Abstract
Background Transcriptomic profiling has been pivotal in better comprehending the convoluted biology of tumors including head and neck squamous cell carcinoma (HNSCC). Recently, mitogen-activated protein kinase-activated protein kinase-2 (MAPKAPK2/MK2) has been implicated in many human diseases including tumors. We recently reported that MK2 is a critical regulator of HNSCC and functions via modulating the transcript turnover of crucial genes involved in HNSCC pathogenesis. Consequently, to expand our insight into the biological relevance of MK2 and intricate cross-talks in tumor milieu, we conceptualized a comprehensive transcriptome analysis of HNSCC.
Results We performed an extensive transcriptomic profiling to ascertain global patterns of gene expression in both in vitro and in vivo experimental models of HNSCC which exquisitely emulates the tumor microenvironment. Transcriptomic characterization substantiated an intrinsic role of MK2 and certain MK2-regulated genes in HNSCC pathogenesis, an outcome that reiterates our recent findings. Annotation and differential gene expression analysis revealed candidate genes whilst pathway enrichment analysis corroborated the biological significance of findings. Furthermore, advanced gene expression assays through nCounter system (primary validation) in conjunction with immunohistochemical analysis (secondary validation) validated the transcriptome profiling outcomes quite robustly.
Conclusions Our results have underpinned the importance of seven differentially expressed MK2-regulated genes which are constitutively involved in HNSCC pathogenesis and could serve as potential candidates for future endeavors pertaining to therapeutic interventions and diagnosis pertaining to HNSCC. Collectively, our findings have paved the way towards the identification and development of new effective tumor markers and potential molecular targets for HNSCC management and improved clinical outcomes.
Background
Multifaceted regulatory networks tend to connect genes within a myriad of cellular processes. The involvement of a plethora of genes in fundamental biological processes viz. cell differentiation, growth and programmed cell death, and their involvement in many diseases is presently known (Spataro et al., 2017). However, the apprehension of their roles at a global level is still incomplete. Gene transcription and regulatory networks in conjunction with new genome-wide approaches have garnered huge attention in the pretext of gene regulation. Nevertheless, post-transcriptional mechanisms like transcript stability and translation are also highly crucial and require intricate regulation via a multitude of intracellular signaling pathways. Particularly, the control of transcript stability through phosphorylation-mediated regulation of RNA-binding proteins (RBPs) by mitogen-activated protein kinases (MAPKs) has been a topic of huge interest (Lasa et al., 2000; Venigalla et al., 2012, Soni1 et al., 2019, Soni2 et al., 2019).
Head and neck squamous cell carcinomas (HNSCCs) having an incidence rate of ~600,000 cases yearly, is the seventh most common cancer worldwide and one of the most lethal with overall mortality rate of 40-50% (Jemal et al., 2011; Ferlay et al., 2015). HNSCCs are classified either histologically (Woolgar et al., 2011) or via analysis of global transcription that employs etiology specific profiles (Martin et al., 2014; Van Waes et al., 2017). However, when these parameters were used for patient clustering, specific differences were observed in clinical behavior of patients as well as their response to therapy (Denaro et al., 2014). The survival rates of HNSCC patients have not improved much, hence, HNSCC has been rightly termed as a malignant tumor having a low survival rate (Narayanan et al., 2015). Consequently, an improved mechanistic insight into the molecular basis of HNSCC pathogenesis is urgently required to help in early diagnosis and development of effective therapeutics aimed at improved clinical outcomes (Jamali et al., 2015).
The role of differentially expressed genes (DEGs) and endogenous RNA networks in HNSCC is not fully deciphered. Past reports on genome and transcriptome studies in various human tumors have revealed aberrant regulatory programs, driver mutations, and disease subtypes (Weinberg, 2014). The cancer genome study has acted as a valuable tool for classification, diagnosis, and prognosis in HNSCC. There have been many past reports pertaining to genomic alterations in HNSCC (Chau et al., 2016; Bossi et al., 2016). Recently, the cancer genome atlas (TCGA) led to a global analysis of the major molecular changes, comprehensive landscape of transcriptomic alterations and pathogenesis-linked signalling pathways in tumors thus contributing to the identification of novel prognostic biomarkers or specific anticancer molecular candidate groups (Cancer Genome Atlas Network, 2015; Campbell et al., 2018). However, there is still need of extensive research insights to decipher the prognostic value attributed by these genomic alterations in tumors like HNSCC. A variety of biomarkers like MAPK phosphatase-1 (MKP-1), p16, p27, p53, tumor necrosis factor-α (TNF-α), and vascular endothelial growth factor (VEGF) have been shown to be linked with HNSCC (Thomas et al., 2005; Soni2 et al., 2019), but they have not proved to be sufficient in accurately defining HNSCC pathogenesis. Single biomarkers have generally proved insufficient in the prediction of therapeutic response thereby necessitating research on combinatorial markers through high-resolution “omic” profiling (Bhatia et al., 2012). Consequently, identification of new reliable molecular biomarkers associated with HNSCC using omics-based analysis is the need of the hour in order to develop them as novel potential therapeutic targets (Fiore et al., 2016).
Lately, the mRNA regulatory networks involved in tumor progression have garnered huge research interest. Recently, many reports have showcased the role of these intricate networks in tumorigenesis (Fang et al., 2018). However, research insights in this area are quite limited, thereby pointing to a pertinent need of comprehensive analysis of mRNAs and regulatory networks. Next generation sequencing (NGS) has fast evolved as an important tool for epigenomic, genomic, and transcriptomic profiling (Cieślik et al., 2018). Technological advances in mining and deciphering the vast cancer transcriptomic data has enabled us to better comprehend the complexity of various tumors and has streamlined efforts to discover novel biomarkers and therapeutic targets aimed at tumor management (Cieślik et al., 2018).
