BEAMing and Droplet Digital PCR Analysis of Mutant IDH1 mRNA in Glioma Patient Serum and Cerebrospinal Fluid Extracellular Vesicles

Introduction

There has been growing interest in cancer diagnostics using circulating tumor-derived nucleic acids as a source for tumor biomarkers. This approach is particularly attractive because it allows detection of genetic mutations in the tumor as well as quantification of mutant sequences present in the sample. Biofluid-based analysis is less invasive than performing tumor biopsy, extends the range of a single biopsy by offering the potential to mirror the genetic diversity within a tumor, and also enables longitudinal measurements to monitor genetic changes in tumor over time.^1,2

A new promising source of biofluid-based biomarkers is extracellular vesicles (EVs); they are actively released from normal as well as cancer cells both in vitro and in vivo,³ with upregulated release from many tumor cells, as compared with normal cells.^3,4,5,6 They include several types of membrane-bound vesicles, such as exosomes, shed EVs, oncosomes and apoptotic blebs.⁷ EVs contain a diverse repertoire of genetic material and protect their cargo from the destructive action of enzymes, such as RNases,⁴ thus enabling a stable enrichment of nucleic acids from biofluids. RNA from glioblastoma (GBM) serum EVs has previously been used to detect the EGFRvIII mutation.⁴ Tumor-derived EVs are also found in the cerebrospinal fluid (CSF),⁸ and may provide valuable biomarker information about the tumor origins and status. Several diagnostic techniques utilize CSF for analysis; however, most of them do not provide any molecular or genetic information on malignancies. In vivo techniques, including CSF cytology, clinical findings and neuro-imaging, as well as in vitro techniques like immunocytochemistry, flow cytometry and PCR have so far only been informative for leptomeningeal cancers, in which tumor cells have entered and disseminated within CSF.⁹

Highly sensitive technologies must be used when detecting mutations in the background of wild-type sequence and the optimal assay may vary according to the type of genetic alteration to be detected. For example, large gene/transcript deletions like EGFRvIII are relatively easy to detect in the background of wild-type sequences using well optimized allele-specific PCR assays.⁴ In contrast, single nucleotide mutations are very difficult to pick up when the fraction of mutated transcripts is low in a background of wild-type sequences, and even well optimized allele-specific PCRs can rarely detect mutations at a frequency of less than 1%.¹⁰ To increase the sensitivity of EV RNAs as tumor biomarkers, we adapted a well established and highly sensitive form of digital PCR known as BEAMing (Beads, Emulsions, Amplification, Magnetics)¹¹ to interrogate the mRNA within EVs for the presence of tumor specific mutations. BEAMing PCR is a highly sensitive and quantitative technique that has been successfully applied to the mutational analysis of circulating DNA even when the mutation is as rare as 0.01%.^12,13,14 We also used a droplet digital PCR (ddPCR) technology, which creates millions of uniform 5 picoliter-volume droplets per sample for counting applications including rare allele detection and validation in tumor and cell-free plasma,^15,16,17 and assessment of DNA quality before sequencing.¹⁸ Neither of these technologies have previously been used to quantify rare mutations at the RNA level or been tested on EV nucleic acids.

The IDH1 mutation was chosen for analysis because of its high relevance in tumor pathogenesis. Analysis was carried out for wild-type (G395) and mutant (A395) sequences in the mRNA transcripts for isocitrate dehydrogenase 1 (IDH1), with the latter resulting in a substitution of arginine for histidine at amino acid position 132 (R132H;).¹⁹ Functional studies have revealed that the H132 mutant protein catalyzes the conversion of α-ketoglutarate into 2-hydroxyglutarate (2-HG), a suspected oncometabolite.^19,20 Mutant IDH1 was first reported in colorectal cancer²¹ and has been identified in acute myelogenous leukemia and in GBM through genome wide mutational analysis.^22,23 A subsequent study found mutations in IDH1/2 in almost 80% of grades II and III astrocytomas, oligodendrogliomas and secondary GBMs,²⁴ with the most common mutant being the G395A substitution. Microarray RNA profile analysis indicates that IDH1 mutations in gliomas occur primarily in the proneural subtype, making them distinctive from the classical, neural, and mesenchymal subtypes which rarely show this mutation²⁵ and suggesting that this mutation is correlated to an improved prognosis and a potential stratifier for treatment options. Indeed, mutations in the IDH1 gene in all low glioma grades, as well as secondary GBMs, predict longer survival and better response to temozolomide treatment.^26,27 Although several studies have reported that levels of 2-HG increase up to ~100-fold in glioma cell lines expressing mutant IDH1,²⁸ a recent study failed to show a correlation between serum levels of 2-HG with IDH1/2 mutation status or tumor size in patients with glioma,²⁹ suggesting that quantifying mutant mRNA copy number rather than the metabolite may provide more accurate information about tumor status.

