Post-mortem ventricular cerebrospinal fluid cell-free-mtDNA in neurodegenerative disease

Cell-free mitochondrial DNA (cfmtDNA) is detectable in almost all human body fluids and has been associated with the onset and progression of several complex traits. In-life assessments indicate that reduced cfmtDNA is a feature of neurodegenerative diseases such as Parkinson’s disease, Alzheimer’s disease and multiple sclerosis. However, whether this feature is conserved across all neurodegenerative diseases and how it relates to the neurodegenerative processes remains unclear. In this study, we assessed the levels of ventricular cerebrospinal fluid-cfmtDNA (vCSF-cfmtDNA) in a diverse group of neurodegenerative diseases (NDDs) to determine if the in-life observations of reduced cfmtDNA seen in lumbar CSF translated to the post-mortem ventricular CSF. To investigate further, we compared vCSF-cfmtDNA levels to known protein markers of neurodegeneration, synaptic vesicles and mitochondrial integrity. Our data indicate that reduced vCSF-cfmtDNA is a feature specific to Parkinson’s and appears consistent throughout the disease course. Interestingly, we observed increased vCSF-cfmtDNA in the more neuropathologically severe NDD cases, but no association to protein markers of neurodegeneration, suggesting that vCSF-cfmtDNA release is more complex than mere cellular debris produced following neuronal death. We conclude that vCSF-cfmtDNA is reduced in PD, but not other NDDs, and appears to correlate to pathology. Although its utility as a prognostic biomarker is limited, our data indicate that higher levels of vCSF-cfmtDNA is associated with more severe clinical presentations; suggesting that it is associated with the neurodegenerative process. However, as vCSF-cfmtDNA does not appear to correlate to established indicators of neurodegeneration or indeed indicators of mitochondrial mass, further work to elucidate its exact role is needed.


Results
vCSF-cfmtDNA is reduced in PD, but not other NDDs. Similar to our previous cell-free mtDNA (cfmtDNA) work on lumbar CSF in PD 37 , we observed a significant reduction of ventricular CSF cfmtDNA (vCSF-cfmtDNA) in PD cases compared to controls (Fig. 1a, Dunnett's p = 1.1 × 10 −4 ). However, we observed no significant difference in vCSF-cfmtDNA levels between any other NDD group and controls (Dunnett's p > 0.05, Fig. 1a) or when NDD patients were grouped and compared to controls (SFigure 1, TTEST p > 0.05). We also found no correlation between age of onset or disease duration and vCSF-cfmtDNA (r2 < 0.2, p < 0.05) (SFigures 2 and 3). We found no correlation between vCSF-cfmtDNA and postmortem delay when analysed as groups seperately or when combined. vCSF-cfmtDNA is increased in patients exhibiting neocortical pathology. To assess the links between vCSF-cfmtDNA and neurodegenerative disease, we next compared vCSF-cfmtDNA levels to NDD cases stratified by tau pathology Braak staging 40,41 . We observed a significant correlation between Braak stage (I to VI) and vCSF-cfmtDNA levels (p = 0.028, Pearson's correlation = 0.31). To simultaneously increase statistical power, and perform a focused regional analysis, we grouped our analysis into three neuropathological regions; early entorhinal (Braak I/II), a limbic (Braak III/IV), and a late neocortical (Braak V/VI) stage 40,41 . This stratification indicated patients with neocortical pathology showed significantly higher vCSF-cfmtDNA than patients with stage I/II pathology (Dunnett's p < 0.05, Fig. 1b). It should be noted that this effect appears to be driven by AD patients (31/33 or 94% of individuals with Braak V/VI were AD). vCSF-cfmtDNA is increased in patients with dementia. Around 80% of NDD patients (n = 65) were diagnosed with dementia before death. Similar to studies in sporadic Creutzfeldt-Jakob disease 42 and AD dementia 27 , we observed significantly higher vCSF-cfmtDNA levels in patients diagnosed with dementia when compared to non-demented cases (TTEST p = 0.024, Fig. 1c). Unlike the elevation of vCSF-cfmtDNA seen in Braak V/VI, where the subset of samples were largely AD cases, demented cases were a cross-section of the NDD cohort (38 AD, 11 DLB, 9 FTD/VD, 3 PD, 3 PSP and 1 MND), indicating that this result is somewhat independent of regional tau pathology.
