A candidate gene approach for virally induced cancer with application to HIV-related Kaposi's sarcoma


Correspondence to: Brahim Aissani, Department of Epidemiology, School of Public Health, University of Alabama, Birmingham, AL 35294-0022, USA, Tel.: 205-975-8663, Fax: 205-934-8665, E-mail baissani@uab.edu


Like other members of the γ-herpesvirus family, human herpes virus 8, the etiologic agent of classic and HIV-related Kaposi's sarcoma (HIV-KS) acquired and evolved several human genes with key immune modulatory and cellular growth control functions. The encoded viral homologs substitute for their human counterparts but escape cellular regulation, leading to uncontrolled cell proliferation. We postulated that DNA variants in the human homologs of viral genes that potentially alter the expression or the binding of the encoded factors controlling the antiviral response may facilitate viral interference. To test whether cellular homologs are candidate susceptibility genes, we evaluated the association of DNA variants in 92 immune-related genes including seven cellular homologs with the risk for HIV-KS in a matched case and control study nested in the Multicenter AIDS Cohort Study. Low- and high-risk gene-by-gene interactions were estimated by multifactor dimensionality reduction and used as predictors in conditional logistic models. Among the most significant gene interactions at risk (OR = 2.84–3.92; Bonferroni- adjusted p = 9.9 × 10−3– 2.6 × 10−4), three comprised human homologs of two latently expressed viral genes, cyclin D1 (CCND1) and interleukin-6 (IL-6), in conjunction with angiogenic genes (VEGF, EDN-1 and EDNRB). At lower significance thresholds (adjusted p < 0.05), human homologs related to apoptosis (CFLAR) and chemotaxis (CCL2) emerged as candidates. This “proof of concept” study identified human homologs involved in the regulation of type I interferon-induced signaling, cell cycle and apoptosis potentially as important determinants of HIV-KS


acquired immunodeficiency syndrom


-C-C chemokine ligand 2 gene


-cyclin D1 gene


CASP8 and FADD-like apoptosis regulator gene


highly active antiretroviral therapy


human herpes virus 8


human immunodeficiency virus


human leukocyte antigen


Hardy Weinberg equilibrium


interleukin-6 gene


Kaposi's sarcoma


Kaposi's sarcoma herpes virus


multicenter AIDS cohort study


men who have sex with men

Kaposi's sarcoma (KS) is a multifocal angioproliferative inflammatory condition well described in its classical, post-transplantation and African endemic forms before the advent of the epidemic of HIV-1/AIDS.[1] Histologically, KS evolves from a small patch of new blood vessels to a vascular tumor, with a sparse inflammatory infiltrate to nodular lesions with proliferating spindle cells.[2] KS cell lines have been shown to express markers of endothelial, lymphatic and smooth muscle lineages.[3] Seroepidemiological studies have clearly established HHV-8 (also called Kaposi's sarcoma herpes virus, KSHV) as the etiologic agent of all clinical forms of KS. These studies have found increased KSHV seroprevalence rates in individuals at high risk for sexually transmitted diseases[4] and in the general populations of several central and east African and Mediterranean countries.[5, 6] Thus, KSHV infection occurs at higher rates among men who have sex with men (MSM) (∼40%) and in Mediterranean populations (5–20%), whereas it is less common in other unselected populations (1–5% in North America and Northern Europe). In the context of HIV/AIDS, KS is the most common malignancy,[7] although its incidence decreased precipitously with the advent of highly active anti-retroviral therapy (HAART). In HIV-1-infected individuals other than MSM (i.e., heterosexually infected women, injecting drug users, patients with hemophilia), the incidence of HIV-KS is as low in KSHV endemic countries of Southern Europe as in countries where KSHV infection is uncommon in the general population.[8] In other words, the high rate of KS in MSM appears to be due to both immunosuppression and shared host or behavioral risk factors.