Using BEAMing and ddPCR on mRNA from EVs, we reliably identified the mutant IDH1 mRNA in CSF-derived EVs collected at the time of brain tumor surgery from patients bearing mutant IDH1 glioma tumors, but not in serum-derived EVs. Under the conditions used, both assays gave similar levels of sensitivity and specificity in CSF samples. We also showed that CSF-derived EVs from patients with tumor have higher levels of IDH1 mRNA than CSF EVs from controls. A higher copy number of IDH1 mRNA was also found in serum from patients with tumor as compared with that from controls. In conclusion, we report that CSF-derived EVs provide a platform for detection of brain tumor specific biomarkers and that digital PCR platforms work well for detection of rare mutations in EV mRNAs.

Results

Discussion

Here, we show that the mutational profile of IDH1 mutated brain tumors can be measured in CSF without the need for a biopsy of the tumor. We can measure the presence as well as the abundance of the mutated transcript found in CSF EVs with digital PCR methods such as BEAMing or ddPCR. This method may be useful as a non-invasive tool to verify and subtype brain tumors in cases where the tumor location makes biopsies risky or impossible. It may also be used to subtype tumors for drug clinical trial enrollment. BEAMing and ddPCR assays had the same sensitivity and specificity levels using CSF. Serum contains vesicles from many normal processes naturally occurring in the blood such as maturation of reticulocytes and platelet activation. The CSF may therefore be a more enriched source of the brain tumor–derived vesicles and therefore easier to detect than in serum. Interestingly, the number of copies of IDH1 mutant mRNA in CSF EVs from patients with tumor was significantly higher than in controls (Table 1). Noteworthy is the difference in sampling site, most CSF samples from patients with tumor were collected upon opening of the dura mater, near the primary tumor, whereas CSF control samples were collected by lumbar puncture. Patient sample 308 was collected by lumbar puncture and shows that mutation detection can be done from lumbar puncture, and this sample actually had the highest percentage of mutant IDH1 transcripts. The high levels of IDH1 mRNA in cisternal CSF from patients with glioma collected during surgery compared with control patients collected by lumbar puncture may be due to a higher number of EVs released by tumor or normal cells in the vicinity of the tumor or from debris from normal cells in the mechanical process of opening the dura mater. Two of the false negative samples (in the EVs) had unusually high copy numbers of WT IDH1 (~400,000 WT copies; Table 1) that could have masked the detection of the mutant. For this purpose, lumbar puncture samples may be preferred for this analysis.

Using the BEAMing assay we determined that between 41 and 181 mutant IDH1 mRNA transcripts were present in 1 ml CSF in four samples, with one sample having as many as 23,912 mutant copies. ddPCR detected roughly twice as many mutated RNA transcripts per ml CSF (ID#s 308 = 536, 2244 = 318, 2160 = 203, 2199 = 102, and 2510 = 57,389). Interestingly, two of the three IDH1 mutant positive tumors that were not detected by either method were grade II CSF samples (IDs# 2287 and 2516, Table 1). The third sample that was not detected was a grade III tumor (ID# 2513) that had the smallest tumor size in grade III category suggesting a possible correlation with grade as well as tumor size (Figure 4c). Further studies with a larger number of samples will address this correlation more thoroughly. The IDH1 mutant detection sensitivity in serum will need further optimization, and the relatively low IDH1 mRNA copy number suggest that increasing the volume of serum input may increase the sensitivity of the assay. All tumor-free controls included in the study were wild-type for IDH1 mRNA as analyzed by FACS (Table 1). Future studies will aim to increase the copy number input to improve the sensitivity of the assay in blood-based assays and to correlate the detection of mutant IDH1 with clinical, demographic, MRI and molecular neuropathologic studies. Other studies have reported detection of other tumor biomarkers in blood on DNA, such as the EGFR gene amplification in lung adenocarcinomas³⁰ and PIK3CA mutations in metastatic breast cancer.³¹ We have previously shown that also DNA can be present in glioma EVs;³ however, this study focused only on the RNA fraction.