We compared protein levels between each NDD and controls, revealing significant associations between 14-3-3 zeta protein and AD, DLB and FTD/VD cases, synaptophysin protein and PD, DLB, FTD/VD and PSP cases, TFAM protein and DLB and FTD/VD cases and 14-3-3 beta and TRKB proteins with PD cases (STable 1a). However, 95% confidence intervals showed great variability within disease groups which may be reflective of modest sample sizes. Thus, to improve statistical power, we combined all NDDs into one group and compared the protein levels to controls (STable 1b).
Although generally, vCSF markers of neurodegeneration were reduced in combined NDD cases compared to controls (STable 1b), only 14-3-3 zeta reached statistical significance (TTEST p = 0.011, Fig. 2a  www.nature.com/scientificreports/ Next, we compared synaptophysin, a synaptic vesicle marker 58 , between combined NDD cases and controls. Synaptophysin was significantly elevated in NDD cases compared to controls (TTEST p = 0.001) ( Fig. 2b and Stable 1b). Finally, we observed no significant association between SDHA or porin and NDD (STable 1b), although TFAM levels were elevated in NDD cases compared to controls (TTEST p = 0.006) ( Fig. 2c and STable 1b). Stratification of cases by the diagnosis of dementia revealed both a significant elevation of NSE (TTEST p = 0.038, Fig. 2d and STable 2a) and a significant reduction of 14-3-3 zeta in NDD cases with dementia when compared to NDD cases without dementia (TTEST p = 0.006, Fig. 2e and STable 2a). However, we observed no significant association between protein levels and Braak staging (STable 2b).
Although several proteins were associated with NDD and dementia, we found no significant correlations between vCSF-cfmtDNA and protein levels in either combined NDD cases or controls (Fig. 2f), when NDDs were separated (STable 3), or when individuals were grouped by dementia status (STable 3). www.nature.com/scientificreports/

Electron microscopy of the vCSF indicates extracellular mitochondria-like structures. Recent
work has suggested that, in addition to cfmtDNA, intact mitochondria can be observed in extracellular fluids such as plasma 59 . To investigate this in vCSF, we performed transmission electron microscopy (TEM) of two vCSF-cfmtDNA samples with large volumes of vCSF available (> 500 µl). TEM indicated the presence of abundant proteinaceous material (Fig. 2g) and various small and larger vesicular structures, potentially microvesicles, (Fig. 2gi), but no cellular debris. Cell-free mitochondria-like structures could be detected with double membranes and remnants of cristae like morphology, however, these were scarce and showed gross morphological changes and partial membrane degradation ( Fig. 2gii-iii).
vCSF-cfmtDNA circulates as intact genomes, with low deletion and heteroplasmy levels. We next assessed the integrity of vCSF-cfmtDNA using both qPCR, to measure mtDNA deletion levels 28 , and nextgeneration-sequencing (NGS), to assess mtDNA mutational load 28 . Only 31 samples (27%), 23 (29%) NDD and 8 controls (23%), harboured mtDNA deletions above the established detection threshold of 10% 60 , suggesting that in the majority of samples (73%) vCSF-cfmtDNA is intact www.nature.com/scientificreports/ and full length. Although mtDNA deletion levels appeared lower in NDD cases compared to controls (Fig. 3a) this did not reach statistical significance (TTEST p > 0.05). Next, we performed NGS of vCSF-cfmtDNA from 72 samples (51 cases, or 71%, and 21 controls, 29%). Although we were able to detect heteroplasmic variants (average = 23.2 in NDD and 22.3 in controls), we observed no significant difference between NDDs and controls (Dunnett's > 0.05) when analysed as total heteroplasmic mutation rate (Fig. 3b) or when comparing non-synonymous heteroplasmic (i.e. putatively functional) variation (Fig. 3c). Combining NDD cases, to improve statistical power, and comparing mutation rates to controls also failed to identify a significant difference (TTEST p > 0.05, SFigure 4a and b). In previous work 28 , we had observed an imbalance in the Ka/Ks ratio (an indication of variant selective bias) of vCSF-cfmtDNA between progressive MS patients and controls However, we observed no significant difference in the Ka/Ks ratio of vCSF-cfmtDNA in NDDs compared to controls when analysed individually (Fig. 3d, Dunnett's > 0.05) or when grouped as NDD disease (SFigure 4c TTEST p > 0.05).