HIV-1-infected MSM have a 50% risk of developing KS within 10 years when KSHV infection occurs before HIV-1 infection and within 5 years when KSHV is acquired following HIV-1 infection.[9] Replication of these data in the MACS cohort showed that those patients who seroconverted for KSHV after HIV-1 had more than twice the risk (relative hazard 2.19; p = 0.057) for developing KS. The risk associated with immunosuppression is apparently independent of HIV-1 infection because KSHV-infected transplant recipients have an increased cumulative risk (8–30%) for KS.[10, 11] Collectively, these data clearly establish the degree of immunosuppression as one of the most important risk factors for KS in KSHV-infected individuals.

Few studies have explored the host genetic determinants other than those of the suspected human leukocyte antigen system. Positive associations of HIV-KS with promoter polymorphism G-174C in the IL-6 gene[12] and with DNA variants in the gene encoding FcγRIIIa[13] have been reported but not replicated. We postulated that in susceptible hosts (HIV-1 and KSHV seropositive individuals), DNA polymorphisms in immune or cell cycle/apoptosis genes that alter the expression or the binding of the encoded factors involved in the antiviral response may provide favorable conditions for viral homologs to interfere with that response. To test this hypothesis, we selected 247 SNPs in a set of 92 human homologue and nonhomologue of KSHV genes involved in cellular pathways targeted by KSHV and evaluated their effects on the risk of HIV-KS. We present supportive data for the effectiveness of the proposed approach to identify candidate susceptibility genes for HIV-KS.

Material and Methods

Study participants

Eligible participants to the Multicenter AIDS Cohort Study, a prospective longitudinal study of the natural history of HIV-1 infection among 5,622 homosexual men recruited in 1984–1985 and 1987–1990 in major metropolitan US cities.[14] The cut-off date for the follow-up is January 1, 1996, when HAART became more widely available.


An immunoblot-confirmed positive ELISA defined HIV-1 seropositivity. Standardized T-cell phenotyping was performed at each follow-up visit. KSHV antibodies against KSHV lytic antigens were determined by use of an indirect immunofluorescence assay using 10-Q-tetradecanoyl phorbol 13-acetate induced body cavity B cell lymphoma-1 cells containing the KSHV genome. For each batch of serum samples tested, known KSHV-positive and -negative sera were assayed. All serum samples were tested twice in a blinded fashion and were assessed microscopically for the presence of whole cell immunofluorescence by the same reader. Sera from the enrollment or the immediately subsequent visit and from the most recent visit were tested. Positivity at either sample defined a KSHV-infected individual and KSHV negativity at both visits defined an uninfected individual.

Study design

The study sample consisted of 360 matched pairs of cases and controls predominantly (88%) of European American descent. Cases were defined as dually (HIV-1 and KSHV) infected individuals who later developed KS and controls as dually infected individuals who were free of KS. Controls were matched to cases by HIV/KSHV serostatus, race, KS-free time and CD4+ T lymphocyte cell counts (henceforth CD4+ counts). To account for the reported influence of the temporal order of HIV-1 and KSHV infections on progression to KS, cases were matched by controls within each of the four different serostatus groups [HIV-1-seroprevalent (SP)/KSHV-SP; HIV-1-SP/KSHV-seroconverter (SC); HIV-1-SC/KSHV-SP and HIV-1-SC/KSHV-SC)]. Time to KS was estimated as the date of KS diagnosis in index minus the baseline date or minus the date of the seroconversion date, defined as the mid-point between the last HIV-1- or KSHV-negative and the first positive date. Case and control pairs were matched for CD4+ counts at the visit within 1 year before index diagnosis.

SNP selection and genotyping

KSHV encodes at least 15 known human homologs implicated in immunoevasive pathways; however, at the time this study was designed, validated SNP assays were not available for all known human counterpart of KSHV genes. We, therefore, expanded the selection to include genes with documented relevance to KS pathogenesis (e.g., genes with upregulated or downregulated expression in KS lesions or KS cell lines and/or genes regulating angiogenesis). Overall, screening of Phase I HapMap resource and other resources (NCI SNP500Cancer, Seattle SNPs, Perlegen and NIEHS SNPs) identified an initial set of 284 SNPs from a selected set of 96 genes.

