogenes HMGA1 and TACSTD2, established a yet unknown link between RRM2 and PLK1, and identified RPS2 and NOL5A as promising potential therapeutic targets in CRC.

Introduction

The application of parallel gene expression profiling techniques to large sets of primary tumor samples has revealed profound alterations in the cancer transcriptome. The catalogs of deregulated genes defined by such approaches are not only important because they provide new insight into tumor biology, but also because they reveal those genes that are specifically upregulated in tumors, some of which may represent promising new anti-tumor molecular targets.

Finding such critical genes within a complex expression signature, however, remains a formidable challenge. One possible approach to this problem is to modulate the expression of these genes in cell line models. For example, loss-of-function (LOF) analysis can be used to identify genes whose reduction of expression may have a direct impact on cancer cell survival.

We recently presented comprehensive gene expression signatures of colorectal cancer and matched normal mucosa.^{1, 2} In order to identify those genes that could directly influence tumor survival, and thus may represent candidate molecular targets, we first identified colorectal cancer cell lines in which the expression of these genes was upregulated accordingly. We then used RNAi-mediated gene silencing to reduce their expression and quantified cell survival. To uncover their roles in cell viability we monitored genome-wide transcriptional consequences of silencing these candidate genes. This allowed the identification of the cellular interaction networks, and, by comparing these signatures with gene expression profiles of independent primary tumors, established the clinical relevance of our approach.

Material and Methods

Cell Culture, RNA Isolation and Real-time PCR

Twenty-four colorectal cancer cell lines were purchased from the American Type Culture Collection (ATCC; Manassas, VA): SW480, HT-29, SW837, SW48, T84, LS 513, HCT 116, SW1463, SW1116, LS 411N, NCI-H716, LoVo, NCI-H508, COLO 201, COLO 320DM, LS 123, LS 174T, DLD-1, Caco-2, LS 1034, SK-CO-1, RKO, SW403, SW620. p53HCT 116 cells were a gift from Bert Vogelstein. All cell lines were cultured in the recommended media containing 10% fetal bovine serum (FBS), and cell line cross-contamination was excluded by short tandem repeat profiling.³ Total RNA was isolated using TRIzol (Invitrogen, Gaithersburg, MD), and purified using the RNeasy Mini kit (Qiagen, Germantown, MD). Detailed protocols are at http://www.riedlab.nci.nih.gov/. Nucleic acid quantity, quality and purity were determined using a spectrophotometer (Nanodrop, Rockland, DE) and a 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Subsequently, real-time PCR was performed as previously described.² Details are available in SI Methods.

RNAi-based Experiments

The target sequences for synthetic siRNA duplexes (Qiagen, Germantown, MD) are listed in Table S1. Lipid-based reverse transfections were performed as previously described.^4,5 Individual Expression Arrest™ lentiviral inducible short hairpin RNA constructs (Open Biosystems, Huntsville, AL) were transfected using Nucleofector technology (Amaxa, Cologne, Germany), and stable single-cell clones were subsequently expanded. For siRNAs, gene-specific RNAi was measured at an mRNA level using either the QuantiGene Reagent System as previously described⁵ or a custom QuantiGene Plex 1.0 Reagent System (Panomics, Fremont, CA) (Table S2). shRNAs knockdown was validated using real-time PCR. Details are available in SI Methods.

Cell Viability Assay

Cell viability was determined using the CellTiter-Blue^® reagent (Promega, Madison, WI). Replicate transfections per siRNA duplex were set up for every time point, and reduction of resazurin to resorufin was measured 72 hours (data not shown) and 96 hours after transfection using a plate reader (Spectramax 2e, Molecular Devices, Sunnyvale, CA). Viability of transfected cells was compared to cells transfected with a negative-control siRNA (Qiagen). Similarly, expression of shRNAmir clones was induced as described above, and viability was determined 72 hours (data not shown) and 96 hours after induction compared to cells expressing a negative control shRNAmir (Open Biosystems).

