Evaluation of putative CSF biomarkers in paediatric spinal muscular atrophy (SMA) patients before and during treatment with nusinersen

Abstract Spinal muscular atrophy (SMA) is a genetic neurodegenerative disorder leading to immobilization and premature death. Currently, three alternative therapeutic options are available. Therefore, biomarkers that might reflect or predict the clinical course of the individual patient with treatment are of great potential use. Currently, the antisense oligonucleotide nusinersen is the prevalent and longest validated therapy for SMA. We analysed CSF candidate biomarkers for degenerative CNS processes (namely phosphorylated heavy chain (pNf‐H), light‐chain neurofilaments (NfL), total tau protein (T‐Tau), neurogranin, β‐secretase BACE‐1 and alpha‐synuclein) in 193 CSF samples of 44 paediatric SMA types 1, 2 and 3 patients before and under nusinersen treatment and related them to standardized clinical outcome scores in a single‐centre pilot study. pNf‐H and NfL correlated with disease severity and activity, emphasizing their relevance as marker of neuronal loss and clinical outcome. T‐Tau was significantly correlated with motor function scores in SMA type 1 making it an interesting marker for treatment response. Additionally, baseline T‐Tau levels were elevated in most SMA patients possibly reflecting the extension of neuronal degeneration in paediatric‐onset SMA. Further investigations of these CSF proteins might be beneficial for paediatric SMA subtypes and treatment modalities as an indicator for clinical outcome and should be analysed in larger cohorts.

dysfunction and later apoptosis of the lower motor neuron resulting in progressive muscular weakness, respiratory insufficiency and bulbar symptoms. However, the SMN1 gene is ubiquitously expressed and the relevance of SMN protein in other tissues is not fully understood. 2 The SMN2 gene as a nearly identical copy of SMN1 only produces about 10% of full-length SMN protein due to altered splicing. 3 The copy number of the SMN2 gene shows remarkable interindividual differences with later onset and milder course (SMA types 1-4) in SMA patients with more SMN2 copies. 4 Neuropathological findings in deceased patients with SMA type 1, the most severe SMA subtype, showed not only neuronal degeneration in lower motor neuron, but also in the cerebral cortex, basal ganglia, brain stem and cerebellum. 5 Nusinersen alters the splicing of SMN2 pre-mRNA increasing the concentration of functional SMN protein. Clinical trials have shown promising clinical efficacy in paediatric SMA patients compared to untreated patients. [6][7][8] To date, individual clinical improvement cannot be predicted by specific clinical or biochemical markers. The presence of biomarkers correlating with the efficacy of treatment or even predicting outcome or clinical success of a therapy is essential considering alternative therapeutic options, for example gene replacement therapy. 9 Until now, neurodegeneration in various diseases has been addressed by different biomarkers in plasma and/or cerebrospinal fluid (CSF) such as neurofilaments and Tau protein in neuronal injury or loss, neurogranin in synaptic loss and β-secretase BACE-1 in hypomyelination. [10][11][12][13] Furthermore, the accumulation of alpha-synuclein has been associated with the pathomechanisms in different neurodegenerative diseases (ie Parkinson's disease, multiple system atrophy). 13 In order to better understand neurodegeneration in SMA, some efforts to characterize the disease at baseline in clinical and biochemical aspects have been made: Kolb et al. found altered electrophysiological and plasma protein parameters and lower SMN mRNA levels in SMA patients than in the control group. However, the course of these parameters during disease progression was not determined. 14 Recently, elevated plasma pNf-H levels were correlated with disease severity and inversely correlated with the age at presentation in 117 nusinersen-treated SMA infants included in the ENDEAR study (NCT02193074). 15 During nusinersen treatment, pNf-H decreased faster and to lower values in the plasma compared to sham control-treated infants. 15 However, data of baseline CSF pNf-H levels, CSF NfL levels and their course during nusinersen treatment in paediatric SMA patients are still limited. 16 As the investigation of neurofilaments only addresses one aspect of neurodegeneration, the course of additional biomarkers before and during nusinersen treatment might be able to further characterize the mechanisms of damage in SMA and to predict clinical outcome under treatment. Therefore, we investigated a number of candidate biomarkers in CSF samples of patients with SMA types 1-3 at baseline and under treatment with nusinersen and aimed to find a specific profile of these proteins under treatment.
The aim of the study was to answer the questions 1. whether there are differences between the SMA subtypes, 2. whether nusinersen treatment affects the candidate biomarkers levels, 3. whether there is a correlation between clinical scores and candidate biomarkers levels and 4. whether candidate biomarkers levels predict clinical response.