High-throughput genotyping was performed on the Illumina BeadArray® platform (Illumina Inc., San Diego, CA) for most SNPs. For a subset of the SNP selection, typing was carried out using custom TaqMan® assays from ABI (Applied Biosystems, Inc.) or bidirectional Sanger sequencing (Polymorphic DNA Inc., Alameda, CA) when several SNPs lie within short DNA segments such as in gene promoters.

Quality control

Reliability in our typing data was assessed by a small set of intra- and inter-plates blind duplicates. SNP calls were checked for adherence to Hardy-Weinberg equilibrium (HWE) in each of the KS outcome categories and only SNPs showing no significant deviation (p > 0.01) from HWE in KS-free controls were included for analyses.

Statistical analysis

Population composition

To minimize type-I error associated with population structure and to allow for a maximum number of cases and controls to be incorporated in our paired analyses, principal component analysis (PCA) and discriminant analysis were used to identify misclassified samples in the self-report of ancestry. A combined dataset based on 743 SNPs (247 SNPs from the present study and 496 SNPs from another investigation of the same study sample) was used to calculate the principal components.


We controlled for the confounding effects of age (age at the time of seroconversion to HIV-1 and KSHV infections), of the degree of immunosuppression and of the rate of immunity decline over the exposure time (time from dual infection to index diagnosis). The degree of immunosuppression was measured by the area under the curve (AUC), which was estimated by the trapezoidal method[15] as the time-weighted sum of the CD4+ counts below the cut-off level of 200/mm3 integrated over the exposure time. The slope of the CD4+ count changes over the exposure time measured the rate of immunity decline. Data points with more than three standard deviations from the mean were considered outliers and were not included in the calculation of the slope.

Association testing

To detect gene by gene interactions, which we have hypothesized to be the most likely determinants of risk because of the multiple cellular pathways targeted by KSHV, we adopted the reduction strategy used in GMDR (Generalized Multifactor Dimensionality Reduction)[16] and in the original MDR approach.[17] We first imputed the missing genotypes separately for the case and control samples on the basis of pair of haplotypes estimated via. the EM-based haplotype inference program HAPLORE [18] (http://www.soph.uab.edu/Statgenetics/People/KZhang/HAPLORE/). We then evaluated the effects of all possible two- and three-way gene interaction models instead of limiting the analysis to the best gene interaction models as in the MDR and GMDR approaches. An in-house algorithm in R was developed to classify SNP genotypes in the interaction terms as high or low risk variables and modeled them as dichotomous predictors in conditional logistic regression analysis. We performed the analysis under an additive genetic model and treated the low risk variable as reference in the estimation of the odds ratios (OR). Baseline age, AUC and the slope of CD4+ counts were used as covariates and the affection status as the outcome variable.

For the purpose of comparisons, we also evaluated the marginal effects of the studied SNPs using conditional logistic regression analysis with adjustments for covariates as indicated above. For each of the above tests, we derived the ORs and 95% confidence limits, and the p-values for the test of significance after adjustment for multiple testing using Bonferroni correction.

System biology

To validate our hypothesis, we tested whether the most significant genes identified in this study act in immune regulatory pathways relevant to HIV-KS pathogenesis. To this end, we used g:Profiler (http://biit.cs.ut.ee/gprofiler/index.cgi), a web based server for functional profiling,[19] to query several ontology databases including but not limited to Gene Ontology (GO),[20] Human Phenotype Ontology (HPO)[21] and the biological pathway databases KEGG[22] and Reactome.[23] We queried ontology databases with two different sets of candidate genes; those from the most significant (Bonferroni-adjusted p = 10−3) gene interaction models or those occurring in at least two significant (Bonferroni-adjusted p < 0.05) interaction models (top 28 genes). A stringent threshold (p = 10−4) adjusted for multiple testing using the Benjamini-Hochberg FDR approach was set to declare statistical significance. The null hypothesis H0 was that given a set of input genes Q as query, and genes associated to GO (KEGG, HPO…etc) term T, the number of common elements to Q and T, i.e., the intersection of Q and T, has appeared by random chance.


