CSF chitinase proteins in amyotrophic lateral sclerosis

Objective To evaluate the classifier performance, clinical and biochemical correlations of cerebrospinal fluid (CSF) levels of the chitinase proteins Chitotriosidase-1 (CHIT1), Chitinase-3-like protein 1 (CHI3L1) and Chitinase-3-like protein 2 (CHI3L2) in amyotrophic lateral sclerosis (ALS). Methods CSF levels of CHIT1, CHI3L1, CHI3L2, phosphorylated neurofilament heavy chain (pNFH) and C-reactive protein were measured by ELISA in a longitudinal cohort of patients with ALS (n=82), primary lateral sclerosis (PLS, n=10), ALS-mimic conditions (n=12), healthy controls (n=25) and asymptomatic carriers of ALS-causing genetic mutations (AGC; n=5). Results CSF CHIT1, CHI3L1 and CHI3L2 were elevated in patients with ALS compared with healthy controls (p<0.001) and ALS-mimics (CHIT1, p<0.001; CHI3L1, p=0.017; CHI3L2, p<0.001). CHIT1 and CHI3L2 were elevated in ALS compared with PLS (CHIT1, p=0.021; CHI3L1, p=0.417; CHI3L2, p<0.001). Chitinase levels were similar in AGCs and healthy controls. Chitinase proteins distinguished ALS from healthy controls (area under the curve (AUC): CHIT1 0.92; CHI3L1 0.80; CHI3L2 0.90), mimics (AUC: CHIT1 0.84; CHI3L1 0.73; CHI3L2 0.88) and, to a lesser extent, PLS (AUC: CHIT 0.73; CHI3L1 0.51; CHI3L2 0.82) but did not outperform pNFH. CHIT1 and CHI3L2 correlated with disease progression rate (Pearson’s r=0.49, p<0.001; r=0.42, p<0.001, respectively). CHI3L1 correlated with degree of cognitive dysfunction (r=−0.25, p=0.038). All chitinases correlated with pNFH. CHIT1 levels were associated with survival in multivariate models. Chitinase levels were longitudinally stable. Conclusions CSF chitinase proteins may have limited value as independent diagnostic and stratification biomarkers in ALS, but offer a window into non-autonomous mechanisms of motor neuronal loss in ALS, specifically in assessing response to therapies targeting neuroinflammatory pathways.


INTrOduCTION
Amyotrophic lateral sclerosis (ALS), the most common form of motor neuron disease, is a fatal neurodegenerative disease resulting in progressive weakness and death due to loss of neurons of the motor cortex, brainstem and spinal cord. 1 Despite two licensed disease-modifying treatments, most patients survive less than 3 years from symptom onset, 2 but there is significant clinical and prognostic heterogeneity. Characteristic proteinaceous inclusions of the DNA and RNA binding protein TDP-43 are found in more than 95% of cases postmortem. 3 A significant proportion of ALS cases are associated with frontotemporal cognitive and behavioural dysfunction. Although most cases occur sporadically, up to 10% are associated with a family history of ALS or frontotemporal dementia (FTD). 4 5 The most common genetic cause of both ALS and FTD is a hexanucleotide repeat expansion (HRE) in an intronic region of the C9orf72 gene. 6 In common with other neurodegenerative diseases, glial pathology is observed in ALS. 7 Although the role of microglia and astrocytes in ALS is unresolved, whether causing, exacerbating or mitigating neurodegeneration, interest in modifying glial mechanisms as a therapeutic approach has led to clinical trials of drugs aiming to suppress microglial activity. Biomarkers quantifying the glial response in patients are, therefore, a priority to assess target engagement in therapeutic trials. Cerebrospinal fluid (CSF) levels of chitinase proteins, in particular Chitotriosidase-1 (CHIT1), Chitinase-3-like protein 1 (CHI3L1 or YKL-40) and Chitinase-3-like protein 2 (CHI3L2 or YKL-39) may be markers of activity of microglial (and possibly other glia, such as astrocytes) in ALS. [8][9][10] Both agnostic and target-driven studies of CSF have highlighted elevations of these three proteins in ALS as well as other neurodegenerative diseases including Alzheimer disease and FTD. 11 12 This study sought to compare levels of CSF chitinase proteins in patients with ALS, those with diseases that clinically mimic ALS, patients with the upper motor neuron-only slowly-progressive variant primary lateral sclerosis (PLS), healthy control subjects and asymptomatic carriers of the C9orf72 HRE. Diagnostic performance of chitinase proteins was examined in relation to CSF levels of one of the leading ALS diagnostic biomarker candidates, namely phosphorylated neurofilament heavy chain (pNFH). Chitinase proteins are also compared with another ALS biomarker candidate, CSF C reactive protein (CRP). This marker has been previously shown to be elevated in CSF from patients with ALS and, though not validated as a measure of intrathecal inflammation, likely reflects a combination of serum CRP level and blood-CSF barrier dysfunction. 13 14