IDH1 is usually a heterozygous mutation,²⁸ so when 5% of the IDH1 transcripts are mutated, approximately 10% of the IDH1 transcripts come from the tumor. It may be possible to enrich the tumor EV fraction by immunoaffinity methods using antigens specific to the tumor cell surface³² or using magnetic resonance detection of mutant proteins to potentially increase the sensitivity of mutation assays.³³

Information on the mutational profiles of different types of gliomas has increased to the point where genomic alterations and expression profiles characteristic of the classical, proneural, neural, and mesenchymal glioma subtypes have been identified.¹² EV-BEAMing or ddPCR on biofluid EVs should have high value in monitoring specific subtypes of gliomas in a minimally invasive manner. Although obtaining CSF is not a routine procedure in patients with brain tumors, it is markedly less invasive than obtaining an intracerebral biopsy of the tumor. Further extension of this work in clinical studies will require focusing on samples derived by lumbar puncture, as this is the only feasible sampling methodology to perform on a routine basis but this study lays the foundation for lumbar puncture–based CSF mutation analysis studies in patients. This technology should enable monitoring of a variety of mutations and transcript levels in tumors using CSF and potentially serum-derived EVs not only when the tumor is located in areas with inherent risk of sampling such as brain, but also for tumors in other parts of the body. This study provides a proof of principle for detection of single base pair mutations in tumor RNA using EVs in CSF samples, combined with a sensitive digital PCR assays.

Materials and methods

Study design and sample collection. This study was approved by the Institutional Review Boards of Massachusetts General Hospital, Beth Israel Deaconess Medical Center, and the University of California, Los Angeles. Serum and CSF samples from patients and healthy volunteers were collected according to approved IRB protocols. Brain tumors were obtained at the Massachusetts General Hospital, Beth Israel Deaconess Medical Center and University of California, Los Angeles and confirmed by histopathology. The mutational status of the IDH1 gene in tumor tissue was determined by genomic sequencing or SNaPShot analysis by the Molecular Neuropathology Laboratory at UCLA and Massachusetts General Hospital, respectively.

Sample processing. Blood was collected from patients undergoing surgical removal of gliomas using BD Vacutainer SST tube (BD Biosciences, San Jose, CA), before opening of the dura mater, maintained for 30 minutes at room temperature and then centrifuged at 1,300g for 10 minutes. The serum was separated from the coagulated blood cells, filtered through a 0.8 μm filter and stored at −80 °C. CSF, also collected at surgery from patients with glioma upon opening of the dura mater, was centrifuged at 300g for 10 minutes to remove cells, and stored at −80 °C. Control CSF samples were collected by lumbar puncture from patients with suspected non-malignant, neurological disease whose CSF profile was later found to be normal, and stored at −80 °C. At the time of analysis, samples were thawed, centrifuged at 100,000g for 80 minutes at 4 °C to collect EVs and RNA was extracted, as previously described.³ To minimize any risk of contamination, tumor tissue was extracted and sequenced in one laboratory, whereas serum and CSF EV RNA was extracted at a different location in a DNA-free, RNA extraction facility, and BEAMing and ddPCR were performed at a third location.

Extravesicular RNA isolation. Frozen serum or CSF samples were thawed (1 ml/sample), diluted to 2.5 ml with sterile PBS and centrifuged at 100,000g for 80 minutes. External DNA on the EVs was removed using 2 µl of DNAse from the Turbo DNA-free kit (Ambion, Foster City, CA), according to manufacturer's recommendations. RNA was purified using the miRNeasy Qiagen kit (Qiagen, Valencia, CA) and eluted in 30 µl RNase-free water. RNA quality was assessed using a 2100 Bioanalyzer (Agilent) with an RNA 6000 Pico Chip kit (Agilent).

