A small noncoding RNA signature found in exosomes of GBM patient serum as a diagnostic tool

Abstract

Background

Glioblastoma multiforme (GBM) is the most frequent malignant brain tumor in adults, and its prognosis remains dismal despite intensive research and therapeutic advances. Diagnostic biomarkers would be clinically meaningful to allow for early detection of the tumor and for those cases in which surgery is contraindicated or biopsy results are inconclusive. Recent findings show that GBM cells release microvesicles that contain a select subset of cellular proteins and RNA. The aim of this hypothesis-generating study was to assess the diagnostic potential of miRNAs found in microvesicles isolated from the serum of GBM patients.

Methods

To control disease heterogeneity, we used patients with newly diagnosed GBM. In the discovery stage, PCR-based TaqMan Low Density Arrays followed by individual quantitative reverse transcriptase polymerase chain reaction were used to test the differences in the miRNA expression levels of serum microvesicles among 25 GBM patients and healthy controls paired by age and sex. The detected noncoding RNAs were then validated in another 50 GBM patients.

Results

We found that the expression levels of 1 small noncoding RNA (RNU6-1) and 2 microRNAs (miR-320 and miR-574-3p) were significantly associated with a GBM diagnosis. In addition, RNU6-1 was consistently an independent predictor of a GBM diagnosis.

Conclusions

Altogether our results uncovered a small noncoding RNA signature in microvesicles isolated from GBM patient serum that could be used as a fast and reliable differential diagnostic biomarker.

biomarkers, diagnosis, exosomes, GBM, sncRNAs

Glioblastoma multiforme (GBM) is the most frequent malignant primary brain tumor in adults. Despite recent advances in treatment, which consists mainly of surgery and combined radiochemotherapy,¹ the prognosis of GBM patients remains dismal, with median overall survival times of <15 months. Although neuroimaging may suggest its diagnosis, other brain lesions, such as abscesses, metastases, and other types of tumors, might share radiological features with GBM.² Hence, histological examination of the tumor tissue obtained by surgery is currently mandatory for definite diagnosis. Therefore, the development of minimally invasive tests for the detection and monitoring of GBM would be clinically meaningful, especially in those cases in which biopsy results are inconclusive or surgery is contraindicated, and would provide a significant advance in the diagnosis of this devastating disease. Although conventional strategies for blood-based biomarker discovery have been shown to be promising, the development of clinically validated cancer detection markers remains an unmet challenge for many common cancers,³ especially GBM.

Exosomes from body fluids have emerged as promising reservoirs of diagnostic and prognostic biomarkers in cancer.^4–6 They are microvesicles (MVs), formed by endosomal membrane invagination, that later fuse to the plasmatic membrane and are released out of the cell.⁷ Their release is increased in tumor cells,^8,9 and it has been shown that cancer patients have a higher amount of circulating exosomes than do healthy controls.¹⁰ Exosomes are a rich source of selectively packaged molecules from the cell of origin, suitable for profiling analysis,^7,11,12 such as small noncoding (snc)RNAs, including microRNAs (miRNAs).^13,14 They play important roles in oncogenesis,¹⁵ and several studies have shown that signatures of miRNA expression differ between normal and tumor tissues, including GBM.¹⁶

Recently, 2 different groups described the release of MVs, and more specifically exosomes, from GBM cells.^11,17 Importantly, they isolated exosomes from GBM patient serum and analyzed their protein and RNA content, pointing to them as potential biomarker reservoirs in GBM.^11,17

The goal of this hypothesis-generating study was to develop a noninvasive and specific diagnostic tool for GBM patients based on miRNA expression in serum exosomes. We hypothesized that a miRNA signature in serum exosomes might be used as such a tool. To address this hypothesis, we screened serum exosome miRNAs by using PCR-based arrays followed by an extensive study that used individual quantitative real time PCR assays in 2 independent cohorts of GBM patients.

