Given that C3+ neurotoxic astrocytes have been shown to be pathological in rodent models of MS (Gharagozloo et al., 2021), and enriched in persons with MS (PwMS) (Liddelow et al., 2017; Gharagozloo et al., 2021), we wished to determine whether induced pluripotent stem cell (iPSC) derived astrocytes from PwMS would respond differently to pro-inflammatory signals compared to NMSCs. Using hiPSCs reprogrammed from peripheral blood mononuclear cells (PBMCs) from seven PwMS and six NMSCs (seven controls were initiated but one failed quality control at the transcript stage) (Table 1), we differentiated them to a neural lineage using previously published methods (Paull et al., 2015; Barbar et al., 2020a; Figures 1A,B). After sorting on CD49f to enrich for astrocytes, we stimulated them with TNF and IL1α (TI) to promote a neurotoxic phenotype. Astrocytes were treated with or without TI for 48 h then RNA was collected and subjected to RNAseq analysis. Principle component analysis of the top 500 most variable genes showed the primary driver of variance was the addition of TI (Figure 1C). Differential expression testing between samples treated with TI and their paired untreated controls found 2228 upregulated genes and 1948 downregulated genes (Figure 1D and Supplementary Table 1). Examining relative expression of genes previously used to distinguish neurotoxic astrocytes from those associated with hypoxia in rodents showed an overlapping but distinct phenotype present in these human astrocytes, although the PwMS and NMSC samples clustered together suggesting no differences between them (Figure 1E). This lack of difference in TI response was further illustrated when we compared the fold change of genes in the NMSC group to the PwMS group relative to their −TI controls. The few genes that appear to respond differently are primarily ribosomal RNA present in an individual sample, likely due to incomplete ribosomal depletion in the RNAseq library preparation (Figure 1F and Supplementary Figure 1). Given women are more likely to develop MS than men, we assessed whether there was a difference in the response to +TI stimulation between astrocytes derived from females relative to those derived from males. Comparing the fold change of genes between the two sexes showed high similarity again except for a few extreme outliers driven by inefficient ribosomal RNA depletion in isolated samples (Supplementary Figure 2A). Seventy-five genes did differ moderately in their response to +TI stimulus in females compared to males (adjusted p < 0.05, Supplementary Figure 2B and Supplementary Table 2). While some of those were sex-linked genes (e.g., NLGN4Y), many were related to immune responses to cytokine signaling and were induced less in females relative to males (Supplementary Figures 2B–F).

To better understand the functional changes induced in human astrocytes following TI stimulation we performed gene ontology (GO) term enrichment analysis on genes found to be up-regulated or down-regulated when comparing the +TI to −TI conditions in all 13 cell lines. As expected, many of the enriched terms in the up-regulated genes were associated with inflammatory and anti-viral responses (Figure 2A and Supplementary Table 3), while many of the terms enriched in the down-regulated genes were associated with axonal support and nervous system development (Figure 2B and Supplementary Table 4). Further examination of individual genes associated with a select set of non-overlapping up-regulated (Figures 2C,D) and down-regulated (Figures 2E,F) terms again supported the prior results that PwMS and NMSC respond similarly to TI.

To compare the effect of TI on hiPSC astrocytes in this current dataset with previously published findings, we retrieved RNAseq data from rat neonatal astrocytes stimulated with TNF, IL1α, and C1q (TIC) for 24 h (Hasel et al., 2021) and hiPSC astrocytes stimulated with TIC for 24 h (Barbar et al., 2020a). As expected, there were many similarities between the three datasets, but also robust differences particularly between human and rat datasets that are not explained by difference in treatment (TI vs. TIC) or time (24 h vs 48 h) with only 42% (379/961) of up-regulated and 34% (468/1366) of down-regulated genes in rat also overlapping with either human dataset (Supplementary Figure 3). We also compared the transcriptomic profile of hiPSC astrocytes with or without TI stimulation to previously published single nucleus RNA sequencing (snRNA-seq) dataset derived from post-mortem human brain tissue including both controls and MS lesions (Absinta et al., 2021). Using non-parametric ranked correlations, we saw that all hiPSC astrocytes, regardless of TI stimulation, most resembled the inflamed and reactive astrocyte clusters (Supplementary Figure 4). The most pronounced effect of TI stimulation was to make them less like the non-reactive and senescent astrocyte clusters.

