Impact of viral presence in tumor on gene expression in non-small cell lung cancer

Florida residents who underwent surgical resection for NSCLC at Moffitt Cancer Center and consented to the Total Cancer Care protocol between 2000 and 2013 were eligible in our previous study for a viral DNA detection in NSCLC tumor tissue samples [3]. Approval for the use of archived tissue and patient information was obtained from the University of South Florida IRB, Protocol No. MCC16765.

We randomly selected 70 archived frozen NSCLC tumor samples based on: 1) having enough volume of frozen tissue to perform the studies, 2) preoperative radiographs showed no pneumonia or distal atelectasis, and 3) no patient had chemotherapy or radiotherapy before resection. Resulting NSCLC samples encompass 10 primary adenocarcinomas, 10 bronchioloalveolar carcinomas (BAC, although currently termed adenocarcinomas with lepidic spread) and 30 squamous cell carcinomas (SCCs). In addition, we selected 10 resected stage IV tumors (three SCCs and seven adenocarcinomas) and their 10 matched surgically-resected distant oligometastases. Anatomic sites of those 10 metastatic tumor specimens were brain (n = 3), soft tissue (n = 2), adrenal (n = 4) and kidney (n = 1). 10 non-neoplastic lung specimens were also obtained as controls for this study.

All primary and metastatic tumor specimens underwent viral DNA screening tests by the Lawrence Livermore PanMicrobial Detection Array (LLMDA) designed to detect all sequenced viral families. The LLMDA was developed at the Lawrence Livermore National Laboratory (LLNL; Livermore, CA, USA) and designed to detect all sequenced viral and bacterial families, with appropriate controls [6, 7]. The 135 K format of the LLMDA (v.5) targets all vertebrate pathogens including 1856 viruses, 1398 bacteria, 125 archaea, 48 fungi, and 94 protozoa [8]. In the development of this microarray, PCR was used extensively to validate the results and verify that the statistical algorithm was accurate [9‐11]. The high-density oligo LLMDA and statistical analysis method has been extensively tested in numerous problems in viral and bacterial detection from pure or complex environmental or clinical samples [9, 12‐15]. A subset of NSCLC tumors [10 squamous cell carcinomas (SCCs)] was evaluated using an oncovirus panel of the International Agency for Research on Cancer (46 HPV types, 10 polyomaviruses, and 5 herpesviruses) [16]. In addition, all 70 NSCLC underwent HPV PCR genotyping using the INNO-LiPA Genotyping Extra Assay which detects 28 HPV genotypes classified as high or low risk, depending on their association with carcinogenesis. Details concerning the detection techniques, patient’s clinical characteristics and virus DNA detection results can be found in our previous study [3]. For the current study, the Moffitt institutional honest broker retrieved 39 available gene expression microarrays from these same tumor samples, of which the platform was Rosetta/Merck Human RSTA Custom Affymetrix Genechip with 60,607 probe sets interrogating 25,285 genes.

All microarray data were normalized using the iterative rank-order normalization algorithm [17], and experimental batch effects and outlying observations were further examined by a two-way hierarchical cluster analysis based on sample-wise correlation matrix as a distance matrix and a principal component cluster analysis. Unsupervised and supervised approaches for differential gene expression analysis and interpretation of resultant genes were utilized to associate virus infection with gene expressions in lung tumor cells according to NSCLC histological subtypes and metastasis status to minimize their potential confounding effects.

The unsupervised approach first used Linear Models for Microarray Analysis (LIMMA) to identify individual genes with significant differential expression between virus-infected and uninfected tumor samples [18]. Thereafter, a functional annotation and gene ontology (GO) analysis of the genes was performed by using Database for Annotation, Visualization and Integrated Discovery (DAVID) [19] and a database of virus-host protein-protein interactions (VirusMentha) [20].