Our results indicate that, in line with our earlier observations in lumbar CSF 25,61 , vCSF-cfmtDNA is reduced in PD compared to controls, and this phenomenon appears to be specific to PD cases, rather than a feature of other NDDs. However, our data suggest that the relationship between vCSF-cfmtDNA and NDD may be more complex, possibly reflecting the diverse and multifaceted underlying pathoetiology of each disease. Intriguingly, a subsequent analysis indicates that elevated vCSF-cfmtDNA is associated with the degree of neuropathology, with higher levels of vCSF-cfmtDNA associated with late neocortical tau aggregations (Braak stage 5/6) and dementia.
In PD, the reduced vCSF-cfmtDNA we observed suggests that reduced cfmtDNA persists throughout the disease course with post-mortem vCSF-cfmtDNA levels comparable to those we observed in early-stage PD lumbar CSF 25,37 . How this relates to the neurodegenerative process in PD remains still unclear as we found no correlation between vCSF-cfmtDNA and neurodegenerative protein markers. However, the answers may reside in the relationship between cfmtDNA and neuronal mtDNA copy number. Previous work has shown that neuronal mtDNA copy-number is reduced in PD 24,62 , particularly in dopaminergic neurons 63 , and is correlated to www.nature.com/scientificreports/ mitochondrial dysfunction 62 . Thus, the reduced vCSF-cfmtDNA in PD may reflect an already depleted cellular mtDNA pool, with PD simply having less mtDNA to release. Further, reduced vCSF-cfmtDNA in PD may be a reflection of the treatment that PD patients have received, as our recent analysis indicates that reduced vCSF-cfmtDNA in PD correlates to levodopa effective daily dose 25 . Thus, as treatment in the latter stages of PD typically moves from symptom relief and prevention of motor symptoms (i.e. L-dopa) towards non-motor treatments and palliative care 64 , it is possible that the reduction of vCSF-cfmtDNA occurs early in the disease course and, for the majority of PD cases, remains reduced. However, this can only be tested in a longitudinal study that allows in-life lumbar CSF-cfmtDNA levels to be correlated with post-mortem vCSF-cfmtDNA levels.
Unlike PD, we did not observe a significant reduction in vCSF-cfmtDNA in the other NDDs. Previous literature 26 suggests that 'probable' AD cases (defined as those with low CSF amyloid beta 1-42 and elevated tau levels) and high-risk for AD individuals (defined as asymptomatic but with low CSF amyloid beta 1-42 ) exhibit significantly lower lumbar CSF cfmtDNA than controls. Conversely, a study of well-characterised AD patients reports significantly higher lumbar CSF cfmtDNA level than that of cognitively healthy controls 27 .
In our study, AD vCSF-cfmtDNA levels were not significantly different from matched controls. However, we did observe increased vCSF-cfmtDNA in NDD cases who exhibited extensive neocortical involvement (Braak V/VI 40,41 when compared to Braak stages I/II) and individuals who had progressed to dementia. These results are likely self-confirmatory as AD cases made up 93% and 59% of all cases from both groups respectively (31/33 Braak stage V/VI cases and 38/65 dementia cases). When taken together, this suggests that in AD cfmtDNA levels may rise with disease severity and neuropathology, beginning at markedly low levels in pre-clinical stages and rising to exceed that of controls when dementia becomes severely debilitating and there is widespread neocortical pathology. However, longitudinal studies in AD would be needed to confirm this.
Elevated cfmtDNA has been observed in patients with traumatic brain injury 65 , relapsing-remitting multiple sclerosis 14 and Anti-NMDAR Encephalitis 66 , suggesting that elevated cfmtDNA may be a component of the inflammatory response 14,65,66 . The innate immune system is involved in the pathoetiology of many NDDs 67 and can be activated by mtDNA. Circulating cfmtDNA is a damage-associated molecular pattern (DAMP) molecule that can trigger the antimicrobial and inflammatory response 68 after shock, injury 68 , infection 69 and cancer 70 . Thus, given the increased neuroinflammation observed in NDDs 71 , our data suggest that the increase in vCSF-cfmtDNA we observed in chronic NDD may be a component of this response and may be a component of the pathogenesis of NDD and future work should attempt to link cfmtDNA to CSF markers of neuroinflammation.
