Background
Rhabdoid tumors (RTs) are aggressive tumors that occur most frequently in children under 2 years old. RTs often occur in the kidney (KRTs) or the central nervous system (CNS), which are termed Atypical teratoid/rhabdoid tumors (AT/RTs). Extracranial RTs were first recognized as a physiological entity nearly 40 years ago [
1]. Later, Haas and colleagues introduced the term rhabdoid in describing KRT, due to the close histological resemblance of the tumor cells to rhabdomyoblasts, although subsequent studies have not confirmed a myogenic origin of these tumor cells [
2]. In 1987, AT/RT was recognized as a discrete clinical entity based on pathologic and genetic characteristics [
3]. Prior to that, it had been mostly classified as either medulloblastoma, primitive neuroectodermal tumor, or choroid plexus carcinoma. Following this description, the World Health Organization (WHO) began to classify AT/RT as an embryonal grade IV neoplasm in 1993 [
4].
Epidemiologic studies of RT have been limited by the fact that this is a rare disease. So far there have been only a handful of epidemiologic reports. In a study conducted in the UK, 106 children under 15 years old were diagnosed with extracranial RT in the UK between in a period of nearly 20 years [
5], resulting in an age-standardized annual incidence of 0.6 per 1 million children. In the US, several studies observed that AT/RT accounted for 1–2% in pediatric brain tumors, and for 4.4% of CNS tumors in children aged zero to 5 years [
6‐
9]. Two more recent surveys conducted in China draw consistent results of a prevalence of AT/RT at approximately 5% in pediatric CNS tumors, which is comparable to that in the US study.
Aside from low incidence rate, there are other factors that poses challenges to the diagnosis and treatment of RTs. Histologically, RTs manifest several characteristic features, including eosinophilic cytoplasm, large nucleoli, and filamentous cytoplasmic inclusions. The tumors may present a host of neural, epithelial, mesenchymal, or ependymal patterns, which makes them rather variable and difficult to diagnose [
10]. Moreover, RTs often progress fast and lead to a high lethality. In the UK study of extracranial RT, 1-year survival was 31% [
5]. The patients usually suffers from metastasis and, to make things worse, the young age of patients limits use of radiotherapy. In an early report of 22 cases of KRTs in children, metastases were found in 82% of cases, either at diagnosis, or developing from 2 weeks to 9 months after diagnosis. Only two patients eventually survived, both with localized disease (stage II) [
11]. Therefore, early diagnosis of this formidable disease is of key importance and in urgent demand.
Currently RTs are diagnosed mainly on immunohistochemistry (IHC) results, specifically, the lack of SMARCB1/INI1 protein expression, or less frequently, that of SMARCA4/BRG1 protein expression [
4]. Initial genetic studies suggested that approximately 75% of RTs are characterized by biallelic inactivation of the
SMARCB1 locus, which indicated a sensitivity of close to 75% [
12]. However, loss of expression of SMARCB1 is not exclusive to RTs, but also has been observed in other types of cancers, including chordoma, epithelioid sarcoma, cribriform neuroepithelial tumor, and medullary renal cell carcinoma [
13‐
19]. Together, these lines of evidence suggest that SMARCB1 expression alone is neither sufficiently sensitive nor specific for diagnosing RTs. Moreover, in particular for CNS AT/RTs, another severe limitation in clinical diagnosis is the potential misdiagnosis as medulloblastomas (MBs) or primitive neuroectodermal tumors (PNETs), owing to the close histological resemblance of the rhabdoid cells and neuroepithelial tissue in these tumors [
3,
20]. In conclusion, diagnostic markers with improved sensitivity and specificity are needed to complement the current practice, to the end of developing a comprehensive diagnostic strategy with enhanced sensitivity and precision.
In this study, we set out to identify diagnostic markers for RTs by employing a molecular profiling approach. Protein coding genes (PCGs) and long non-coding RNAs (lncRNAs) showing aberrant expression in AT/RT and KRT cases were identified, respectively, and the co-expression between these significantly dysregulated genes was evaluated. Through further comparison of differentially expressed genes, the dysregulated PCG-lncRNA pairs, and commonly known cancer genes, candidate diagnostic markers for AT/RT were identified and subjected to Receiver Operating Characteristic analysis to assess the performance of these candidates. Two PCGs, RPL5 and PRL10, exhibited high sensitivity and specificity not only in diagnosis of AT/RT but also differential diagnosis of AT/RT and KRT, as therefore show considerable promise for AT/RT diagnosis, and warrants further investigation.
Methods
Data analysis
The raw data were downloaded from the NCBI GEO database (GSE15641, GSE11482, GSE30946, GSE64019, GSE28026, GSE35493, GSE64019, GSE70421, GSE35493). The limma package was used to deal with the raw data in CEL format, with MAS5 algorithm, to quantify expression level and to identify the difference of gene expression. The biomaRt package was used to convert the probe ID to Ensembl ID. Genes were categorized as “protein coding” and “long non-coding” based on an Ensembl annotation file in the GTF format. Among non-coding genes, rRNAs, tRNAs, miRNAs, snoRNAs and other known classes of RNAs were excluded, and lncRNAs were defined as all non-coding genes longer than 200 nucleotides and not belonging to other RNA categories.
Pearson’s correlation coefficient
Pearson’s correlation coefficient (PCC) was calculated by in-house R- scripts and was utilized to evaluate the co-expression relationship between lncRNA and PCG. Co-expressed pairs were defined with a cutoff of |PCC| ≥ 0.7 and P < 0.001.
Data visualization
Unsupervised hierarchical clustering was done by R software (version 3.3.2,
http://www.r-project.org/). The receiver operating characteristic (ROC) and the area under the ROC curves (AUC) values were obtained from the pROC package. Unless otherwise specified, data were analyzed and visualized using R software (version 3.3.2).