In addition to their ability to kill neurons and mature oligodendrocytes, neurotoxic astrocytes have also been shown to suppress rodent OPC differentiation (Liddelow et al., 2017). In addition, hiPSC derived astrocytes with GFAP mutations have been shown to suppress human OPC differentiation in an in vitro model of Alexander’s disease (Li et al., 2018). To examine whether there was any difference between NMSC and PwMS in their oligo suppressive capabilities, we utilized a recently reported genetically modified reporter hESC line (Li et al., 2022). The reporter cells (from hereon called hOPC reporter cells) express rodent surface marker Thy1.2 and tdTomato under the control of the endogenous human PDGFRA promoter and a secreted Nanoluciferase (NLuc) under the control of the endogenous human myelin basic protein (MBP) promoter. This line allows for the enrichment of OPCs by selecting on Thy1 and then monitoring differentiation into oligodendrocytes by quantifying NLuc activity in the culture media (Li et al., 2022). We tested the ability of TI stimulated human astrocytes to suppress human OPC differentiation by collecting conditioned media (CM) from hiPSC derived astrocytes following 48 h of stimulation and treating hOPC reporter cells with that CM for 5–8 days (Figure 3A). To control for the possible effects of residual TI acting directly on hOPC reporter cells, media only controls with or without TI that were never in contact with astrocytes were used. The response to +TI CM did not differ between PwMS and NMSC (p = 0.48) or between females and males (p = 0.33). Across PwMS and NMSC, we found that CM from +TI treated astrocytes robustly suppressed hOPC reporter cell differentiation (Figure 3B, β_+TI = −0.786, standard error = 0.052, p < 0.0001). To determine whether the decreased amount of secreted luciferase was due to hOPC reporter cell death, we performed automated nuclear counts in a subset of the hOPC wells following treatment with astrocyte CM (Figure 3C). Performing similar analyses, we found no evidence of a different response between PwMS and NMSC (p = 0.47). We also saw no evidence that +TI CM reduced cell survival compared to −TI controls. In fact, we saw the opposite pattern, where hOPC wells treated with CM from +TI astrocytes had more nuclei (β_+TI = 88.0, standard error = 19.3, p < 0.0001), although the effect was modest.

Since the +TI CM was suppressing differentiation of human OPCs without killing them, we next performed RNAseq on the hOPC reporter cells following 72 h of treatment. Principle component analysis of the top 500 variable genes showed that hOPC reporter cells treated with CM from +TI treated astrocytes segregated separately from wells treated with CM from −TI treated astrocytes and the media only controls (Figure 3D). We then performed differential expression analysis to identify transcripts that are significantly altered in the +TI CM treated OPCs when compared to −TI CM as compared to the media only conditions. This analysis found 673 genes that were up-regulated and 205 genes that were down-regulated in +TI CM treated OPCs (Supplementary Table 5). Examining relative expression levels of a selection of genes we and others have previously reported to be differentially expressed in human and rodent OPCs following IFNγ stimulation (Chew et al., 2005; Kirby et al., 2019; Morales Pantoja et al., 2020), we saw consistency with upregulation of anti-viral response pathways, NF-κB signaling, and cell cycle progression and downregulation of myelin genes and glial differentiation (consistent with above luciferase reporter results). Indeed, GO term enrichment analysis in the up-regulated (Supplementary Table 6) and down-regulated (Supplementary Table 7) further supported these findings (Figures 3F,G).

Peripheral blood mononuclear cells were isolated from whole blood samples collected using gradient density separation using SepMate tubes (STEMCELL Technologies, 85450) and Lymphoprep (STEMCELL Technologies, 07801). PBMCs from PwMS were obtained from donors with the approval of the Johns Hopkins University Institutional Review Board. PBMCs from NMSC were obtained with approval by the New York Stem Cell Foundation’s external Institutional Review Board, WCG IRB (#20112091). All participants signed a consent and/or a repository consent as required by the overseeing IRB to allow their data and biospecimens to be repurposed.

