CHI3L1 protein, human

C57BL/6 and BALB/c WT mice were obtained from The Jackson Laboratory. Standard targeting approaches and homologous recombination were used to generate BRP-39–null embryonic stem cells and knockout mice (Fig. S1). Mice with null mutations in BRP-39 were generated on a mixed 129/C57BL/6 background and subsequently bred for >10 generations onto a C57BL/6 or BALB/c background. CC10–rtTA–IL-13 Tg mice were generated in our laboratory (20 (link)), bred onto a C57BL/6 background, and used in these studies. These mice use the Clara cell 10-kD protein (CC10) promoter and the rtTA (reverse tetracycline transactivator) to target IL-13 to the lung in a dox-inducible manner. Tg mice in which human YKL-40 was tightly and inducibly overexpressed (CC10-rtTA-tTS-YKL-40) in a lung-specific manner were generated using constructs and approaches that have been previously described by our laboratory (Fig. S7) (42 (link)). Mice that lacked BRP-39 and produced YKL-40 only in pulmonary epithelial cells (CC10-rtTA-tTS-YKL-40/BRP-39^−/−) were generated by breeding the CC10-rtTA-tTS-YKL-40 and BRP-39^−/− mice. Animal protocols were approved by the Yale University Institutional Animal Care and Use Committee (IACUC), and all experiments were performed according to the guidelines of the Yale University IACUC.

For each of the CSF biomarkers, we excluded the extreme values defined as either those that fell outside of three times the interquartile range below the first quartile (Q1) or those above the third quartile (Q3). In the main text, all the analyses were performed excluding extreme values, but including them rendered similar results (see Tables S2–S4 in supporting information). We tested for normality of the distribution for each biomarker using the Kolmogorov‐Smirnov test and visual inspection of histograms. CSF Aβ42, p‐tau, t‐tau, NfL, neurogranin, YKL‐40, GFAP, IL6, S100, α‐synuclein, and the p‐tau/Aβ42 ratio did not follow a normal distribution and were thus log₁₀‐transformed. CSF Aβ40, Aβ42/40 ratio, and sTREM2 followed a normal distribution and were not transformed.
We conducted a one‐way analysis of variance (ANOVA) to test statistically significant differences on age and education among AT groups. The mean levels of CSF biomarkers among AT (Aβ/tau pathology) groups were assessed by a one‐way analysis of covariance (ANCOVA) adjusting for age and sex. Cognitive performance (Mini‐Mental State Examination [MMSE] and Free and Cued Selective Reminding Test [FCSRT]) was assessed by an ANCOVA adjusting for age, sex, and education. These comparisons were followed by Tukey corrected post hoc pairwise comparisons. Differences in the frequencies of sex and apolipoprotein E (APOE)‐ε4 categories were assessed by the Pearson's χ² test. Correlations between CSF biomarkers were tested by partial correlations adjusted by age.
To study the association of each CSF biomarker with demographic characteristics and APOE‐ε4 status, we computed a linear regression model with age, sex, and APOE‐ε4 status as predictor variables. We conducted these analyses stratifying by Aβ status and, additionally, including in the linear regression an “Aβ42/40 ratio x age” interaction term.
We plotted CSF biomarkers levels as a function of CSF Aβ42/40, p‐tau, and p‐tau/Aβ42. We corrected each of the CSF biomarker values by age and sex and computed the mean and standard deviation of each biomarker in the A–T– group (reference group). We next converted CSF biomarker values to z‐scores by subtracting the mean and dividing by the standard deviation of the reference group. The relationship between each CSF biomarker and the proxies of disease progression (ie, CSF Aβ42/40, p‐tau, and p‐tau/Aβ42) were modelled using a robust local weighted regression method (rlowess; “smooth” function in Matlab and a span of 300) and we plotted the resulting model.²², ²³ Moreover, we calculated the linear regression slopes for each of the biomarkers in each of the negative or positive groups, using the previously described cutoffs. We computed the P‐values testing the null hypothesis of whether the regression slope being equal to zero. In addition, the probability of the two regression slopes being equal was also calculated.
For all the analyses, we applied a false discovery rate (FDR) multiple comparison correction following the Benjamini‐Hochberg procedure.²⁴ All tests were two‐tailed, with a significance level of α = 0.05. Statistical analyses were performed in SPSS IBM 20.0 and R software (http://www.r-project.org/). Figures were built using R and Matlab (v2018b).