Materials and Methods

Patients

For this study, serum was used from 75 patients with newly diagnosed, untreated, histologically proven GBM, according to the World Health Organization classification. Serum samples were collected between 2009 and 2012 at the Pitié-Salpètriere Hospital, Paris, France; Clínica Universidad de Navarra, Hospital de Navarra, and the Tumor Biobank of the Servicio Navarro de Salud-Osasunbidea, Pamplona, Spain; and Hospital Universitario Donostia and the Basque Research Biobank-O+Ehun-nodo Gipuzkoa, San Sebastian, Spain. The control group consisted of 55 healthy donors from the Basque Research Biobank-O + Ehun and subjects recruited at Clínica Universidad de Navarra who showed no history of cancer.

The study protocol was approved by the institutional review board, and all of the participants signed the informed consent form approved by the respective institutional review boards or ethical committees.

For the initial training group, serum samples from 25 GBM patients (14 men and 11 women; median age 59.6 ± 10.6 y, range 30–75) and 25 healthy controls (14 men and 11 women; median age 60 ± 10.9 y, range 45–78) paired for sex and age were included (see Supplementary Data).

For the validation study, sera were analyzed from an additional 50 GBM patients (30 men and 20 women; median age 61 ± 12.7 y, range 17–79) and 30 healthy controls (14 men and 16 women; median age 54.3 ± 11.8 y, range 27–70) (see Supplementary Data).

Microvesicle Isolation From Human Subjects

MVs were isolated from serum using Exoquick^18,19 precipitation solution according to the manufacturer's recommendations (System Biosciences). Serum was filtered with a 0.22-μm pore filter before Exoquick precipitation, and the obtained pellet was resuspended in water.

RNA Extraction

MVs precipitated with Exoquick were treated with RNase A (Ambion) and DNase I (New England Biolabs) for 30 min at 37°C to eliminate extramicrovesicular RNA and DNA. Total RNA was isolated from MVs using Trizol reagent (Invitrogen). In order to increase the yield of small RNAs, glycogen (Ambion) was added. RNA was quantified in a Nanodrop ND-1000 spectrophotometer (Thermo Scientific).

MiRNA Profiling by Quantitative Real-time PCR

MiRNA expression was analyzed using the Applied Biosystems TaqMan Human microRNA A Array v2.1 (TaqMan Low Density Array [TLDA]) to profile 381 mature miRNAs by qRT-PCR. Raw cycle threshold (Ct) values were calculated using SDS software v2.4 with automatic baseline settings and a threshold of 0.05.

Specific miRNAs were validated by qRT-PCR using individual TaqMan miRNA assays. All reactions were performed using a 7900HT RT-PCR instrument (Applied Biosystems), and triplicate samples were used throughout.

TLDA Data Analysis

TLDA data were analyzed with R/Bioconductor.²⁰ The analysis consisted of a filtering process in order to eliminate miRNAs that were not detected in the experiment. We considered Ct scores >35 to be nonspecific.^21,22 Therefore, miRNAs that had a raw Ct value >35 in more than 50% of the patients or control samples were excluded. Using this filtering criterion, we considered for statistical analysis only 48 miRNAs.

LIMMA (Linear Models for Microarray data) was used to find differentially expressed miRNAs.²³ A false discovery rate (FDR) of 0.05²⁴ and an expression change (−▵▵Ct) of at least 2 Ct were established as selection criteria. MiRNA expression normalization was performed following 2 different approaches: using the RNU48 endogenous control and using median normalization analysis.²⁵ Both normalization methods identified almost the same miRNAs deregulated with a significant FDR supporting each other. We decided to use RNU48 as endogenous control for further validation analysis because of the feasibility to implement one gene versus the median of the whole PCR array in the following validation studies.

Validation of TLDAs and the Independent Validation Group

The normalization for the Ct values of the TLDA validation and the independent validation group of samples was performed using RNU48 as endogenous control. Fold change was calculated as the difference of the mean normalized expression values between GBM patients and controls (−▵Ct = mean (▵Ct GBM patients) – – mean (▵Ct controls)). A t-test analysis was performed with R/Bioconductor,²⁰ and miRNAs were considered differentially expressed using a threshold of P < .01.