In our recent study, we elucidated the role of mitogen-activated protein kinase-activated protein kinase-2 (MAPKAPK2 or MK2) in HNSCC pathogenesis using clinical tissue samples, cell lines and heterotopic xenograft mice model (Soni2 et al., 2019). We reported that MK2 is critically important in regulating HNSCC, and acts by modulating the transcript stability of crucial pathogenesis-related genes. We also established that knockdown of MK2 attenuates tumor progression in a xenograft mice model. Thereupon, in an attempt to delve deeper into the mechanistic insights of the role of MK2 and to decipher the molecular markers responsible for MK2-mediated changes in HNSCC pathogenesis, we conceptualized to perform a comprehensive transcriptome profiling of HNSCC cell line and xenograft-derived tumor samples evaluated in our recent study.
In the present study, we have evaluated the global mRNA expression profiles of our previously established HNSCC experimental model sets using transcriptome analysis on NovaSeq 6000 system (Illumina Inc., USA). The in vitro HNSCC cell line model including CAL27-MK2WT (wild-type) and CAL27-MK2KD (knockdown) cells cultured in both normoxic as well as tumor microenvironment mimicking hypoxic conditions comprised the first set. The in vivo heterotopic HNSCC xenograft bearing tumors from CAL27-MK2WT and CAL27-MK2KD cells in immunocompromised mice as previously described (Soni2 et al., 2019) formed the second set. The findings of the transcriptomic profiling corroborated with our recently published results, hence, ascertaining a crucial role of MK2 in HNSCC pathogenesis via transcript stability regulation (Soni2 et al., 2019). Thus, by comprehensive transcriptome analysis we have elucidated certain specific MK2-mediated DEGs and regulatory networks that play an integral role in HNSCC pathogenesis in our experimental models. Furthermore, we performed gene expression assays on the nCounter system (NanoString Technologies, Inc., USA) in order to obtain a sensitive, highly multiplexed, and reliable detection of the specific mRNA targets based on our initial transcriptomic profiling. Finally, we cross validated the nCounter gene expression analysis results in in vitro setting using immunohistochemical analysis and found that the findings are in consonance and reiterate our recent discovery that MK2 is critically involved in regulating HNSCC progression and functions by modulating the transcript stability of crucial pathogenesis-related genes. The assays yielded highly precise and reproducible data which confirmed transcriptomic findings and yielded seven candidate MK2-regulated genes intrinsically involved in HNSCC pathogenesis which could potentially aid in discovery of new molecular markers for HNSCC management and diagnostic benefit.
Results
Qualitative assessment of the generated cDNA library followed by filtering and assembly of reads depicted optimum alignment
The RNA samples used for sequencing were isolated from appropriate cells/tissues using RNeasy Mini kit (Qiagen, Germany) and the quality assessment of the isolated total-RNA as well as the cDNA library formed was performed using Bioanalyzer (Agilent 2100, Agilent Technologies, USA). The RNA integrity number (RIN) value of all the RNA samples was found to be >5 (exclusion criteria for this study). It is quite evident from the outcome of the quality check analysis that all the RNA samples utilized, as well as the cDNA library constructed were of apt quality for their use in downstream experiments (Figure 1a, S1 and S2).
Deep sequencing of RNA obtained from NovaSeq 6000 platform resulted in 349 million raw reads (~58.2 million raw reads per sample) of average insert size 210bp. Figure 1a and S3 summarizes the quality check (QC) results of the sequencing experiment. The raw FASTQ sequences were filtered using NGSQC tool kit to obtain high quality (HQ) reads based on the set parameters, generating 258 million filtered HQ reads (~43.1 million HQ reads per sample), amounting to 74.1% of total raw reads, hence, implying that the obtained data was of good quality. A total of 25.7 giga bases of data was generated which foretells the enormity and wide complexity of the human genome.
The HQ reads obtained were further considered for downstream analysis and mapped over the human reference genome. The alignment performed employing the Kallisto pipeline was optimum with approximately 88.7% of the HQ reads on an average getting mapped to the human reference genome (Figure 1a, 1b and S4).
Annotation and differential gene expression analysis revealed candidate genes
The present study was conducted using two distinct experimental datasets as mentioned in Figure 1a and Table 1. Using the criteria of fragments per kilobase of transcript per million mapped (FPKM)≥0.1, we identified on an average around 62791 expressing transcripts in all the samples, representing thousands of genes. A-D datasets depicted the in vitro HNSCC cell line samples while F vs E illustrated the in vivo xenograft dataset. As detailed in Figure 1a and Table 1, we performed analysis in MK2KD vs MK2WT scenario under both normoxic and hypoxic conditions to obtain a comprehensive picture of the changes in the global gene expression pattern in the backdrop of tumor core emulating hypoxic niche. Firstly, we evaluated the differential gene expression in the various studied datasets and found a large pool of DEGs, precisely 1403 in B vs A, 924 in D vs C, 1360 in D vs B, 1456 in C vs A, and 984 in F vs E as assessed by the pre-set cut-off values. Figure 1c represents the total number of upregulated (fold change (FC)>2) and downregulated (FC<-2) genes among all the DEGs in the analyzed datasets.
Pathway enrichment analysis exhibited the biological significance of our findings
The multitude of DEGs in various experimental datasets in the transcriptomic profiling are implicated in hundreds of significant biologies/biological processes as summed up in Figure 1d. To gain further insight into the biological significance of changes in gene expression and to attain a global picture of the molecular pathways possibly contributing to HNSCC pathogenesis, pathway enrichment analysis was performed using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. It yielded an integrated network analysis and revealed the top five biological processes enriched in the experimental datasets. The top gene ontology pathway analysis for the DEGs is depicted in Figure 1e. It was notably found that a significant percentage (~5%) of the total DEGs in the various studied combinations belonged to the pathways involved in cancer progression (Figure 1e). The DEGs showing significant changes between various groups were then selected, followed by generation of heat maps to assess clustering of gene expression profiles among the experimental datasets (Figure 2a). Figure 2b further depicts the distribution of all the transcripts on the two dimensions of −log(P) and FC by way of volcano plots with differentially expressed transcripts highlighted in blue.
