Cerebrospinal fluid liquid biopsy for detecting somatic mosaicism in brain

Abstract Brain somatic mutations are an increasingly recognized cause of epilepsy, brain malformations and autism spectrum disorders and may be a hidden cause of other neurodevelopmental and neurodegenerative disorders. At present, brain mosaicism can be detected only in the rare situations of autopsy or brain biopsy. Liquid biopsy using cell-free DNA derived from cerebrospinal fluid has detected somatic mutations in malignant brain tumours. Here, we asked if cerebrospinal fluid liquid biopsy can be used to detect somatic mosaicism in non-malignant brain diseases. First, we reliably quantified cerebrospinal fluid cell-free DNA in 28 patients with focal epilepsy and 28 controls using droplet digital PCR. Then, in three patients we identified somatic mutations in cerebrospinal fluid: in one patient with subcortical band heterotopia the LIS1 p. Lys64* variant at 9.4% frequency; in a second patient with focal cortical dysplasia the TSC1 p. Phe581His*6 variant at 7.8% frequency; and in a third patient with ganglioglioma the BRAF p. Val600Glu variant at 3.2% frequency. To determine if cerebrospinal fluid cell-free DNA was brain-derived, whole-genome bisulphite sequencing was performed and brain-specific DNA methylation patterns were found to be significantly enriched (P = 0.03). Our proof of principle study shows that cerebrospinal fluid liquid biopsy is valuable in investigating mosaic neurological disorders where brain tissue is unavailable.


Introduction
Cell-free DNA is externalized, fragmented DNA found in bodily fluids and is the product of programmed cell death. Fragments are predominantly short with a predominant peak at $167 base pairs (bp) (Chandrananda et al., 2015). Analysis of cell-free DNA, typically from peripheral blood, is now clinically used as a liquid biopsy for diagnosis and monitoring of cancer (Crowley et al., 2013), obstetric practice (Bianchi and Chiu, 2018) and as a biomarker for allograft rejection after organ transplantation (Goh et al., 2017). Interrogation of CSF-derived cell-free DNA allows detection of mosaic somatic mutations in malignant brain tumours (Pan et al., 2015;Wang et al., 2015;Garcia-Romero et al., 2019), but its application to diagnosis of non-malignant brain diseases has not been explored.
While a significant proportion of patients with neurological and neurodevelopmental disorders have disease-Graphical Abstract causing germline mutations (Allen et al., 2013;Iossifov et al., 2014), many remain unsolved after routine genetic testing. Recent studies indicate that some of these unsolved cases may be explained by somatic mutation (Thomas and Berkovic, 2014). Mosaic somatic mutations have been identified in patients with neurological disorders including brain malformations and epilepsy (Poduri et al., 2012;D'Gama et al., 2017;Winawer et al., 2018;Sim et al., 2019;Ye et al., 2019) and autism spectrum disorder (D'Gama et al., 2015). Somatic mutations occur post-zygotically and, if confined to brain, are difficult or impossible to detect by conventional sequence analysis of lymphocyte-derived DNA (Poduri et al., 2013). At present, the only established route to discovery of somatic mutations in brain is via the privileged situation of having brain tissue from surgical or autopsy specimens; this is not generally applicable to common epilepsies and other neurological and neurodevelopmental disorders. An alternative route is needed to detect the brain-only somatic mutations when brain tissue is not available. Here, we provide proof of principle that CSF liquid biopsies can be used as a surrogate to detect somatic mosaicism, without the need for brain tissue.

Study approval
The Human Research Ethics Committees of The Royal Children's Hospital (Project No. 29077F) and Austin Health, Melbourne, Australia (Project No. H2007/02961) approved this study. Written informed consent was obtained from all participants or their parents or legal guardians according to the Declaration of Helsinki.

