Profiling metabolome of mouse embryonic cerebrospinal fluid following maternal immune activation

Abstract

The embryonic cerebrospinal fluid (eCSF) plays an essential role in the development of the central nervous system (CNS), influencing processes from neurogenesis to lifelong cognitive functions. An important process affecting eCSF composition is inflammation. Inflammation during development can be studied using the maternal immune activation (MIA) mouse model, which displays altered cytokine eCSF composition and mimics neurodevelopmental disorders including autism spectrum disorder (ASD). The limited nature of eCSF as a biosample restricts its research and has hindered our understanding of the eCSF's role in brain pathologies. Specifically, investigation of the small molecule composition of the eCSF is lacking, leaving this aspect of eCSF composition under-studied. We report here the eCSF metabolome as a resource for investigating developmental neuropathologies from a metabolic perspective. Our reference metabolome includes comprehensive MS¹ and MS² datasets and evaluates two mouse strains (CD-1 and C57Bl/6) and two developmental time points (E12.5 and E14.5). We illustrate the reference metabolome's utility by using untargeted metabolomics to identify eCSF-specific compositional changes following MIA. We uncover MIA-relevant metabolic pathways as differentially abundant in eCSF and validate changes in glucocorticoid and kynurenine pathways through targeted metabolomics. Our resource can guide future studies into the causes of MIA neuropathology and the impact of eCSF composition on brain development.

Keywords: developmental metabolomics; embryonic cerebrospinal fluid; embryonic metabolism; maternal immune activation; metabolic composition of the embryonic cerebrospinal fluid.

Figure 1

An approach to generate an embryonic CSF (eCSF) compound database as a tool for future developmental and neurological research.A, schematic depicting experimental workflow for embryonic CSF sample preparation. CSF from embryos (eCSF) from a single pregnant mouse was pooled and 3 to 7 μl extracted for analysis. Each pregnant mouse is considered a biological replicate. B, schematic depicting the analytical workflow for the generation of an eCSF polar compound database. Polar chromatography was performed in HILIC mode (ZIC-HILIC: zwitterionic hydrophilic interaction column) and metabolites were detected using a high-resolution (HiRes) mass spectrometer (Thermo Orbitrap QEactive). Data was processed using two separate approaches depending on the compound identification strategy. Contaminants were filtered by either “Mock filtering” - based on an extensive set of mock samples, or by “IROA” credentialing – based on labeled metabolome correspondence (Fig. S1, B–C). Signals (features) that passed the abundance filter (mock filtered features) and quality control were then used to generate compound databases at MS¹ and MS² levels. C, Representative results from untargeted polar metabolomics on eCSF from CD-1 mice (n = 7) for mock-filtered dataset outlined in (B). Negative- and positive-mode analyses are shown separately. Mock and eCSF samples are compared using a volcano plot. The signal was normalized based on targeted analysis of polar metabolites and mean-centering per sample (see methods for details). Metabolites highlighted in orange are >2-fold significantly higher in eCSF than in mock samples. D, representative results from untargeted polar metabolomics of eCSF from CD-1 mice (n = 7) for IROA credentialed dataset as outlined in B. Relevant biological features were identified using IROA-“LTRS” – a 1/1 mix of 95:5/5:95 unlabeled (¹²C) and ¹³C-labeled reference yeast metabolome (¹² C/¹³C reference, IROA TruQuant Yeast Extract Semi-targeted QC Workflow). IROA-credentialed features were identified from eCSF (¹²C, unlabeled) mixed with a reference internal standard (“IS”) that was composed of 5:95 unlabeled (¹²C) and ¹³C-labeled metabolome. Non-credentialed ¹²C signals (¹²C reference or ¹²C eCSF), with no matching IROA ¹³C signals, are depicted as indicated in the legend. E, Strategy for the implementation of the eCSF library into an untargeted analysis workflow combining in-house and online databases at MS¹ and MS² levels. Levels of identification certainty (1 through 4) are depicted following recommendations from the metabolomics community. F, breakdown of the composition of the eCSF-specific filtered features from the mock-filtered dataset. Negative- and positive-mode are presented separately. The left panels present the fraction of total filtered features at different identification levels (with or without database matches). Right panels (colored Venn diagrams) represent the breakdown of database-matched features identified either at Level 1 (exact mass, retention time, and MS² match or exact mass and RT match only) or Level 2 (exact mass and external database MS² match only). G, Breakdown of the composition of the eCSF-specific filtered features from the mock-filtered CD-1 untargeted dataset for which we have obtained MS² level information. Negative and positive mode are presented separately. Matches to in-house (House MS² DB) and external (mzCloud MS²) libraries are depicted. The remaining spectra were manually quality controlled and consolidated into the CD-1 MS² database.