Biomarker Discovery Using a Classifier Approach

A machine learning algorithm based on logistic regression was applied to classify patients and identify the optimal separating miRNAs between GBM patients and healthy controls.²⁶ The performance of the classifiers was evaluated using receiver operator characteristic (ROC) analysis.²⁷

Results

Characterization of Microvesicles Isolated From GBM Patient Serum

The aim of this hypothesis-generating study was to determine whether a specific miRNA profile found in exosomes isolated from the serum of GBM patients could serve as a biomarker of the disease. First, we isolated MVs from serum of GBM patients and healthy controls using an adsorption method (Exoquick). Exoquick is a compound of undisclosed composition that has been proposed on the market to precipitate by a single step exosomes from small volumes of serum or cell culture supernatant. This method is less time-consuming than the traditional method of gradient ultracentrifugation and therefore would be easier to implement in a clinical setting. In order to confirm their nature, we followed an extensive characterization. We assessed the morphology and size using transmission electron microscopy and dynamic light scattering. These analyses showed 2 sets of membrane-like structures: one with a diameter ranging from 12.86 to 23.81 nm and another set of MVs with sizes ranging from 157.9 to 210.8 nm (Fig. 1A and B). We further evaluated the expression of several exosome markers by western blot in total serum and in MVs isolated from healthy controls and GBM patients. We found that markers such as CD9, Lamp1, TSG 101, and Alix²⁸ were enriched in the MVs of both controls and patients compared with total serum (Fig. 1C). Importantly, the endoplasmic reticulum marker calnexin was not detected in these samples, suggesting that our samples were enriched with exosome-like MVs. In addition, using flow cytometry, we detected that the expression of CD9 and of CD63, another exosome marker, was enriched on GBM patients' MV surface (CD9 = 73.18% ± 10.02 and CD63 = 28.25 ± 11.3) compared with healthy controls (CD9 = 52.27 ± 13.34, P < .002, and CD63 = 10.16 ± 2.7, P < .001) (Fig. 1D and E). Altogether our data showed that we were able to isolate exosome-like MVs from the serum of GBM patients in a fast and reliable manner by using an adsorption method such as Exoquick.

Fig. 1.

Characterization of MVs isolated from GBM patients’ serum. (A) Transmission electron microscopy analysis. Representative micrograph depicting MVs found in the serum of GBM patients. Arrows point to membrane-like structures with different sizes. (B) Determination of MV size distribution by dynamic light scattering. (C) Evaluation by western blot of the expression levels of several MV/exosome markers in total serum or in MVs isolated from serum of either healthy controls (C) or GBM patients (P). (D) Determination by flow cytometry of CD9 or ECD63 expression on the surface of MVs isolated from healthy controls or GBM patients’ serum. (F) Study design scheme.