Gene Expression Profiling Following RNAi

Triplicate transfections were independently performed with two different siRNA duplexes corresponding to each gene target. As a negative control, we used a siRNA duplex (AllStar siNegative, Qiagen) that has been experimentally validated as generating minimal non-specific effects on global gene expression levels and cellular phenotypes. Total RNA was isolated from SW480 cells transfected with siRNAs corresponding to HMGA1, RRM2, and RPS2 48 hours post-transfection, and 72 hours post-transfection for siRNAs corresponding to TACSTD2 and NOL5A. Different time points were chosen because maximal reduction of viability upon inhibition of HMGA1, RRM2, and RPS2 was already observed after 72 hours, whereas inhibition of TACSTD2 and NOL5A reached maximal levels after 96 hours. Subsequently, a reduction in target mRNA levels was confirmed by real-time PCR (Table S1), and gene expression profiling was performed as previously described.⁶ The respective gene expression data have been deposited in the NCBI Gene Expression Omnibus at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=vbufpakkuakoqze&acc=GSE15212

Statistical Analysis of Microarray Data

Array data were log₂-transformed, normalized, and corrected for multiple testing. Gene lists for the different comparisons (each experimental siRNA separately versus negative-control siRNA) were generated based on stringent cutoff criteria (q-value of < 0.05 and linear fold change of > ∼1.5). Since two different siRNAs were used to target the same gene, the overlap of these lists was generated using the additional criterion that the regulation should point in the same direction. A detailed description is available in SI Methods.

Estimation of Signal Flow Networks and Pathway analysis

Signal flow networks from RNAi effects in gene expression were reconstructed with our Nested Effects Models approach⁷ using those genes that were selected according to the conditions described above. To further determine the functional relationships between genes within the RNAi signatures Ingenuity Pathway Analysis tools (Ingenuity Systems Inc., Redwood City, CA) were used. Details are available in SI Methods.

Comparison of Cell Line and Tumor Gene Expression Profiles via Enrichment Analysis

To compare these RNAi signatures to gene expression profiles recently generated from an independent set of 65 primary rectal cancers and matched normal mucosa samples (Gaedcke et al., in preparation), we computed lists of differentially expressed genes (cutoff criterion of FDR < 0.05). We then determined the overlap of these lists and the sets of probes deregulated after silencing of a given candidate gene, and we evaluated the significance of an over-representation using a one-sided Fisher's exact test. Further details are available in SI Methods.

Immunohistochemistry

Protein expression levels of RPS2 in colorectal cancers and normal mucosa were assessed on Tissue Microarrays (TMA). Technical details are available in SI Methods.

Results

CRC Cell Lines Recapitulate the Gene Expression Pattern of Primary Tumors

The general outline of our experimental strategy is presented in Fig. 1A. We recently established gene expression signatures of primary colorectal cancers and identified many genes that were highly overexpressed in the tumors.^1,2 In order to identify model systems to study the relevance of such overexpressed genes for cell viability and hence as potential therapeutic targets, we first measured the expression levels of 25 highly upregulated genes in 25 colorectal cancer cell lines using real-time PCR (Table S3). Eleven of these 25 genes were highly overexpressed in the majority of tumor cell lines, and were therefore carried forward for functional validation (Fig. 1B and Table 1). Together with the maintenance of specific patterns of genomic imbalances, the preservation of gene expression changes in CRC cell lines is consistent with data from breast cancer cell lines⁸ and from other tumor models, which suggest that they are credible models for meaningful functional analysis.^9,10

Figure 1. Overview of the experimental strategy and gene expression patterns in CRC cell lines.

(A) Pictorial representation of the experimental strategy used. (B) Heat map of real-time PCR expression data for 25 genes in 25 colorectal cancer cell lines. Red = upregulated genes, blue = downregulated genes.

Table 1.

List of 11 genes that were consistently overexpressed in primary CRC, and in CRC cell lines, respectively. Values represent linear fold changes. These genes were subsequently silenced using RNAi. *Unpublished data (Gaedcke et al., in preparation).