Figure 2

Application of eCSF library in untargeted eCSF metabolomics in two mouse strains and at two developmental time points.A, breakdown of the composition of the eCSF-specific filtered features based on overlapping features between CD-1 and C57Bl/6 eCSF untargeted metabolome at E14.5. Two independent C57Bl/6 eCSF cohorts and one CD-1 cohort were treated as distinct batches because sample collection was performed on different days (minimum of three biological replicates per cohort). Filtered untargeted features from C57Bl/6 with either an in-house database match (House MS¹ DB) or a CD-1 database match (CD-1 MS¹ DB) are shown for MS¹ or MS² databases as indicated. The remaining features, with or without MS²-level information, which were not present in our databases, were identified at Level 3 and 4 and are also indicated. B, heatmap for the combined positive- and negative-mode data in (A). Mock samples were included for comparison. Further 5% (interquartile range) filtering step, log-transformation and Pareto scaling were performed within the MetaboAnalyst platform. C, Volcano plot for the combined positive and negative mode data in (A). For clarity, only significantly changed metabolites between CD-1 and C57Bl/6 eCSF with in-house library match were annotated, and negative ion adducts were further specified. In addition, all detected amino acids are indicated in black, with only significantly altered (alanine and proline) annotated. D, Volcano plot, depicting a comparison between polar metabolites detected by untargeted metabolomics from eCSF (C57Bl/6 samples) of E12.5 and E14.5. Compounds from Levels 1 to 3 were used to generate the plot. For clarity, only significantly changed metabolites with in-house library match are annotated. The label [M-H] depicts ions that were detected in negative mode.

Figure 3

Application of eCSF library in the study of eCSF in a maternal immune activation (MIA) model.A, schematic depicting the collection strategy of eCSF for untargeted LC-MS metabolomics 3 h post induction of MIA by injection with polyI:C. B–E, altered metabolites 3 h post polyI:C injection. Shown are heatmaps of top 25 changed metabolites (B and C) or PLSDA and corresponding important features (VIP) plots (D and E) of untargeted metabolomics analysis of polar eCSF metabolites from treatment and control samples (minimum of three biological replicates per condition). Two independent experiments are presented separately (B vs C and D vs E). Metabolites of interest are highlighted in green (glucocorticoid pathway), and magenta (kynurenine pathway). Metabolites overlapping between the top 25 analysis and VIP analysis are in bold. Plots were generated using the online MetaboAnalyst tool, after log-transformation and Pareto scaling of the combined data from normalized negative and positive modes analyses. Level 1 (exact mass, RT match or RT and MS² match to in-house databases) confidence metabolites were used for this analysis. For a comparison between saline and polyI:C conditions based on Level 1 to 4 metabolites (see Fig. S3, E and F).F, schematic depicting the collection strategy of eCSF for untargeted LC-MS metabolomics 48 h post-induction of MIA by injection with polyI:C. G–J Same as for (B–E) except data is from samples collected 48 h post-induction of MIA. (deg), fragment; HyP, hydroxyproline; GCP, glycerophosphocholine; GSA, guanidinosuccinate; [M-H], ion detected in negative mode.

