Demographic and clinical characteristics of ALS patients and controls

A cohort of 210 ALS patients (86 females and 124 males) with median age at onset of 61 [54–68] years and 218 controls (112 females and 106 males) were included in this study. Three groups of controls were set up: CTRL-1 included 106 individuals, with a median age of 58 [45–71] years, CTRL-2 was made up of 38 patients, with a median age of 58 [47.5–72] years, while CTRL-3 was formed by 74 subjects, with a median age of 76 [71–79] years. Table 1 shows demographic features and biomarker values of all subjects involved in this study. The numerical breakdown of individuals comprising groups CTRL-1, CTRL-2 and CTRL-3 is shown in Supplementary Table 1.

Table 1 Demographic and biochemical features of ALS patients and controls

Sex distribution was significantly different between ALS patients and controls (p = 0.030), due to the preponderance of males in the ALS cohort, while controls had an older age at evaluation compared to ALS counterparts (p = 0.013) (Table 1).

Demographic, clinical and biochemical features of patients with ALS are reported in Table 2. When the DPR was available, ALS subjects were grouped into slow (N = 102) and fast progressors (N = 102), depending on whether DPR was below or above the median value in this sub-cohort. While sex was homogeneously distributed in fast progressors, male patients were overrepresented in slow progressors (p = 0.007). ALS patients with a higher DPR had a significantly older age at onset (p = 0.021), a shorter disease duration (p < 0.0001), a lower ALSFRS-R score (p < 0.0001) and a lower FVC (p = 0.0007) compared to slow progressors, while age at evaluation and site of onset did not significantly differ (Table 2). FVC was available for 135 patients. Fast progressors were more significantly impaired at neuropsychological assessment (p = 0.026; N = 194). No significant differences were found between slow and fast progressing patients in terms of clinical phenotypes, BMI (N = 148), or presence of a causative genetic mutation in one of the four main ALS genes.

Table 2 Demographic, clinical and biochemical features of ALS patientsCSF biomarkers in ALS patients and controls

CSF levels of NfL, CHIT1 and miR-181b were measured for N = 390, N = 226, and N = 176 patients, respectively.

CSF levels of NfL, CHIT1 and miR-181b were significantly higher in ALS compared to all 218 control individuals (p < 0.0001, p < 0.0001, and p = 0.001, respectively). The difference was still significant when comparing the CSF levels of the three biomarkers between the ALS and CTRL-3 group (NfL, p < 0.0001; CHIT1, p < 0.0001; miR-181b, p = 0.016) (Fig. 1 A–C), while only CSF NfL and CHIT1 concentrations, and not miR-181b, were significantly increased in ALS patients compared to CTRL-1 group (NfL, p < 0.0001; CHIT1, p < 0.0001). Conversely, only CSF NfL levels were significantly higher in ALS compared to the CTRL-2 group (p < 0.0001), while CSF CHIT1 and miR-181b levels showed no significant difference between the two cohorts. This finding may be at least in part due to the considerable number of patients with an inflammatory disease included in the CTRL-2 group and the relatively small number of individuals with measurements of CSF miR-181b levels (Supplementary Table 1).

Fig. 1

CSF NfL, CHIT1 and miR-181b distribution in ALS and controls. A–C CSF NfL, CHIT1 and miR-181b levels in ALS compared to all control individuals. D–F CSF NfL, CHIT1 and miR-181b levels in ALS compared to patients with AD and synucleinopathies (CTRL-3 group). Scatter dot plot values represent median and interquartile range. Symbols of statistically significant differences: *p < 0.05; **p < 0.01; ****p < 0.0001 (Kruskal–Wallis test). G–I ROC curves of CSF NfL, CHIT1, miR-181b in ALS patients vs. all control individuals

In order to assess whether the biomarker profile differed among different neurodegenerative disorders, we directly compared ALS with patients with AD and with patients affected by synucleinopathies (Parkinson’s disease, multiple system atrophy, dementia with Lewy bodies). Strikingly, we found that ALS patients had significantly elevated concentrations of CSF NfL, CHIT1 and miR-181b compared to AD (p < 0.0001, p < 0.0001, and p = 0.002, respectively) and increased CSF levels of NfL and CHIT1 compared to synucleinopathies (p < 0.0001 and p = 0.017, respectively; Fig. 1 D–F).

