Unravelling neuronal and glial differences in ceramide composition, synthesis, and sensitivity to toxicity

Neurons and glia have distinct ceramide profiles

Previous studies have established that mouse CNS cell-types differ in their general lipidomic profiles³⁰. However, the regulation of ceramide homeostasis in these cell types is not understood. Specifically, it is not known whether they share common ceramide species or whether ceramide generation occurs through distinct pathways (Fig. 1A). Furthermore, although sphingolipids play important roles in nuclear signaling and metabolism³¹, the subcellular composition of ceramides in CNS cell-types has not been investigated. Therefore, we utilized liquid chromatography/mass spectrometry (LC/MS) to profile nuclear and cytoplasmic fraction ceramides in human iPSC-derived cells (Fig. 1B–E). This allowed us to establish whether there are basal differences in human neuronal and glial ceramide profiles. Cytoplasmic C16- and C18-ceramide were highest in motor neurons (MNs), whereas astrocytes had the highest C24-ceramide and microglia the highest C24:1 (Fig. 1B). Astrocytes and microglia had significantly higher levels of nuclear fraction long-chain ceramides (C22, C24) than MNs (Fig. 1C). Consistent with the cytoplasmic fraction, MNs had significantly higher levels of nuclear C18 than glial cells (Fig. 1C). In vivo studies have shown that CerS1 is neuronally localized and generates C18-ceramide³². This illustrates that the differing ceramide profiles in our iPSC-derived neuronal cell types may match in vivo profiles.

Ceramide profiles in different neuronal subtypes also vary

As neurons are often uniquely affected in diseases with sphingolipid dysregulation, we next determined the relative levels of ceramides in glutamatergic and GABAergic iPSC derived cortical neurons. Consistent with MNs, C18-ceramide levels were high in cortical neurons relative to the levels seen in glia (Fig. 1D, E). Glutamatergic neurons had the highest levels of C16- and C18-ceramide in both nuclear and cytoplasmic fractions (Fig. 1D) and had the highest levels of all ceramide chain lengths in the nuclear fraction (Fig. 1E). MNs had uniquely low levels of C24:1 in both fractions and C14 in the cytoplasmic fraction (Fig. 1D, E). This establishes that, while neurons have some commonalities in their ceramide profiles, the neuronal-subtype ceramide signatures are significantly different in ways that could be relevant to normal physiology and disease.

Bulk RNA-Seq characterization of ceramide pathway members within MNs and glia

Our lipidomics data establish that neurons and glia have distinct ceramide profiles. To identify the underlying genetic expression driving our lipidomic profile differences, we performed RNA sequencing on iPSC-derived astrocytes, MNs, and microglia. Due to differences in cellular viability, RNA libraries from neurons and glia were prepared differently. We are not comparing cellular expression directly by RNA-sequencing, we only assessed the relative expression levels within each cell-type. These data suggest that both glia and MNs depend on SPTLC1, CERS2, DEGS1, and SPTSSA, for ceramide production and that MNs might be distinctly dependent on SPTSSB and DEGS2 (Fig. 2A). MNs, astrocytes, and microglia have relatively lower CERS3 expression relative to other ceramide synthesis pathway members (Fig. 2A). Microglia had uniquely low CerS1 expression as compared to other CerS. ORMDLs, negative regulators of SPTLC³³ were expressed in all cell types with ORMDL1 and ORMDL3 levels being relatively higher than ORMDL2 in astrocytes and MNs (Fig. 2A). Expression of enzymes comprising the sphingolipid anteome, metabolic pathways (e.g., NADPH, L-serine and Acyl-CoA) that directly converge on ceramide synthesis³⁴, showed similar patterns of expression across all cell types with a few exceptions (Supplementary Fig. 1C).