For the supervised approach, Gene Set Enrichment Analysis (GSEA) was utilized to determine whether genes on a known biological or functional pathway have jointly concordant differential expressions between virus-infected and uninfected lung tumor specimens. Fifty hallmark, 189 oncogenic, and 4872 immunologic gene sets annotated in Molecular Signatures Database were subjected to GSEA [21]. For a multiple testing correction, false discovery rate (FDR)-adjusted p-values were estimated in all LIMMA, GO, and GSEA analyses, and a statistical significance was defined when the FDR-adjusted p-value was less than 0.2.

Sample-wise Spearman’s rank correlation coefficients of 39 microarrays over all probe sets ranged from 0.866 to 0.944. These high correlation coefficients indicated that a majority of genes had similar expression patterns across NSCLC tumor tissue samples regardless of their diverse histological subtypes and disease progression status. However, a two-way hierarchical cluster analysis and a principal component analysis of all microarrays revealed that all three brain metastatic NSCLC tumor tissue specimens (one SCC sample harboring the Y73 sarcoma virus (Y73SV) DNA, one uninfected SCC, and one uninfected adenocarcinoma samples) were clustered far apart from other primary lung tumor samples and the non-brain metastatic tumor samples, showing a brain-specific biological variation independent of viral infection status in gene expression profiles. We therefore excluded these three brain metastatic tumor samples from a differential gene expression analysis.

To identify differentially expressed genes (DEGs), LIMMA analysis was applied to a total of 36 NSCLC tumor samples: 21 samples harboring viral DNA of at least one viral type (Virus(+)) and 15 samples without any viral DNA (Virus(−)). This analysis identified 338 overexpressed and 301 underexpressed genes in Virus(+), as compared to Virus(−) (FDR p < 0.2; Additional file 1: Figure S1). For instance, PCYT1A was the most significantly overexpressed gene in Virus(+) (fold change (FC) = 2.24) whereas JMJD1C and CTNNB1 were the top two underexpressed genes with FC of 0.44 and 0.45, respectively. (Additional file 2: Table S1). Li et al. (2015) reported that PCYT1A catalyzes the rate-limiting step in synthesis of phosphatidylcholine that is required for replication of HBV. They also confirmed that PCYT1 was up-regulated at the transcriptional level in HBV-infected human hepatoblastoma cells [22]. In addition, Vaezi et al. (2014) found that PCYT1A is the dominant determinant of 8F1 immunoreactivity in lung SCC samples and that high expression of PCYT1A was found to be prognostic of longer disease-free survival [23]. JMJD1C is a component of DNA-damage response pathway with implications for tumorigenesis by demethylating MDC1 to regulate the RNF8 and BRCA1-mediated chromatin response to DNA breaks. Chen et al. (2016) identified that the gain of the miRNA regulation of MIR141 to JMJD1C resulted in the gain of protein-protein interaction between JMJD1C and GADD45A in hepatocarcinogenesis that is a multistep process mainly associated with persistent infection with HBV or hepatitis C virus [24]. CTNNB1 is considered the cancer drivers for hepatocellular carcinoma development with variable frequencies depending on the etiology. A recent genome-wise RNAi screen revealed that a role of a WNT/CTNNB1 signaling pathway as negative regulator of virus-induced innate immune responses [25]. Nakayama et al. (2014) also reported that pharmacological inhibition or conditional deletion of CTNNB1 inhibited lung tumor formation in transgenic mice [26]. A functional annotation of those DEGs was performed using DAVID tool to gain insight into their biological functions. For the 338 overexpressed genes in Virus(+), the protein catabolic process was the most significantly over-represented function (29 hit genes, p < 0.001) followed by cytoskeleton and proteasome core complex. For the 301 underexpressed genes in Virus(+), six biological processes, such as Ras GTPase binding and RNA polymerase II promoter, were significantly enriched (Fig. 1a and b). GSEA was next performed to identify predefined sets of hallmark, oncogenic, and immunologic genes having concordantly differential expressions between the Virus(+) and Virus(−) NSCLC samples. As a result, three oncogenic gene sets, CSR_late.v1.up, mTOR_up.v1_up, Rb_P107_dn.v1_up, were found to be significantly altered with positive enrichment scores, meaning that a majority of genes in those gene sets were simultaneously overexpressed in the Virus(+) (Fig. 1c, d, and e). Among them, the CSR_late_up.v1_up gene set comprises 172 genes up-regulated in late serum response of human foreskin fibroblasts and associated with increased risk of metastasis and death in human lung, breast, and gastric cancer [27].