Determination of tumor volume. For high grade gliomas with significant contrast enhancement, T1-weighted sequences after injection of contrast agent were used to define the “tumor volume” as the area of contrast enhancement including central necrotic tissue. For low grade tumors and tumors without significant enhancement, volume was determined using T2 areas of hyperintensity. To determine the volume, we measured the tumor area in each slice using a region-of-interest method (BrainLAB Software; BrainLAB, Feldkirchen, Germany). The volumes were calculated by multiplying these resulting areas by the slice thickness. The total tumor volume is the sum of the volumes calculated for each of the multiple parallel slices.³⁴

Nanosight tracking analysis. Nanosight tracking analysis was used to count and size EVs. Serum was processed as above and stored at −80 °C. At the time of the analysis the serum was diluted 1:5,000 and analyzed using the Nanosight. CSF was also stored at −80 °C and diluted 1:50 for Nanosight analysis. Both biofluids were diluted in PBS for nanosight tracking analysis.

Electron microscopy. CSF samples (500 μl) were thawed and ultracentrifuged at 100,000g for 80 minutes. The EV pellet was then resuspended in 50 μl of PBS and analyzed by EM. 2.5 μl of the samples was adsorbed for 30 minutes on carbon coated grids that had been made hydrophilic by a 30-second exposure to a glow discharge. Excess liquid was removed with a Whatman filterpaper (GE Healthcare, Piscataway, NJ) and the samples were stained with 0.75% uranyl formate for 30 seconds. After removing the excess uranyl formate with a filter paper the grids were examined in a JEOL 1200EX Transmission EM or a TecnaiG² Spirit BioTWIN and images were recorded with an AMT 2k CCD camera.

Cryo-transmission EM. CSF and plasma samples were preserved in vitrified ice supported by holey carbon films on 400-mesh copper grids. Each sample was prepared by applying a 3 µl drop of sample suspension to a cleaned grid, blotting away with filter paper, and immediately proceeding with vitrification in liquid ethane. Grids were stored under liquid nitrogen until transferred to the EM for imaging. EM was performed using an FEI Tecnai T12 EM, operating at 120keV equipped with an FEI Eagle 4k x4k CCD camera. Vitreous ice grids were transferred into the EM using a cryostage that maintains the grids at a temperature below −170 °C. Images on each grid were acquired at multiple scales to assess the overall distribution of the specimen. After identifying potentially suitable target of areas for imaging at lower magnifications, pairs of high magnification images were acquired at normal magnifications of 110,000×, 52,000×, and 21,000×.

Quantitative RT-PCR. mRNA was reverse transcribed into cDNA using the SuperScript VILO cDNA synthesis kit (Invitrogen, Carlsbad, CA). To determine the copy number of IDH1 transcripts, we performed quantitative PCR on the cDNA. Reactions were done in a 25 µl reaction using Power SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA) and 160 nmol/l of each primer. Amplification conditions consisted of: (i) 1 cycle of 50 °C, 2 minutes; (ii) 1 cycle of 95 °C, 10 minutes; (iii) 40 cycles of 95 °C, 15 seconds; and 60 °C, 1 minute, and (iv) a dissociation stage consisting of 1 cycle of 95 °C, 15 seconds; 60 °C, 20 seconds; and 95 °C, 15 seconds on the 7000 ABI Prism PCR system (Applied Biosystems). Ct values were analyzed in auto mode and manually inspected for accuracy. Primer dimers were excluded by evaluation of dissociation curve and agarose gel electrophoresis. The following IDH1 primers were used: Forward: CGGTCTTCAGAGAAGCCATT and Reverse: AGGCCCAGGAACAACAAAAT.