We also found cell type differences in the expression of enzymes regulating the generation of ceramide via its salvage from the dephosphorylation or hydrolysis of more complex lipids, the use of ceramide as a substrate in lipid synthesis, ceramide hydrolysis, and ceramide transport (Fig. 2B–F). PLPP1–3 are enzymes that dephosphorylate ceramide-1P to ceramide. All cells had relatively low PLPP2 expression and microglia had low PLPP3 as compared to PLPP1 (Fig. 2B). Lysosomal enzymes (i.e., GALC, ARSA, GLB1, HEXB and SMPD1) that generate ceramide through lipid hydrolysis had high levels of expression in glia relative to other pathway members (Fig. 2C and F). SMPD3, a non-lysosomal hydrolytic enzyme of sphingomyelin, was the most highly expressed SMPD gene in MNs, while being amongst the lowest in glia (Fig. 2F). Similarly, enzymes regulating ceramide hydrolysis also varied by cell type with ASAH1 being enriched in glia and ACER1 in MNs (Fig. 2D). The expression pattern of lipid synthases that use ceramide to make more complex lipids were comparable across cell types except for UGCG, which was enriched in microglia relative to other synthases (Fig. 2C).

Because we were not able to directly compare gene expression of different cell-types by RNA-sequencing, we performed qPCRs on ceramide synthesis pathway members to perform more direct comparisons of CerS. The SPT complex is an essential rate limiting step in ceramide synthesis, so we also assessed the expression levels of complex members SPTLC1, SPTLC2, SPTSSA, and SPTSSB. Microglia had higher expression of multiple synthesis genes (CERS1, SPTLC1, SPTLC2, SPTSSA) as compared to astrocytes and MNs (Fig. 2G, Supplementary Fig. 1A and B). Microglia also expressed ceramidases (ASAH1 and ACER3) at significantly higher levels (Fig. 2K and L). Notably, SPTSSA was highest in microglia, as this preferentially uses shorter chain acyl-CoAs to produce long chain bases, whereas MNs were highest for SPTSSB (Supplementary Fig. 2B). SPTSSB prefers longer chain acyl-CoAs such as C18 to produce long chain bases³⁵. By qPCR, MNs also expressed significantly higher levels of CerS1 than astrocytes (Fig. 2G). CerS1 generates C18-fatty acid chain length and is known to be enriched in neurons in vivo. Astrocytes were highest for expression of CerS4 and CerS5 (Fig. 2H).

Neurons and glia differ in ceramide synthesis enzyme activities

RNA expression levels do not always correlate well with protein expression or protein function in the CNS, as might be indicated by our RNA-seq, qPCR, and lipidomics results. For example, CerC18 was highest in MNs relative to glia, CERS1 expression measured by RNA-seq was low in MNs relative to other ceramide synthesis pathway members, and CERS1 expression measured by qPCR was higher than astrocytes but lower than microglia (Figs. 1 and 2). To clarify the synthesis activity of each cell type, we characterized the functional activity of the CerS enzymes (Fig. 3A) by utilizing a ceramide synthesis activity assay³⁶. Microglia had significantly higher levels of CerS1/4 (Fig. 3B), CerS2 (Fig. 3C), and CerS5/6 activity (Fig. 3D), than did astrocytes and MNs. Astrocytes had significantly higher CerS5/6 activity than MNs (Fig. 3D), whereas MNs had significantly higher CerS1/4 activity than astrocytes (Fig. 3B). This is consistent with our data showing high CerC18 levels in MNs (Fig. 1), and in vivo reports suggesting that neurons have uniquely high CerS1 expression and CerC18 levels³². This is additional confirmation that our iPSC derived cells maintain features of in vivo cell types. Based on comparing the different activity assays, astrocytes might be uniquely dependent on CerS5/6 activity for ceramide synthesis, while neurons uniquely dependent on CerS1. All microglial CerS activity was higher than the other cell types. Additionally, the unique dependence of astrocytes on CerS5/6 does not match the relative levels of astroglial CerC14 and CerC16 compared to neurons and microglia. These discrepancies between RNA levels and CerS activity, indicate that functional outputs for ceramide synthesis are needed. The lack of alignment between lipid levels and CerS activity is also likely due to cell type dependent differences in the generation of ceramide from alternative pathways.