| Patients and Study Samples
In the Department of Pediatrics of the University Medical Center Hamburg-Eppendorf, treatment with nusinersen was initiated within the early access programme (EAP) in January 2017 for patients with SMA type 1 Patients with SMA types 2 and 3 have been included since July 2017 after the medication was approved by the European Medicines Agency (EMA). All patients had a documented 5q-associated SMA. Written informed consent from the parents or guardians were obtained for the lumbar puncture; the intrathecal administration of nusinersen and the storage of CSF samples followed institutional guidelines. The biomarker investigation in CSF samples was approved by the local ethics committee (PV5865), and parents or guardians and children who were able to read provided written informed consent.
In accordance with the recommended dosing schedule, the in- The interval between (reported) symptom onset and first nusinersen treatment (=disease duration) was collected.

| Laboratory Analysis
The CSF samples were analysed with ELISAs commercialized The Neurofilament light (NfL) was still a prototype ELISA developed by EUROIMMUN, which is not commercially available yet.

| Statistical Analysis
Sociodemographic variables were reported with mean and standard deviation as well as absolute and relative frequencies depending on the scale level of the variables.
For determining the course of the motor function and the CSF candidate biomarkers, we used a linear mixed model in each case with the changes from baseline as the dependent variable, time point, age (in respect to candidate biomarkers), SMA type, baseline value of the respective dependent variable and interaction between time point and SMA type as fixed effects and patient as random effect. If the interaction was not significant, we eliminated it from the model.
That was the case for neurogranin, β-secretase BACE-1 and alphasynuclein, HFMSE and RULM. To assess the association between motor function and the biomarkers, we used a linear mixed model with motor function as dependent variable, biomarkers as fixed effect and patient as random effect. For every combination, we conducted a single model. To assess the association between the changes from baseline of motor function and the changes from baseline of biomarker levels, we calculated a linear regression in each case with motor function on day 180 and day 300, respectively, as the dependent variable, SMA type, baseline value of the respective dependent variable, changes from baseline of the preceding times and the interaction between SMA type and changes from baseline of the preceding times as fixed effects. Including an age adjustment in the model was not possible due to insufficient sample size. To evaluate the association between the changes from baseline of motor function and the biomarker levels at baseline, we calculated a linear mixed model with changes from baseline of motor function as the dependent variable, SMA type (in case of RULM and HFMSE), time point, baseline value of the respective dependent variable and the biomarker level at baseline as fixed effects and patient as random effect. To assess the association between motor function at baseline respectively biomarkers at baseline, and the duration of the disease, we evaluated a linear model with motor function, and biomarker, respectively, as dependent variable, SMA type, duration and the interaction between both variables as fixed effects. The model used for the biomarker included age as fixed effect. To assess the association between the changes from baseline of biomarker levels and the duration of the disease, we calculated a linear mixed model with changes from baseline as the dependent variable, SMA type, time point, duration, the baseline value of the respective biomarker and the interaction between SMA type and duration as fixed effects and patient as random effect. In models with CHOP INTEND as dependent variable, we omitted SMA type and possible interactions. Due to the exploratory character of our study, we did not correct for multiplicity. For all tests (each with a two-sided hypothesis), level of significance was set at p < 0.05. All analyses were conducted with SAS, version 9.4.

| Patients and Study Samples
In total, 193 CSF samples of 44 5q-associated SMA patients were obtained. Sixteen patients with SMA type 1 (male 5, female 11, average age at treatment start: 16.0 months (range: 2-68 months)), 16 patients with SMA type 2 (male 5, female 11, average age at F I G U R E 1 The course of motor function scores before (baseline) and with nusinersen treatment in SMA patients measured by CHOP INTEND, HFMSE and RULM. Time points are shown on x-axis and changes of scores from baseline on y-axis. SMA types are presented separately with different colours (blue: type 1, orange: type 2; green: type 3) treatment start: 78.8 months (range 21-152 months)) and 12 patients with SMA type 3 (male 5, female 7, average age at treatment start: 119.9 months (range 38-208 months)) were initially included. Finally, 15 patients with SMA type 1, 15 patients with SMA type 2 and 10 patients with SMA type 3 were included in the statistical analysis determining the course of CSF proteins during nusinersen treatment. Samples of 1 SMA type 1 patient, 1 SMA type 2 patient and 2 girls with SMA type 3 had to be excluded because of a missing baseline (ie before treatment) sample.
Only patients with a valid baseline value and at least one valid follow-up value in our analysis population were included. Further, 7 samples had to be excluded from analyses because of blood contamination.
The initial nusinersen dose was applied in patients between the age of 9.0 weeks and 17 years and 8 weeks.
Clinical and demographic data are shown in Table 1.