In A-D datasets, the DEGs belonged to a large number of biological processes thereby limiting the information that could be harnessed. Henceforth, in an effort to filter down data and fulfill the aim of extracting valuable leads, we selected the 77 elements/biological processes that were common in A-D combinations using Venny 2.1.0 (http://bioinfogp.cnb.csic.es/tools/) (Figure 2c). These processes belonged to numerous categories which are listed in Table S1. On the background of this knowledge, we further filtered down the common genes in these 77 biological processes (Figure 2d). Consequently, we obtained five genes common in A-D combinations namely EH domain binding protein 1 (EHBP1), SMC5-SMC6 complex localization factor 2 (SLF2), death associated protein 3 (DAP3), inositol hexakisphosphate kinase 2 (IP6K2) and runt related transcription factor 1 (RUNX1) as listed in Table S2. Notably, these genes play specific roles in HNSCC pathogenesis possibly via MK2-mediated regulation.
Gene regulatory networks and pathways depicted the importance of candidate genes in HNSCC pathogenesis
Gene regulatory networks for these five common genes in A-D datasets furnished a detailed overview of the various inter-connections and the biological processes affected by them. Collectively, the results showed that IP6K2 plays an intrinsic role in nucleotide binding, while DAP3 in apoptosis and poly(A) RNA binding as confirmed in past reports (Thomas et al., 2014; Wazir et al., 2015). Interestingly, these genes showed differential regulation in various experimental datasets (Figure S5-S8), hence, clearly signifying that MK2-knockdown as well as hypoxic tumor microenvironment affects the genes and pathways in HNSCC via differential regulation, pointing to a central role of MK2 in transcriptional regulation of HNSCC.
Furthermore, to retrieve the information about the MK2-regulated candidate genes intrinsic to HNSCC, we narrowed down our search to cancer related processes for further analysis. This filtered down our dataset to 16 cancer-specific biological processes that were common in A-D datasets (listed in Table S3). In these 16 processes only two genes, DAP3 and RUNX1, were found to be common as per the analysis performed using Venny 2.1.0 (Figure 2e and Table S4). Collectively, these findings have indicated that these two genes are intrinsically involved in HNSCC pathogenesis and warrants further investigation in this regard. Notably, the 16 processes were found to be clustered in 5 major biological pathways (Figure S9), clearly showcasing that MK2 portrays an intrinsic role in the hypoxic tumor microenvironment by regulating processes like apoptosis and transcription, hence, potentiating and strengthening our latest in vitro findings (Soni2 et al., 2019).
Similarly, on comparing the mice bearing CAL27-MK2KD tumors with those bearing CAL27-MK2WT tumors (F vs E xenograft dataset), we found that the DEGs were clustered in 14 biological processes of relevance to tumor pathogenesis as listed in Table S5. The genes involved in these biological processes were then filtered out and Figure S10 shows the gene regulatory network for the same. This analysis provided us with certain candidate genes like TRAF2 (apoptosis); EPB41L1 (cytoskeleton); FOXO3, H2AFY and YAP1 (transcriptional regulation); and DIDO1 (RNA binding) that are involved in key cellular processes in the xenograft model. These findings can be explored further to decipher the putative role of these potential candidate genes in HNSCC pathogenesis with an aim to define probable therapeutic targets for HNSCC management.
MK2 mediated master of regulatory network functions by modulating the transcript stability
The prime objective of this study was to gather comprehensive information about the various DEGs and pathways rendering essential roles in HNSCC pathogenesis in the presence and absence of MK2 in both normoxic as well as the tumor core mimicking hypoxic conditions. Thereupon, keeping MK2 at the nexus of further analysis, we shifted the focus on elucidating in detail, the regulation of MK2 pathway and its downstream targets in various experimental datasets. It was found that MK2 is involved in regulation of six major biological pathways as shown in Figure S11. Recent in vitro findings by our team have asserted that MK2 controls transcript stability of critical genes involved in HNSCC pathogenesis via RBP-mediated regulation (Soni2 et al., 2019). Hence, we analyzed the transcriptomic data in regard to deciphering the role of MK2 in the regulation of mRNA stability. Interestingly, MK2 was found to accomplish the task of regulating the transcript stability through RBP-mediated regulation with HuR (ELAVL1) and TTP (ZFP36) playing intrinsic roles (Figure 3a), thus clearly affirming the hypothesis and corroborating our previous findings (Soni2 et al., 2019). The levels of expression of these RBPs and hence regulation was different in various experimental datasets clearly suggesting that tumor environment, in association with presence/absence of MK2, plays an important role in HNSCC pathogenesis (Figure 3a).
Next, to attain a clear picture of what was happening at the transcriptional level, we narrowed down our analysis by filtering down to selected genes of the MK2 pathway that were previously studied in MK2-centric studies and were also analyzed in our recent in vitro study. The genes selected for this analysis were p38, MK2, AUF1, TTP, CUGBP1, CEBPδ, HuR, MKP-1, p27, TNF-α and VEGF, which have been known to be involved in a number of key cellular pathways. The analyzed genes were clustered into eight major biological processes as shown in Figure S12. Elucidation of the MAPK signaling cluster in detail indicated that in the background of MK2-knockdown in normoxia (B vs A dataset), VEGF and TNF-α tend to show downregulation (Figure 3b) which is in complete consonance with our previously established findings. Collectively, transcriptomic analysis results were quite promising and corroborated very well with previous findings, thereby, strongly validating our hypothesis that MK2 is the master regulator of transcript stability of genes critical to HNSCC pathogenesis.