A Schematic of the different fatty acid carbon chain lengths produced by different CerS enzymes. B Assay of CerS1/4 (18:0 palmitoyl-CoA and NBD-sphinganine added to lysate) activity in astrocyte, microglia, and motor neuron (MN) cell lysates. MNs had significantly higher CerS1/4 than did astrocytes (One-way ANOVA, Tukey’s multiple comparisons test). C Assay of CerS2 (24:1 palmitoyl-CoA and NBD-sphinganine added to lysate) activity in astrocyte, microglial, and MN cell lysates. D Assay of CerS5/6 (16:0 palmitoyl-CoA and NBD-sphinganine added to lysate) activity in astrocyte, microglial, and MN cell lysates. B–D Microglia had the highest CerS activity for all ceramide synthases measured (One-way ANOVA, Tukey’s multiple comparisons test). D Astrocytes had significantly higher CerS5/6 activity than did MNs (One-way ANOVA, Tukey’s multiple comparisons test). All synthesis activity was normalized to cell number (activity per 200 k cells) by taking nuclear cell counts and subtracting dead cells using nuclear cell-death stains from representative culture wells (activity assay is n = 3–6 wells of plated cells as shown in graphs, with n = 3 CerS activity reactions performed per well).

Pharmacological inhibition of CerS has cell-type specific effects on ceramide profiles

While basal ceramide levels and RNA expression data are informative, it is important to determine if CerS activity drives basal ceramide levels in differentiated neurons and glia. By treating with the pan-CerS inhibitor fumonisin-B1 (FB1), we isolated the portion of the ceramide profile driven by de novo CerS synthesis, versus lipids that are generated through alternative pathways (Fig. 1A). This might clarify differences observed in CerS-activity, lipidomics, and RNA expression. A variety of ceramide chain lengths were reduced after 24 hours of FB1 treatment of glial cells (Fig. 4A, B, and C, Supplementary Fig. 4C, 4E). Notably, the magnitude of reduction in astrocyte ceramides was larger than changes in microglial ceramide in response to FB1 inhibition (Fig. 4A, B, and C, Supplementary Fig. 4B and C). This could indicate differences in the levels of ceramide synthesis between astrocytes and microglia, or a greater compensatory change in microglia to keep ceramide levels stable via other ceramide generating pathways. Microglia did have less significant ceramide changes in the cytoplasmic fraction (Fig. 4C), indicating that cellular compartments such as lysosomes might give microglia unique capacities to compensate for changes in lipid synthesis. Interestingly, some of the largest changes in astrocyte ceramides were in CerC14 and CerC16, consistent with the high levels of CerS5/6 activity in these cells (Figs. 3D, 4B, and E). Microglia had some of the largest changes in longer-chain ceramides, consistent with their high level of CerS2 activity.

Motor neuron (MN) (A, D) astrocyte (B, E) and microglial (C, F) cytoplasmic and nuclear ceramide profiles after 72 hours of CerS inhibition. A–F Ceramides analyzed by 2-way ANOVA (3 μM FB1 treatment group shown, see Supplementary Fig. 4 for complete comparisons). G–I By introducing a traceable sphinganine (d17:0) precursor, in the presence or absence of 3 μM FB1, we were able to track the active CerS-dependent ceramide synthesis in each cell type over a 2-hour period. A significant percentage of astroglial (H) and microglial (I) ceramides were labeled (bars with green outlines) after the two hours. Notably, MNs (G) had a lower percent of each ceramide labeled. Further labeling data in Supplementary fig. 3 and 4. G–I Each ceramide chain length, vehicle vs FB1 treatment, was analyzed by Student’s t-test (****P < 0.0001, ***P < 0.001, *P < 0.05; n = 4 replicates).

In MNs, there were minimal changes in ceramide in response to FB1 inhibition – even after 72 hours of treatment. MNs showed some small changes in CerC18 and CerC24:1 (Fig. 4A and Supplementary Fig. 5A). By contrast, in response to FB1 treatment, glutamatergic and GABAergic cortical neurons both showed more robust significant reductions in ceramides (Supplementary Fig. 5). This may suggest that MNs have some distinct properties in maintaining ceramide stability, though we cannot exclude that differences in ceramide signatures could be partially driven by differences in cellular sizes and morphologies.