Quality control of the data indicated that 11 samples (0.9%) failed typing and among the remaining 709 typed samples, the call rate was greater than 98% in more than 92% of the sampled individuals. The overall concordance rate between the duplicates was 99.4%, implying that false finding due to typing errors or mishandling of the samples is unlikely. Of the 284 SNPs assayed, 12 (∼4%) did not meet the HWE criterion in controls and were excluded.

Concordant with the self-report of race, PCA yielded two distinct clusters representing populations of European and African American descents (Supporting Information Fig. S1; panel A). Discriminant analysis identified 22 misclassified individuals who were then reclassified on the basis of the first two principal components (Supporting Information Fig. S1; panel B). The majority of these individuals self-reported as Hispanic Europeans indicating that population admixture may explain the misclassification. Rematching of the cases and controls within this subset of 22 individuals resulted in a loss of three additional individuals because no matched cases or controls could be identified.

With our study sample composed of men only, a subset of 23 SNPs from chromosome X and two SNPs from chromosome Y were not eligible for MDR analysis because this approach uses autosomal genotype data. Based on a final set of 247 eligible SNPs (Supporting Information Table S1) in 92 genes, we show that no SNP main effect or two-way gene interaction remained significant at p = 0.05 after Bonferroni correction (not shown). However, 60 gene interaction models reached significance in the three-way analysis, of which the most significant 20 models (Bonferroni-adjusted p ≤ 0.01) are depicted in Supporting Information Table S2. None of these 20 interaction models involved any of the eight negative control SNPs composed of null Ancestry Informative Markers or SNPs located in large gene deserts. Examination of gene identity in the 11 top gene interaction models (Bonferroni-adjusted p ≤ 0.001) revealed CCND1 (cyclin D1 gene) and IL-6 (interleukin-6, alias interferon, beta 2, gene), two homologue genes expressed by KSHV, as important determinants of risk for HIV-KS (OR = 2.84–3.92; p = 9.9 × 10−3–4 × 10−3) (Table 1). In several instances, the associated gene interactions implicate nonsynonymous and functional noncoding SNPs (highlighted in bold, by parentheses or in italics in Table 1) with known variants at risk; for example, the IL-6 high-producing gene promoter variant G-174C (rs1800795) and the G870A splice variant (rs603965, merged to rs9344) of CCND1 influencing the relative production of Cyclin D1b. Consistent with the above findings is the observation that the gene encoding IL-6R, the cognate receptor of IL-6, also emerged in the interaction model 11 implicating the same CCND1 variant. As can be reasonably predicted for this tumor of endothelial origin, among the most significant interaction models identified in this study are those (models 3, 8, 9 and 11) implicating genes encoding angiogenic factors (VEGF, EDN1, EDNRB) central to HIV-KS.

Table 1. Best three-way gene interaction models at risk for HIV-related Kaposi's sarcoma
Model #SNP #SNP descriptionaBest gene modelsbORcLower 95% CLUpper 95% CLBonferroni-corrected p
  1. a

    Non-synonymous SNPs (in bold), SNPs in gene promoters (in parentheses) and SNPs in 3′ UTR (in italics).

  2. b

    Cellular genes acquired by KSHV (in bold). IL18, interleukin 18 (interferon-gamma-inducing factor) gene; HIVEP1, human immunodeficiency virus type I enhancer binding protein 1 gene; TNFAIP3, tumor necrosis factor, alpha-induced protein 3 gene; AKAP13, AKAP13 A kinase (PRKA) anchor protein 13 gene; ID4, inhibitor of DNA binding 4, dominant negative helix-loop-helix protein encoding gene; SERPINB9, serpin peptidase inhibitor gene; EDN1, endothelin 1 gene; TDP2, tyrosyl-DNA phosphodiesterase 2 gene; F13A1, coagulation factor XIII, A1 polypeptide gene; CCND1, cyclin D1 gene; NFKBIE, nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, epsilon gene; IL6, interleukin 6 (interferon, beta 2) gene; CXCR5, chemokine (C-X-C motif) receptor 5 gene; LPA, lipoprotein gene; HFE, hemochromatosis gene; EDNRB, endothelin receptor type B gene; LTA4H, leukotriene A4 hydrolase gene; PRL, prolactin gene; VEGF, vascular endothelial growth factor A gene; IL6R, interleukin 6 receptor gene.