3’-untranslated region-based filtering furnished information regarding important MK2-regulated downstream target genes
In the present study, our prime focus was role of MK2 and MK2-regulated genes in HNSCC pathogenesis. MK2 can potentially regulate the transcript stability of only those targets that possess the binding regions for RBPs in their 3’-untranslated region (3’-UTR). Practically, only these genes could be the probable candidates of MK2 downstream targeting via RBP-mediated regulation. Hence, we narrowed down the transcriptomic analysis to only those DEGs that harbored RBP binding regions in their 3’-UTRs. To accomplish this task, we fetched out the 3’-UTR regions of all DEGs using Ensembl (http://www.ensembl.org/) (Aken et al., 2016). Next, the domain sequences of RBPs were assessed using catalog of inferred sequence binding preferences of RBPs (CISBP-RNA) database. Lastly, the transcripts that harbour RBP specific regions in their 3’-UTRs were filtered out using RBPmap v1.1 web tool (http://rbpmap.technion.ac.il/). Using this approach, we analyzed the top two upregulated and downregulated genes in previously selected 16 cancer specific pathways (Table S3) in the transcriptomic profiling of the in vitro HNSCC cell line model (A-D datasets) which resulted in 34 MK2-regulated genes listed in Table S6. Similarly, for the in vivo heterotopic HNSCC xenograft experimental dataset (F vs E dataset), we filtered down the topmost upregulated and downregulated genes in all the cancer specific pathways which provided us 48 MK2-regulated genes that are listed in Table S7. This specific filter-based transcriptomic analysis brought into limelight possible MK2-downstream target genes that could be integral in HNSCC pathogenesis. Furthermore, to cross-validate findings of the transcriptomic profiling, these filtered candidate genes along with the 5 common genes in the 77 common elements in A-D datasets listed in Table S2 were used for further in vitro experimentation. One gene (H2AFY) was common in the transcriptomic analysis for both A-D and F vs E datasets resulting in a total of 86 genes (34 MK2-regulated genes and 5 common genes for A-D datasets, 48 MK2 regulated genes in F vs E dataset, and 1 gene was common in both) that were selected for further validation.
Highly efficient and precise detection of gene expression via nCounter gene expression assays potentiated transcriptomic outcomes
Routinely, the findings of transcriptomic analysis are generally validated in in vitro setting via gene expression analysis employing qRT-PCR and using the same RNA sample in order to maintain homogeneity. In lieu of the high-throughput nature of our validation, qRT-PCR analysis could have been prone to numerous errors; hence, we employed a latest and highly precise gene expression assay-based nCounter system approach (NanoString Technologies, Inc.) to validate the transcriptome analysis. To accomplish this assay, 90 specific custom designed molecular probes corresponding the selected candidate genes were procured from NanoString Technologies, Inc. These probes aided in imaging and detection of multiple transcripts (90 in this case) in single reaction with high fidelity rate. The gene set comprised of the 86 selected genes from the transcriptomic profiling as well as four housekeeping genes (HKG). The HKG were selected based on extensive literature survey which ascertained that their levels remain unchanged in HNSCC (listed in Table S8 and S9, respectively).
The assay was performed using standard procedure as highlighted in Figure S13 and S14 and detailed in the methods. Briefly, the custom synthesized probes were hybridized to the target RNA samples followed by washing of the excess probes. Further, immobilization of the probe/RNA complexes on the nCounter cartridge was performed, samples were run on the nCounter instrument and finally data was collected. The statistical analysis (p<0.05) indicated that some of these genes were differentially expressed in the various experimental datasets. We next performed pathway enrichment analysis of the nCounter gene expression assay data to determine the biological significance of changes in gene expression. KEGG enrichment analysis revealed the top five biological processes (Figure 4a and 4b), while heat map analysis deciphered the clustering of gene expression profiles among various combinations (Figure 4c). The results from the nCounter assays correlated with the transcriptomic analysis (Table 2 and 3), hence substantiating our findings.
Furthermore, considering the individual CodeSet of 39 genes for the in vitro HNSCC cell line model (A-D datasets) and 48 genes for the in vivo heterotopic HNSCC xenograft experimental dataset (F vs E dataset), the heat-map and box-plot representation of the results as analyzed by nCounter gene expression assay have been shown in Figure 4d and 4e, respectively. These results indicated the expression profile and variation in FC among DEGs. Figure 4d clearly depicts the upregulated and downregulated DEGs in the nCounter assays and the results were in consonance with transcriptomic profiling with a high percentage of genes showing similar pattern of expression and even matching fold change values as showcased in Table 2 and 3. For B vs A dataset, a total of 39 genes were analyzed, out of which 24 matched with the transcriptomic analysis (61.6% matching score). The matched genes were then analyzed for FC where, BRD2 was found to be the only upregulated gene (FC>2) and CLK2 was the only downregulated gene (FC<2). Similarly, after analysis of all the datasets we found that 12 DEGs were common in our transcriptomic and nCounter analysis (the results have been summed up in Table 4). Notably, these genes are key players in important processes like cell cycle progression, apoptosis and transcription regulation, hence, potentiating their role as important MK2-regulated genes involved in HNSCC pathogenesis. Thus, these results strengthen our recent findings that MK2 is critically important in regulating HNSCC and functions by modulating the transcript stability of crucial pathogenesis-related genes. Further, detailed statistical analysis accentuated that expression of only 7 (MELK, ZNF662, BMP7, CREB3L1, IGFBP2, MUC4, PRKAR2B), out of the 12 candidate genes were significantly changed (FC>2 or <-2, p<0.05) (Table 2 and 3). MELK was the only gene belonging to the C vs A dataset (cell line comparison) while the other 6 genes were from the F vs E dataset (xenograft comparison). Hence, we next analyzed these 6 genes (ZNF662, BMP7, CREB3L1, IGFBP2, MUC4, PRKAR2B) in vitro by immunohistochemistry (IHC) to ascertain our transcriptomic and nCounter findings in an experimental HNSCC xenograft model (Figure 5a).