Labeling with d17:0-sphinganine confirms glial and neuronal ceramide synthesis rates

Perturbation of CerS and measures of overall ceramide levels may fail to adequately detect small changes in synthesis or the overall de novo synthesis capacity of each cell type. For a more sensitive measure of acute ceramide synthesis by CerS, we introduced a sphinganine precursor with an unnatural 17-carbon chain length (d17:0-sphinganine). This label is then incorporated into ceramide produced through CerS/de novo ceramide synthesis pathway. Across cell types, label incorporation was higher in dihydro-ceramides (dhCer) than in ceramides, consistent with the newly synthesized lipids incorporating the label, but labeled dihydroceramide not yet converted to ceramide by DEGS activity (Fig. 4–I). This, along with the lower label incorporation in FB1 treated cells, is indicative that the d17:0-sphinganine label is labeling ceramide generated through de novo synthesis as expected. There was a greater reduction in labeled ceramides with FB1 treatment than in unlabeled ceramides, consistent with labeled ceramides acting as a metric of acute ceramide synthesis (Fig. 4G–I).

Labeled ceramide was a more significant proportion of total ceramide in glia than MNs after two hours of incubation (Fig. 4G–I, Supplementary Fig. 2). This is consistent with our observation that MNs have significantly lower activity from multiple CerS than glial cells and show a minimal change in ceramide following FB1 treatment. This may also suggest that the high level of MN CerS1/4 activity in our enzyme assay may not directly parallel CerS1 kinetics in MNs in vitro, or account for ceramide incorporation into more complex lipids or its breakdown. Notably, newly synthesized lipids tracked cellular CerS activity in most cases. MNs had high CerS1/4 activity and low activity for other CerS (Fig. 3B), and correspondingly the limited newly synthesized ceramides were mostly dhCerC18 (Fig. 4G, Supplementary Fig. 3E, and F) which results from CerS1 activity. Consistent with these findings CerS5/6 (Fig. 3D) was maximal in astrocytes, which correspondingly had higher amounts of labeled dHCer16 and Cer16 (Fig. 4H). Microglia were high for activity of all CerS enzymes (Fig. 3) and had high levels of newly synthesized Cer16 and Cer24:1 (Fig. 4I). The main inconsistency between CerS-activity and ceramide labeling is the high level of microglial CerS1/4 (Fig. 3B) activity but no detectable levels of newly synthesized Cer18 (Fig. 4I). Future experiments will need to determine the relative contribution of CerS1 and CerS4 in regulating Cer18 levels in microglia. If CerS1 plays a more limited role than Cers4, which is suggested by our RNA-seq findings, this would be consistent with the essential role for CerS4 in regulating immune cells and the context specificity of the ceramide chain lengths generated by CerS4³⁷. This discrepancy could also be due to microglia having an increased capacity for hydrolyzing CerC18 or incorporating it into more complex sphingolipids and glycosphingolipids compared to other cell types.

Inhibition of glucosylceramide synthesis alters ceramide profiles in neurons and glia

The de novo ceramide synthesis pathway is highly dependent on CerS generation of dHCer and its subsequent conversion to ceramide by DEGS. FB1 was a logical tool to reduce ceramide synthesis and the addition of sphinganine allowed us to track de novo synthesis in these cell types. To induce ceramide accumulation in cells without the addition of an exogenous lipid, we treated cells with a glucosylceramide synthase inhibitor (GCSi; 667161) to prevent ceramide from being used to synthesize glucosylceramide (Fig. 1A). Given the differences in glial and MN dependence on de novo ceramide synthesis, we predicted that cellular differences in ceramide accumulation would likely occur after GCSi treatment. In astrocytes, GCSi treatment significantly increased CerC14 and CerC16 (Fig. 5B and E), consistent with high CerS5/6 activity and the reduction in the same ceramides with FB1 treatment. Microglia also had small accumulations in CerC14 and CerC16 in response to GCSi (Fig. 5C and F). The most dramatic increase in MNs after GCSi treatment was C18-ceramide in the cytoplasmic fraction (Fig. 5A and D), the expected chain length based on MN CerS1 activity and which ceramides were reduced after FB1 treatment. We note that microglia did not significantly accumulate CerC24 or Cer24:1(although there was a trend) in response to GCSi treatment as we would have expected from CerS-activity, ceramide labeling, and FB1 treatment. This observation also supports the idea that microglia are more readily adaptive to elevated ceramide levels (e.g., by increasing hydrolysis or increasing complex lipid synthesis) compared to astrocytes and MNs.