LIMMA was performed to identify DEGs between 20 virus-infected primary NSCLC tumor specimens (Virus(+)) and 10 uninfected specimens (Virus(−)). Seven hundred seventy-seven genes were significantly overexpressed in Virus(+) and 751 genes underexpressed (Additional file 3: Figure S2). To understand biological meaning behind these DEGs and discover enriched functional-related gene groups, a functional annotation enrichment analysis was performed using the DAVID tool. Table 1 showed their functional annotation results. For the overexpressed genes, the cell cycle was the most significantly overrepresented function (23 hit genes, p < 0.001). Strikingly, two virus-related biological processes, 1) viral carcinogenesis (22 hit genes) and 2) human T-cell lymphotropic virus Type I (HTLV-1) infection (23 hit genes), were also significantly overrepresented, along with several NSCLC tumorigenesis-related pathways, including proteasome pathway [28] and p53 signaling pathway [29]. In particular, HPN (hepsin), ACTN4 (actinin alpha 4), and GP130 (interleukin 6 signal transducer) on the viral carcinogenesis pathway were known to be host cellular targets of three viral oncoproteins (HBx, Tax, and vIL-6) that lead to cell proliferation/survival, regulation of actin cytoskeleton, and proliferation angiogenesis, respectively (Fig. 2; Additional file 4: Figure S3) [30‐32]. On the other hand, cAMP signaling, vascular smooth muscle contraction, and metabolic pathways were the most representative pathways for the underexpressed genes in Virus(+) (FDR p < 0.05). A recent study reported that the cAMP signaling augments radiation-induced apoptosis in NSCLC cells [33].

Using GSEA of hallmark gene sets, e2F transcription factor target, G2/M-checkpoint, mTORC1 signaling, and mitotic spindle assembly were found to be concordantly over-expressed gene sets in Virus(+) (all FDR p < 0.2; Additional file 5: Figure S4A). The GSEA analysis of 189 oncogenic gene sets also revealed that Rb/P107_DN.v1.up, Rb_down [34], e2F1 target [35], GCNP/SHH [36], and mTOR_up.v1_up [37] were significantly enriched again with all positive enrichment scores (all FDR p < 0.2; Additional file 5: Figure S4B).

Noticeably, all SCC tumor specimens bear viral DNA of at least one viral type and thus differential gene expression analyses were performed for each viral type except HBV, with which only one SCC specimen was uninfected. Table 2 shows the list of 12 DEGs from the LIMMA analysis of the SCC specimens with and without viral DNA of each of Bovine leukemia virus (BLV), Panthera leo persica Papillomavirus Type 1 (PlpPV1), HPV, Simian T-cell leukemia virus Type 1 (STLV1), Type 2 (STLV2) and Type 6 (STLV6).

For BLV, PSG4 and CPB2 were significantly underexpressed in BLV(+) (n = 8), as compared to BLV(−) SCC specimens (n = 2). Of these, CPB2 is an extracellular matrix-regulated gene and has been considered an indicator for an impaired lung function [38] (Additional file 6: Figure S5A).

The GSEA of hallmark gene sets resulted in two significantly altered cell cycle-pathways. One was the G2/M checkpoint (FDR = 0.019) and the other was the e2F targets that encode cell cycle-related targets of e2F transcription factors (FDR p = 0.054) (Fig. 3a). Successively, in the GSEA of oncogenic gene sets, seven significant signatures came up and the most significant signature was a set of genes up-regulated in primary keratinocytes with knockout of both Rb1 and Rbl1 (FDR p = 0.038) (Fig. 3b) [34]. Above all, the GSEA of immunologic gene sets revealed that 133 immunologic signatures were significantly enriched. Among these, the top significant signature was WT_vs_NFATC1_KO, which comprises 200 genes up-regulated in B lymphocytes stimulated by anti-IgM under knockout of NFATc1 (ES = 0.674, FDR p = 0.026; Additional file 6: Figure S5B). This suggested that BLV in SCC tumor cells might interact closely with NFATc1, which is an oncogene involved in various functions in cancer [39, 40].