Pre-amplification of IDH1 cDNA. We performed IDH1 PCR on the cDNA derived from EV RNA using forward (5′-tcccgcgaaattaatacgacCGGTCTTCAGAGAAGCCATT-3′) and reverse (5′-gctggagctctgcagctaAGGCCCAGGAACAACAAAAT-3′) primers flanking nucleotide position 395 of the IDH1 transcript, where lower case letters denote universal primer sequences and upper case letters denote IDH1 specific primer sequences. Ten μl of cDNA was split into two PCR reactions, each performed in 50 μl volumes containing 5 μl cDNA, 1 U of Phusion Hot Start II DNA Polymerase (New England Biolabs, Ipswich, MA), 1X Phusion HF buffer (New England Biolabs), 0.2 mmol/l dNTP mix (Invitrogen), and 0.2 μmol/l of each primer. Amplification was done in a Veriti Thermal Cycler (Applied Biosystems) using the following conditions: 98 °C for 3 minutes; 10 cycles of 98 °C for 10 seconds, 68 °C for 15 seconds (−0.5 °C/cycle), 72 °C for 10 seconds; 30 cycles of 98 °C for 10 seconds, 62 °C for 15 seconds, 72 °C for 10 seconds; 1 cycle of 72 °C for 5 minutes. The PCR products for each sample were then pooled and purified using a QIAquick PCR purification column (Qiagen) and the concentration determined using a 2100 Bioanalyzer with a DNA 1000 LabChip kit (Agilent).

EV-BEAMing PCR. PCR amplicons from each sample were used as templates for each BEAMing analysis. We performed the BEAMing PCR using a protocol slightly modified from that previously described.¹¹ We set up 150 μl PCR reactions containing 10 μl 20 pmol/l template DNA, 2X TITANIUM Taq DNA Polymerase (Clontech, Mountain View, CA), 1X TITANIUM Taq buffer (Clontech), 1 mmol/l dNTP mix, 8.3 mmol/l MgCl₂ (Invitrogen), 0.05 μmol/l universal forward primer (5′-tcccgcgaaattaatacgac-3′), 8 μmol/l nested reverse primer (5′-AATCAGTTGCTCTGTATTGATCC-3′), and 6 × 10⁷ magnetic streptavidin beads (MyOne, Invitrogen) coated with a modified universal forward primer (5′-dual biotion-T-Spacer18-tcccgcgaaattaatacgac-3′). The aqueous reaction mixture was then combined with 600 μl of an oil/emulsifier mixture of 7% w/v ABIL WE09 (Degussa, Essen, GE), 20% v/v M3516 mineral oil (Sigma-Aldrich, St Louis, MO) and 73% v/v Tegosoft DEC (Degussa), and one 5 mm steel bead (Qiagen) in one well of a 96-deep-well plate (Abgene, Lafayette, CO). The plate was shaken in a TissueLyser (Qiagen) for 10 seconds at 15 Hz and then 7 seconds at 17 Hz, and 80 μl of the emulsions were distributed into each well of a MicroAmp 96-well PCR plate (PE Biosystems, Carslbad, CA). The emulsion PCR was done in a Veriti Thermal Cycler set to a 20% temperature ramp rate to prevent disruption of the emulsions, with the following cycling conditions: 94 °C for 3 minutes; 12 cycles of 94 °C for 10 seconds, 65 °C for 45 seconds (−0.5 °C/cycle), 70 °C for 75 seconds; 50 cycles of 94 °C for 10 seconds, 59 °C for 45 seconds, 70 °C for 75 seconds; 1 cycle of 70 °C for 5 minutes. We broke the emulsions by adding 150 μl breaking buffer (10 mmol/l Tris–HCl, pH 7.5, 1% Triton X-100, 1% SDS, 100 mmol/l NaCl, 1 mmol/l EDTA) to each well and mixing the samples with a TissueLyser (Qiagen) at 20 Hz for 30 seconds. The samples were then centrifuged at 3,200g for 2 minutes and the oil phase was removed. We performed this breaking procedure twice and then combined and resuspended all of the beads for each respective sample in 100 μl wash buffer (20 mmol/l Tris–HCl, pH 8.4, 50 mmol/l KCl). The beads were incubated with 0.1 mol/l NaOH to remove non-biotinylated DNA, and then washed and resuspended in 100 μl wash buffer.