A–F To alter ceramide production through inhibition of downstream glucosylceramide production, MNs (A, D), astrocytes (B, E) and microglia (C, F) were treated with a GCSi (Genz-667161) and ceramides were measured in nuclear and cytoplasmic fractions. Each ceramide chain length vehicle vs GCSi treatment was analyzed by Student’s t-test, (****P < 0.0001, ***P < 0.001, *P < 0.05; n = 4 wells).

Bulk RNAseq of cells treated with FB1 to reduce ceramide synthesis and GCSi to induce ceramide accumulation

While we observed different gene expression profiles across cell-types by qPCR and RNA-seq, this did not determine which of their cellular functions depend on ceramide synthesis. Ceramide labeling, FB1 treatment, and GCSi treatment all revealed cell-type differences in ceramide homeostasis. To assess how ceramide accumulation influences cellular function we treated each cell type with the pan-CerS inhibitor FB1 or GCSi (Genz-667161), then performed RNA-seq. As in previous experiments, cells were cultured for 1-week post thaw then treated with 3 μM FB1 or Genz-667161 for 72 hours before lysis. After CerS inhibition, we observed many differentially expressed genes (DEGs) in both astrocytes (1820 genes) and microglia (1734 genes) (Fig. 6B–D). Notably, there were no DEGs in MNs after FB1 treatment, consistent with our data showing minimal reliance on ceramide synthesis in MNs relative to glia (Fig. 6A). GCSi induced significant DEGs in all three cell types (astrocytes 409 DEGs, Microglia 920 DEGs, and MNs 576 DEGs) (Fig. 6F–I).

A-E Motor neurons (MNs), astrocytes, and microglia were cultured for 7-days, then treated with 3 μM FB1 for 72-hours, as with previous protocols. We then isolated RNA and samples were sent for bulk sequencing. A–C Volcano plots of RNA-seq data in MNs, astrocytes, and microglia after FB1 treatment. B Astrocytes showed 1820 significant differentially expressed genes (sDEGs) in response to FB1 treatment. C Microglia showed 1,734 sDEGs in response to FB1 treatment. A MNs had no sDEGs. D In response to FB1 treatment, astrocytes and microglia had 143 common sDEGs. E The top 10 common IPA pathway changes in both astrocytes and microglia after FB1 treatment graphed by -log(p-value). F–H Volcano plots of RNA expression in astrocytes, microglia, and MNs after 72 hours 3 μM GCSi treatment. I Venn diagram of common gene expression changes after GCSi in MNs, astrocytes, and microglia. J, K Venn diagrams of genes that both reduce expression after FB1 treatment and increase expression after GCSi treatment in astrocytes or microglia.

Ceramide is implicated in many cellular functions, including glial reactivity, so we first looked at Ingenuity Pathway Analysis (IPA) to determine if ceramides had different functions across cell types³⁸. Our RNA-seq data indicate that ceramide levels may act as a key regulator of reactivity and inflammation. Within the top 10 differentially regulated IPA pathways in response to FB1, astrocytes had multiple pathways involved in inflammatory regulation (Supplementary Data 1: astrocyte FB1 treatment pathways). Microglial inflammatory pathways were similarly impacted by FB1 treatment (Supplementary Data 1: microglia FB1 treatment pathways). Additionally, consistent with their role in brain lipid regulation, FB1 significantly altered the Superpathway of Cholesterol Biosynthesis in astrocytes. Of interest, astrocytes also had pathway changes that could relate to axon guidance. Common pathway changes in response to FB1 treatment in both astrocytes and microglia showed themes in lipid synthesis, NAD regulation, and inflammation (Fig. 6E).