Sequence Description	Avg. Rectal Cancer/ Avg. Mucosa	Avg. Colon Cancer/ Avg. Mucosa	Avg. Cell Lines/ Avg. Mucosa Pool
Tumor-associated calcium signal transducer 2 (TACSTD2)	8.95	1.53	23.2
Pleckstrin homology-like domain, family A, member 1 (PHLDA1)	3.33*	2.20	4.74
Histone 1, H2aj (HIST1H2AJ)	1.43*	6.23	3.06
V-myc myelocytomatosis viral oncogene homolog (MYC)	2.11	5.49	5.13
High-mobility group AT-hook 1 (HMGA1)	3.27*	5.35	4.77
Ribosomal protein S2 (RPS2)	1.61*	5.91	4.17
Ribonucleotide reductase M2 polypeptide (RRM2)	2.55	5.14	4.48
CDC28 protein kinase regulatory subunit 2 (CKS2)	2.28	3.11	8.18
Nucleolar protein 5A (56kDa with KKE/D repeat) (NOL5A)	2.28	3.12	2.37
Member RAS oncogene family (RAN)	1.76	1.97	1.76
Protein kinase, DNA-activated, catalytic polypeptide (DNAPK)	2.02	3.18	5.92

Loss-of-function Analysis Identifies Genes Required for Cell Viability

The 11 genes selected for LOF analysis using RNA interference were studied in SW480 using two synthetic siRNA duplexes per gene. Transfection of 15 of the 22 siRNAs (68%) decreased mRNA levels of the corresponding target gene by more than 70 percent relative to a control siRNA (siNeg) 48 hours post transfection (Table S1). Silencing of seven genes, HMGA1, MYC, NOL5A, TACSTD2, RAN, RPS2, and RRM2, mediated a 20% or greater decrease in cellular viability (Fig. S1A). Obviously, we were reassured to identify MYC because it validates our general experimental approach. MYC is frequently amplified, overexpressed, and activated via chromosomal translocations in human cancer.¹¹ However, due to the wealth of knowledge on MYC signaling, we did not include MYC in further experimentation. Accordingly, the five final candidate genes were TACSTD2, HMGA1, NOL5A, RPS2, and RRM2 (Fig. 2A).

Figure 2. The viability of CRC cell lines following RNAi against candidate genes.

(A) Relative viability of SW480 cells transfected with two different siRNAs corresponding to RRM2, RPS2, TACSTD2, HMGA1, or NOL5A. (B) Relative viability of SW480 cells expressing shRNAmirs corresponding to RRM2, RPS2 and NOL5A. (C) Relative viability of HT29 and DLD-1 cells transfected with siRNAs targeting RRM2, RPS2, or NOL5A.

siRNA duplexes have the potential to interact with other transcripts as a result of microRNA (miRNA) like interactions.^12,13 One approach to reducing the probability that an observed phenotype is due to non-specific effects is to show that additional siRNA duplexes with different sequences lead to similar phenotypic consequences. Following this approach, we observed reduction of viability with at least three siRNAs (Fig. S1B).

Since the RNAi machinery can also be harnessed to suppress gene expression using short-hairpin RNA (shRNAs), we wished to reproduce our siRNA results employing shRNA expression cassettes (with different sequences from that of the siRNAs). We therefore established stable SW480 clones that can be induced to express shRNAmirs targeting RRM2, RPS2, and NOL5A (the genes for which shRNAs were available). In all instances, the shRNA-mediated decreased cell viability was very similar to that mediated by synthetic siRNAs (Fig. 2B).

To ensure that the effect on cell viability is not limited to SW480, we transfected the colorectal cancer cell lines HT-29 and DLD-1, for which we had confirmed overexpression of the candidate genes. Silencing of NOL5A, RPS2, and RRM2 reduced the viability of both cell lines (Fig. 2C).