Immunohistochemical analysis indicated probable role of candidate genes in HNSCC pathogenesis
To add another of confidence to the dataset and probe the role of candidate genes in HNSCC pathogenesis immunohistochemical analysis was performed in tumor xenograft tissues. The six filtered candidate DEGs that were common in transcriptomic and nCounter data analysis in F vs E dataset were then analyzed in tumor sections from xenografted animals using IHC to evaluate the expression levels of these proteins (Figure 5). Results obtained from IHC largely strengthen our findings about the common DEGs in our transcriptomic and nCounter analysis. The in situ expression of IGFB2, MUC4, PRKARB2 was mostly present in the cellular cytoplasm and stroma of squamous cell carcinoma (Figure 5). The tumor sections with the expression have cellular pleumorphism, mitotic figures and formation of nests of tumor cells. It is clearly evident from the IHC outcome that expression levels of three MK2-regulated candidate genes (IGFB2, MUC4, PRKARB2) were upregulated in tumor xenografts created using CAL27-MK2KD cells as compared to CAL27-MK2WT cells (Figure 5). These genes are widely considered imperative to processes like cell cycle progression, apoptosis and transcription regulation. There was no significant change in the protein expression levels of the three other analyzed genes (data not shown). Findings of the Immunohistochemical analysis clearly implied the elevated expression level of IGFB2, MUC4, and PRKARB2 in tumor xenografts, it corroborates the obtained datasets from transcriptome analysis and nCounter gene expression validation. These finding clearly upholds the importance of these MK2 regulated candidate genes in HNSCC pathogenesis.
Discussion
To improve understanding of the convoluted biology and leverage the outcomes to optimize the management of HNSCC, there have been many efforts to characterize HNSCC at the transcriptome level. Methodological breakthroughs in the recent past have revolutionized the area of transcriptome profiling by providing a link between the molecular mechanisms and cellular phenotypes (Cheng et al., 2018; Cieślik et al., 2018). In recent times, a comprehensive landscape of genomic and transcriptomic alterations in squamous tumors including HNSCC had emerged from the TCGA network (Lawrence et al., 2015; Campbell et al., 2018). Cellular models that can comprehensively characterize metastatic HNSCC are still lacking and hence translationally relevant transcriptome profiling underlying the basis of HNSCC metastasis has proved to be a powerful tool for preclinical research (Byron et al., 2016; Nisa et al., 2018). In the context of the present study, this link presented us with an opportunity to thoroughly dissect the huge complexity and large heterogeneity in head and neck tumors in an attempt to discern novel biomarkers and potential therapeutic targets.
Generally, there are two well characterized ways to analyze cancer transcriptomic data. First is the differential approach, in which tumor expression profiles relative to the normal patient-matched or even unmatched tissues are analyzed. Second is the relative approach, in which transcript levels across different tumors or other samples are compared. The simplest example of the differential approach is identification of the up/down-regulated DEGs in a particular sample. There are many established methods that help in detection of DEGs for both microarray-based approach as well as RNA-sequencing (Gentleman et al., 2006; Love et al., 2014). A typical transcriptome profiling result is generally a never-ending list comprising of thousands of DEGs, hence, it has always been very difficult to interpret this data without additional filtering via functional annotations. A large variety of methods are available for analysis of DEGs and for obtaining critical understanding of the pathways, gene regulatory and co-expression networks involved (Ackermann et al., 2009; Mitrea et al., 2013).
Keeping in mind the critical findings from the previous studies, we performed transcriptome profiling of both the in vitro cell line as well as in vivo xenograft tumor samples resulting in thousands of DEGs. These genes were segregated based on their clustering in various biological processes (Table S1, S3 and S5). In line with our primary goal, we filtered the processes on the basis of relevance in cancer and ended up with five DEGs that were common in cell line data sets (A-D) namely EHBP, SLF2, DAP3, IP6K2 and RUNX1 as listed in Table S2. Eps15 homology domain binding protein is encoded by the EHBP1 gene and this protein has been shown to portray a role in actin reorganization and endocytic trafficking (Guilherme et al., 2004). Polymorphism in this gene at a single nucleotide level has been reported to cause prostate cancer (Sun et al., 2011). SLF2 is a DNA damage response pathway gene and functions by regulating genomic stability by postreplication repair of damaged DNA (Räschle et al., 2015). DAP3 has been shown to be involved in mediating interferon (IFN)-γ induced cell death in addition to its role in organelle biogenesis as well as maintenance and mitochondrial translation (Kissil et al., 1995; Tang et al., 2009). DAP3 has been characterised by its pro-apoptotic function as a prognostic factor in gastric cancer (Jia et al., 2014). The protein encoded by IP6K2 gene has been shown to affect growth suppression and apoptotic action of IIN-β in physiologic regulation of apoptosis in ovarian cancers (Morrison et al., 2002). It has been reported that deletion of IP6K2 leads to HNSCC predisposition (Morrison et al., 2009). Lastly, the protein encoded by RUNX1 has been shown to be involved in the process of normal haematopoiesis with chromosomal translocations related to this gene showing association with leukemias (Sood et al., 2017). Further, it has been reported that RUNX1 depletion in human HNSCC cells caused growth arrest (Taniuchi et al., 2012). Collectively, it is quite evident that all these MK2-regulated genes are playing vital role in tumor pathogenesis, hence showing consistency with our finding of their involvement in HNSCC pathogenesis. Therefore, they can be further explored as candidates for the development of novel biomarkers and utilized as potential therapeutic targets in HNSCC management.