GCSi treatment altered many inflammatory pathways in astrocytes, similar to FB1 treatment (Supplementary Data 1: astrocyte pathways in GCSi treatment). Astrocytes also showed changes in lipid/cholesterol regulation after GCSi treatment (Hepatic Cholestasis, LXR/RXR Activation). GCSi induced a variety of inflammatory changes in microglia as well, many involved in inflammatory cell migration and infiltration (Supplementary Data 1: microglia pathways in GCSi treatment).

While MNs did not respond to FB1, there were a variety of pathway changes in response to GCSi. The changes centered around amino acid degradation, lipid/cholesterol regulation (LXR/RXR Activation and FXR/RXR Activation), and inflammatory responses. Consistent pathway changes across these cell-types indicate that ceramides may have key roles in regulating inflammation and cholesterol homeostasis.

To assess which pathways were commonly altered in response to ceramide changes, we analyzed the pathways that changed the most in response to FB1 and GCSi across all three cell types. Of the 20 most highly affected pathways, many were involved in inflammatory functions (including IL-33 Signaling Pathway, IL-17 Signaling). Many of these pathways were modified in opposing directions in response to FB1 and GCSi in astrocytes but changed in the same direction in microglia (Supplementary Fig. 7). Notably, IL-33 is known to be an astrocyte-derived cytokine capable of regulating microglial-synaptic pruning, and IL-17 similarly regulates synaptic activity^39,40.

Of interest were common and unique gene changes between cell types. In response to FB1 treatment, there were 143 DEGs in common between astrocytes and microglia (Supplementary Data 7). Of the 143 shared DEGs, 38 genes increased expression. These related to ER stress/apoptosis (STC2, CHAC1), inflammatory regulation (AREG, BEX2, TREML3P, FOXD1), glycolysis (GCKR), antioxidant activity (CHAC1), NAD metabolism/mitochondrial function (ALDH1L2, NMNAT2), and cellular adhesion/cytoskeletal dynamics/cellular motility (PCDH19, PAPPA). There were 41 downregulated genes in astrocytes and microglia in response to FB1 treatment. These included genes involved in cholesterol and fatty acid regulation (MSMO1, HMGCS1, CYP2U1-AS1), inflammatory regulation (RNASE1, CXCL14, TGFB2-AS1, PTGES3L), oxidative stress (MAOB), axon and synaptic development (CBLN2, TMEM108), sulfatation (ARSI), and cellular adhesion/cytoskeletal dynamics/cellular motility (DNAH6, MYH2, ADAM19, ADAMTS12, CLDN5, TPM2).

In response to GCSi treatment, there were no common DEGs amongst all three cell-types. However, GCSi treatment induced 24 common DEGs between astrocytes and microglia, astrocytes and MNs shared 2 DEGs, and microglia and MNs had 16 common DEGs (Supplementary Data 8). The 16 common DEGs between microglia and MNs had functions in inflammatory regulation (IL18RAP, GIMAP4, CD1B, TAFA3) and ubiquitin degradation of proteins (ASB15). CD1B is of particular interest as it functions to present lipid and glycolipid self-antigens⁴¹. The 24 common DEGs between microglia and astrocytes had functions in the extracellular matrix and blood clotting regulation (TFIP2, SERPINE1, PCED1B, CLDN5, SPON1), inflammatory regulation (IL11, CLCF1), and apoptosis (DTHD1). It is notable that both IL11 and CLCF1 had increased expression in response to GCSi and both promote oligodendrocyte survival^42,43,44. It is possible that lipid stress induced by GCSi treatment is detected by glial cells as a sign of dysmyelination or demyelination. In response to GCSi astrocytes and MNs had only 2 DEGs, PRRG and KCNE1B.