Participants and sampling
Participants were recruited from a tertiary ALS referral clinic at the John Radcliffe Hospital, Oxford, UK. Patients were diagnosed with ALS and other motor disorders by two experienced neurologists (KT, MRT). Consecutively-consenting patients with ALS and PLS were recruited as well as patients with non-ALS causes of motor weakness in whom the diagnosis of ALS was considered a plausible differential diagnosis (designated ALS CSF samples were obtained by lumbar puncture directly into polypropylene collection tubes. Venous blood was obtained using the BD Vacutainer Safety-Lok blood collection set into 5.0 mL serum separator tubes. Biofluid samples were centrifuged at 3000 rpm for 10 min at 4°C within 1 hour of sampling and stored at −80°C until measurement. Samples were obtained at the first visit and then at intervals of 3-6 months. Patients with ALS and PLS were followed until death.

Neurodegeneration
Clinical parameters were obtained contemporaneously on the day of biofluid sampling. Symptom onset was defined as first weakness or relevant neurological symptom reported by patients. Disability was assessed using the revised ALS functional rating scale (ALSFRS-R) at each visit, 15 and the Edinburgh Cognitive and Behavioural ALS Screen (ECAS) at first visit. 16 Disability progression rate (PR) was calculated by dividing the gradient of descent of the ALSFRS-R from symptom onset to sampling. A clinical upper motor neuron (UMN) burden score was calculated by the presence of hyperreflexia in each of 15 reflexes obtained at the first visit. 17 Ethical approval for the study was obtained from NRES Central Committee South Central-Berkshire (14/SC/0083 and 10/H0505/71). All participants provided written consent (or gave permission for a carer to sign on their behalf).

biochemical assays
All ELISAs and activity assays were performed in duplicate. Plates were read using a FLUOstar Omega plate reader (BMG LABTECH, UK). Standard curves were fitted with 4-parameter logistic regression using MARS data analysis software (BMG LABTECH, UK). CSF samples were thawed on ice prior to measurement. Measurements of CSF and serum protein concentrations and activity were performed using commercially available assays according to the manufacturer's instructions (CircuLex human Chitotriosidase ELISA, CircuLex human YKL-39 ELISA, CycLex Chitotriosidase Fluorimetric Assay Kit, MBL, UK; Human Chitinase-3-like 1 Quantikine ELISA kit, R&D systems, UK; Abcam CRP human ELISA kit, Abcam, UK; pNf-H ELISA, Euroimmun AG, Germany).

statistical analysis
All hypothesis tests were two-sided, using p<0.05 to denote statistical significance. Analysis of demographics was performed using a χ² test with post hoc Fisher's exact test for categorical variables and Kruskal-Wallis H test with Dunn's post hoc for continuous variables. Group comparisons of level were performed using pairwise Mann-Whitney U tests with multiple comparison correction using the Benjamini-Hochberg step-up procedure. Additional analysis of covariance (ANCOVA) of log-transformed analyte concentration was performed to correct for age differences between groups, again using Benjamini-Hochberg correction for multiple comparisons. Classifier performance was examined using receiver operator characteristic (ROC) analysis. Comparison between ROC curves was performed using the DeLong method, with paired analysis to account for the use of multiple markers in the same samples. Optimal combinations were established by the highest ROC of summed log abundances of all combinations of chitinase proteins.
Univariate and multivariate correlation analysis was performed using simple and multiple linear regression and associations between predictor and outcome variables were tested using a Student's t-test. Chitinase protein concentrations were log transformed prior to inclusion in linear regression models. Survival analysis was performed using a Log-rank test for individual variables and Cox proportional hazards modelling incorporating clinical variables that have been previously associated with survival in ALS and for which sufficient information was available. 18 Longitudinal analysis was performed using a random intercept, random  slope linear mixed effects model, with individual participants specified as random effects. Statistical analysis was performed in R.