| Motor function assessment
Motor function scores at baseline and their course during treatment are shown in Figure 1 and as mean total values and as values reflecting the change from baseline in Table 2a In SMA type 2 and 3 patients, motor function scores improved as well during treatment, but only upper limb function (measured by RULM) was significantly different between baseline and day 300 of treatment.

| Correlation of biomarkers with motor function scores
Since CHOP INTEND was performed in SMA type 1 patients and RULM and HFMSE was performed in SMA types 2 and 3, correlation of biomarkers and motor function scores had to be evaluated separately for type 1 and type 2/3, respectively. Results are shown in Table 2b. A significant inverse correlation between CHOP INTEND scores and Tau and pNf-H concentration could be found in SMA type 1 after treatment with nusinersen ( Figure 2). In SMA type 2 and 3 patients, there was a significant correlation of RULM scores and Neurogranin concentration as well as a significant inverse correlation of pNf-H-concentration with RULM and HFMSE scores and of NfL and HFMSE scores.

| Correlation between disease duration and CSF candidate biomarkers
Only in SMA type 1 patients, longer disease duration significantly correlated with lower baseline levels of T-Tau, NfL and p-NfH and lower changes of NfL levels from baseline to treatment day 63, 180 and 300. Additionally, longer disease duration was correlated with lower changes of alpha-synuclein concentrations from baseline to treatment day 63, 180 and 300 (p = 0.003). With age adjustment to decrease ageing effects on the correlation, the effect was still significant for the baseline levels of NfL in SMA type 1 (Table 3).

| CSF candidate biomarkers
The mean concentrations of putative CSF biomarkers in all statistically analysed samples and the adjusted mean of the concentration differences compared to baseline levels are reported in Table 4 and   Figure 3).

| Total Tau (T-Tau) protein
Highest baseline T-Tau levels were found in SMA type 1 with lower levels in types 2 and 3, but baseline levels did not differ significantly between groups. In both SMA type 1 and 2 patients, T-Tau levels significantly decreased from baseline to day 300 of treatment (Table 4; Figure 3).

| Neurogranin
Significantly higher baseline levels were found in SMA type 3 compared to SMA type 2 and type 1. Only in SMA type 3, a significant increase in concentrations from baseline to treatment day 63, 180 and 300, respectively, were observed with significant group differences between SMA type 3 and the other SMA types (Table 4; Figure 3).

| BACE-1
At baseline, mean concentrations were lowest in SMA type 1 patients, we measured higher levels in SMA type 2 patients and highest levels in SMA type 3 patients. There were no significant differences in baseline concentrations between the SMA types. The course of BACE-1 levels under nusinersen treatment in the different patient groups is shown in Figure 3. A relevant decrease in concentration (p = 0.056) was found from baseline to day 300 during treatment only in SMA type 1 (Table 4; Figure 3).