Ceramide precursors are differentially toxic between glia and neurons

We showed that ceramide synthesis is significantly different between iPSC-derived cell types. Additionally, glia and neurons incorporated sphinganine into ceramide chain lengths differently. In our RNA-seq data, CerS inhibition and GCSi-induced gene changes related to ER stress, ferroptosis, and apoptosis. Sphingolipids are dysregulated in a variety of neurological diseases, and we showed that sphinganine can be turned into ceramide when applied to glia and, to a lesser extent, neurons. We therefore treated cells to alter ceramide and/or sphinganine levels, and analyzed the resulting toxicity by Cell-Titer Glo (CTG), DAPI/cytotox staining, and neurite analysis.

We first sought to determine the toxic doses of GCSi (Genz-667161), as these may directly relate to the ceramide accumulation seen in Fig. 5. Despite having large changes in nuclear ceramides with GCSi treatment (Fig. 5E), astrocytes were more resistant to GCSi toxicity than MNs and microglia (Fig. 7A–C). We did see small decreases in astroglial viability at 72 hours (Fig. 7B), but both microglia and MNs showed more significant toxicity (Fig. 7A and C).

A–C Astrocytes, microglia, and motor neurons (MNs) were cultured for 7-days, then treated with increasing doses of GCSi. D–F Sphinganine was dosed into cells at increasing concentrations and Incucyte live imaging and analysis was performed. Representative images show MN cell-death and disintegrating neurites, microglia showed signs of cell death, while astrocytes remained largely resilient to 10 μM sphinganine treatment (scale bars 200 μm). G–I Cell titer glo (CTG) analysis of MN cell viability was consistent with neurite findings that toxicity starts at 10 μM doses with similar levels of toxicity in microglia, but resistance to 10 μM dose toxicity in astrocytes. J Live imaging of MNs and analysis of neurites by phase microscopy revealed acute toxicity at 10 μM and higher (transparent and opaque datapoints correspond to the individual replicate and mean values, respectively). Data analyze by one-way ANOVA with Bonferroni’s multiple comparisons test (n = 4 or 6 technical replicates as depicted in graphs).

To alter ceramide levels by a different mechanism, we treated cells with increasing sphinganine concentrations to determine the effects on the viability of different cell-types. Both sphinganine (Fig. 7D, G, and J, Supplementary Fig. 8A–C) and sphingosine (Supplementary Fig. 8D–G) were significantly toxic to neurons at 10 μM doses, as assessed by CTG, DAPI/cytotox staining, and neurite analysis. High doses were acutely toxic and 10 μM sphinganine treatment showed more acute toxicity than 10 μM sphingosine.

Sphinganine was toxic to astrocytes as well, but only starting at the 30 μM dose (Fig. 7E and H, Supplementary Fig. 9A, C, and D). This illustrates the potential resistance of astroglial cells to this sphingolipid driven toxicity. We therefore treated microglia to see if other glia had similar resistance to toxicity. Microglia had higher mRNA expression of ceramidases, so we anticipated that microglia might have heightened resistance to sphinganine toxicity. In contrast to our expectations, and similar to MNs, microglia underwent cell death starting at 10 μM doses (Fig. 7F and I, Supplementary Fig. 9E–G). We also observed significant acute morphological changes in microglial cells in response to sub-toxic doses (3 μM) of sphinganine (Supplementary videos 1 and 2). These morphological changes normalized over time, though there were signs of intracellular inclusion (possibly lipid droplets). It’s known that lipid accumulation can alter microglial activity⁴⁵, but future studies will need to ascertain whether sphinganine similarly alters their activity.

Liposomal delivery significantly alters sphinganine delivery to neurons

Bulk application of free lipids is not the most physiological delivery and therefore may not recapitulate in vivo lipid-related toxicity. There is evidence that astrocytes can release lipoparticles and/or exosomes containing toxic lipids including ceramides^28,46. Therefore, we repeated our sphingolipid treatment assay with the delivery of sphinganine, but in a lipoparticle/liposomal format. NBD-sphinganine was mixed with POPC in a 5:95 ratio and delivered to cells as liposomes or liposomes bound to APOE. We compared this delivery to liposomes lacking sphinganine (with and without APOE) as well as the delivery of free NBD-sphinganine at doses equivalent to that of NBD-sphinganine delivered by liposomes. NBD (green) allowed us to track the rate of incorporation and localization of the NBD-sphinganine within MNs, though we can’t rule out that the sphinganine could be incorporated into other lipids and/or metabolized in some way.