  3. c

    Estimates of the odds ratio (OR) and confidence limits (CL) from conditional logistic models adjusted for the effects of age at the time of seroconversion to HIV-1 and KSHV infections and of immunosuppression (area under the curve and slope of CD4+ counts) from the time of dual seroconversion to the time of diagnosis of Kaposi's sarcoma in the matched cases and back imputation in the controls.

1100-20-27(rs2043055)-rs220022-rs598493IL18-HIVEP1-TNFAIP33.332.314.792.6 E-04
239-52-22rs745191-rs1047033-rs318477AKAP13-ID4-SERPINB93.172.224.535.2 E-04
379-165-244rs1794849-rs3181245-rs9504743EDN1-TDP2-F13A13.062.154.351.2 E-03
428-36-66rs603965-rs730775-rs1474347CCND1-NFKBIE-IL63.112.174.451.4 E-03
528-36-86rs603965-rs730775-(rs1800795)CCND1-NFKBIE-IL62.992.104.252.8 E-03
63-125-238rs3922-rs2228212-rs9457935CXCR5-HIVEP1-LPA3.072.144.423.1 E-03
728-36-107rs603965-rs730775-rs2069833CCND1-NFKBIE-IL62.942.074.184.0 E-03
828-73-215rs603965-rs1629862-rs6918586CCND1-EDN1-HFE3.922.516.145.4 E-03
9182-39-22rs3818416-rs745191-rs318477EDNRB-AKAP13-SERPINB93.022.104.345.5 E-03
1028-145-41rs603965-rs2540489-(rs849877)CCND1-LTA4H-PRL2.982.074.299.9 E-03
1128-40-284rs603965-rs833069-rs4129267CCND1-VEGF-IL6R2.842.014.039.9 E-03

Because our study is not statistically powered to detect interactions beyond the three-way interaction order, we used system biology to search for cellular pathways or functions shared by the interacting genes in models 1 to 11. Queries of GO databases with the 20 genes defining models 1 to 11 (Table 1) showed that the most likely cellular pathways are in the order of decreased importance those involved in (i) regulation of smooth muscle cell proliferation (Q&T/T = 0.057; FDR-adjusted p = 9.0 × 10−6), (ii) leukocyte chemotaxis (Q&T/T = 0.04; FDR-adjusted p = 5.6 × 10−5), and (iii) negative regulation of endopeptidase activity (Q&T/T = 0.038; FDR-adjusted p = 5.3 × 10−5) (Table 2). For increased specificity, we then restricted the search to cellular pathways defined by T (the number of genes associated to GO) of less than 100 and found that only cellular pathway # 3 (regulation of smooth muscle cell proliferation) remained significant at FDR-adjusted p < 10−4.

Table 2. Gene ontology for the genes in the most significant three-way gene interaction models
Obsp-valueTQQ&TQ&T/QQ&T/TIDNameGene list
  1. Gene ontology databases Gene Ontology (GO), Human Phenotype Ontology (HPO) and the biological pathway databases KEGG and Reactome were queried via g:Profiler, a web server for functional profiling, using as input gene query (Q) the set of 20 genes identified in the eleven most significant (Bonferroni-adjusted p<10−2) gene interaction models shown in Table 1. Only gene ontology pathways predicted at FDR-adjusted p<10−4 are shown.

  2. T denotes the number of genes associated to GO (KEGG, HPO…etc) terms. Q&T is the number of genes shared between Q and T (listed in the last column).

  3. The most important parameter is Q&T/T which represents the percentage of query genes found in a given GO pathway. Note that only pathway #3 (regulation of smooth muscle cell proliferation) remains significant when a more stringent cut-off (T = 100) for the number of functional categories was used.