Detection of Serum Exosome MiRNAs and Their Association With GBM Diagnosis

In order to accomplish our study objective, we followed the scheme depicted in Fig. 1F. We profiled the expression of 381 known miRNAs by using microfluidic TLDA miRNA qRT-PCR in 25 GBM patients matched by age and sex with healthy donors (see Supplementary Data). We generated a list of likely exosome-based miRNA biomarker candidates for GBM patients that satisfied the following criteria for additional individual qRT-PCR validation: (i) to be differentially expressed by at least 2 Cts and (ii) to have a significant FDR (P < .05). Overall, we found 7 miRNAs (miR483-5p, miR-574-3p, miR-320, miR-197, miR-484, miR-146a, and miR-223) and 1 small nuclear RNA (RNU6-1) that met both criteria (Table 1). These 8 sncRNAs were further subjected to single qRT-PCR analyses to validate their expression in the same samples. This new analysis confirmed the overexpression of RNU6-1, miR-320, and miR-574-3p in GBM patients. As shown in Fig. 2A, the expression levels of the 3 sncRNAs were significantly different between the GBM patient group and the healthy controls (P < .0001 for RNU6-1; P = .007 for miR-320; P < .003 for miR-574-3p). Intriguingly, the small nuclear RNA RNU6-1 was the most upregulated in GBM patients (fold change, 2-ΔCt = 387 ± 1370) compared with healthy controls (fold change = 11.73 ± 23.30, P < .001; Table 2). Next, a machine learning algorithm based on logistic regression was applied to the expression of each single sncRNA or the expression of the 3 sncRNAs to interrogate whether they could discriminate between GBM patients and healthy controls. The obtained classifier had a good diagnostic performance, as shown by ROC analyses, with areas under the curve of 0.926 (95% confidence interval [CI], 0.84–1; P < .0001) for the 3 sncRNAs together, 0.852 (95% CI, 0.74–0.96; P < .0001) for RNU6-1, 0.720 (95% CI, 0.56–0.87; P = .0067) for miR-320, and 0.738 (95% CI, 0.58–0.89; P = .0055) for miR-574-3p (Fig. 2B). At a cutoff value of 0.349 for the 3 sncRNAs, sensitivity was 87% and specificity was 86%. At a cutoff value of 0.454 for RNU6-1, sensitivity was 73% and specificity was 70%. At a cutoff value of 0.477 for miR-320, sensitivity was 65% and specificity was 65%. At a cutoff value of 0.454 for miR-574-3p, sensitivity was 59% and specificity was 59%. These results suggested that either RNU6-1 alone or the miR-320/miR-574-3p/RNU6-1 combined signature could have diagnostic value in GBM patients.

Table 1.

List of differentially expressed sncRNAs in serum MVs of 25 GBM samples compared with 25 healthy controls paired by age and sex, identified using real-time PCR–based miRNA profiling assays (change >2 Ct and FDR <0.05 as a cutoff level)

MicroRNA	−ΔCt	P	FDR
RNU6	5.0	<.0001	<0.001
MiR-483-5p	2.7	.0011	0.011
MiR-574-3p	3.0	.0039	0.024
MiR-320	2.1	.0050	0.027
MiR-197	2.9	.0068	0.030
MiR-484	2.3	.0070	0.030
MiR-146a	2.1	.0080	0.032
MiR-223	2.1	.0110	0.041

Table 2.

Differentially expressed sncRNAs in serum MVs of 25 GBM samples compared with 25 healthy controls paired by age and sex, after validation using single qRT-PCR analyses (change >2 Ct and t-test P < .01 as a cutoff level)

MicroRNA	−ΔCt	P
RNU6	4.5	<.0001
MiR-320	2.5	.0076
MiR-574-3p	3.1	.0031

Fig. 2.

Expression of sncRNAs in MVs and their association with GBM diagnosis. (A) SncRNA validation by individual quantitative reverse transcription PCR in the training cohort. Expression levels of the individual sncRNAs are normalized to RNU48. Fold change was calculated as the difference of the mean normalized expression values between patients’ and controls’ MVs (−ΔCt = mean (ΔCt GBM patients) – –mean (ΔCt controls)). A t-test analysis was performed with R/Bioconductor. (B) ROC curve showing the true positive and false positive rates for the training cohort for the 3 sncRNA signatures or for each individual one. Validation of the diagnostic value of the sncRNA signature in an independent cohort. AUC, area under the curve. (C) SncRNA expression by individual quantitative reverse transcription PCR in an independent cohort. Expression levels of the individual sncRNAs are normalized to RNU48. Fold change was calculated as the difference of the mean normalized expression values between patients’ and controls’ MVs (ΔCt = mean (ΔCt GBM patients) – –mean (ΔCt controls)). A t-test analysis was performed with R/Bioconductor. (D) ROC curve showing the true positive and false positive rates for the training cohort for the 3 sncRNA signatures or for each individual one.