We observe that treatment of MNs with NBD-sphinganine (NBD-SA) in both a free and liposomal (LP) format led to NBD localization in MNs, with no green signal detected in POPC treated neurons (Fig. 8B–E and Supplementary Fig. 10A). A large portion of NBD signal in free NBD-sphinganine treatment localized to cell bodies and dead cell clumps. It is difficult to remove these clumps without negatively impacting live MN health. We can’t exclude that a large portion of free NBD-sphinganine signal in our analyses could localize to dead cells. Notably, we saw many neurites that were NBD⁺ in the liposomal treated cells (NBD-SA + POPC + APOE LP; Fig. 8B, C, and E).

A Diagrams illustrating the treatment of motor neurons (MNs) with liposomes with NBD-labeled (green) sphinganine and liposomes lacking sphinganine. These were delivered either bound or not bound to APOE. These treatments were compared to the delivery of free NBD-sphinganine. B, C To quantitatively assess if liposomal delivery increased NBD-label incorporation into neurites, we performed an analysis for green-labeled neurites. Graphs illustrate the delivery of 1.0 μM and 10.0 μM of NBD-sphinganine in both free and liposomal forms, with increased NBD⁺ neurite area in liposomal delivery regardless of APOE status (transparent and opaque data points correspond to the individual replicate and mean values, respectively). Dotted line in B is the timepoint images taken for D, E 8 h post-treatment, showing significant NBD labeling in neurites when NBD-sphinganine was delivered in liposomes (scale bars 200 μm) (E) but did not see the same extent of labeling in the free NBD-sphinganine (D) delivery condition. No green labeling was observed in neurons treated with liposomes lacking NBD-sphinganine (Supplementary Fig. 10). We noted F, G Cell titer glo (CTG) analysis showed that NBD-sphinganine had a similar toxicity curve when delivered freely or delivered in liposomes. Notably, liposomes without NBD-sphinganine showed minimal toxicity (Supplementary Fig. 10), showing that APOE and the carrier lipid POPC (19x the concentration of NBD-sphinganine in the liposome) are not contributing significantly to lipid toxicity. Diagrams generated in BioRender.

After 72 hours of treatment, CTG was performed for cell viability. POPC and APOE were kept in consistent proportion to the sphinganine. NBD-sphinganine was similarly toxic when delivered freely or in liposomal formulations, regardless of whether the liposomes had APOE bound or not (Fig. 8F and G, Supplementary Fig. 10C and D). We note that POPC delivered in the same doses, but lacking sphinganine, did not show significant toxicity (Supplementary Fig. 10A and B). As POPC was a significantly higher proportion of the liposome than sphinganine, this indicates that bulk lipid application alone does not drive toxicity.

Based on the finding that neurites were NBD⁺, we utilized Incucyte neurite analysis for NBD⁺ (green) neurites to assess lipid incorporation. At multiple doses, liposomal delivery of NBD-sphinganine induced significantly more NBD incorporation into neurites than free NBD-sphinganine delivery (Fig. 8B–E, Supplementary Fig. 10A and E). APOE did not significantly alter this labeling. This might indicate better incorporation of the lipid into the whole cell or that delivery of the lipid through a different pathway might alter neurite incorporation.

We additionally analyzed neurites by phase microscopy in cells treated with liposomal NBD-sphinganine, as a measure of neuronal health. Consistent with our previous free sphinganine treatment experiment (Fig. 7 and Supplementary Fig. 8), there was acute neurite loss in response to sphinganine treatment and then a plateau, as dead neurites were not retracted and were still picked up by the analysis software. Surprisingly, the analysis detected more neurite loss in neurons treated with NBD-sphinganine liposomes bound with APOE (Supplementary Fig. 10C and D). In both instances, the neurites appeared dead but without APOE the neurite debris stayed in place, whereas there might have been more neurite retraction with APOE. This was a subtle effect picked up by coarse phase analysis, so future studies will need to better characterize the effect and if APOE alters neurodegeneration mechanisms.