Clinical cohorts
For initial CSF cell-free DNA quantitation, we studied 28 patients with epilepsy (mean age, 7.8 6 5 years old) and 28 controls without epilepsy (mean age, 46.9 6 15.6 years old). They were recruited through Austin Health, Royal Children's Hospital, and Queensland Children's Hospital, Australia. All patients had drug-resistant, lesional, focal epilepsy requiring surgical treatment (Supplementary  Table 1). Controls had been referred for non-epilepsy presentations (e.g. persistent headache). CSF was collected in Cell-Free DNA BCT V R (Strek, La Vista, NE) tubes via dural puncture immediately after craniotomy (epilepsy patients) or lumbar puncture (controls). Mean CSF volume collected was 2 ml (range, 0.25-10.5 ml).
For molecular diagnosis, CSF samples were obtained by lumbar puncture (patient 1, separate from the 28 patients used for the quantitation) and peri-operative dural puncture (patients 2 and 3, from the cohort of 28 patients used for quantitation) in patients with known or suspected brain mosaic mutations.

DNA extraction
Cell-free DNA was extracted from CSF using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany): CSF was centrifuged at 1000 Â g at 4 C for 10 min then extracted according to the manufacturer's protocol. For blood and brain, genomic DNA was extracted using standard methods (Qiagen QIAamp DNA Maxi Kit or AllPrep DNA/RNA Kit, Hilden, Germany).

Droplet digital PCR
For quantitation of total cell-free DNA in CSF, we used a PrimerPCR TM Mutation Assay (Catalog: 10049047; Bio-Rad, Hercules, CA) that we previously optimized inhouse (Hildebrand et al., 2018) to detect the GNAQ c.548G wild-type allele (NM_002072.5:c.548G) as a reference for cell-free DNA copies (Supplementary Table 2). Total cell-free DNA copies and cell-free DNA concentration (copies/ml CSF and ng/ml CSF) were calculated using reported formulas (Supplementary Methods).

Whole genome bisulphite sequencing and cell-of-origin analysis of cell-free DNA
To determine origin of CSF cell-free DNA, we examined cell-free DNA for brain-specific DNA methylation patterns. We analysed four control CSF samples and patient 1's CSF sample by whole-genome bisulphite sequencing (WGBS) followed by methylK analysis (Supplementary Methods). Sequencing libraries were prepared using Accel-NGS Methyl-Seq DNA Library Kit (Swift Biosciences, USA) according to the manufacturer's instructions for low-input methylomes. Pronex beads were used to deplete adapter dimers and enrich library fragments prior to sequencing. We sequenced 2-5 million paired-end reads for each CSF sample. All samples had 0.024-0.051x coverage for cytosines in the CG dinucleotide sequence context and 0.025-0.055x coverage for cytosines in the CH dinucleotide sequence context. In addition, we acquired five WGBS data sets from Swift Biosciences that were produced from peripheral blood plasma cell-free DNA in which the libraries were made with the same library preparation kit and sequenced to a comparable depth to CSF samples.
To identify the cell-of-origin of CSF cell-free DNA, we established an analytical method for assigning single WGBS reads. Briefly, publicly available WGBS from 11 cell types-of-interest [e.g. blood Cells (ENCODE) and central nervous system (CNS) cell types (Lister et al., 2013)] provided reference cell-type data sets from which DNA methylation fractional measurements were converted into cell-type specific reference sequences (FASTA). Sub-sequences (k-mers) were indexed from all cell-type references (n ¼ 11) using Kallisto software (Bray et al., 2016). Trimmed and truncated WGBS reads from CSF (n ¼ 5) and plasma (n ¼ 5) cell-free DNA were assigned (pseudoaligned) to their cell-of-origin by hash table lookup using Kallisto software (Bray et al., 2016) and only high-quality reads assigned to DNA methylation reference sequences of a unique cell-type were counted and used for statistical analysis (Supplementary Methods, Supplementary Fig. 4).

Statistical analysis
Student's t-tests were used to compare counts of WGBS reads assigned to CNS cells and blood cells between CSF cell-free DNA and plasma cell-free DNA. All statistical tests were performed using SPSS Statistics (SPSS 25, SPSS Inc., Chicago, IL, USA). A P-value of <0.05 was considered statistically significant.