Independent Validation of the Diagnostic Value of the SncRNA Signature

To further verify the diagnostic value of the sncRNA signature identified in the previous cohort, the expression of the 3 sncRNAs was assessed in an independent group of 80 serum exosome samples, from 50 GBM patients and 30 healthy controls (see Supplementary Data). In this new cohort, only the expression of RNU6-1 was significantly elevated in GBM patients (fold change: 1732 ± 11 590) compared with healthy controls (fold change: 63.79 ± 210, P < .001; Fig. 2C). We did not find significant differences in the expression levels of miR-320 and miR-574-3p (P = .592 and P = .173, respectively). However, ROC curve analyses revealed that the expression levels of either RNU6-1 alone or the combination of the 3 sncRNAs found in the exosomes of the validation cohort were useful and robust biomarkers for differentiating GBM patients from healthy controls, with areas under the curve of 0.722 (95% CI, 0.60–0.84; P = .0007) for RNU6-1 and 0.775 (95% CI, 0.65–0.90; P < .0001) for the 3 markers together (Fig. 2D). Importantly, at a cutoff value of 0.372 for RNU6-1, sensitivity was 66% and specificity was 68%. Meanwhile, at a cutoff value of 0.374 for the 3 sncRNA signatures, sensitivity was 70% and specificity was 71%. Altogether our results support the notion that the expression levels of RNU6-1, miR-320, and miR-574-3p found in exosomes isolated from serum have the power to discriminate healthy individuals from GBM patients and therefore could serve as diagnostic biomarkers.

Discussion

In this hypothesis-generating study, we demonstrate for the first time that an sncRNA signature of 2 miRNAs (miR-320 and miR-574-3p) and 1 small nuclear RNA (RNU6-1) found in exosomes isolated from the serum of GBM patients could serve as a potential diagnostic biomarker. Although previous studies suggested the utility of plasma or serum miRNAs in GBM,^29–31 to our knowledge this is the first report on the quantitative assessment of miRNAs in exosomes isolated from GBM patients' sera.

There is a paucity of robust biomarkers in GBM that allow early detection and monitoring of response to treatment. In the last years, circulating miRNAs have emerged as ideal candidate biomarkers because they are stable in plasma and serum,³ their expression is deregulated in cancer, and they appear to be tissue specific.³² Moreover, the notion that solid tumors, including GBM,^11,17 shed large quantities of small, membranous MVs that serve as biocargos for proteins^17,33 as well as DNA³⁴ and RNA¹² into the circulation has opened the possibility to use these MVs as potential sources of biomarkers to predict both diagnosis and response to therapy. Therefore, in this study we explored the possibility of finding a specific miRNA signature in serum MVs that would allow discernment of GBM patients from healthy controls.

Several studies have identified aberrantly expressed miRNAs in GBM tumor tissue,^35,36 including circulating miRNAs^29–31 that suggest different sets of miRNAs to be of prognostic and/or diagnostic value in different tumors and in glioblastoma. This fact could be due not only to the type of tumor but also to the way that the sample has been collected, method of extraction, and method of detection (eg, real time, deep sequencing). In addition, other variables are whether the patients had already undergone surgery and the type of treatment they had received. On many occasions the patients had not been treated homogeneously. One advantage of our analyses is that the sera of both sets of patients were collected at initial diagnosis, immediately before surgery and without any prior treatment. This fact makes our series very homogeneous and adds extra value to the data.

In our study, results suggest that either RNU6-1 alone or the miR-320/miR-574-3p/RNU6-1 combined signatures are biomarker candidates to distinguish between GBM patients and healthy controls. The most intriguing result of the miRNA analysis was the finding that RNU6-1 was the main biomarker candidate to distinguish between GBM patients and healthy controls, since this RNA is part of the splicing machinery and resides in the nucleus.³⁷ Interestingly, it has been shown that RNU6-1 is synthetized by RNA polymerase III, which is negatively regulated by tumor suppressors such as retinoblastoma protein (Rb) and phosphatase and tensin homolog (PTEN),^38,39 and its enhanced activity has been shown essential for tumorigenesis.⁴⁰ Therefore, it could be possible that in the context of Rb or PTEN pathway dysfunctions, which are prevalent in GBM, RNU6-1 is overexpressed. Supporting this notion, the U6 : SNORD44 ratio was found to be consistently higher in the sera of breast cancer patients, regardless of estrogen receptor status.⁴¹ Furthermore, RNU2, a counterpart component of RNU6-1 in the spliceosome, has recently been detected circulating in serum and has been proposed as a novel diagnostic biomarker for pancreatic ductal adenocarcinoma and colorectal carcinoma.⁴² Nevertheless, further studies would be needed to elucidate the molecular mechanism of RNU6-1 upregulation in GBM serum MVs.