resuLTs demographic data
Demographic data are given in  figure 1A-C). Levels of all three chitinase proteins were similar between healthy control subjects and AGCs (CHIT1 p=0.844, CHI3L1 p=0.657, CHI3L2 p=0.578). Using ANCOVA to control for the differing age of participants across groups yielded similar results, with the exception of CHI3L1 levels which were higher in patients with ALS than patients with PLS and higher in healthy controls compared with AGCs (ALS vs PLS p=0.027, healthy control vs AGC p=0.027; online supplementary table 2).

Classifier performance
Chitinase protein diagnostic performance is outlined in figure 1D-F. All chitinase proteins performed well in distinguishing ALS from healthy controls and mimics. Individual chitinase proteins did not outperform pNFH alone, nor did the optimal combination of chitinase proteins and pNFH (ALS vs healthy controls:

Correlation with clinical parameters Disease progression rate
In first-visit samples from patients with ALS and PLS, linear regression models of log 10 PR were constructed. Higher levels of CHIT1 and CHI3L2 were modestly correlated with higher PR in both simple linear regression and multiple linear regression models controlling for age of onset, site of onset and gender (CHIT1 simple regression Pearson's r=0.49, p<0.001, multiple regression p=0.049; CHI3L2 r=0.42, p<0.001, multiple regression p=0.048; figure 2A and C. Multiple regression models are detailed in table 3). CHI3L1 levels did not correlate with the rate of disease progression in simple or multiple regression models (simple regression r=0.13, p=0.241; multiple regression p=0.176; figure 2B and table 3). Levels of pNFH were more closely correlated with log 10 PR (simple regression r=0.69, p<0.001; multiple regression p<0.001; table 3, figure 2D).

Cognitive function
Regression analysis was also performed on a subgroup of patients (n=39) with ALS and PLS who had undergone testing using the Neurodegeneration Figure 3 Longitudinal stability of the csF chitinase proteins. Thick, shaded line indicates the overall linear fit of high and low disease progression rate strata. aLs, amyotrophic lateral sclerosis; chI3L1, chitinase-3-like protein 1; chI3L2, chitinase-3-like protein 2; chIT1, chitotriosidase-1; csF, cerebrospinal fluid; pR, disease progression rate. pearson's r given. aLs, amyotrophic lateral sclerosis; chI3L1, chitinase-3like protein 1; chI3L2, chitinase-3-like protein 2; chIT1, chitotriosidase-1; csF, cerebrospinal fluid; ecas, edinburgh cognitive and behavioural aLs screen; pLs, primary lateral sclerosis; pNFh, phosphorylated neurofilament heavy chain; pR, disease progression rate. ECAS. Higher levels of CHI3L1 were found to be weakly associated with a greater degree of cognitive dysfunction as measured by the ALS-specific ECAS score in both simple linear regression and multiple linear regression models controlling for the effect of disease duration and the degree of disability as measured by the ALSFRS-R (Pearson's r=−0.25, p=0.038; multiple regression p=0.044; figure 2D) figure 1F-H).

Longitudinal analysis
Longitudinal CSF chitinase protein levels in patients with ALS and PLS were fitted to a random intercept random slope linear mixed effects model, beginning at the baseline visit. The number    figure 4A-C). No significant difference between serum levels of patients with ALS and healthy controls was observed for any of the three chitinase proteins ( figure 4D-F).

