Our findings demonstrate the feasibility of genome wide gene expression analysis of RNA extracted from the cell-free microvesicle fraction (exoRNA) of frozen biobanked serum samples, and that clear differences can be observed between the exoRNA expression profiles from GBM patients and normal controls. The two major differences between exoRNA from GBM patients vs. normal controls were found to be: 1) a significantly reduced level of mRNAs encoding ribosomal protein genes; and 2) an overall significant up-regulation of RNA amounts in the serum of GBM patients, which seems to be primarily due to a larger fraction < 300 nt and could not be attributed to any of the mRNAs investigated by qRT-PCR.
ExoRNA from tumour microvesicles in serum will always be diluted into the background of exoRNA coming from normal non-malignant cells. Depending on the purpose of the study, normal exoRNA may also generate useful information. Previous studies have indicated that RNA expression patterns in blood cells can change as a response to tumours [7
]. To look specifically at the genes dysregulated in the tumour, one would need to enrich the tumour specific microvesicle fraction before extracting the RNA.
Microvesicles are shed by many different cell types into the blood stream, including circulating WBCs, platelets, endothelial cells, and dendritic cells [17
]. Shedding has been shown to be higher from tumour cells in culture compared to normal fibroblasts [28
] and tumour-derived microvesicle concentrations in serum increase as a function of increased malignancy [29
]. However, microvesicles of tumour origin circulating in the blood are mixed with microvesicles from other cell sources and it is likely that tumour microvesicles in many cases constitute a relatively small fraction of the total population.
There is a large contribution of microvesicles from normal cells and platelets in blood. In control plasma, for example, platelet-derived microvesicles have been reported to constitute the majority of all microvesicles [27
] and although tumour-derived microvesicles increase in serum as a function of degree of malignancy in ovarian cancer [30
] and that the total number of microvesicles in blood increase with disease progression [29
], it is not clear what proportion of the increased number of vesicles is coming from the tumour vs. from normal cells as a response to the tumour. Thus, down-regulated genes in GBM exoRNA could either be caused by altered biological processes in normal cells as a response to or consequence of the disease, or by large numbers of tumour microvesicles with low levels of particular RNAs.
A major fraction of the genes we observed to be significantly down-regulated in serum microvesicles from GBM patients in this study code for various ribosomal proteins (e.g. RPL11, RPS29, RPLP1, RPS27A
, etc.). However, these genes were not found to be among the most abundant transcripts in platelets in a study by [31
], which we also confirmed in our analysis of the public dataset from GEO (GSE11524). We speculate that these RNAs would therefore also not be abundant in the platelet-derived microvesicles, which constitute a substantial fraction of the plasma microvesicles [27
]. However, these genes have been shown to be very highly expressed in lymphocytes relative to other blood cells [23
] and were also found to be up-regulated in our analysis of the public dataset for PBMC (GSE22224) relative to GBM. This led us to speculate that the lower abundance of these mRNAs in circulating microvesicles from GBM patients may be a consequence of a reduced level of lymphocyte-derived microvesicles in patients as compared to normal controls, since cancer patients are known to often be immune-compromised [26
]. Evaluation of the GBM patients in this study confirmed that their lymphocyte counts were lower than the normal reference level interval, but because we were unable to measure lymphocyte counts in the normal controls, it is impossible to completely correlate transcript down-regulation with lymphocyte depletion in the GBM patients. None of the GBM patients had received chemotherapy prior to diagnosis, but several of them were on steroids and seizure medications at the time of blood draw, which are two potential sources of lymphocytopenia. Evaluation of gene expression differences between lymphocytopenia caused by cancer and by medication would require a much larger study of more carefully chosen subjects.
