Towards Stratification of Lymphomas with Involvement of the Central Nervous System Using miRNA-Seq from Small Volumes of Cerebrospinal Fluid

Early diagnosis and therapy monitoring have improved overall survival of lymphoma patients. Biomarkers from body fluids are an active field of research for diagnosing and monitoring malignant diseases. MicroRNAs (miRNAs) are present in all body fluids and have already been described as biomarkers in lymphoma patients. However, only few studies addressed the impact of miRNAs as diagnostic tools for central nervous system (CNS) involvement of systemic lymphomas (CNS positive lymphomas). In addition, due to technical limitations such as low cerebrospinal fluid (CSF) starting amounts, translation into clinical application has been very limited.

In order to facilitate development of an applicable diagnostic assay of circulating miRNAs from low CSF starting volumes (400ml), we compared quantitative realtime PCR (qRT-PCR) arrays with next generation sequencing (miRNA-Seq) and established a protocol to robustly quantify a miRNA signature for CNS positive lymphomas from diagnostic volumes of CSF. First, we could show that circulating miRNAs are highly stable in CSF and that miRNA levels can be assessed robustly with qRT-PCR and miRNA-Seq. For this purpose, miRNAs were extracted from 400µL of CSF using an optimized miRNA extraction protocol. MiRNAs were quantified by qRT-PCR with the Exiqon Human miRNome Panel v2 which assays 752 miRNAs or by miRNA-Seq of libraries prepared with the Seqmatic TailorMix v2 miRNA Sample Preparation Kit using the Illumina HiSeq.

In total, eight patients with diffuse large B-cell lymphoma (DLBCL) that showed involvement of the central nervous system (CNS) were analyzed. Of each patient, two paired CSF samples were available both from initial diagnosis and complete remission after treatment. The presence of CNS manifestation was established via magnetic resonance imaging (MRI) and flow cytometry at the Erasmus MC Daniel den Hoed Cancer Center in Rotterdam. CSF from ten patients with cephalalgia was used as a control.

After inter-plate calibration and normalization, 260 miRNAs were detected on average with the Exiqon Human miRNome Panel v2. In order to assess the probability of deriving false positive signals from e.g. primer dimers, two miRNome Panels (four 384-well plates) were processed without template. Surprisingly, after 40 cycles of qRT-PCR, 10% of assays showed amplicons in duplicates. Furthermore, false positive amplicons were not reproducible using this method, underlining that qRT-PCR had a low overall specificity that could not be adjusted by exclusion of erroneous amplicons.

In contrast, miRNA seq produced a median of ~40,000 mappable reads per sample (range 2x10⁴ - 1x10⁶) with a Phred score of >20. Prediction analysis using all detectable miRNAs segregated CSF samples from CSF positive lymphoma patients before treatment and after treatment based on a 14-miRNA signature with a misclassification error of <0.1. Unsupervised clustering of the 14-miRNA signature correctly stratified 7/8 samples after treatment i.e. CSF negative lymphomas, while 4/5 samples from patients with CSF positive lymphomas clustered separately. Thus, using the expression levels of only 14 miRNAs determined with miRNA-Seq, the two disease stages could be differentiated with each group containing one misclassification.

The robust quantification of miRNAs from small volumes of CSF with next generation sequencing has potential to monitor minimal residual disease diagnosis or even to stratify other pathologic entities such as dementias or inflammatory diseases where only unreliable markers or no biomarkers at all are currently available.