Additionally, gene regulatory networks in the transcriptome profiling provided information of the various biological processes regulated by these candidate genes. It supplied us with a wealth of information which can be further utilized to study the pathogenesis of HNSCC in detail especially in the background of MK2-knockdown and varied tumor microenvironment (normoxia/hypoxia). Figure 3a is a representation of gene regulatory network showing the role of MK2 in regulation of mRNA stability in various experimental dataset (F vs E dataset). The figure clearly demonstrates that MK2 regulates transcript stability via RBP-mediated regulation with HuR (ELAVL1) and TTP (ZFP36) playing integral part. Similarly, Figure 3b portraying the regulatory network that represents MAPK signaling cluster of the selected MK2 pathway genes (p38, MK2, AUF1, TTP, CUGBP1, CEBPδ, HuR, MKP-1, p27, TNF-α and VEGF) in the transcriptome profiling data of the in vitro HNSCC cell line dataset (B vs A, normoxic microenvironment) indicates VEGF and TNF-α down regulation. Interestingly, these transcriptomic profiling results are in complete consonance with our recently published findings and hence succeed to potentiate and validate the fact that MK2-knockdown destabilized VEGF and TNF-α in normoxia via RBP-mediated interaction.
Transcriptome analysis techniques are being commonly utilized in endeavors to decipher various molecular mechanisms of tumorigenesis and to fetch out novel prognostic and therapeutic markers (Bergers et al., 2008; Trapnell et al., 2010). In this study, we aimed to assess the MK2-regulated candidate genes playing prominent role in HNSCC pathogenesis. To accomplish this, we filtered out the DEGs based on the presence of adenine uridine rich elements (ARE)-regions in their 3’-UTRs where RBPs could bind and modulate their function via MK2-mediated regulation. This provided us with a list of 34 genes in A-D datasets while 48 genes in the xenograft dataset (listed in Table S6 and S7). The data was further validated using the nCounter gene expression assay system (NanoString Technologies, Inc.).
nCounter gene expression assays enable the digital quantification and single molecule imaging of multiple target RNA molecules using multicolor-coded molecular barcodes (Figure S2 and S15). This system provides discrete and accurate counts of RNA transcripts at a high level of sensitivity and precision (Brumbaugh et al., 2011). We chose four commonly used reference genes viz. ABCF1, GAPDH, POLR2A, and RPL19, in nCounter data analysis because of their baseline expression in HNSCC tumorigenesis as well as in MK2-knockdown conditions, as listed in Table S9 (Lallemant et al., 2009; Palve et al., 2018).
Gene expression assays are independent of any enzymatic reactions or amplification protocols and have no reliance on degree of fluorescence intensity to determine target abundance. As a result of these characteristics, and also the highly automated nature of bar-coded sample processing, these assays result in highly precise and reproducible outcomes. On an average, we obtained around 52% matching score of transcriptome profiling data with nCounter gene expression assay-based validation which is considered a good percentage match. Filtering of the DEGs in the matched data revealed a list of 12 genes (6 upregulated and 6 downregulated in various experimental datasets) that were common in our comprehensive nCounter system-based validation of transcriptomic profiling (Table 4). Intriguingly, these genes portray crucial roles in processes like apoptosis (CLK2, MELK, MUC4), cell cycle regulation (CLK2, MELK) and transcription regulation (BRD2, H2AFY, SAMD4B, ZNF662) suggesting that they could be potentially used as molecular candidates for further investigations in regard to design of molecular markers and therapeutics for HNSCC management. Hence, transcriptome analysis followed by nCounter assay-based validation of our data has provided us with valuable findings that can aid in extending the field of HNSCC research.
Six candidate genes (ZNF662, BMP7, CREB3L1, IGFBP2, MUC4, PRKAR2B) that showed statistically significant up/downregulation in F vs E dataset were further analyzed in tumor sections from xenografted animals using IHC to evaluate the expression levels of these proteins. Interestingly, our IHC findings confirmed the validity of our in-silico findings. Consistent with in silico findings the expression levels of three genes (IGFB2, MUC4, PRKARB2) were found to be upregulated, suggesting the roles of these MK2-dependent candidate genes in pathogenesis and progression of HNSCC. Insulin like Growth Factor Binding Protein 2 (IGFBP2) has been shown as a growth promoter gene in several tumors and considered as a central hub of oncogenic signaling network governing transcriptional regulation and promoting epithelial to mesenchymal transition, invasion, angiogenesis and metastasis (Lindström 2019). Mucin 4 (MUC4) serves as a major constituent of mucus secreted by epithelial cells and found overexpressed in a variety of cancers like papillary thyroid carcinomas. It is known for promotion of tumor growth, proliferation and migration (Yu et al., 2020). Recent insights have been made into the transcriptional regulation of protein kinase cAMP-dependent type II regulatory subunit beta (PRKAR2B) by miRNAs and X-Box binding protein 1 leading to a better understanding of PRKAR2B-driven prostate cancer progression (Xia et al., 2020). Consistent to our findings these genes have been suggested to be prognostic indicators and therapeutic targets in various cancers including HNSCC (Park et al., 2015; Zhao et al., 2018; Lindström 2019; Xia et al., 2020; Yu et al., 2020). It is worth mentioning that all of these genes are MK2-regulated and play specific roles in pathogenesis. In conclusion, present study vindicates involvement of more MK2-dependent genes as critically important in regulating HNSCC progression by modulating the transcript stability of pathogenesis-related genes via RBP-mediated regulation. This study has made it possible to filter down from thousands of DEGs to a few potential candidate genes using a comprehensive transcriptomic and in vitro approach. In an attempt to delve deeper into the mechanistic insights and to decipher the molecular markers responsible for MK2-mediated changes in HNSCC pathogenesis, the role of these potential therapeutic targets warrants further detailed investigation.