14.4 E-056472080.40.012GO:0033993Response to lipidIL-18,TNFAIP3,AKAP13, SERPINB9,EDN1, CCND1,IL6,IL6R
25.3 E-051311950.2630.038GO:0010951Negative regulation of endopeptidase activityTNFAIP3,SERPINB9,IL6,LPA, VEGF
39.0 E-06872050.250.057GO:0048660Regulation of smooth muscle cell proliferationTNFAIP3,EDN1,IL6, VEGF,IL6R
45.6 E-051252050.250.04GO:0030595Leukocyte chemotaxisEDN1,IL6,EDNRB, VEGF,IL6R
59.8 E-067812090.450.012GO:0009725Response to hormone stimulusIL18,AKAP13,SERPINB9,EDN1,CCND1,IL6,LTA4H,PRL,IL6R
68.9 E-057092080.40.011GO:0008284Positive regulation of cell proliferationIL18,TNFAIP3,ID4, EDN1,CCND1,IL6, VEGF,IL6R

To ensure that no candidate genes were missed in our stringent sampling of only the most significant (FDR-adjusted p ≤ 10−2) gene interaction models, we extended the input query genes (Q) to include all genes that occurred in at least two interaction models with FDR-adjusted p < 0.05. This led to the identification of eight additional genes including two human genes acquired by KSHV, CCL2 (C-C chemokine ligand 2 gene) and CFLAR (CASP8 and FADD-like apoptosis regulator gene) (Supporting Information Table S3). Querying GO databases with the entire set of 28 genes ordered in decreased importance (frequency of occurrence in significant interaction models) or unordered identified a third pathway (positive regulation of JAK-STAT cascade, pathway #3, Table 3) in addition to pathways #3 and #4 of Table 2. It is worth emphasizing that at a stringent cut-off (T ≤ 100), genes like CCND1 that function far downstream in the signal transduction cascade are not among the listed genes in the candidate pathways; this is consistent with the JAK-STAT cascade being an early event in IFN signaling.

Table 3. Gene ontology for the most frequently occurring genes in gene interactions at risk for HIV-KS
Obsp-valueTQQ&TQ&T/QQ&T/TIDNameGene list
  1. Gene ontology databases Gene Ontology (GO), Human Phenotype Ontology (HPO) and the biological pathway databases KEGG and Reactome were queried via g:Profiler, a web server for functional profiling, using as input gene query (Q) the set of 28 genes (Supporting Information Table S2) that occurred more than once in 60 significantly associated (Bonferroni-adjusted p < 0.05) gene interaction models. In querying g:profiler, input genes were entered either as an ordered list of genes with decreasing order of frequency (pathways 1, 2 and 3) or unordered (pathways 4 and 5). Only gene ontology pathways predicted at a FDR-adjusted p < 10−5 are shown. T denotes the number of genes associated to GO (KEGG, HPO…etc) terms. Q&T is the number of genes shared between Q and T (listed in the last column). The most important parameter is Q&T/T which represents the percentage of query genes found in a given GO pathway.

14.3 E-06872560.240.069GO:0048660Regulation of smooth muscle cell proliferationIL6,IL6R,VEGF,EDN1, TNFAIP3,IFNG
22.6 E-06551650.3120.091GO:0002690Positive regulation of leukocyte chemotaxisIL6,IL6R,VEGF,EDN1, CCL2
33.8 E-07522860.2140.115GO:0046427Positive regulation of JAK-STAT cascadeIL6,IL6R,PRL, TNFRSF1A,IFNG,CD40
44.1 E-07522860.2140.115GO:0046427Positive regulation of JAK-STAT cascadeIL6,IL6R,PRL, TNFRSF1A,IFNG,CD40
59.9 E-06872860.2140.069GO:0048660Regulation of smooth muscle cell proliferationIL6,IL6R,VEGF,EDN1, TNFAIP3,IFNG