The diagnostic performance of these biomarkers was assessed using ROC curves. CSF NfL levels displayed a high accuracy in discriminating ALS from controls with an AUC of 0.899 (95% confidence interval (CI): 0.868 to 0.929, p < 0.0001), corresponding to a sensitivity of 80% (95% CI: 73.6–85.2%) and a specificity of 84.3% (95% CI: 78.6–88.6%) at a cut-off of 2079 pg/mL (Fig. 1G). Conversely, CSF CHIT1 had a lower ability to predict the diagnosis of ALS vs. controls with an AUC of 0.688 (95% CI: 0.615–0.762, p < 0.0001), showing poor sensitivity (46.8%; 95% CI: 37.1–56.8%) and high specificity (90.2%; 95% CI: 83.9–94.2%) at a cut-off of 1564 pg/mL (Fig. 1H). CSF miR-181b had low sensitivity (41.4%; 95% CI: 31.6–51.9%) and high specificity (85.4%; 95% CI: 76.6–91.3%) at a cut-off of 0.424. The AUC for CSF miR-181b was 0.640 (95% CI: 0.558 to 0.722, p = 0.001) (Fig. 1I).

In order to assess whether combining the three biomarkers improved the diagnostic performance, we computed values of a z-score-based composite biomarker for the sub-cohort for which levels of all three biomarkers were available (N = 58 ALS patients and N = 76 controls). We compared ALS patients to the whole control group (N = 76) and to the neurodegenerative group (CTRL-3, N = 56), assessing combinations of two biomarkers at a time. For both comparisons, the best diagnostic performance was obtained when combining NfL and CHIT1 levels (ALS vs. all controls: AUC = 0.848; 95% CI: 0.785–0.911; p < 0.0001; ALS vs. CTRL-3: AUC = 0.826; 95% CI: 0.753–0.899; p < 0.0001). ROC analyses are shown in Supplementary Fig. 1 (Figure S1A–F). However, the combination of the three biomarkers does not seem to significantly improve the discrimination between ALS and controls (AUC = 0.829; 95% CI: 0.761–0.898; p < 0.0001) or neurodegenerative disorders (AUC = 0.809; 95% CI: 0.731–0.886; p < 0.0001) (Figure S1 G, H).

Indeed, NfL alone has the best diagnostic performance in differentiating ALS from controls (AUC = 0.889; 95% CI: 0.836–0.942; p < 0.0001) and neurodegenerative disorders (AUC = 0.871; 95% CI: 0.808–0.933; p < 0.0001).

ALS patients show elevated CSF CHIT1 levels despite higher frequency of CHIT1 polymorphism

In a substantial subset of our cohort (N = 163 subjects, including N = 86 ALS patients and N = 77 controls), we assessed the presence of the 24-bp duplication polymorphism in CHIT1, which was reported to lower the levels of CHIT1 protein in the biofluids [41]. In this subgroup, CHIT1 measurement in the CSF was available for 153 individuals.

We split our cohort according to the median CHIT1 value (390 pg/mL) and we demonstrated that heterozygous and homozygous carriers of the mutated allele were significantly more represented among patients with CHIT1 levels lower than or equal to the median value (p = 0.004), confirming literature data [41]. In terms of allelic frequency, 52 mutated alleles were present in patients with lower CHIT1 levels, compared to only 14 mutated alleles in those with CHIT1 concentrations above the median value (p < 0.0001; Fig. 2A).

Fig. 2

Analysis of CHIT1 polymorphism. A distribution of wild-type and mutated alleles in individuals with CHIT1 levels equal to 390 pg/mL vs. above 390 pg/mL (****p < 0.0001). B distribution of wild-type and mutated alleles in ALS patients and controls (*p = 0.021). C comparison of CSF CHIT1 levels in wild-type homozygotes vs. heterozygous/homozygous polymorphism carriers in ALS cohort (*p = 0.014)

Thus, we investigated whether CHIT1 polymorphism was responsible for the decreased CHIT1 levels in the control group. Interestingly, we found that ALS patients had an over-representation of the mutated allele (ALS: 127 wild-type alleles and 45 mutated alleles; controls: 130 wild-type alleles and 24 mutated alleles; p = 0.021; Fig. 2B). Analysis of the distribution of CSF CHIT1 concentrations across ALS patients with different genetic backgrounds demonstrated that patients with wild-type homozygosity had significantly higher CHIT1 levels compared to heterozygous and homozygous polymorphism carriers (2037 vs. 390 pg/mL; p = 0.014; Fig. 2C). These findings suggest that ALS patients showed significantly increased levels of CHIT1 in the CSF despite a higher prevalence of CHIT1 polymorphism.

