Chemokines are a large superfamily of small, structurally-related peptides originally discovered as neutrophil attractants. Interleukin-8 (IL-8) is a potent member of the supergene family of CXC chemokines with ELR motif (ELR+). ELR⁺ CXC chemokines are potent angiogenic factors, whereas ELR^- CXC chemokines are potent angiostatic factors. Studies have shown that these angiogenesis-related activities are correlated with tumorigenesis in many tumor types and are distinct from their ability to recruit neutrophils [1, 2].

Interleukin-8 (IL-8) is inducible in a wide range of cells including lymphocytes, monocytes, endothelial cells, fibroblasts, hepatocytes, and keratinocytes [3–5]. IL-8 is also found to be constitutively expressed in several tumor tissues including bronchogenic carcinoma, non-small cell lung cancer, colorectal carcinoma, breast cancer, melanoma, prostate cancer, gastric carcinoma, and ovarian cancer. While IL-8 has been implicated in growth-potentiation [6], angiogenesis [7], metastasis [3, 8], and tumorigenesis [3, 6] of various tumors, its specific role remains unclear. Several observations support the assertion that a complex interaction between IL-8 and several other growth factors, cytokines, or other proteins is responsible for these tumor-related events [5, 9–13]. Growth regulated protein alpha (GRO-α), beta (GRO-β), and ENA-78 have been reported to be co-induced with IL-8 in A549 cells stimulated with two proinflammatory cytokines, IL-1β and TNF-α [14]. This result is consistent with the presence of nuclear factor kappa B (NF-κB) consensus binding sites in the promoter regions of all three genes [4, 15]. In addition, GRO-α and -β are also members of the ELR⁺ CXC cytokine family that are reported to be important mediators of tumorigenesis through their angiogenic properties [2, 16]. They both share one of the IL-8 receptors, CXCR2, which has been postulated to regulate the ELR⁺ CXC chemokine-mediated angiogenesis and resulting-tumorigenesis [1]. These observations suggest that GRO-α and -β may exhibit expression profiles similar to IL-8. Coordinated expression of these and other factors with IL-8 in certain tumor tissues suggests that they may be functionally associated with IL-8 in these tumors. In order to fully understand IL-8's role in these events, it is important to investigate these coordinately expressed genes. A systematic, unbiased approach to identification and ranking of proteins related to IL-8's expression in tumor tissues/cells may help to define the scope of IL-8's role in tumorigenicity and important related interactions with other factors.

The vast amount of available EST data allows expression analysis based solely on computational methods. It is possible to construct a "virtual tissue" based on the expression pattern of a particular gene or group of genes (e.g. tumor marker genes). It is also possible to group genes by tissue type through combining genes from the same source tissue (e.g. tumor, brain, etc.). In this study we have generated "virtual tissues" based on (a) expression of IL-8 in all tissues and (b) expression of IL-8 in tumor tissues. Next we generated lists of genes that are most highly co-expressed with IL-8 in these two virtual tissues. Since the method is not limited to genes previously correlated with IL-8 expression nor to known genes, we have the opportunity to identify previously overlooked correlations with known or novel genes. In addition, the relative strength of the correlations is measured by a Z-score based on the number of clones representing IL-8 or IL-8-tumor tissue compared with the total number of clones. Assuming that co-expression is related to function, these Z-scores provide a basis for ranking genes according to potential involvement in IL-8's function. This reference database for genes co-expressed with IL-8 can be cross-referenced with other large scale expression analyses, such as microarray experiments, to help decipher the regulation network and the functions of IL-8 and IL-8 co-expressed genes.

Among the genes most highly correlated with IL-8 in tumor tissues are KLF2, PS1, VEGF, and tumor necrosis factor receptor-1 (TNFR-1), lumican (keratin sulfate proteoglycan), claudin-3, and perlecan. Cytokines (IL-2 and IL-7) induce the expression of KLF2 in activated T cells which correlates with their survival [20]. They suggest that KLF2 may be involved in avoiding activation-induced cell death. It is possible that IL-8 may also mediate similar actions. This type of action could increase the ability of tumor cells to survive under conditions where cells normally apoptose.