To investigate the links between cfmtDNA and neurodegeneration further, we compared vCSF-cfmtDNA levels to the levels of reported CSF protein markers of NDD. Similar to others 72 , we hypothesised that vCSF-cfmtDNA levels would correlate to indicators of neuronal cell death, which may, in turn, explain the origins of vCSF-cfmtDNA. However, although we observed associations between protein levels and disease, we did not observe associations between these proteins and vCSF-cfmtDNA. This suggests that vCSF-cfmtDNA levels are not wholly coupled to apoptosis or derived from cellular debris. It must be noted, however, that we chose these proteins based on a priori assumption regarding their role in disease pathogenesis or pathology. Thus, further work should utilise an unbiased proteomic approach (i.e. mass spectrometry), potentially establishing novel protein interactions and further developing the understanding of underlying mechanisms involved in this biological phenomenon.
Previous work in serum indicates that cfmtDNA may be derived from circulating intact mitochondria 59 and work in C. Elegans identified mitochondria in extracellular vesicles extruded under neurotoxic stress 73 . Similarly, we were able to detect the presence of cell-free mitochondria-like structures within vCSF. However, unlike the 'in-life' observations of intact and functionally competent mitochondrial made by Al Amir Dache et al. (2020), the mitochondrial structures we identified in NDD vCSF were sparse, grossly degraded and lacking defined membranes and cristae shape. Given that mitochondrial dysfunction is a component of many NDDs 4,62,63 , mitochondrial expulsion may be linked to pre-existing dysfunction, but it is also possible that the degraded morphology is a result of the post-mortem nature of the vCSF. To investigate further, we measured the abundance of the mitochondrial proteins SHDA 55 , TFAM 56 and porin 57 . Although these mitochondrial markers were detectable in both NDD and controls they did not correlate to vCSF-cfmtDNA, further suggesting vCSF-cfmtDNA levels are independent of cell-free mitochondrion abundance. TEM of post-mortem CSF suggested the presence of extracellular vesicles within the vCSF, in line with similar reports 74,75 , which may point towards a possibility of an active release of cellular mitochondria to CSF via the extracellular vesicles route. Elevated CSF synaptophysin a synaptic vesicle marker, has been observed in patients with traumatic brain inury 75 , and its release into CSF is associated with neuronal cytotoxicity 76 . Similarly, in our study synaptophysin was significantly enriched in vCSF of NDDs, however, we found no correlation between vCSF-cfmtDNA and synaptophysin protein levels. Taken together, these data suggest that further work, specifically focusing on vCSF extracellular vesicles and quantifying their ccf-mtDNA content, could be vital in elucidating the origin and release of ccf-mtDNA.
MtDNA lesions and mutations are a hallmark of many NDDs 77 , with high levels of mtDNA deletions and somatic (heteroplasmic) mutations identified in disease-specific brain regions of PD 78,79 and AD 80 patients. Similar to previous work 81 , we used qPCR and NGS to interrogate the integrity of vCSF-cfmtDNA under the hypothesis that increased vCSF-cfmtDNA release linked to mtDNA maintenance mechanism, where mutated mtDNA is expelled from cells to maintain a 'wild-type' mtDNA pool. Our analysis indicates that although mtDNA deletions and point mutations were present in NDD vCSF-cfmtDNA, they were not significantly more frequent than in controls. Whilst this does not necessarily negate the hypothesis that vCSF-cfmtDNA is a mechanism for maintaining a healthy cellular mtDNA pool, our results do suggest that this mechanism is independent of disease status. However, it must be noted that a potential reason for this could be neuronal 'survivor bias'; where, due to the end-stage of disease, only the cells with low-level mtDNA defects remain. This is somewhat supported by brain mtDNA heteroplasmy analysis 82 , whereby post-mortem, end-stage NDDs and controls show similar brain mtDNA heteroplasmy. www.nature.com/scientificreports/ Our work is not without limitations. Firstly, the availability of post-mortem vCSF samples is limited and future work should expand the cohort sizes, particularly of non-AD NDDs. Secondly, although we were able to identify several protein-NDD associations, we were unable to link vCSF-cfmtDNA to established hallmarks of neurodegeneration. Thus, future work should take a more unbiased proteomics approach, for example using mass-spectrometry to identify all proteins with the vCSF. Finally, as discussed, neuroinflammation is demonstrated to be a key feature of many NDDs, most likely contributing to the pathogenesis of these disorders (as reviewed in 83 ). Interestingly, cfmtDNA is also implicated in the inflammatory process, and previous studies have shown direct interplay between levels of this inflammatory molecule and important inflammatory cytokines in both human plasma 84 and CSF 66 . This is a particularly crucial area for future study as it may reveal functional attributes of the cfmtDNA and its potential role in the mediation of the immune response in NDD.