Generation of Loss-of-function-specific Transcriptional Signatures

To explore the cellular pathways in which the candidate genes operate we conducted whole transcriptome expression analyses of siRNA transfected SW480 cells to identify an “RNAi signature” for each gene of interest. These RNAi signatures were defined as the overlap of genes with altered expression following transfection of SW480 with two different siRNA duplexes compared to a negative control. We then systematically explored whether any of the gene expression changes that we observed could be due to non-sequence-specific or sequence-specific miRNA-like “off-target” interactions,^14-16 and removed genes with potential interactions from our lists (details are available in SI Results). Based on these criteria, the number of deregulated probes amounted to 324 for RRM2, 187 for NOL5A, 136 for RPS2, 123 for HMGA1, and 76 for TACSTD2 (Table S4 A-E). In all instances, the target gene corresponding to each siRNA was successfully silenced as evidenced by a profound reduction in target transcript levels measured by both real-time PCR (Table S1) and expression microarray (Table S4). The cluster analysis (Fig. 3A) visualizes the results and illustrates experimental reproducibility. Furthermore, the remarkably high correlation in the changes in gene expression that we observed following RNAi with two different siRNA duplexes suggests that the vast majority of these modulations are a functional consequence of gene specific silencing and not due to non-specific effects (Fig. 3B).

Figure 3. Visualization of gene-specific RNAi signatures.

(A) Unsupervised hierarchical clustering of gene expression data obtained from SW480 cells transfected with two different siRNAs corresponding to RPS2, HMGA1, RRM2, NOL5A or TACSTD2, and gene expression profiles from control experiments. Each row represents the results of individual array experiments conducted using RNA obtained from separate transfections. Red = upregulated genes, blue = downregulated genes. (B) Correlation of changes in gene expression observed following gene silencing with two different siRNAs. The red circles represent those probes corresponding to the gene targeted by the stated siRNAs.

Prediction of Functional Relationships from Gene-specific RNAi Signatures

While the established or proposed functions of RPS2, NOL5A, RRM2, HMGA1 and TACSTD2 are diverse, we were interested to see if the respective RNAi signatures contained common transcriptional features that indicate functional relationships between these overexpressed CRC genes, and thus reflect critical processes altered in CRC. To assess this we used Nested Effects Models^17,18 to estimate signal-flow network relationships. The principal idea of this approach is that the silencing of a gene that is high in a signaling cascade will influence the expression of many downstream genes, while the silencing of a gene that acts lower in the cascade will generate effects in only a subset of these genes. 740 unique probes were selected for analysis (Table S4 A-E). The results are shown in the left part of Fig. 4, and the genes that contributed to the NEM are listed in Table S5. This analysis suggests that TACSTD2 is below the other genes because its signature has the most overlap with the other RNAi signatures. The RNAi signatures for HMGA1, RRM2 and RPS2 suggest these influence separate processes as each display substantial independent effects. The NOL5A RNAi signature overlaps with the RRM2 RNAi signature but many of these changes are not seen within the TACSTD2 RNAi signature. The Nested Effect Models predicted relationships for the CRC overexpressed genes are displayed on the right part of Fig. 4. The placement of HMGA1, RRM2 and RPS2 high in the hierarchy of this model is consistent with the observation that these genes were associated with the most pronounced reduction of cell viability (Fig. 2A).

Figure 4. Prediction of functional relationships from gene-specific RNAi signatures.

Visualization of the Nested Effects Model (NEM) analysis of the RNAi signatures for the five candidate genes. Red = upregulated genes, blue = downregulated genes. The right part of the figure represents the hierarchy in which these genes were found (the numbers on the edges represent the corresponding bootstrap values).

Biological Networks Perturbed by Gene-specific Loss-of-function

Silencing of NOL5A, RPS2, HMGA1, TACSTD2, and RRM2 in CRC cell lines induced a substantial reduction of cell viability albeit with different kinetics and to a different degree. The broad mechanism for the effect on cell survival following LOF is firmly established for RRM2, as it is required for DNA synthesis,¹⁹ and for HMGA1, which acts as transcription factor and oncogene.²⁰ However, it is less well defined or unknown for the other genes. We therefore used computational pathway analysis tools to interrogate the LOF-RNAi signatures for each gene to identify critical biological networks and to identify commonalities. It should be noted that, while our phenotypic endpoint was reduction of cell viability, we saw minimal changes in the expression of genes involved in apoptotic or other cell death related pathways. This suggests that we have captured a picture of change in gene expression prior to cell death.