In this first attempt to test the hypothesis that human homologue of KSHV genes are potential susceptibility genes for HIV-KS, four (CCND1, IL-6, CCL2 and CFLAR) out of seven assayed cellular homologs, showed specific gene interactions conferring up to threefold increase of risk for KS. No simple model can predict how these gene variants favor or abrogate viral interference. Plausible models include different types of competition for common cognate receptors or binding partners, with the control of the immune response by the virus or the host being dictated by the polymorphisms affecting the expression or binding affinity of target host homologs or their binding partners (receptors or cofactors). Furthermore, study of system biology suggested that regulation of smooth muscle cell proliferation and leukocyte chemotaxis are the most likely target cellular processes in HIV-KS, consistent with the current belief that the proliferating spindle-shaped cells in KS lesions are of endothelial, lymphatic and smooth muscle lineages.[3]

The lack of significant main and two-way gene interaction effects in the presence of three-way effects deserves further comments. A straightforward explanation would be that the joint action, additive or combinatorial, of undetectable and numerous weak main effects predispose to higher risks for HIV-KS. Although this is currently considered to be more the rule than the exception in the etiology of cancer and complex diseases in general, understanding how genetic interactions lead to the expression of disease phenotypes is elusive. Nonetheless, because gene effects are context dependent[24] and molecular mimicry, as a pathogenic mechanism, interferes and abrogates the regulatory activity of numerous cellular genes, an alternative explanation for the absence of detectable main effects is locus heterogeneity (nonallelic heterogeneity) that diminishes the power to detect them. Under the proposed model, the affected cases carry risk variants in different combinations of homologue and nonhomologue genes but acting in the same cellular pathways antagonized by KSHV or relevant to KS such as in angiogenesis. This model is not inconsistent with our detection of several three-way gene interactions in which the interacting gene(s) function in specific cellular pathways targeted by KSHV. For example, CCND1 was observed in four distinct gene interactions, some of which implicate genes involved in the regulation of type I IFN-induced signaling, an immune response heavily antagonized by KSHV through expression of pleiotropic viral IRF homologs (vIRF-1-3).

Members of the IRF family play central roles in one of the early immune responses to viral infection through transcription of type I IFN genes, IFN-stimulated genes and other cytokines and chemokines. Although none of the four cellular IRF genes tested in this study (IRF-1, IRF-4, IRF-5 and IRF-8) was found in the top 11 gene interaction models, two of them (IRF-5 and IRF-8) occurred in more than two significantly associated interaction models. Human IRF-8 (hIRF-8) is the closest homolog (22% amino-acid identity) of vIRF-1 encoded by ORFK9; it is considered as a tumor suppressor and mutations in hIRF8 cause immunodeficiency and chronic myelogenous leukemia-like syndrome in mice.

The classification in latent and lytic replication patterns is relevant to our study because latency is generally assumed to be the state leading to cell proliferation, whereas lytic replication results in cell death and is, therefore, by definition antitumorigenic. Interestingly, viral homologs of CCND1, vCYC (ORF72), of IL-6, vIL-6 (K2) and of CFLAR, vFLIP (K13) encode latently expressed molecules. If replicated, this would imply that cellular homologs of latently expressed viral genes are potentially important determinants of HIV-KS.

We have hypothesized that the gene encoding the pleiotropic cytokine IL-6 is a strong candidate susceptibility gene because vIL-6 is present only in the Rhadinoviruses and KSHV is the sole human representative of this γ-2 herpesvirus subfamily. In-vitro studies have shown that, despite their similar structure (26% amino-acid identity) and downstream signaling activity, hIL-6 and vIL-6 have different receptor requirements and utilization. The subversive activity of vIL-6 resides in its independence of the IL-6α receptor (IL-6R, gp80 or CD126), which is downregulated by IFNs during KSHV infection.[25] In classical signaling, hIL-6 first forms a complex with the membrane-bound gp80 subunit which in turns associates with and activates the signal-transducing β-receptor chain gp130 (encoded by IL-6ST). In contrast, vIL-6 bypasses IL-6R and directly binds gp130. vIL-6 inhibits Tyk2 and STAT2 phosphorylation by IFN-αR thereby blocking downstream signaling and IFN induction of the tumor suppressor p21. The presence of IFN-stimulated response elements (ISREs) in the promoter of vIL-6 further enhances the expression of vIL-6 by IFN-α and provides a novel negative feedback mechanism. hIL-6 cannot reproduce this effect because IFN-α down-regulates surface expression of IL-6R.