Peripheral blood mononuclear cells were thawed and recovered in StemPro-34 SFM Complete Medium (Thermo Fisher, 10639-011) supplemented with cytokines SCF (200 ng/μL, Thermo Fisher, PHC2111), Flt3 (200 ng/μL, Thermo fisher, PHC9411), IL3 (40 ng/μL, Thermo Fisher, PHC0034), and IL6 (40 ng/μL, Thermo Fisher, PHC0065), GlutaMAX (Thermo fisher, 35050-061) overnight as previously described (Zhou et al., 2015). The next day, a 96-well flat bottom plate was coated with Cultrex (HESC qualified Cultrex, Trevigen, 3434-0001-02) at 1:10 dilution and warmed for 1 h at 37°C. Cells were transferred into Cultrex coated 96-well flat bottom plates at 60K and 100K seeding density for reprogramming using CytoTune-iPS Sendai Reprogramming v2.0 Kit (Thermo Fisher, A16517) per the manufacturer’s recommendations, modified for cell number and plate format. After infection, cells were gradually transitioned to Freedom media (DMEM-F12 with Freedom-1 Supplement, Life Technologies, Custom) for 5 days post-infection with daily media changes on the NYSCF Global Stem Cell Array platform (Paull et al., 2015). Live cell surface staining using the iPSC marker Tra-1-60 (Tra-1-60 Antibody, Life Technologies) was performed 12–14 days post-transfection to identify reprogrammed cells. Successfully reprogrammed cell lines are consolidated in a 96-well Cultrex coated plate and stored in LN2 upon reaching confluency. Sendai reprogrammed iPS cell lines undergo enrichment and monoclonalization. This process utilizes FACS sorting on +CD56, +CD13, +Tra-1-60, +SSEA4 to first bulk sort newly reprogrammed lines to increase the iPSC population (enrichment) followed by single-cell sorting and machine learning augmented clonality assessment (Fischbacher et al., 2021) in order to select and consolidate monoclonalized iPSC lines. Monoclonalized iPSCs were then expanded via automation on the NYSCF Global Stem Cell Array platform for further quality control assays and then frozen into barcoded Matrix tubes in Synth-a-Freeze Cryopreservation Media at R500K cells/vial. All iPSC lines undergo rigorous quality control that includes a sterility check, mycoplasma testing, viability, karyotyping via Illumina Global Screening Array, SNP ID fingerprinting via Fluidigm SNPTrace, pluripotency and embryoid body scorecard assays via NanoString. iPSCs were maintained using Freedom media.

Astrocytes were differentiated from iPSC precursors as previously described (Barbar et al., 2020a). Briefly, iPSCs were induced to neural stem cells (NSCs) within the first 12 days of differentiation. NSCs were expanded as floating neurospheres until day 30, when they were picked and plated onto poly-L-ornithine (0.1 mg/ml) and laminin (10 μg/ml)-coated dishes, at about 40 neurospheres per well of a 6-well plate. Cultures were maintained in PDGF medium for 4 weeks to allow for progenitor cells migration and differentiation toward astrocytes. PDGF medium composition: DMEM/F12, PenStrep (100×), 2-mercaptoethanol (1000×), MEM non-essential amino acids, N2 supplement (100×), B27 without VitA (50×), human insulin solution (25 μg/ml), PDGFaa (10 ng/ml), IGF-1 (10 ng/ml), HGF (5 g/ml), NT3 (10 ng/ml, T3 (60 ng/ml, biotin (100 ng/ml, cAMP (1 μM). Around day 70, cultures were dissociated through enzymatic digestion (Accutase, Thermo Fisher; A1110501) for 30 min at 37°C, and then filtered using a 70 μm filter (STEMCELL Technologies; 27260). Cells were spun at 300 g for 5 min at 4°C, resuspended in FACS buffer (PBS, 0.5% BSA, 2 mM EDTA, 20 mM Glucose) and incubated for 20 min with CD49f antibody (1:50; BD bioscience, 555736) for FACS purification. Sorted CD49f⁺ astrocytes were replated onto poly-L-ornithine/laminin-coated 24-well plates at 250K cells/well. Astrocytes were maintained in Glial medium (PDGF medium without the growth factors PDGF, IGF-1, HGF, NT3) for 24 h and then stimulated with TNFα (30 ng/ml; R&D system, 210-TA-020) and IL1α (3 ng/ml; Sigma, I3901) in Brainphys medium (STEMCELL Technologies, 05790) with B27 supplement minus antioxidants (Thermo Fisher, 10889038) for 48 h. CM from stimulated cultures and unstimulated controls were collected and frozen at −80°C until use, while cells were lysed for RNA isolation. For further details on astrocyte culture and stimulation to a neurotoxic reactive state see our previous publications (Barbar et al., 2020a,b).