The RNAi signature for RRM2 stood out from the other signatures because almost 90% of the deregulated genes were upregulated as a consequence of RRM2 silencing. The RRM2 RNAi signature genes were connected through ERK and three members of the cyclin family (Fig. 5A), or through CTNNB1 (Fig. 5B). Additional networks centered on HNF4A, TGFB1, MAPK14 and NFkB, TNF and STAT3, SMARCA4, ERBB2, SMAD2, and TP53 (Fig. S2).

Figure 5. Pathway analysis of predicted functional relationships from gene-specific RNAi signatures.

Ingenuity pathway analysis of genes deregulated as a consequence of silencing RRM2 (A and B), RPS2 (C and D), and NOL5A (E and F). Red = upregulated genes, green = downregulated genes.

Within the TACSTD2 signature the next most down-regulated gene after the target itself was RRM2 (Fig. S3). Networks that included TACSTD2 itself showed HIF1A and NHF4A as connecting focus genes (Fig. S3). Two other networks reflect TP53 signaling and the ras-PI3k-Akt axis (Fig. S3).

We identified five networks associated with the HMGA1 RNAi signature: in the first, the focus genes are connected through TP53, the second was enriched in IGF signaling, and the third was defined by the ras-PI3k-Akt axis (Fig. S4). In these three networks, HMGA1 was included as a downregulated focus gene. In addition to that, we identified two networks that were connected solely through genes that were deregulated as a consequence of HMGA1 LOF, i.e., HMGA1 did not appear as a focus gene (Fig. S4). One of these networks was connected via the focus genes NFkB and STAT3, the second showed HNF4A as the central node. These two networks therefore describe yet unknown HMGA1 associated pathways (Fig. S4).

Interestingly, many of the transcripts within the RPS2 RNAi signature have minimal annotation and/or functional data. However, pathway analysis showed major nodes of network connections centered around MAPK1 in one network (Fig. S5), HNF4A in another one (Fig. 5C), and TGFB1 in a third network (Fig. S5). This third network was also enriched for multiple members of the matrix metalloproteinase gene family ADAM. The network in which RPS2 was downregulated as a focus gene was connected through CCNE1, ERK, and the translation initiation factor EIF4EBP1 (Fig. 5D).

When studying the deregulated genes associated with NOL5A silencing, we identified a network connected through ERK, MAPK14, CCNE1, and SERPINE1 (Fig. 5E). As observed for other target genes as well, another network was connected through HNF4A (Fig. 5F). Other networks included TGFB1, NFkB and SMARCA4, CTNNB1, or TP53 (Fig. S6). The latter was the only network, which included the downregulated NOL5A as focus gene. Interestingly, we observed that another network exhibited many focus genes as central hubs including JUN, FOS, SP1, HIF1A, VEGFA, and ERBB2 (Fig. S6).

While most of the transcriptional changes within each RNAi signature were specific for each silenced gene, there was an interesting overlap in the biological networks (Table S6), consistent with the Nested Effect Models analysis. For instance, for many RNAi signatures we generated networks with connecting nodes that included the focus genes TP53, CTNNB1 (b-catenin), NFKB, TGFB, and ERK, all of which are relevant for colorectal carcinogenesis. In addition, HNF4A played a dominant connecting role for all five genes, confirming recent findings demonstrating that inhibition of HNF4A activity inhibits colorectal carcinoma cell line growth in vitro and in vivo. ²¹

Comparison of RNAi Signatures with Expression Profiles of Primary Rectal Cancer

If overexpression of a gene in primary CRC suppresses the transcription of a network of genes, we speculated that silencing of this gene should lead to the upregulation of at least some of these genes within the RNAi signature. Conversely, if overexpression of a gene in primary CRC activates the transcription of a network of genes, then silencing of this gene should lead to the downregulation of these genes within the RNAi signature (Fig. S7A). We tested this hypothesis by performing an enrichment analysis (see Materials and Methods). Figs. S7B and C show a statistically significant inverse correlation for NOL5A, RRM2, and RPS2 RNAi signatures versus tumor profiles, i.e., upregulation of signature genes that are downregulated in primary carcinomas in cell lines after RNAi, and vice versa, validating our approach by lending weight to their biological significance.