Conclusions
In conclusion, present study vindicates involvement of more MK2-dependent genes as critically important in regulating HNSCC by modulating the transcript stability of pathogenesis-related genes possibly via RBP-mediated regulation and these results are in perfect consonance and augmentation with our recent findings (Soni2 et al., 2019). Comprehensively, few crucial MK2-regulated putative candidate genes were identified in this study and their plausible mode of action in HNSCC pathogenesis was elucidated, which could be further explained as possible targets in the pathway for HNSCC management (Figure 6). This study has made it possible to filter down from thousands of DEGs to a few potential candidate genes using a comprehensive transcriptomic and in vitro approach. In an attempt to delve deeper into the mechanistic insights and to decipher the molecular markers responsible for MK2-mediated changes in HNSCC pathogenesis, the role of these potential therapeutic targets warrants further detailed investigation.
Methods
Culturing and maintenance of cell lines
Homo sapiens tongue squamous cell carcinoma cell line CAL27 (CRL-2095™, ATCC, USA) and Homo sapiens palatal mesenchyme cell line HEPM (CRL-1486™, ATCC) were grown in specific media supplemented with 10% fetal bovine serum and 1% antibiotic-antimycotic (Gibco, USA). The cells were cultured in normal conditions (37°C, 5% CO2 incubator with 95% humidity) and were free from any kind of contamination. Further, MK2-specific short hairpin RNA-green fluorescent protein (shRNA-GFP) constructs were used to stably knockdown MK2 from cultured CAL27 cells, thus forming CAL27-MK2KD cells (as previously described in Soni2 et al., 2019). For hypoxia experimentation, the cultured cells plated in petriplates were incubated for 48 hours in 0.5% O2 at 37°C in a hypoxia chamber (Bactrox, Shel-Lab, USA).
Xenograft mice model generation
In order to mimic the tumor microenvironment more closely to the human tumors, a biologically relevant heterotopic xenograft model of HNSCC was developed in non-obese diabetic/severe combined immune deficiency (NOD/SCID) mice. The immunocompromised mice were randomly assigned into control (CAL27-MK2WT) and experimental (CAL27-MK2KD) group based on specific cell-type injected (as previously described in Soni2 et al., 2019). Briefly, for xenograft generation, one million cultured cells suspended in 100μl of 1x phosphate buffered saline were injected subcutaneously in the right flank of mice. Seven weeks after graft inoculation, the mice were euthanized by CO2 asphyxiation; tumors were aseptically excised, weighed and used for tissue embedding/RNA isolation.
RNA extraction and sample preparation
CAL27 cells (both WT and KD) cultured in petriplates and grown in both normoxic and hypoxic environment as well as tumors resected from xenografted mice were used for isolation of total cellular RNA using RNeasy Mini kit (Qiagen) following manufacturer recommended protocol (sample details are provided in Figure 1a and Table 1). Consequently, the quality as well as concentration of all the RNA samples was determined by employing NanoDrop 2000C spectrophotometer (Thermo Fisher Scientific, USA) and Bioanalyzer (Agilent 2100, Agilent Technologies). RNA samples having RIN>5 was used for cDNA library preparation. For each sample, RNA was isolated from at least three individual plates/tumors and was pooled for library construction and further analysis.
cDNA library preparation and sequencing
Total RNA (5μg) from each sample was used to isolate poly-A mRNA and to prepare cDNA library using TruSeq mRNA sample preparation kit v2 (Illumina Inc.). Each sample was tagged with a unique TruSeq index tag in order to prepare multiplexed libraries. Six paired-end adapters with unique six base index sequences were used for the library preparation. The index tag permits accurate differentiation between each sample. The quantification of prepared libraries was performed on Qubit fluorometer using Qubit dsDNA BR assay kit (Life Technologies, USA) and size of the libraries was checked on Bioanalyzer DNA 1000 series II chip (Agilent Technologies). The flow chart of the sequential steps involved in TruSeq library preparation is given in Figure S15. The insert size and purity of libraries were then validated on Bioanalyzer DNA 1000 series II chip (Agilent Technologies). All the libraries (six in total, 4 from cell line model and 2 from animal model) had an average insert size of 210bp and were pooled by taking 10μl from each library. Final pool was loaded in one lane of a S2 flow cell using NovaSeq XP protocol. Cluster amplification and sequencing data generation were performed on NovaSeq 6000 system (Illumina Inc.) using 2×100 paired-end cycles (Figure 1a). Raw data quality control was performed using NGSQC tool kit v2.3 using default parameters (Patel et al., 2012).
Reference based assembly and homology search
Raw FASTQ files with low-quality reads of sequencing data were filtered to obtain high quality reads. High quality filtered data was aligned with the reference genome (genome reference consortium human build 38 patch release 12, GRCh38.p12). Kallisto pipeline was used for alignment and identification of transcript coding regions followed by quantitation and annotation using default parameters (Huang et al., 2009). Furthermore, we performed the removal of multi-mapped reads and finally converted them to read counts for annotated genes. Figure 1b is a flowchart representation of the various steps involved in the sequencing data analysis.
Annotation, differential gene expression and pathway analysis
Expression of the commonly found transcripts in all the samples was analyzed based on their FPKM values. Transcripts were given a score for their expression by Cufflinks-based maximum likelihood method and values of FPKM≥0.1 were considered significant for downstream analysis. Transcripts uniquely expressed in each sample were considered specific and were analyzed separately. False discovery rate (FDR) was employed to correct the statistical significance of the p-values for multiple tests. DEGs in multiple combinations were identified by DESeq analysis pipeline using a fold change threshold of absolute log2 fold change (FC)≥2 and a statistically significant Student’s t-test p-value threshold adjusted for FDR<0.001. Hence, transcripts having FC<-2 were considered as downregulated while those with FC>2 were deemed upregulated. Statistically significant enriched functional classes with a p-value adjusted for FDR>0.05 derived using the hypergeometric distribution test corresponding to DEGs were determined using Student’s t-test with Benjamini Hocheberg FDR test.