dIsCussION
This study compared CSF glial markers from a cohort of patients with ALS, disease mimics and relevant control groups including longitudinal data, and in relation to clinical parameters of ALS and other CSF markers. CHIT1 and CHI3L1 in ALS: CHIT1 and CHI3L1 were elevated in patients with ALS compared with both healthy controls and asymptomatic carriers of genetic mutations predisposing to ALS. CHIT1 and CHI3L2 showed the best performance in distinguishing ALS from disease mimics and PLS. In common with previous studies, however, they did not improve on the performance of neurofilament proteins. The size of the disease mimic and PLS groups and differences in the sampling latency with respect to PLS limits extrapolation of this finding to the clinical diagnostic setting. Truly accurate ascertainment of diagnostic accuracy would require prospective recruitment of a large number of subjects in whom the diagnosis is not established at the time of sampling. Although a 2015 study using dried blood spot CHIT1 activitydemonstrates differences between healthy controls and correlation with progression rate, the poor correlation between CSF and serum levels of the chitinase proteins in this study agrees with recent analysis suggesting that measurement of serum chitinase proteins is unlikely to be useful as an ALS biomarker. 9 11 19 Lower levels (and activity) of CHIT1 have been shown in subjects with a common duplication in the CHIT1 gene, found in around a third of Europeans. 11 20 CHIT1 genotype was not available for this cohort, but the effect of this polymorphism on CSF CHIT1 levels might favour measurement of CHI3L2 instead, which showed similar diagnostic performance and clinical correlations but is without a similar confounding polymorphism.
Intriguingly, the analysis of clinical and biochemical correlations described here demonstrates differences in the relationship between the three chitinase proteins and different clinical and biochemical Neurodegeneration parameters. Specifically, CHIT1 and CHI3L2 correlated with the rate of disease progression, whereas CHI3L1 correlated with clinical upper motor neuron burden and cognitive impairment. This is in keeping with previous studies demonstrating elevated CSF CHI3L1 in patients with sporadic FTD compared with ALS (although this was not recapitulated in familial FTD) and elevated CHIT1 in ALS compared with familial FTD. 11 These findings hint at differences in the source of CHIT1, CHI3L2 and CHI3L1 within the central nervous system in ALS. Although CHIT1 colocalises with microglia in the spinal cord of patients with ALS, 9 similar findings have not been demonstrated in brain regions affected in ALS. In addition, although mRNA expression data show that CHI3L1 and CHI3L2 are upregulated in the motor cortex of patients with ALS, the cellular origin of this elevation has not been demonstrated. Studies in other diseases give a conflicting picture. In Alzheimer disease, CHI3L1 colocalises with astrocytes but not microglia, 21 whereas in multiple sclerosis, CHI3L1 is produced in microglia and astrocytes. 22 The secretion of CHI3L2 within the central nervous system has not been studied, though this work would suggest it is likely to be similar to CHIT1. Further work is required to establish the regional and cellular expression of these proteins in order to delineate their relevance to ALS pathogenesis.
This study did not demonstrate significant longitudinal change in the CSF levels of chitinase proteins. This is consistent with measurement of CSF and serum neurofilaments and the rate-of-change of the ALSFRS-R, which also remain relatively stable over time, and suggests that this reflects a relatively stable immune response through the course of the disease. 23 Whether this response is a primary driver of progression, a contributory or protective factor requires additional work.
The levels of all three chitinase proteins were low in AGCs, in keeping with studies examining CSF CHIT1 levels in this group (though it must be noted that in the current study the number of subjects is small and younger than the control group). 11 Together, these data suggest that CSF glial activity is likely to be normal in the years before the onset of symptoms in ALS. Greater insight into the glial response just prior to the symptomatic phase would be gained from examination of longitudinal samples from subjects transitioning from asymptomatic gene carriage to symptomatic ALS, as has been undertaken using neurofilament proteins. 24 Similarly, although a detrimental role for excessive glial activity is suggested by some animal models of ALS, it remains unresolved whether the correlation between disease severity and microglial activity in humans is a factor modulating the underlying progression of ALS or is simply a bystander effect. 7 17 The finding of a correlation between survival and CHIT1 level independent of pNFH lends limited support to a modifying role of microglia, though additional validation of this is necessary particularly given the omission of certain prognostic factors for which data were unavailable.
Although this study suggests that CSF chitinase proteins do not improve on the performance of neurofilaments for the diagnosis of ALS, it that they might find adjunctive use in the prediction of progression and survival and in particular that they represent a useful biochemical measure of the effect of therapeutic strategies targeting glial responses.