Figure 4

Targeted metabolomics of MIA eCSF reveals involvement of kynurenine pathway and a corticoid stress response.A, schematic of glucocorticoid synthesis in rodents. In orange – corticosterone, was detected by targeted metabolomics in MIA eCSF. B, Bar graphs depicting mean ± standard deviation of two corticosteroids identified at Level 2 confidence (including m/z match to the online database with no RT) by untargeted metabolomics 3 h post polyI:C injection. Two independent experiments (each with a minimum of three biological replicates) were combined for the analysis (leading to a minimum of 7 biological replicates per condition). Missing values for individual metabolites (in cases where measurements were under our limit of detection) were omitted from the analysis. ∗∗ = p < 0.001; ∗∗∗ = p < 0.001 (Unpaired t test with Welch's correction). C, Bar graphs depicting mean and standard deviation levels post polyI:C injection from eCSF or maternal serum (mSer) for different glucocorticoid-related metabolites detected by our LC-MS ∗ = p < 0.05; ∗∗ = p < 0.005 (one-way ANOVA, Šídák's multiple comparisons test). Graphs are presented relative to eCSF abundance in control. “OH” denotes “hydroxy”. D, schematic of tryptophan degradation pathways in mammals. Metabolites present in our in-house library and analyzed via targeted metabolomics are highlighted in orange. “OH” denotes “hydroxy”. E and F, Mean and standard deviation of indicated metabolites 3 h (D) or 48 h (E) post polyI:C injection. Data were analyzed using targeted metabolomics approach. Two independent experiments were combined for each time point. 3h data includes exp 1 and 2: exp 1 with n = 3 for saline group and n = 4 for polyI:C group, and exp 2 with n = 4 in each group. 48h data includes exp 1 and 2: exp 1 with n = 4 for saline group and n = 3 for polyI:C group, and exp 2 with n = 3 in the saline group and n = 5 in the polyI:C group. ∗ = p < 0.01; ∗∗ = p < 0.001; ∗∗∗ = p < 0.001 (unpaired t test, two-tailed). Of note, our method could not differentiate 5- and 3-hydroxykynurenine, labeled here simply as “OH-kynurenine”.

Figure 5

Metabolic changes in glucocorticoid-related intermediates as assessed by untargeted metabolomics in embryonic liver and CSF or maternal liver and serum.A, schematic depicting the collection strategy of embryonic CSF (eCSF) and liver (eLiv) and maternal liver (mLiv) and serum (mSer) for untargeted LC-MS metabolomics 3 h post-induction of MIA by injection with polyI:C. B, PLSDA and corresponding important features plot of untargeted metabolomics analysis of polar eCSF, eLiv, mLiv, and mSer metabolites from polyI:C- and saline-treated samples at 3 h post injection (each with a minimum of three biological replicates). Plots were generated using the online MetaboAnalyst tool, after log-transformation and Pareto scaling of the combined data from normalized negative- and positive-mode analysis. Levels 1 to 3 (exact mass, RT match, MS² match to in-house or online databases, and match to eCSF CD-1 library) confidence metabolites were used for this analysis. eCSF data from experiment 2 from previous figures is from the same mouse cohort and was re-analyzed for this set of graphs. Metabolites of interest are highlighted in green (glucocorticoid pathway) and magenta (kynurenine pathway). “OH” denotes “hydroxy”. C and D, Bar graphs depicting mean and standard deviation levels of indicated metabolites post polyI:C injection for different glucocorticoid-related metabolites detected by our LC-MS ∗ = p < 0.01; ∗∗ = p < 0.001; ∗∗∗ = p < 0.001; ∗∗∗∗ = p < 0.0001 (one-way ANOVA, Šídák's multiple comparisons test). Graphs are either presented relative to eCSF abundance (C) or relative to the corresponding control saline sample set (D). “OH” denotes “hydroxy” CD.

Figure 6

Metabolic changes in kynurenine pathway metabolites as assessed by untargeted metabolomics in embryonic liver and CSF or maternal liver and serum.A, Bar graphs depicting mean and standard deviation levels of indicated metabolites post polyI:C injection for indoxyl and kynurenine pathway intermediates by untargeted metabolomics. ∗ = p < 0.01; ∗∗ = p < 0.001; ∗∗∗ = p < 0.001; ∗∗∗∗ = p < 0.0001(one-way ANOVA, Šídák's multiple comparisons test). Graphs are relative to eCSF abundance. B, Volcano plot comparing embryonic (eLiver) to maternal (mLiver) liver samples at 3 h post saline injection (each with a minimum of three biological replicates). Untargeted metabolomics at Levels 1 to 3 (exact mass, RT match, MS² match to in-house or online databases and match to eCSF CD-1 library) confidence metabolites were used for this analysis. Metabolites of interest are highlighted (blue – lower in eLiver, red – higher in eLiver). Only identified kynurenine pathway intermediates are annotated. C, Same as A, but graphs are relative to the corresponding control saline sample set.