Many of the biological activities of hIL-6 are mediated by trans-signaling through a naturally occurring soluble IL-6R (sIL-6R), which is either a splice variant of IL-6R or a product of a proteolytic cleavage. In the trans-signaling mode, signal transduction by hIL-6 relies solely on the extracellular concentration of sIL-6R, whereas that of vIL-6 is independent of it. Because trans-signaling is counteracted by a soluble form of gp180 (sgp130, sIL-6ST), which inhibits the activity of the [sIL-6R/hIL-6] complex, competition between vIL-6 and hIL-6 for the membrane-bound IL-6ST is a possible pathogenic mechanism of KS. In this scenario, genetic polymorphisms inducing low expression of hIL-6 and/or high levels of sIL-6R would favor trans-signaling by vIL-6. Our study revealed the -174 G > C promoter polymorphism of IL-6 (rs1800795) and the IL-6R SNP rs4129267, which is associated with a high level of serum level of sIL-6R (p = 10−57),[26] as gene interaction variants conferring excess risk for HIV-KS. In line with our hypothesis is a report of an experimental evidence for trans-signaling and expression of sgp130 in HIV-KS cell cultures.[27]

The observation of several independent gene interactions comprising CCND1 is appealing. First, this gene encodes the proto-oncogene cyclin D1, a powerful regulator of cell cycle progression extensively investigated in human cancer.[28] Second, polymorphism rs603965 (SNP 28) is the splice donor site G870A (Pro241Pro) that produces cyclin D1b, an alternate splice variant overexpressed in several cancers and associated with poor disease outcome.[29] Third, regulation by cyclin D1 is a late event in the mitogenic signal transduction cascade; this suggests that dysregulation of cyclin D1 activity is a determinant and necessary event before engagement of the cell in the neoplastic process.

As outlined earlier, our study was not powered to detect higher order gene interactions; nevertheless, several identified gene interactions are predicted by system biology to take place in the same cascade of signaling events or in distinct cellular pathways connected by pluripotent molecules. The pluripotent IL-6 cytokine is important in both physiological and pathological angiogenesis and its role in VEGF production in some cancer have been reported[30, 31]; thus, IL-6 may bridge IFN-mediated cell cycle arrest and apoptosis (abrogated by vIL-6), and angiogenesis in KS lesions.

Endothelin-1 (ET-1), the product of EDN1, plays a crucial role in migration and metastasis of human cancer cells[32, 33] and is likely to be upregulated in multifocal angioproliferative conditions such as KS. Indeed, overexpression of this angiogenic factor and its receptor ETBR, which is encoded by EDRNB, is a hallmark of KS development.[34, 35]

It is not quite clear how the other interaction models identified in this study relate to each other; our data suggest possible cross-talk between the identified signaling pathways, with the angiogenic ET-1 and IL-6 possibly as key molecules linking the pathogenic pathways. The identification of IL-18, IL-6, CCL2 and CXCR5 as potential candidate susceptibility genes suggests a skewed antiviral immune response toward the less effective TH2-cell response as a possible mechanism of KS pathogenesis.

Implicit in this approach to selecting candidate genes for HIV-KS is the evaluation of a strategy, a proof of concept for investigating cancer induced not only by the viruses of the herpesviridae family but also by those of other families that have similarly captured and evolved cellular genes for immune evasion.


Data in this manuscript were collected by the Multicenter AIDS Cohort Study (MACS) with centers (Principal Investigators) at The Johns Hopkins University Bloomberg School of Public Health (Joseph B. Margolick, Lisa Jacobson), Howard Brown Health Center and Northwestern University Medical School (John Phair), University of California, Los Angeles (Roger Detels), and University of Pittsburgh (Charles Rinaldo). Website located at http://www.statepi.jhsph.edu/macs/macs.html.