Mutational analysis. We interrogated the mutational status of the bead-bound DNA using three fluorescent oligonucleotides: an Alexa Fluor 488 (AF488)–labeled oligonucleotide (5′-AF488-ATAAGCATGACGACCTATGAT-3′) specific to the wild-type IDH1 sequence, an Alexa Fluor 647 (AF647)–labeled oligonucleotide (5′-AF647-ATAAGCATGATGACCTATGAT-3′) specific to the A395 IDH1 variant, and a PB-labeled control oligonucleotide (5′-PB-TTTTACCCATCCACTCACAA-3′) specific to a sequence region shared by both wild-type and mutant IDH1 variants. The oligohybridization was in a 100 μl hybridization mixture of 30 μl beads, 66 μl hybridization buffer (3 mol/l tetramethylammonium chloride (Sigma-Aldrich), 50 mmol/l Tris–HCl, pH 7.5, 4 mmol/l EDTA) and 4 μl of a mixture of the oligonucleotides, each at 5 μmol/l. Samples were heated to 70 °C for 10 seconds, cooled to 35 °C at 0.1 °C/second, incubated at 35 °C for 2 minutes, and finally cooled to 25 °C at 0.1 °C/second using a Veriti Thermal Cycler. The beads were collected, washed in 100 μl hybridization buffer at 48 °C for 5 minutes, and then 100 μl wash buffer at room temperature, before a final resuspension in 400 μl wash buffer. We analyzed all oligohybridized beads using an LSR II flow cytometer (BD Biosciences) and processed the data using FACSDiva software (BD Biosciences). Events were gated to exclude doublets, aggregates, and beads without IDH1 amplicons. At least 100,000 events were counted and the mutant signal was calculated using events that had either wild-type or mutant fluorescence, but not both.

Digital droplet PCR methods.

Reaction preparation: Pre-amplified PCR product was diluted and added to the reaction mixture. Primers and fluorescent Taqman MGB probes (FAM-labeled G395A-5′- ATAAGCATGATGACCTATGAT-3′ and VIC-labeled WT- ATAAGCATGACGACCTATGAT-3′) were designed using Applied Biosystems' Custom TaqMan Assay Design Tool. The final IDH1 assay mix contained 900 µmol/l forward and reverse primers, 200 nmol/l VIC-labeled WT IDH1 probe, and 200 nmol/l FAM-labeled G395A probe. The PCR reaction was prepared in 25 µl final volume, using 12.5 µl TaqMan Genotyping Master Mix (Life Technologies), 2.5 µl Droplet Stabilizer (RainDance Technologies, Lexington, MA), 2.5 µl Assay Mix (10×) and 7.5 µl cDNA template in water.

Digital droplet assay: Droplets containing PCR reaction components were generated using a hydrodynamic flow-focusing microfluidic chip (RainDrop Source chip; RainDance Technologies), deposited into PCR tubes as 5 pl aqueous droplets suspended in inert fluorinated oil (REB Carrier Oil; RainDance Technologies), and subjected to PCR amplification in a thermal cycler (Mastercycler pros; Eppendorf, Hamburg, Germany). PCR cycling parameters: 10 minutes 95 °C, then 44 cycles of 95 °C for 15 seconds and 60 °C for 1 minute. After PCR completion, the emulsion was injected into a second microfluidic chip (RainDrop Sense chip; RainDance Technologies) for fluorescence measurement. Following 488 nm excitation, droplet fluorescence was detected through filters (FAM and VIC centered emission) with photomultiplier tubes recording fluorescence intensity (“height”) and duration (“width”). Spectral crosstalk-corrected data from each sample or control were converted to a two-dimensional (FAM and VIC intensity) histogram (electronic and photonic noise was removed, droplet data were width filtered and normalized to background fluorescence). Custom software (RainDrop Analyst; RainDance Technologies) was used to define graphical areas or “gates” associated with each allele type and count the number of droplets within each gate.

SUPPLEMENTARY MATERIAL Figure S1. RNA profiles from CSF and serum EVs.

Supplementary Material