Cells (hiPSC derived astrocytes and reporter hOPCs) were lysed in RLT plus buffer and RNA was isolated using RNEasy Plus Micro kit (Qiagen 74034). Quality and quantity of RNA was determined using a NanoDrop and Fragment Analyzer. Of the 14 pairs of samples from iPSC derived astrocytes, 13 were sufficient to proceed to library preparation. Of the 10 pairs of samples from hOPC reporter cells, all 10 were sufficient to proceed. RNAseq libraries were prepared using Illumina Stranded Total RNA Prep with Ribo-Zero Plus kit (Illumina 20040525). Libraries were sequenced using 100 bp paired end configuration on Illumina NovaSeq. Quality of sequencing data was checked with FastQC (Andrews, 2017) then MultiQC (Ewels et al., 2016) and no data was discarded or trimmed. Transcript level estimated counts were acquired with Salmon (v1.4.0) (Patro et al., 2017) using default options but with the gcBias flag enabled. For selective alignment, the entire genome was used as a decoy sequence (Srivastava et al., 2020). RefSeq GRCh38.p12 was used as the reference genome and transcriptome (O’Leary et al., 2016). Estimated transcript level counts were then imported into R (v4.1.2), adjusted for transcript length, and aggregated to gene level using tximeta (v1.12.3) (Love et al., 2020). Count normalization and differential expression testing was done with DESeq2 (v1.34.0) (Love et al., 2014). For the astrocyte experiment, the contrast was (+TI to −TI) with blocking for individual in the model. For the OPC experiment, the contrast was [(astrocytes +TI to astrocytes −TI)–(media_only +TI to media_only −TI)] again with blocking for individual in the model. Log fold change was adaptively shrunk using the ashr package (v2.2-47) in DESeq2 (Stephens, 2017). Genes were considered differentially expressed if they had an adjusted p-value less than 0.05 and a Log₂ Fold Change greater than 1 or less than −1 in the astrocyte experiment and greater than 0.5 or less than −0.5 in the hOPC reporter experiment. Volcano plot in Figure 1 was generated using the EnhancedVolcano package (v1.12.0) (Blighe et al., 2021). Gene Ontology enrichment analysis was conducted with clusterProfiler (v4.2.1) (Yu et al., 2012) using the enrichGO function with default options against the “Biological Process” GO library (Ashburner et al., 2000; Gene Ontology, 2021). Heatmaps were generated with pheatmap (v1.0.12) using the rlog scaled data derived from DESeq2. Unsupervised clustering in the heatmaps was performed using complete linkage method.

RNAseq data for rat astrocytes treated with TIC was obtained from the Gene Expression Omnibus (GEO) under accession GSE165069 (Hasel et al., 2021). All conditions were retrieved and included in the model but differential expression testing was only done to compare TIC vs. vehicle. RNAseq data from previously published hiPSC derived astrocytes treated with TIC was obtained from the Synapse open source platform under accession syn21861229 (Barbar et al., 2020a). Analysis of both previous datasets followed same method as above. Rat data was quantified using mRatBN7.2 assembly (Howe et al., 2021). For comparing between datasets, genes with 1:1 orthologs from human to rat were identified with biomaRt (Durinck et al., 2009) and only those genes were used in the cross-species comparisons. For pathway enrichment of genes differentially expressed exclusively in rat or human, genes were only included if they were differentially expressed in both human datasets (for those exclusive to human) or in neither (for those exclusive to rat). To avoid including genes that were differentially expressed but that did not meet the fold change threshold, a gene was only considered exclusive if the ortholog in the other species either (1) did not have an adjusted p-value less than 0.05 or (2) the Log2 Fold Change was in the wrong direction, i.e., >0 if being compared to down regulated genes in other species or <0 if being compared to up regulated genes in other species. The snRNA-seq data was obtained from GEO under accession GSE180759 (Absinta et al., 2021). Using Seurat (v4.1.0) (Satija et al., 2015), cells previously identified as astrocytes in the original publication were subsetted and the top 1000 highly variable genes were identified using log-normalized data. Pseudobulk gene expression levels for each cluster were calculated based on the previously published cluster identities for each cell. Spearman correlation coefficients were calculated for each hiPSC astrocyte sample and snRNA-seq cluster pairing using all 20,622 genes detected in both datasets or only the top 1000 highly variable genes in snRNA-seq astrocyte dataset with R (v4.1.2).