Discussion

Robust array technologies allow global surveillance of tumor transcriptomes and matched normal tissues. Differentially expressed genes are likely to be involved in tumorigenesis, and can therefore serve as potential molecular targets for rational cancer therapy. However, the interpretation of global expression signatures in terms of their functional relevance remains a major challenge.

We therefore explored a loss-of-function approach to systematically investigate the role of overexpressed genes derived from previously established gene expression signatures of primary colorectal cancers for viability of colorectal cancer cells.^1,2 Our experimental strategy led us to focus on HMGA1, RRM2, TACSTD2, NOL5A, and RPS2 as important candidate genes because their downregulation diminishes the viability of CRC cell lines. Systematic mRNA expression analyses following RNAi-mediated gene silencing unveiled unique RNAi expression signatures pointing to critical transcriptional responses. The underlying deregulation of cellular pathways within these RNAi signatures could be recapitulated in an independent data set of 65 rectal carcinomas. To the best of our knowledge, we are the first to systematically combine gene expression profiling of primary tumors and extensive expression profiling after RNAi mediated gene silencing. This approach proved exceedingly robust.

While the independent identification of RRM2, and TACSTD2, and HMGA1 provides elegant proof-of-concept for our experimental strategy, the suggested role of NOL5A and RPS2 as potential therapeutic targets is a novel finding. RRM2 confers the enzymatic activity of the ribonucleotide reductase complex (RNR), which represents the rate-limiting enzyme in DNA synthesis and as such is required for DNA replication and DNA repair.¹⁹ It was previously shown that RRM2 is upregulated in colorectal cancers,²² and that its inhibition reduced the proliferation of cancer cells, triggering the development of RRM2 targeted siRNA-containing nanoparticles.^23,24 Our observations confirm that this gene has enormous potential as a therapeutic target. Importantly, we are the first to report a yet undiscovered link between RRM2 and PLK1, which was one of the most drastically down-regulated genes following RRM2 silencing (Table S4B and Fig. 5A). PLK1 encodes the Polo-like kinase 1 protein, a key regulator of cell division. Inhibition or depletion of PLK1 is associated with activation of apoptosis and it is actively being pursued as an anti-cancer target.²⁵ Since both proteins are in the focus of current anti-cancer therapeutic development efforts, this may open a new entry-point for combinatorial treatment.

Our data for TACSTD2 are consistent with the emerging role for this cell-surface receptor in CRC and potentially other cancers.^26,27 Recently, Wang and coworkers demonstrated that expression of TACSTD2 was required for the tumorigenic and invasive potential of colorectal cancer cells,²⁸ and TACSTD2 has been proposed as a marker for prostate basal cells with stem cell characteristics.²⁹ Of note, a therapeutic antibody targeting TACSTD2 was described recently.³⁰ Our data would warrant exploring the use of this antibody in CRC. Very interestingly, we identified RRM2 as one of the highly downregulated genes within the TACSTD2 RNAi signature, suggesting a potential functional link.

HMGA1 is a known transcription factor and colorectal oncogene.³¹ However, despite the wealth of information on this promising target, there are still no specific compounds in pre-clinical development. Itself a target of c-Jun, caseine kinase 2, protein kinase, Ubc9, the ATM kinase and – most recently - miR-16, it has been described from biochemical studies as a modulator of the Ras/ERK mitogenic signaling pathway, including KIT ligand and caveolins 1 and 2, and consequently a negative regulator of mTOR phosphorylation (additional pertinent references can be found in SI References). Since this is the first report elucidating the transcriptional pathways modulated by HMGA1, we anticipate that this might help to identify downstream targets that affect the same pathway and that can be inhibited by more traditional means.