Regarding miR-320 and miR-574-3p, both have been found circulating in serum or plasma.^43,44 MiR-320 has important roles in different cancer types, such as breast and colon,^44,45 and miR-574-3p has an oncogenic function in bladder and gastric cancer.^46,47 However, we cannot rule out that miR-574-3p detection in serum is due to secretion of myeloid cells rather than tumor cells.⁴⁸

One limitation of the present report is the lack of a control group of patients with other CNS disorders, such as cerebrovascular diseases, multiple sclerosis, and abscesses, and other brain tumors such as primary CNS lymphoma or metastases that may mimic GBM on neuroimaging. Further studies incorporating patients with these malignancies would be of interest in order to confirm the diagnostic accuracy of the sncRNA signature.

In the fullness of time, our study provides for the first time an sncRNA signature found in serum MVs that discriminates GBM patients from healthy controls and that could hence serve as a noninvasive diagnostic tool for early detection and monitoring of the disease. Moreover, it could constitute an ideal candidate biomarker for point-of-care diagnostic testing.

Funding

This work was supported by the European Union (Marie Curie IRG270459 to M.M.A.), the Spanish Ministry of Health (PI10/00399 to M.M.A., PI10/01069 to A.A., and Sara Borrell contract CD06/0275 to L.M.), the Spanish Ministry of Science and Innovation (Ramón y Cajal contract RYC-2009-05571 to M.M.A.), Diputación Foral de Gipuzkoa DFG 09/003 to L.M.), and the Basque Government (SAIO11-PC11BN002 to A.M.).

Acknowledgments

We thank Dr Charles Lawrie (Instituto Biodonostia, San Sebastian, Spain) and Dr Rubén Pío (Center for Applied Medical Research, Pamplona, Spain) for the critical reading of the manuscript and their helpful comments.

Conflict of interest statement. No potential conflicts of interest were disclosed. The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Appendix

³⁴

Table 1.

Training set patient characteristics

Code	Age	Sex
1	72	F
2	68	F
3	61	M
4	56	F
5	48	M
6	57	M
7	57	M
8	45	F
9	62	F
10	72	M
11	75	F
12	47	M
13	67	F
14	70	F
15	30	M
16	60	M
17	66	M
18	57	M
19	65	M
20	60	F
21	50	F
22	64	F
23	55	M
24	75	M
25	52	M

Table 2.

Validation set patient characteristics

Code	Age	Sex
V1	37	F
V2	78	F
V3	17	M
V4	65	F
V5	78	F
V6	78	M
V7	79	M
V8	77	M
V9	73	F
V10	63	M
V11	75	F
V12	60	M
V13	40	F
V14	68	M
V15	67	M
V16	29	M
V17	65	M
V18	62	F
V19	68	F
V20	51	M
V21	65	M
V22	55	F
V23	48	M
V24	65	F
V25	73	M
V26	67	M
V27	69	M
V28	57	M
V29	62	M
V30	65	F
V31	60	M
V32	42	M
V33	54	M
V34	63	M
V35	44	F
V36	65	M
V37	63	F
V38	49	M
V39	60	M
V40	57	M
V41	56	M
V42	75	F
V43	54	F
V44	71	M
V45	70	F
V46	61	F
V47	69	F
V48	63	M
V49	58	M
V50	78	F

Author notes

This work was presented at the EANO 10th Annual Conference, Marseille, 6–9 September 2012, and the SNO 17th Annual Conference, Washington, 15–18 November 2012.