Unsupervised hierarchical clustering of DEGs was performed using Cluster 3.0 and visualized using Java TreeView (Huang et al., 2009) (Figure 1b). Gene ontologies and pathways that harbour expressed transcripts were identified using DAVID functional annotation tool (http://david.abcc.ncifcrf.gov/home.jsp) (Dennis et al., 2003; Sherman et al., 2009). For DEGs, heat maps and volcano plots were generated using ‘gplots’ and ‘heat map’ packages. Gene ontology (GO) as well as KEGG pathway analysis was accomplished for the assembled transcripts with reference to UniProt database. The data of total DEGs (upregulated and downregulated) was explored further using Cytoscape v3.5.0 (http://www.cytoscape.org/) to better understand the gene regulatory networks and for mapping of the results (Montojo et al., 2010). In figures depicting gene regulatory networks, gene nodes are sized according to their p-values and colored according to their FC, where red shows upregulation, green shows downregulation and yellow shows baseline expression; processes are showed in rectangular boxes and colored in blue.
nCounter gene expression assay
In order to validate the leads obtained from transcriptomic profiling, we utilized custom designed molecular barcodes (NanoString Technologies, Inc.) for single molecule imaging, thereby making it possible to detect and count hundreds of different transcripts in one reaction (Figure 1a and S13). RNA quality was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies) (Figure 1a and S1). Gene expression was analyzed on the nCounter system (NanoString Technologies, Inc.) following manufacturer’s recommendations. Briefly, the custom synthesized probes were hybridized overnight to target RNA followed by washing away of excess probes, immobilization of CodeSet/RNA complexes in the nCounter cartridge and finally data collection on the nCounter system. Gene expression levels were measured in triplicate for total RNA from the cell line and xenografted tumor samples, normalized to the four HKG and analyzed using the nSolver software (NanoString Technologies, Inc.). Each nCounter assay contained synthetic spike-in controls in the preparatory mix. These controls are necessary to allow correction of sample-to-sample variation arising due to common experimental errors such as differences in amount of input transcripts or reagents (Brumbaugh et al., 2011). The counts were normalized with the positive controls and averaged for the samples of each mRNA type. Normalization involved spiked-in positive and negative control probes for background correction in addition to four reference or housekeeping genes. Data analysis was performed on nSolver 3.0 analysis and advanced analysis software (NanoString Technologies, Inc.).
Immunohistochemical analysis
The levels of expression and activation status of specific proteins in xenografted tumor sections were analyzed using IHC to validate the overall findings of the transcriptomic and the nCounter gene expression analysis in an in vivo xenograft model. 5 μm thin sections were fixed on poly-l-lysine coated slides, followed by deparaffinization and rehydration. Antigen retrieval was performed using sodium citrate buffer (pH 6.0) followed by quenching of endogenous peroxidases using BLOXALL™ blocking solution (Vector Laboratories, Inc., USA). Further, incubation of the sections with 2.5% normal horse serum blocked the exposed sites. Sections were incubated with appropriately diluted specific primary antibody overnight at 4°C followed by horseradish peroxidase-conjugated secondary antibody for 1 hour at room temperature. Rinsed sections were then incubated with 3,3’-diaminobenzidine substrate and Mayer’s hematoxylin served as a counterstain. Five field views were obtained from each slide having tissue sections for designated antibody and used for analysis. Observations and imaging of the sections were carried out by a pathologist in a blinded fashion using BX53 bright field microscope (Olympus Corporation, Japan). Protein expression was quantified using ImageJ software 1.8.0 (https://imagej.nih.gov/ij/).
Availability of Data
The generated datasets (raw reads) from NovaSeq 6000 have been deposited to the National Centre for Biotechnology Information-Sequence Read Archive (NCBI-SRA) repository (https://www.ncbi.nlm.nih.gov/sra). The SRA BioProject accession number for the submitted bio projects are PRJNA646850 and PRJNA646851.
Competing interests
The authors declare that they have no competing interests.
Funding
The authors acknowledge CSIR for financial support (CSIR-IHBT; Project MLP0204).
Authors’ contributions
YSP conceptualized the work and designed the manuscript. SS and PA performed experiments and data analysis. MKS helped in RNA sequencing run on NovSeq and data analysis, VP performed IHC imaging and analysis, NVT helped in the xenograft model generation. SS, PA, VP and YSP wrote and edited the manuscript. All the authors read and approved the final version of the manuscript
Acknowledgements
The authors would like to thank the Director, CSIR-IHBT, Palampur for his consistent support and encouragement. SS is immensely thankful to CSIR and PA to DST for providing Ph.D. fellowship and Academy of Scientific and Innovative Research (AcSIR), Ghaziabad for Ph.D. registration. The authors would like to thank Bionivid Technology Private Limited, Bengaluru for help in transcriptome data analysis. IHBT communication number of this manuscript is 4682.
Footnotes
Sourabh Soni: sourabhsoniplp{at}gmail.com, Prince Anand: princeanandsaharsa4u{at}gmail.com, Mohit Kumar Swarnkar: mohitks{at}ihbt.res.in, Vikram Patial: vikrampatial{at}ihbt.res.in, Narendra V. Tirpude: narendra{at}ihbt.res.in
References
- [1].↵
- [2].↵
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].↵
- [44].↵
- [45].↵
- [46].↵
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