Statistical analysis and visualization
of metabolite concentrations, glycoproteins, lipoproteins, NFL, YKL-40,
metal elements, and demographic data were performed using R, SPSS
v.20, and Metaboanalyst v.5.0. The gender distribution was analyzed
with Pearson’s Chi-Square, while the age at inclusion was analyzed
with one-way analysis of variance. EDSS was analyzed with an independent
sample t-test with the assumption of equal variance
based on Levene’s test for equality of variance, while equal
variance was not assumed for the disease duration for the same test.
All concentrations were Pareto scaled prior to multivariate analysis.
Due to the relatively high number of variables relative to the number
of observations, sparse partial least squares discriminant analysis
(sPLSDA) with leave-one-out cross-validation was used to identify
the metabolites with the greatest contribution to group separation.
Since some of the concentration data were not normally distributed
based on Kolmogorov–Smirnov and Shapiro–Wilk tests,
nonparametric tests, such as the Kruskal–Wallis test for multiple
comparisons and the Mann–Whitney U test for
pairwise comparisons, were performed. Descriptive statistics for the
biomarkers are reported as the median and interquartile range (IQR),
while those for the demographic data are reported as the mean and
standard deviation (SD). Due to multiple comparisons and unless otherwise
stated, p-values lower than 0.01 were considered
to be statistically significant, while p-values higher
than 0.01 but less than 0.05 were considered as trends. Spearman’s
correlation coefficients were also calculated for relevant metabolites.
Receiver operator characteristic (ROC) curves were also constructed
for significant metabolites using SPSS v.20. The multidimensional
scaling method ALSCAL was used to project the significant features
in two dimensions with the dissimilarity matrix based on the Euclidean
distance.

CSF
and serum samples were collected from 22 SPMS, 18 PPMS, at the time
of disease progression confirmation, and 13 controls at the Neurology
Unit of the “Virgen de la Macarena” University Hospital
in Seville, using standardized protocols. All samples were collected,
coded, and stored in the hospital biobank, from where they were sent,
anonymized, to the NMR Unit for analysis. All study subjects signed
the informed consent form for research on biobanked samples, which
was approved by the Ethics Committee of the Hospital, prior to participation
in the study. MS patients were classified according to the 2017 revised
McDonald criteria.^{2 (link)} Control subjects were
patients whose diagnosis required the collection of CSF samples but
whose conditions were not considered to be neurodegenerative in nature.
The absence of unrecognized neurodegeneration or neuroinflammation
among controls was subsequently confirmed with low concentrations
of neurofilament light (NFL) and YKL-40 in their CSF. All samples
were stored at −80 °C until analysis.
None of the
PPMS patients were under any treatment at the time of sample collection,
and only four of them were taking a combination of B vitamin supplementations
(B12, B6, and B1). As for the SPMS patients, only 4 of the 22 patients
were under treatment, 1 with interferon beta-1b and 3 with natalizumab.
Furthermore, one of the patients under natalizumab was also supplemented
with B vitamins, and another patient was only supplemented with the
same combination of B vitamins. The rest of the SPMS patients had
been without any type of treatment for at least a year.

Quantification of CSF NFL was performed with
the Uman Diagnostics NF-Light ELISA kit (Umea, Sweden) according to
the manufacturer’s protocol using a final dilution factor of
1:2. CSF and serum YKL-40 levels were also quantified according to
the manufacturer’s protocol for YKL-40 (Invitrogen, Thermo
Fisher Scientific) with dilution factors of 1:200 and 1:150 for CSF
and serum, respectively. All samples were subjected to the same freeze–thaw
cycle to ensure comparability of the results. In addition, samples
were randomized to nullify potential differences due to a batch effect
in all analyses and an edge effect during ELISA.