CSF NfL, CHIT1 and miR-181b levels correlate with clinical variables in ALS patients

No significant difference was found in CSF NfL, CHIT1 and miR-181b with respect to site of onset. CSF CHIT1 levels were significantly increased in male compared to female ALS patients (median values, 1783 pg/mL vs. 390 pg/mL; p = 0.038), while CSF NfL and miR-181b levels were equally represented in both sexes. However, when comparing male patients to male controls and female patients to female controls, CSF CHIT1 levels were higher in the ALS groups, irrespective of sex (male ALS patients vs. male controls, median values: 1783 pg/mL vs. 390 pg/mL; p < 0.0001; female ALS patients vs. female controls, median values: 390 pg/mL vs. 390 pg/mL; p = 0.005). No differences were identified between males and females in the control category. Altogether, the above findings suggest that the presence of higher CHIT1 concentrations in ALS patients is not simply ascribable to the preponderance of men in the ALS group.

When we analyzed the distribution of the three biomarkers across different clinical phenotypes (classic, bulbar, flail arm, flail leg, respiratory, pyramidal, pure lower motor neuron [PLMN], pure upper motor neuron [PUMN]), we found a trend towards higher NfL in classic and bulbar forms compared to others (p < 0.011), but no statistically significant differences were evident after multiple comparisons. No difference in the levels of the three biomarkers was identified across different neuropsychological phenotypes (purely motor ALS, ALS with cognitive impairment (ALSci), ALS with behavioural impairment (ALSbi), ALS with cognitive and behavioural impairment (ALScbi), and ALS-frontotemporal dementia (ALS-FTD)). However, when comparing the concentrations of the three biomarkers among patients with different genetic backgrounds (C9ORF72 repeat expansion carriers, SOD1 mutation carriers, TARDBP mutation carriers, patients tested but not carrying any causative genetic mutations in these genes), CSF miR-181b levels were significantly higher in SOD1 carriers vs. non-mutated patients (p = 0.004). However, patients with SOD1 mutations had a younger median age at evaluation (51.5 vs. 62 years, p = 0.0007), therefore, the higher miR-181b levels found in these patients could be at least partly explained by this age difference (see below).

In order to investigate the relationships between biomarkers and clinical variables within the ALS cohort, we performed a correlation analysis using a Spearman non-parametric test. A positive correlation was observed for CSF NfL levels with both CSF CHIT1 (r = 0.568, p < 0.0001) and CSF miR-181b levels (r = 0.272, p = 0.029); on the contrary, no significant correlation between CSF CHIT1 and miR-181b was found. In the ALS group, there was a significant inverse correlation of age at evaluation with both CSF NFL (r =  – 0.178, p = 0.017) and CSF miR-181b levels (r =  – 0.233, p = 0.030); a negative trend was also observed for CSF CHIT1, albeit without statistical significance (Fig. 3 A–C). Conversely, age at evaluation positively correlated with CSF NfL levels in the control group (r = 0.525, p < 0.0001), while no correlation was found with CSF CHIT1 and miR-181b levels. Moreover, CSF NfL levels inversely correlated with disease duration (r =  – 0.331, p < 0.0001) (Fig. 3D), while CSF CHIT1 and miR-181b concentrations did not (Fig. 3E–F). An inverse correlation existed between ALSFRS-R scores and both CSF NfL levels (r =  – 0.219, p = 0.004) (Fig. 3G) and CSF CHIT1 levels (r =  – 0.265, p = 0.012) (Fig. 3H), while there was no significant correlation with CSF miR-181b levels (Fig. 3I). FVC showed a significant inverse correlation only with CSF NfL (r =  – 0.218, p = 0.019) (Fig. 3L–N). No correlation was identified between BMI and any of the three biomarkers.