Lumican expression has previously been correlated with higher tumor grade, lower estrogen levels in the tumor and younger age of patients in human breast cancer [21]. In addition, lumican is structurally related to both biglycan which is among the top IL-8-tumor related genes and the human embryonal carcinoma marker antigen TRA-1-60 [22]. These relationships support the hypothesis that lumican may be a tumor related protein involved with IL-8's tumorigenic function.

PS1 was also found to be highly correlated with IL-8 in both the entire list of tissues and specifically in tumor tissues. Based on EST distribution, presenilin 2 (PS2) is less related to IL-8-tumor libraries (Z-score = 0.3, IL-8-tumor clone/Total Clone = 10%). In the breast tumor microarray experiments, unlike PS1, the peak level of expression for PS2 is not in BT-549. PS1 is a gene involved in early-onset familial Alzheimer's disease. There is accumulating evidence that mutations in PS1 accelerate neurodegeneration and facilitate apoptosis, and some researchers [23] suggest an association with the p53 signal transduction pathway. It was suggested that down-regulation of PS1 by wildtype p53 and also p21^WAF-1 may be independent mechanisms leading to apoptosis and tumor suppression [24]. Our data correlating PS1 with IL-8-tumor related tissue also suggest a potential anti-apoptotic role for PS1 in tumorigenesis.

Vascular endothelial growth factor (VEGF), like IL-8, is a potent mediator of angiogenesis. VEGF has been shown to regulate angiogenesis and metastasis of bladder cancer [25] and several recent studies have reported correlations with IL-8 in several cell types including bladder cancer (TCC), non-small cell lung cancer (NSCLC) [26, 27], human brain microvascular endothelial cells (HBMECS) [28] and monocytes [29]. These data collectively demonstrate the correlation of VEGF with IL-8 in tumor tissues.

TNF-α is a well-characterized, potent inducer of IL-8 transcription and an anti-apoptotic agent [30–32]. Roebuck provides a detailed analysis of IL-8 promoter structure including a promoter recruitment mechanism which involves TNF-α and the cooperativity of NF-kB and NF-IL-6 binding sites [4, 32]. TNF-α exerts many of its effects through TNFR-1 and TNFR-2 receptors, thus co-expression of these receptors with IL-8 might be anticipated. Our analysis does reveal a correlation between TNFR-1 and IL-8 expression. Using another bioinformatic approach, Eidelman et al. report a close relationship between IL-8 secretion in cystic fibrosis cells and expression of genes from the TNFR-1/NFkB pathway. As is the case for several other genes found in this study, the TNFR-1/NFkB pathway is associated with p53/p21^WAF-1 tumor suppressor systems [33–37].

Our analysis identified several interesting genes co-expressed with IL-8 in tumor tissues which have not previously been associated with IL-8. Along with lumican, claudin-3, KLF2, PS1, VEGF and TNFR-1 mentioned above, these include several secreted proteins (secreted apoptosis related protein 1 and NPDC1) which are also likely to function in coordination with IL-8; several genes from the well-known family of complement factors (C1 esterase, human complement C4A and complement factor B); and several unknown or hypothetical genes. Interestingly, several of the genes correlate with the p53/p21^WAF-1 tumor suppressor network. We expect that some of these genes may present important therapeutic targets at one or more levels in the network of IL-8 mediated interactions related to tumorigenesis.

Like all EST-based expression analysis, there are limitations for this in silico method. First, EST-based expression analysis is not suitable for transcripts with low abundance (no or few representative ESTs in the database), which precludes statistical analysis. Second, EST-based expression analysis can only indicate the co-expression of genes, but can not accurately measure the expression level. Third, to generate a comprehensive reference database, we decided not to exclude cDNA libraries based on the sequence depth or special manipulation (e.g. subtraction and normalization). The depth of sequencing and special manipulation are not consistent across cDNA libraries, which could also influence the sensitivity of the analysis.

The computational expression analysis methods developed here are not limited to the identification of IL-8 related genes, but can also be applied to many other proteins of interest. This method is complementary to other large-scale expression analysis methods (e.g. microarray) in that it is not limited by the physical presence of a gene on microarray, thus it offers an unique approach to discovering potential functional links between genes through expression profiling.