The observation that many ribosomal protein genes are down-regulated in GBM serum microvesicles may prove to be a valuable contributory marker for GBM and other pathological states. Sharma et al. [5
] also observed a reduction in many of the same genes in PBMCs including lymphocytes, from breast cancer patients compared to normal controls. The down-regulation of ribosomal protein genes is therefore unlikely to provide a specific GBM component for a diagnostic classification, but may serve as a more generic pathological indicator and can provide confirmatory support when combined with other exoRNA profiles. Interestingly, many ribosomal proteins have extra-ribosomal functions that go beyond the function of protein biosynthesis. Ribosomal protein genes have been shown to have important regulatory functions also in cancer cells. Some ribosomal proteins are tumour suppressors [32
], oncogenes or have regulatory functions in tumour progression, invasion and metastasis [33
The "ribosome" GO-clusters from down-regulated genes had very high enrichment scores from DAVID (> 50) compared to any of the clusters obtained with up-regulated genes (< 2.5), confirming that the patterns of down-regulated genes were much more significant. For gene ontology analysis it is important to analyze a relatively large number of genes (e.g. 100 [22
]) in order to avoid stochastic errors, e.g. if a list of 10 genes is analyzed, the presence of a single gene (10%) of a certain class or belonging to a certain pathway will appear overrepresented compared to the prevalence in the genome, when in fact this is just a sampling artifact. The lack of strong GO associations between the up-regulated genes in our study does not mean that individual genes might not be strongly dysregulated, significant and predictive in GBM exoRNA, but simply that the up-regulated genes do not appear to be associated by already known relationships. Up-regulation of specific genes in exoRNA from GBM patients could be a biological response from normal cells to the presence of the tumour, but they could also be specifically derived from the tumour cells and as such be useful as markers of the tumour. However, identifying uniquely elevated levels of tumour-derived exoRNAs in a high background of normal exoRNA from other serum microvesicle sources is more challenging than identifying systemic changes, such as the depressed lymphocyte count and ribosomal protein RNA levels.
The observation that the up-regulated genes chosen for qRT-PCR validation was not verified may be attributed to the fact that we used two different sets of serum samples for the microarray analysis and the qRT-PCR validation, making borderline significant increases difficult to confirm. Similarly, the positive validation of the down-regulated genes in a separate set of samples makes this observation stronger.
We observed that, coinciding with the significantly increased amounts of microvesicle RNA in GBM patient serum compared to normal controls, the expression levels of all the genes we investigated by qRT-PCR were lower than in normal controls when normalized to total RNA amount (Figure ). It is likely that this is a general trend across most mRNAs, and since all these genes are down-regulated other transcripts must be up-regulated to account for the overall increase in RNA amounts.
The Bioanalyzer profiles of the different RNA serum samples suggest that this up-regulation of RNA in the GBM patients mainly falls in the range of < 300 nt. This could explain why the majority of the tumour exoRNA transcripts appeared down-regulated on the array. If the < 300 nt RNAs do not contain sequences belonging to the coding genes, they would not have been picked up by the microarray and the mRNA-fraction recognized by the capture probes would be relatively smaller since the same amount of input RNA was used for hybridization. There are many RNA species that are not covered by the microarray, including miRNA (mature and precursors), repetitive elements and other non-coding RNAs, as well as single stranded DNA [28
Balaj et al. [28
] showed that human endogenous retroviruses and other transposable elements, as well as single stranded DNA fragments are very abundant in microvesicles from cancer cells, as compared to normal fibroblasts, including Alu, LINE and HERV sequences and it is likely that these and other non-coding RNAs are contributing to the increased amounts of RNA we observe in GBM patient serum (single stranded DNA is also detected on the Bioanalyzer RNA chip). We tested a single transposable element Alu-Y by qRT-PCR and found it to be extremely abundant in the microvesicles, but did not observe any differential expression between the GBMs and normal controls. The human genome consists of about 40% retrotransposon sequences [35
], and there is an increasing number of publications showing the dysregulation of transposable elements and other non-coding RNAs in cancer [36
The very existence of the peak of < 500 nt RNAs in GBM serum exoRNA make it a fertile ground for biomarker discovery, and it warrants further investigation to establish the exact nature and distribution of the nucleic acid species contained in this fraction. A number of regulatory non-coding RNAs are transcribed off coding elements [39
]. Collectively, down-regulation of specific RNAs and up-regulation of RNA levels in exosomes/microvesicles from serum of GBM patients, as compared to controls provides promising biomarkers. In addition, in other studies, tumour mutant mRNAs [18
] and elevated oncogene mRNA [28
] have been detected in serum exosomes/microvesicles from human GBM patients and mice bearing medulloblastoma tumours, respectively. Tumour mutant RNA from prostate cancer patients can also be found in urine exosomes/microvesicles [40
]. More detailed analysis of exoRNA released by tumour cells into serum should be possible as isolation methods are developed with tumour specific surface markers for different types of cancer, e.g. with magnetic activated sorting [30
] and microfluidic capture [41