| DISCUSS ION
Spinal muscular atrophy is mainly associated with injury and loss of lower motor neurons, but there is also evidence for brain neuronal degeneration. 5 The significance of this observation is unclear but of increasing importance considering new therapeutic options and an increasing life span of treated SMA patients. Different CSF proteins have been proposed to reflect CNS neurodegeneration and have been used as biomarkers in various neurological diseases mainly in adults but sporadically in children. 17 In SMA patients, the spectrum of useful biomarkers for disease staging or monitoring disease progression is still under discussion. Therefore, we investigated a broad range of potential CSF proteins that might reflect the different proposed pathomechanisms in SMA and might serve as potential biomarkers in our pilot study. Overall, our data were most interesting for both neurofilaments and T-Tau.
Neurofilaments are uniquely expressed in neurons, and neuronal injury and/or degeneration (as eg in amyotrophic lateral sclerosis (ALS)) lead to elevated neurofilament levels in CSF and plasma. 10,18,19 In ALS patients, high concentrations have been found to correlate with increased disease severity and progression. 20 The course of pNf-H in plasma of nusinersen-treated SMA type 1 patients was investigated in the ENDEAR treatment study. 15 In the latter, higher baseline pNf-H levels were associated with disease severity and a more rapid decline in pNf-H levels was found in treated patients compared to those in the sham procedure control group. However, changes in motor function under nusinersen treatment were not reported.    Alzheimer's disease) but was also suggested to contribute to motor neuron degeneration in SMA. 28 Increased CSF T-Tau could be correlated with degeneration of cortical axons 29,30 and decreased CSF Tau levels were found during miglustat treatment in Niemann-Pick type C patients, a therapy known to slow down disease progression. 31 Recently, CSF T-Tau levels below 200 pg/ml were found in 11 adult SMA type 3 patients before and during loading with nusinersen. 22 However, in most of our SMA patients, baseline T-Tau levels were above the normal range for adults (reference <300 pg/ml), 32

F I G U R E 3
Age-adjusted mean concentration changes from baseline of CSF BACE-1, total Tau protein, alpha-synuclein, neurogranin, NfL, pNf-H before and with nusinersen treatment, separately shown for each SMA type (blue: type 1, orange: type 2, green: type 3). Time points are shown on x-axis and changes of protein levels from baseline values on y-axis median levels in healthy children 25 and in children with leukaemia (244.84 ± 98.96 pg/ml). 33 Tau levels dropped in all SMA patients during nusinersen treatment, but changes were only significant in SMA type 1 patients (after 2 months of treatment) and in SMA type 2 patients after 10 months of treatment. Statistical age adjustment should diminish the probability that the course of CSF T-Tau levels in our SMA type 1 patients simply reflect ageing in infancy.
Additionally, age-dependent changes have previously been reported merely in the first 3 months of life. 34 As baseline levels in our SMA patients were higher than normal, adult-level T-Tau protein might also reflect or contribute to the neuronal degeneration in earlyonset SMA. However, these findings have to be verified by comparing these data with a paediatric control group in additional studies.
In SMA type 1, higher motor function scores significantly correlated with lower T-Tau levels with nusinersen treatment making this parameter suitable as an outcome marker in severely affected patients.
Other investigated CSF proteins showed less encouraging results compared to neurofilaments and T-Tau.
We chose to investigate the post-synaptic protein neurogranin as its increase was linked to cognitive decline in patients with Alzheimer's disease possibly reflecting synaptic dysfunction and degeneration [35][36][37] Cognitive function in children and adolescents with all SMA types has been reported within the normal range, 38  Finally, we investigated the CSF ß-secretase 1 (BACE-1) as an increase of the protein has been suggested to reflect the intensity of axonal degeneration in adult patients with Alzheimer's disease. 46 Although we found a similar correlation between clinical outcome and course of BACE-1 concentrations after 6 months of treatment as aforementioned for alpha-synuclein, our overall data regarding BACE-1 provide no convincing evidence that this protein might reflect pathogenic mechanisms important for SMA.
In conclusion, our findings suggest that pNf-H and NfL are cor- dividual changes of the parameters and considering age adjustment in the statistics thereby addressing ageing effects within our study population. As nusinersen is supposed to slow down but not to stop disease progression, we cannot fully exclude that the latter leads to a decreasing amount of motor neurons contributing to the decline of some CSF proteins (NfL, p-NfH, T-Tau) over time but increasing motor function scores in our cohort contradicts this consideration.
We acknowledge the limitations of our data, due to the small sample size in our pilot study, especially in the group of SMA type 3 patients after 10 months of treatment. Moreover, lacking a paediatric control group provides another limitation considering we had to refer to normal values mentioned in the literature for the CSF proteins NfL, T-Tau and alpha-synuclein. Additionally, higher baseline levels of the investigated CSF proteins in SMA type I patients in the youngest age group might reflect an age-appropriate finding. Therefore, our data should be interpreted as explorative results of a single centre to encourage additional studies.

ACK N OWLED G EM ENTS
Not applicable.

CO N FLI C T S O F I NTE R E S T
The authors DW, LS, AD and JD declare that they have no competing

DATA AVA I L A B I L I T Y S TAT E M E N T
The data sets used and/or analysed during the current study are available from the corresponding author on reasonable request.