Fig. 3

Correlation between biomarkers and clinical variables. Correlation analyses of CSF NfL, CHIT1 and miR-181b with age at evaluation (A-C), disease duration (D-F), ALSFRS-R (G-I) and FVC (L-N) (r = Spearman’s coefficient). ALSFRS-R Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised. FVC forced vital capacity (expressed as % of predicted value)

DPR significantly correlated with both CSF NfL (r = 0.393, p < 0.0001) and CSF CHIT1 concentrations (r = 0.274, p = 0.009), but not with CSF miR-181b levels (Fig. 4A–C). Accordingly, CSF NfL levels were significantly higher in fast vs. slow progressors (median values: 8508 pg/mL vs. 3856 pg/mL; p < 0.0001) (Fig. 4D). The same was observed for CSF CHIT1 levels (median values: 2153 pg/mL vs. 390 pg/mL; p = 0.029), while no significant difference was found between fast and slow progressors in terms of CSF miR-181b levels (Fig. 4E, F). Using the three biomarkers as variables in a multiple linear regression model, CSF NfL was the only independent predictor of DPR (OR: 9.70*10–5; 95% CI, 7.13*10–5–1.23*10–4; p < 0.0001). However, when adding age at onset, site of onset, FVC and BMI as covariates to the model, no variable showed significant association with DPR. In linear regression models assessing each individual biomarker together with the previously mentioned other covariates, CSF NfL and FVC were able to independently predict DPR (NfL: OR = 2.90*10–5; 95% CI: 9.74*10–6 to 4.84*10–5; p = 0.004; FVC: OR =  – 0.009; 95% CI: – 0.02 to 2.10*10–3; p = 0.013) (Supplementary Table 2).

Fig. 4

Correlation between biomarkers and disease progression rate. A-C correlation of CSF NfL, CHIT1 and miR-181b with disease progression rate (DPR) (r = Spearman’s coefficient). D-F CSF NfL, CHIT1 and miR-181b levels in fast progressing ALS cases compared to slow progressing patients (*p < 0.05; ****p < 0.0001; Mann–Whitney test). Scatter dot plot values represent median and interquartile range

CSF NfL levels, but not CHIT1 and miR-181b levels, are associated with survival in ALS patients

Median survival time was 33 months [21.0–60.5, N = 144]. We investigated a possible association between CSF biomarkers and survival time using Kaplan–Meier analysis. CSF NfL levels higher than or equal to 5620 pg/mL (median value in ALS patients) were associated with a significantly shorter survival time (chi-square 39.79, p < 0.0001) (Fig. 5 A); conversely, neither CSF CHIT1 levels above or equal to/below 1042.3 pg/mL (median value in ALS patients) nor CSF miR-181b levels above or equal to/below 0.133 (median value in ALS patients) were associated with survival (p = 0.155 and p = 0.951, respectively) (Fig. 5B, C). In Cox proportional hazards models considering age at onset, site of onset (bulbar vs. spinal), ALSFRS-R, BMI and each biomarker separately (NfL, CHIT1 and miR-181b) as covariates, CSF NfL levels and ALSFRS-R independently predicted survival in the first model (i.e. that including NfL as biomarker) and all variables, including CHIT1, were significantly associated with survival in the second model (i.e. that including CHIT1) (Table 3).

Fig. 5

Survival analysis. A-C Kaplan–Meier curves according to CSF NfL, CHIT1, and miR-181b levels. D survival estimates according to CHIT1 levels in patients with CSF NfL levels below the median. Survival time was calculated from disease onset and median values of these biomarkers were used as cut-offs

Table 3 Cox proportional hazards models with biomarkers and clinical variables

To further explore the prognostic role of CSF CHIT1 in ALS patients not captured by an increased value of CSF NfL, we analyzed CSF CHIT1 measurements in patients with CSF NfL values below the median. Within this cohort (N = 55), no correlation was found between CSF CHIT1 measurements and DPR (p = 0.965). Accordingly, no difference in CSF CHIT1 values between slow and fast progressors was observed (p > 0.999). Finally, we considered patients with low CSF NfL (i.e. CSF NfL below the median) and subdivided them according to the median CHIT1 level recalculated in this group (390 pg/mL). However, among these patients with low NfL, the presence of high vs. low levels of CHIT1 was not significantly associated with survival (p = 0.841) (Fig. 5D).