For PlpPV1, GCM1 and SMR3A genes were significantly overexpressed in PlpPV1(+) (n = 2) in comparison with PlpPV1(−) specimens (n = 8). As for STLV1, comparisons of gene expression between STLV1(+) (n = 2) and STLV1(−) specimens (n = 8) yielded five significantly overexpressed, such as FMN2 and MYEOV, and one down-regulated gene (SPRR3) in STLV1(+). A comparison of gene expressions between STLV2(+) (n = 3) and STLV2(−) specimens (n = 7) detected only one gene, CRISP2 (cysteine rich secretory protein 2), that was significantly down-regulated in STLV2(+) (Table 2).

Interestingly, although the LIMMA analysis for HPV57 and STLV6 viral types resulted in no significant DEG, GSEA yielded several significantly enriched gene sets. For HPV57, five hallmark gene sets, such as oxidative phosphorylation and Myc targets, were significantly enriched (Fig. 3C). Additionally, it was a unique oncogenic gene set that TBK1.DN.48 enriched significantly (ES = 0.58, p = 0.049; Additional file 6: Figure S5C). This gene set comprises 50 genes down-regulated in epithelial lung cancer cells upon over-expression of the proto-oncogene KRAS and knockdown of TBK1, and induced apoptosis [41].

Lastly, the GSEA of hallmark gene sets for comparing STLV6(+) (n = 3) with STLV6(−) specimens (n = 7) recaptured G2/M checkpoint (ES = − 0.581, FDR p = 0.053) and e2F transcription factor target (ES = − 0.622, FDR p = 0.084), both of which were also significantly altered in the aforementioned comparison between BLV(+) and BLV(−), but showed reversed expression changes with negative ESs, which seemed obvious because 6 of 7 STLV(−) SCC specimens were BLV(+) (Fig. 3d).

Eight of 10 (80%) primary adenocarcinoma specimens bear viral DNA of four different viral types (four with Y73SV only, two with HBV only, one with both HPV57 and Y73SV, and one with both Y73SV and STLV2). In the LIMMA analysis of any virus-infected (Virus(+)) (n = 8) versus uninfected primary adenocarcinoma specimens (Virus(−)) (n = 2), ACTC1 and PCSK2 were found to be significantly down-regulated in Virus(+) (Table 2; Additional file 7: Figure S6). ACTC1 is a member of actin cytoskeletons and was recently reported to be a potential candidate contributing to the enhanced lung tumor development [42].

A viral type-specific subgroup analysis was subsequently performed using the LIMMA for each of HBV and Y73SV with at least two infected samples. When HBV(+)(n = 2) and HBV(−) specimens (n = 8) were compared, five genes (CEACAM8, PRSS1, CALB2, NUDT4, and NAP1L6) were significantly over-expressed in HBV(+) whereas CEP170B, MTND6P4 and NUP62 were down-regulated (Table 2). Of these, PRSS1, CALB2, and NUDT4 shared a common molecular function of metal ion binding according to a DAVID gene ontology analysis. Noticeably, VirusMetha interactome analysis of the genes revealed that NUP62 protein has interactions with three viral proteins: P0C206, Q85601, and TAX. Additionally, NUP62 is an essential component of the nuclear pore complex and plays a novel role in centrosome integrity. A recent study noted that knockdown of NUP62 induced G2/M phase arrest, mitotic cell death, aberrant centrosome, and centriole formation [43].

Next the LIMMA analysis of six Y73SV(+) and four Y73SV(−) adenocarcinoma specimens identified one up-regulated gene, MARCH3, and three down-regulated genes, NMNAT2, ANKRD29, and QSER1, in Y73SV(+) specimens. NMNAT2 is involved in nicotinate and nicotinamide metabolism, and is a novel regulator of cell proliferation and apoptosis in NSCLC by binding with SIRT3 [44].