Data availability
The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary Material. Raw data of this study are available from the corresponding authors, upon reasonable request.

Results
First, to establish that CSF cell-free DNA can be reliably quantified, we studied 28 patients with focal epilepsy and 28 controls. We quantified cell-free DNA in CSF samples by interrogating the GNAQ c.548G wild-type allele as a reference for cell-free DNA copies using ddPCR. Cell-free DNA copies were successfully detected and reliably quantified in all patient CSF samples (examples of ddPCR output are shown in Supplementary Fig. 5). The median value of cell-free DNA concentration was 502 copies/ml CSF ($1.5 ng/ml CSF) in paediatric patients with epilepsy and 61 copies/ml CSF ($0.18 ng/ml CSF) in adult controls, respectively (Supplementary Table 3 and Fig. 6). This demonstrated that cell-free DNA could be reliably detected in CSF from subjects with epilepsy and controls.
Next, to establish molecular diagnosis, we investigated CSF cell-free DNA from three patients. Two patients had a known mosaic pathogenic variant in brain tissue which was also found in CSF cell-free DNA. In the third patient, the mutation was first identified by CSF cell-free DNA analysis and later confirmed in brain tissue.
Patient 2 was a 3-year-old girl with a solitary, left inferior temporal gyrus focal cortical dysplasia type IIB, but no other clinical or imaging features of tuberous sclerosis. She had the brain-only somatic TSC1 mutation c.1741_1742delTT (p.Phe581His*6) detected at 2.8% VAF by deep sequencing in the resected lesion. We obtained 1.5 ml of CSF via dural puncture prior to resective surgery and isolated cell-free DNA (789 copies/ml). The TSC1 p. Phe581His*6 mutation was detected at 7.8% VAF in CSF; using the same ddPCR assay, we reconfirmed this mutation in brain tissue (2.3% VAF), and it was undetectable in blood (  Patient 3 was a 2-year-old boy with a left medial temporal ganglioglioma who had received no molecular testing. We obtained 3 ml of CSF via dural puncture prior to resective surgery and isolated cell-free DNA (1058 copies/ ml). Since 50-60% of patients with ganglioglioma have a brain-only recurrent somatic BRAF mutation c.1799T>A (p.Val600Glu) (Koelsche et al., 2013;Berghoff and Preusser, 2014), we screened for this mutation using a targeted ddPCR assay. We detected this mutation at 3.2% VAF in CSF cell-free DNA; subsequent testing confirmed that the mutation was present at 20.4% VAF in his ganglioglioma, and undetectable in his blood (Table 1, Fig. 1C and Supplementary Table 4 and Fig.  7C).
Finally, to confirm that cell-free DNA in CSF is brainderived, we examined cell-free DNA for brain-specific DNA methylation patterns. We showed for CSF cell-free DNA that a greater proportion of WGBS reads pseudoaligned to brain cells than blood cells ( Fig. 2A and Supplementary Fig. 4). We assessed brain cell-derived (NeuNþ neurons, NeuN-glia) and blood cell-derived (CD34þ myeloid progenitors, T-cells, B-cells, Natural Killer cells and CD14þ monocytes) contributions to cellfree DNA within CSF and plasma. We observed that CSF-derived cell-free DNA was significantly (P ¼ 0.03) enriched for unique reads pseudo-aligned to brain cells, whereas plasma-derived cell-free DNA had a significantly (P ¼ 0.04) higher number of unique reads pseudo-aligned to blood cells (Fig. 2B). The small fraction of NeuN kmer within plasma likely reflects a lack of specificity in established methods for deconvolution of CpH methylation in the pseudo-aligned reads from blood genomes. Our findings indicate that the cell-free DNA in CSF is largely brain derived.