Enrichment analysis
For enrichment analysis to explore their biological effects, PCGs were analyzed using the clusterProfiler package. The GO terms and KEGG pathways with p values or FDR of < 0.05 were considered as significantly enriched function annotations.
Differential RPL5/10 expression analysis across Affymetrix datasets
We downloaded GSE85217 and GSE2712 from GEO dataset. GSE85217 contains 762 medulloblastoma patients expression data, and GSE2712 contains 18 Wilms’ tumors and 14 clear cell sarcoma of the kidney. The former used Affymetrix Human Gene 1.1 ST Array, the latter used Affymetrix Human Genome U133A Array. So in order to make the data comparable, we used the Array Generation based gene Centering (AGC) method to compare the expression value of RPL5/10 between different datasets [
21]. The AGC method scaled datasets with a scaling factor that is defined based on the housekeeping genes.
Discussion
Rhabdoid tumors are highly lethal cancers that most frequently observed in young children. Research into the diagnosis and treatment has been hampered by the rare nature of this disease despite its urgency. In a recent study, Chun et al. performed a molecular dissection of Malignant rhabdoid tumors (MRT, mainly KRT) using RNA sequencing [
33]. Expression profiles of 40 primary extra-cranial malignant rhabdoid tumors, three human embryonic stem cell lines, and four fetal cerebellum samples were collected and screened for aberrantly expressed genes. Through compare RTs gene expression with genes expressed both in cell lines and fetal cerebellum samples, Author identified 398 up-regulated genes and 615 down-regulated ones. These genes may be used as diagnosis markers of MRT, but this study did not focus on identifying marker candidates for KRT diagnosis. More specifically, similar investigations were conducted in AT/RT over the past few years. Based on patterns in the transcriptional profile, Torchia and colleagues [
34] classified AT/RT into three subgroups with distinct genomic profiles, implicated cellular processes, and clinicopathological and survival features. These findings were consistent with those of an independent study [
35]. All three reports, however, focused on the classification and prognosis and AT/RT. Chakravadhanula et al. [
36] evaluated the performance of HOTAIR and HOXC as diagnostic markers of AT/RT, however, the authors found that both genes are not sufficient for distinguishing AT/RT from several other forms pediatric brain tumors. In an interesting study, Ho et al. [
37] proposed three oncogenes, FGFR2, S100A4 and ERBB2 (HER2/neu), as markers for diagnosing AT/RT, based on the aberrant high expression in tissue samples expressing SMARCB1. Overexpression of these genes may be used as novel markers that complement the current criteria, lack of SMARCB1expression. However, as these results were derived from a limited number of samples, further research is warranted to validate these candidates.
Regulation of PCG expression have been known to occur through a number of mechanisms. Upstream regulators include microRNAs and lncRNAs. In an interesting study into the role of microRNAs in Grupenmacher et al. [
38] analyzed the expression profiles of microRNA and PCGs in 13 AT/RT and 10 KRT cases, as well as two human RT cell lines. They found 122 genes significantly differentially expressed between AT/RT and KRT, about 76.22% (93/122) of which down regulated in AT/RT, which was in accordance with our result (Table
1). However, the authors reported a general lack significantly altered expressions in microRNAs between AT/RT and KRT, Therefore, we focused on elucidating the potential of significantly altered lncRNA expression in our investigation, rather than miRNA, as lncRNAs have recently been established as key regulators in cancer. Through identifying differentially expression lncRNAs and constructing lncRNA-PCG co-expression network, 19 PCGs were selected based on co-expression relationship. Further screening, based on numbers of co-expressing lncRNAs, provided a final list of eight candidate markers.
Both
RPL5 and
RPL10 encode members of the 60S subunit of the ribosome [
39,
40]. The protein expression of both genes is relative low in the normal brain [
41]. RPL5 binds 5S rRNA and forms a stable complex, the 5S ribonucleo protein particle, which is necessary for the 5S rRNA transport, where cytoplasmic 5S rRNA is transported to the nucleolus to be assembled into ribosomes. RPL5 may inhibit tumorigenesis through the activation of downstream tumor suppressors and the down-regulation of oncoprotein expression. A study showed that impaired ribosomes induce a p53-dependent cell cycle arrest [
42]. RPL5 has also been reported to play tumor suppressor roles in breast tumors [
43].
The functions and significance of RPL10 is largely unknown so far. Existing literature mainly focused on its association with autism and is still in debate [
44,
45]. There is one report implicating RPL10 in T cell acute lymphoblastic leukemia (T-ALLs). Exome sequencing analysis identified mutation of RPL5 and RPL10 in 12 of 122 (9.8%) pediatric T-ALLs, with a recurrent mutation of Arg98 in RPL10 [
46]. Together, these studies point to a potential role of RPL5 and RPL10 in tumorigenesis, although the relevance of both genes in the KRT and AR/RT has not been elucidated.
In this study, we examined the transcriptome profiles to identify novel prognostic markers for RTs, a rare, lethal, mostly pediatric cancer. After identifying differentially expressed lncRNAs and PCGs, we found intense dysregulation in lncRNA-PCG co-expressed pairs in AT/RT and KRT. Among the key cancer-related PCGs in the co-expression network, RPL5 and RPL10 showed high levels of sensitivity and specificity AT/RT and KRT. After comparison with other common pediatric tumors, RPL5 and RPL10 can also be used to distinguish AT/RT from medulloblastoma. To our knowledge, this study is the first in associating RPL5 and RPL10 with AT/RT diagnosis. Our results therefore identify two novel promising diagnostic markers for AT/RT, and provide the basis for work to further assess the performance, and to develop a robust diagnosis practice using these markers.