Unlike the GSEA results of SCC specimen data, however, no oncogenic and immunologic gene set was enriched in the GSEA analysis of primary adenocarcinoma data, indicating that viruses in SCC tumors might interact with human genes more strongly than those in adenocarcinoma.

There were only two BAC specimens bearing viral DNAs of Porcine circovirus type 2 (PCV-2) and Y73SV, exclusively. Several immune-related genes had high expression levels in the both virus-infected specimens, but not statistically significant (e.g. IGKV1–5 with FC = 96 and FDR p = 0.47; IGKV1D-13 with FC = 228 and p = 0.99) (Additional file 8: Figure S7A). Likewise, no gene set was enriched in the GSEA even though notch-signaling hallmark gene set and CRX_NRL oncogenic signatures were highly overexpressed in the two virus-infected specimens (Additional file 8: Figure S7B and S7C).

Y73SV was the unique viral type detected in two of nine metastatic tumor specimens. Therefore, all differential gene expression analyses were performed over all histological subtypes and metastatic disease sites. Using LIMMA, 69 DEGs including CRCT1 and MAGE9 were found (Additional file 9: Figure S8A). Based on DAVID annotation results of the DEGs, the top represented biological process was the positive regulation or activation of cell proliferation (FDR p = 0.13). It involved four overexpressed genes, BCL6, NTN1, NAMPT, and PBX1, and two underexpressed genes, MVD and VEGFC, in Y73SV(+) metastatic NSCLC specimens. The melanoma associated tumor antigen (MAGE) pathway was also significantly over-represented biological process and encompasses three genes (MAGEA9, MAGEA9B, and MAGEB2) overexpressed in Y73SV(+). Furthermore, 39 down-regulated genes were in part associated with the zinc finger binding process as well as the integral component of membrane (Additional file 9: Figure S8B, C, and D).

The VirusMentha interactome analysis next revealed that nine over-expressed genes (BNIP3, BNIP3L, ENY2, BCL6, TMEFF2, ZNF655, TMA7, SSR3, and GART) and five under-expressed genes (CLTA, RPS21, SUB1, ITSN2, and TSNARE1) in Y73SV(+) had significant virus-host protein-protein interaction with human adenovirus 2 (FDR p = 0.019) and human immunodeficiency virus type 1 (FDR p = 0.027), respectively (Fig. 4). Among them, SSR3 has the most complex interaction network with viral proteins, in terms of the number of interacted viral proteins, followed by ENY2 and GART. In particular, BNIP3 is an apoptosis-inducing protein that can overcome Bcl-2 suppression and plays an important role in the calprotectin (S100A8/A9)-induced cell death pathway [45]. Both BNIP3 and BNIP3L interact with small T-antigen E1B viral protein that is a putative adenovirus Bcl-2 homolog that inhibits apoptosis induced by TNF or FAS pathways, as well as p53-mediated apoptosis [46]. It was also reported that without E1B function, virus production is compromised because of premature death of the host cell [47].

Lastly, in order to explore metastasis-specific genes, we compared the 69 DEGs with the 1527 DEGs, which was identified in the above primary tumor analysis comparing Virus(+) to Virus(−). Consequently, eight genes (C17orf80, SUB1, PYROXD2, IFT57, PAM, ARMC8, SSR3, and BNC2) were in common. The first five genes showed expression changes in the same direction with negative fold changes for C17orf80 and SUB1, and positive for PYROXD2, IFT57, and PAM in both Virus(+)-primary and -metastatic specimens (Additional file 9: Figure S8D). Since the primary tumor analysis involves various viruses, the 69 DEGs were further compared to the four genes that had differential expressions between Y73SV(+) and Y73SV(−) primary NSCLC tumors. No overlapping gene was found. Collectively, these observations suggested that Y73SV might have different host cellular gene targets and differently influence their expressions in primary and metastatic tumor cells.