conclusions
We conclude that vCSF-cfmtDNA is reduced in PD, but not other NDDs. Although its utility as a prognostic biomarker is severely limited, our data indicate that higher levels of vCSF-cfmtDNA are associated with more severe clinical presentations; suggesting that it plays a role in the neurogenerative process. However, as vCSF-cfmtDNA does not appear to correlate to established indicators of neurodegeneration or indeed indicators of mitochondrial mass, further work to elucidate its exact role is needed.

Methodology
Ventricular CSF sample cohort. We used 115 ventricular cerebrospinal fluid (vCSF) samples, obtained from the Newcastle Brain Tissue Resource (NBTR) and the Imperial College Tissue Bank (80 Neurodegenerative disease cases: Alzheimer's disease (AD) = 40, dementia with Lewy bodies (DLB) = 11, progressive supranuclear palsy (PSP) = 5, motor neurone disease (MND) = 7, frontal temporal/vascular dementia (FTD/VD), Parkinson's disease (PD) = 8, and 35 controls). vCSF was collected post-mortem from hemisected brains and immediately stored at -80 °C in polypropylene cryovials, which do not influence biomarker outcome 85 . Samples were prepared via methods described previously 86,87 . Briefly, vCSF was centrifuged (2000 g for 10 min) at room temperature, aliquoted into CSF collection tubes and stored upright at − 80 °C. All cases were confirmed in life by a local clinician and re-assessed neuropathologically post-mortem. All controls were negative for hallmarks of neurodegeneration or inflammation.

VCSF transmission electron microscopy (TEM).
Transmission electron microscopy was performed on a subset of vCSF samples. Briefly, vCSF was pelleted, fixed, dehydrated and then embedded into Taab medium epoxy resin, as described previously 88 . After polymerisation, ultrathin Sects. (70 nm) were collected, stained with uranyl acetate and lead citrate and then imaged on a Philips CM100 Transmission Electron Microscope (TEM). vCSF-cfmtDNA sequencing and bioinformatic analysis. Deep sequencing was performed on vCSF-cfmtDNA extracts from a subset (where volume and mtDNA concentration permitted), of vCSF samples (53 NDD patients and 21 controls) as described previously 28,89 . Briefly, mtDNA was extracted from vCSF using Ultrapure™ and ethanol precipitation, as per manufacturers guidelines, and enriched using three amplicon long-range PCR, covering the entire mtDNA genome. PCR products were purified, pooled and prepared using Illumina Nextera XT DNA Library Preparation Kit. Sequencing was performed using the Illumina Miseq v3.0 platform in paired-end, 250 bp reads. Post-run FASTQ files were analysed using an established in-house bioinformatics pipeline, as described previously 28 . Statistical analysis. Normality of vCSF-cfmtDNA distributions were assessed by Shapiro-Wilks and could not be rejected at the 0.05 level. Thus, all vCSF-cfmtDNA levels are expressed as log [10] copy-number per microliter. Data were analysed in R (v4.0) 90 using data appropriate tests (detailed in the text). Statistical significance was set at p < 0.05. All tests were two-tailed with α = 0.05. Correlograms of protein levels were created using corrplot (v0.84) in R. Ethics approval. Donors or next of kin provided informed consent to donate tissue and all procedures were approved by the local UK National Health Service Research Ethics Committee. All procedures were approved by each Local Research Ethics Committee and appropriate informed consent was obtained from donors or next of kin for sample donation. Newcastle Brain Tissue Resource has been approved as a Research Tissue Bank by

Data availability
All data, genetic and proteomic, is available upon request.