Previously published human ESC reporter cell lines for OPC (Chamling et al., 2021) and oligodendrocyte (Li et al., 2022) differentiation were used for the study. The reporter cell line used for this study contains PDGFRa-P2A-tdTomato-P2A-Thy1.2 and MBP-P2A-secNLuc reporters. In these reporter cells, upon PDGFRa expression, tdTomato as well as Thy1.2 protein product are produced. Since Thy1.2 is a surface protein, it migrates to the cell surface, allowing the PDGFRa expressing cells to be immunopurified via Thy1.2 antibody conjugated magnetic microbeads (Chamling et al., 2021; Li et al., 2022). In addition, since the endogenous MBP promoter drives the expression of NLuc (Promega) that is secreted into the cell culture media, MBP expression can be quantitated by measuring NLuc activity in the cell culture media (Li et al., 2022).

The hESC reporters were differentiated into OPCs by following the previously published detailed protocol (Douvaras and Fossati, 2015). Briefly, hESCs were plated on Matrigel and maintained in mTeSR plus. Neural differentiation and spinal cord patterning was induced through dual SMAD inhibition (SB431542, 10 μM and LDN193189, 250 nM) and 100 nM all-trans retinoic acid for 8 days. From day 8 to day 12, differentiating cells were maintained in neural induction media supplemented with RA (100 nM) and SAG (1 mM). At day 12, cells were lifted and cultured in low-attachment plates to support sphere aggregation. At day 30, spheres were plated into poly-L-ornithine/laminin-coated dishes in a media supplemented with B27 (Thermo Fisher, 12587010), N2 supplement (Thermo Fisher, 17502048), PDGF-AA (221-AA-10, R&D systems), neurotrophin-3, HGF (294-HG-025 R&D systems), and T3. Following this differentiation protocol, PDGFRa-tdTomato/Thy1.2+ cells are visible as early as day 45 of differentiation, PLP1-GFP and NanoLuc activity is detected from day 60 onward.

The hOPC Nluc reporter assay was performed as previously described (Li et al., 2022). Day 85 differentiating culture was MACS purified with Thy1.2 (CD90.2 microbeads, Miltenyi Biotech) to enrich for PDGFRA-tdTomato+ OPCs (Chamling et al., 2021). 1.5K cells/well of the OPCs were plated in PLO-laminin coated 384 well plates in 50 μL of glial differentiation media. Two days after plating the cells, the culture media was replaced with glial media containing 1/3 of the astrocyte condition media (CM) (i.e., 30 μL of astrocyte CM plus 60 μL of glial media, total 90 μL per well of a 384 well plate) using integra ViaFlow 384.

Myelin basic protein-Nluc activity was measured using NanoGlo luciferase assay reagents (Promega N1150). Briefly, 20 μL of cell culture media from an hOPC culture was collected in a 384 well plate. NanoGlo reaction mix was prepared by mixing Nano-Glo^® Luciferase Assay Substrate and Nano-Glo^® Luciferase Assay Buffer (1:50), and the reaction mix was further diluted (1:1) with water. To measure Nluc activity, 5 μL of the diluted reaction mix was added to each well containing the 20 μL culture media. Nluc activity was measured as relative light unit (RLU) with a microplate reader (ClariOstar, BMG LABTECH) using the preloaded settings for Nano luciferase and 1 s exposure.

Cells were fixed with 4% PFA, washed with PBS and either stored in 4°C or immediately imaged. Prior to imaging, Hoechst 33342 nucleic acid stain (Thermo Fisher) was diluted to 1:10,000 in PBS and added to each well containing the cells. Image was captured with High-Content Imager (Cellomics CX7) using 10× magnification and nine fields to cover the entire well. A built-in algorithm of the ArrayScan image analysis software, which is a part of the Cellomics CX7 (Thermo Fisher Scientific), was used to count the total number of nuclei per well.

Statistical analyses were performed with R version 4.1.2. Normalized NLuc and nuclear count results (Figure 3) were analyzed with linear mixed-effects models. Normalized NLuc readings or nuclear counts were included as the response, condition (+TI or −TI) and group (NMSC or PwMS) as fixed effects, and technical replicates nested within individuals as random effects. To test whether there was a difference in the effect of +TI between the NMSD and PwMS groups, models were also fitted included an interaction term between condition and group, but no significant difference was observed (likelihood ratio test p > 0.05).