The NOL5A gene encodes the nucleolar protein 5A, which is thought to be involved in the assembly of the 60S ribosomal subunit.³² Little is known about this gene or the druggable characteristics of its protein product, though several intriguing observations have recently been made. As a component of the ribonucleoprotein methylation complex, it is known to co-localize with the putative methyl transferase fibrillarin, and appears to associate with the nuclear phosphoprotein treacle specifically in telophase, but not during mitosis. It is also one of several genes whose downregulation was correlated with acquired chemotherapy resistance in vivo and in multicellular tumor spheroids, but not in monolayers (additional pertinent references can be found in SI References). To the best of our knowledge NOL5A has not previously been linked to human malignancies.

The ribosomal protein S2, RPS2, comprises a component of the 40S ribosomal subunit. Its exact function remains to be determined, though RPS2 has been identified as a substrate of the ribosomal protein methyltransferase, protein arginine methyltransferase 3 (PRMT3).³³ Recently, elevated mRNA levels have been associated with increased cellular proliferation, and overexpression of RPS2 has been detected in liver tumors, astrocytomas and prostate cancers.^34-36 Additionally, RPS2 has been identified as a novel tyrosine kinase substrate in breast cancer.³⁷ Its important role for the viability of colorectal cancer cells is a novel finding, which is substantiated by the fact that the RPS2 protein is not expressed in colon mucosa, but in the majority (72%) of tumor samples based on immunohistochemical staining (Fig. S8).

Interestingly, NOL5A and RPS2 both carry the conserved E-box canonical MYC binding sequence in their promoter region. MYC has been considered for many years as an attractive therapeutic target. However, its activity has been hard to inhibit specifically,³⁸ and targeting critical downstream proteins may be a more realistic approach. Importantly, we observed that the expression levels of MYC and both RPS2 and NOL5A are positively correlated in 65 primary rectal carcinomas (Fig. S9). While there is no formal proof yet that MYC actually binds to regulatory elements of these two genes in CRC, our analysis provides the first experimental evidence that they might be functionally linked. Furthermore, the expression levels of MYC and RPS2 show a positive correlation in lymphomas,³⁹ and it has been reported that the expression levels of NOL5A increase in the presence of MYC. ^40-42

Taken together, these results demonstrate the power of our strategy combining a functional genomics and a systems biology approach. We confirmed the involvement of HMGA1, RRM2 and TACSTD2 and defined their signaling pathways in the context of CRC, which may now enable identification of alternative means for targeting these genes. In addition, our RNAi signatures revealed other attractive candidates that have not yet been considered as target genes in CRC. Among the genes commonly deregulated after gene silencing, we identified the plausible upregulation of ANGPTL4, BRMS1L, and RND3 (RhoE). ANGPTL4 has been previously shown to block tumor cell invasiveness and metastases,⁴³ BRMS1L represents a metastasis suppressor gene,⁴⁴ and RND3 inhibits Ras-induced cellular transformation.⁴⁵ We also observed the downregulation of HAS2 and ID2, overexpression of which triggers tumorigenesis and progression of breast cancer cells,⁴⁶ and promotes survival of colorectal cancer cells.⁴⁷ Most importantly, we identified NOL5A and RPS2 as potential novel therapeutic targets.

Significance

Despite major improvements in elucidating the genetic changes underlying the initiation and progression of colorectal cancer (CRC), there remains a clinical need to implement novel targeted therapeutic strategies. Using cell based model systems that recapitulate the genomic and gene expression changes that we have previously observed in primary CRC we applied a combined functional genomics and systems biology approach to identify such anti-CRC targets. Characteristic loss-of-function signatures were generated for genes that we identified to be required for the viability of CRC cells. Subsequently, these defined gene expression changes could be recapitulated in the expression profiles of primary rectal cancers, suggesting that this is a biologically relevant phenomenon. In summary, we propose a strategy for the functional validation of colorectal cancer genes, and, implementing this approach, we identified potential therapeutic targets, including two genes that encode the ribosomal proteins NOL5A and RPS2.