Discussion
Somatic mutations have been well studied in malignant tumours, but their role in neurological and neurodevelopmental disorders remains poorly understood. With the improving throughput and depth of genetic screening technologies, the contribution of somatic mutations is increasingly being recognized.
There are many unsolved brain diseases for which a genetic aetiology has been invoked. In epilepsy, however, only a minority of patients have a known molecular cause. By study of privileged brain tissue samples, usually from epilepsy surgery, some cases, including but not limited to brain malformations, are now known to be due to somatic mutation in the brain (Poduri et al., 2012;D'Gama et al., 2015D'Gama et al., , 2017D'Gama and Walsh, 2018;Winawer et al., 2018). The discovery of these mutations in post-mortem tissue in individuals with autism spectrum disorder (D'Gama et al., 2015) and Alzheimer's disease (Park et al., 2019) implicates this mechanism more broadly in neurodevelopmental (D'Gama and Walsh, 2018) and neurodegenerative disorders (Lodato and Walsh, 2019). It has even been repeatedly hypothesized that somatic mutations contribute to the pathogenesis of Parkinson disease (Engeholm and Gasser, 2013;Proukakis et al., 2013). Analysing brain tissue in living patients with autism spectrum disorder, Alzheimer's disease or Parkinson disease is not feasible as these patients do not have surgery (except deep-brain stimulation in Figure 2 Enrichment of CNS-derived cell-free DNA in CSF samples. (A) Unique pseudo-alignment of cell-free DNA from CSF and plasma to reference sequences/k-mers from 11 tissues. (B) Unique sequence pseudo-alignments of cell-free DNA from CSF and plasma to brain and blood cells. CSF-derived cell-free DNA has significantly higher number of unique reads pseudoaligned to brain compared to plasma-derived cell-free DNA (P ¼ 0.03), whereas plasma-derived cell-free DNA was significantly enriched for unique reads pseudo-aligned to blood cells (P ¼ 0.04). rare cases where tissue is not resected) as part of their clinical treatment.
Our results show that there is enough cell-free DNA in CSF to make a targeted liquid biopsy possible. We found cell-free DNA concentration is higher in the patients with epilepsy compared to the controls. This might be because seizures can trigger an increase in apoptosis of brain cells, leading to shedding of more cell-free DNA into CSF. But there are multiple confounding factors that may contribute to the difference: (i) age: the patients were children while the controls were adults and apoptosis is elevated in the brain during development; (ii) CSF collection: patient CSF was collected via dural puncture whereas control CSF was collected via lumbar puncture. It would have been ideal to obtain CSF via dural puncture from some controls, but we did not have institutional ethics approval for this.
Our findings suggest that CSF liquid biopsy is a viable surrogate for brain tissue to establish molecular diagnosis of brain somatic mosaicism, with implications for clinical care. These findings may provide new insights into the molecular genetic factors behind epilepsy and brain malformations, as well as other neurodevelopmental and neurodegenerative diseases potentially linked to brain somatic mutations where neurosurgery is not standard of care.
There are three main challenges to clinical implementation of CSF liquid biopsy for diagnostic purposes. First, the technical issue of obtaining high-quality DNA from CSF. We have shown that the level of CNS-derived cellfree DNA in the CSF of epilepsy patients is sufficient to enable targeted molecular interrogation of somatic mutations. Second, the biological issue of detection of low concentrations of mutant DNA in CSF, especially if the mutation is presumed localized to a small lesion; detection of mutations in patients 2 and 3 shows that this challenge may be overcome. Third, we performed only targeted mutational analysis in three patients to establish the proof of principle of this approach. The remaining patients were not tested due to lack of a candidate variant from tissue analysis for targeted assays. In the future, unbiased, high depth screening, with optimized low-input genomic capture for CSF cfDNA, will be required to identify a novel somatic mutation genome-wide in establishing broader clinical utility.

Conclusion
As precision treatments for neurogenetic disorders are being investigated, such as mTOR inhibitors for cortical dysplasia, implementing CSF liquid biopsy for aetiological diagnosis will become increasingly important as a surrogate for brain-derived DNA to diagnose brain-specific somatic disorders.

Supplementary material
Supplementary material is available at Brain Communications online. Bionomics Inc., and licensed to various diagnostic companies. The remaining authors declare no competing interests.