CRISPRi-mediated gene knockdown in iAstrocytes

To investigate the effect of genetic perturbations on inflammatory reactivity in iAstrocytes, we generated iAstrocytes from hiPSCs stably expressing CRISPRi machinery (Fig. 2a), which enables specific and non-toxic knockdown of genes targeted by single guide RNAs (sgRNAs) delivered to the cell, and thus enables large-scale pooled genetic screens²⁴. We found that these iAstrocytes maintained robust CRISPRi activity as measured by near 100% knockdown of cell-surface TFRC (Fig. 2b–c).

Open in a separate window

Fig. 2 |

CRISPR interference (CRISPRi) platform in iAstrocytes.

a, Schematic of CRISPRi machinery cassette integrated into the CLYBL safe-harbor locus. b, Workflow schematic of experiments involving lentiviral sgRNA transduction of iAstrocytes for CRISPRi knockdown. c, Cell-surface TFRC levels (measured by flow cytometry using an anti-TFRC antibody or isotype control; MFI: median fluorescence intensity) in iAstrocytes transduced with a non-targeting control (NTC) sgRNA or a sgRNA targeting TFRC, treated with vehicle control or IL-1α+TNF+C1q (n = 4 wells for anti-TFRC, n = 2 wells for isotype control).

CRISPRi screening and computational master regulator analysis

Having established CRISPRi in iAstrocytes, we aimed to systematically interrogate IL-1α+TNF+C1q-induced inflammatory reactivity in iAstrocytes with two orthogonal approaches – pooled CRISPRi screening^{15, 25} and computational master regulator analysis (MRA)^26–28 (Fig. 3a,​,b;b; see Supplementary Text for details).

Open in a separate window

Fig. 3 |

CRISPRi screening and master regulator analysis uncover regulators of inflammatory reactivity.

a, Workflow schematic of synaptosome phagocytosis and cell-surface VCAM1 CRISPRi screens (n = 2 independent screens). b, Workflow schematic of bioinformatic master regulator analysis (MRA). c, f, Clustering of transcription factors (c) or kinases and phosphatases (f) predicted to regulate inflammatory reactivity based on regulon overlap (see Methods); the activity score and regulon mean log-fold-change (LFC) associated with each predicted regulator (see Methods) are shown below the dendrogram. d, g, Scatterplot of gene scores (see Methods) of transcription factors (d) or the druggable genome (g) from synaptosome phagocytosis screens on iAstrocytes treated with vehicle control vs. IL-1α+TNF+C1q. e, h, Volcano plot of phenotype scores and associated log-scaled P values (see Methods) of transcription factors (e) or the druggable genome (h) from cell-surface VCAM1 screens on iAstrocytes treated with IL-1α+TNF+C1q. P values in panels e and h were calculated using the two-sided Mann-Whitney U test (see Methods).

We performed MRA using the human transcription factors²⁹ (Fig. 3c) or kinases and phosphatases^{30, 31} (Fig. 3f) as upstream regulators, and similarly conducted pooled CRISPRi screening using a custom sgRNA library targeting the human transcription factors (“TF library; Fig. 3d–e) or a library targeting all kinases and phosphatases as well other genes representing the “druggable genome”³² (“H1 library”; Fig. 3g–h). Together, these two approaches uncovered both expected regulators of inflammatory reactivity and potentially novel regulators.

For example, the top hits from both MRA and CRISPRi screens (Supplementary Tables 2–3) included the entire set of genes encoding all potential subunits of the NF-κB transcription factor (REL, RELA, RELB, NFKB1, NFKB2; Fig. 3c–e) as well as genes encoding the upstream kinases of the canonical NF-κB pathway (IKBKB, CHUK, IKBKG; Fig. 3f–h)³³. For all the above genes (except for NFKB2), the directionality of their screening phenotypes or MRA activity scores (see Methods) was consistent with activation of the canonical NF-kB pathway being required for the induction of inflammatory reactivity (see Supplementary Text).

Given that the canonical NF-κB pathway controls cellular responses to inflammatory stimuli³⁴, its strong enrichment in the top hits from the screens and MRA serves as a positive control for the technical quality of the screens (see Extended Data Fig. 4a for quality control data) and the validity of the MRA pipeline. As a further control for the technical quality of the screens, we validated the phenotypes of selected top hits in independent experiments (Extended Data Fig. 4c–d).

In addition to the canonical NF-κB pathway, we also recovered genes involved in numerous other cellular pathways known to mediate inflammatory processes. One group of genes (STAT3, CEBPB, CEBPD) consisted of transcription factors classically associated with the acute phase response in systemic inflammation^{35, 36} which also play important roles in the CNS^{35,37–39,40, 41} (see Supplementary Text). We found that knockdown of STAT3, CEBPB, and CEBPD with IL-1α+TNF+C1q treatment surprisingly promoted the induction VCAM1 while having divergent effects on phagocytosis (Fig. 3d,​,e;e; Extended Data Fig. 4b).

Another group of genes consisted of those involved in the response to interferons (IRF1, STAT1, IKBKE)^42–44. For example, we found that knockdown of IRF1 rescued decreased phagocytosis and reduced VCAM1 upregulation (Fig. 3d,​,ee,​,h;h; Extended Data Fig. 4b). Given that interferons have been shown to potentiate the induction of VCAM1 by inflammatory cytokines^{21, 45}, the reduction of VCAM1 upregulation with IRF1 knockdown suggests that interferons are released by iAstrocytes after IL-1α+TNF+C1q treatment. Interestingly, we observed that knockdown of genes encoding kinases responsible for signal transduction from the IL-1 receptor (IRAK2, IRAK4)⁴⁶ strongly increased VCAM1 induction (Fig. 3h), consistent with the fact that IL-1 signaling is known to antagonize the action of interferons⁴⁶.

Lastly, we found genes involved in cellular pathways that have been relatively less studied in the context of inflammatory astrocyte reactivity, such as genes involved in or related to the mTOR pathway (MTOR, LAMTOR3, LATS2, FOXK1)^{47, 48}, the glucocorticoid receptor pathway (NR3C1), the actin cytoskeleton (ARPC3, ACTR2), and also relatively uncharacterized genes (FOXC2 and ZBTB7A).

Overall, the results from CRISPRi screening and MRA demonstrate that numerous cellular pathways regulate distinct aspects of inflammatory reactivity induced by IL-1α+TNF+C1q. We noted that knockdown of the acute phase response-related genes STAT3, CEBPB, or CEBPD increased the induction of VCAM1 by IL-1α+TNF+C1q, whereas knockdown of the interferon-related gene IRF1 decreased VCAM1 induction. The opposing phenotypes of these genes suggested to us that VCAM1 upregulation may only be capturing one particular aspect of IL-1α+TNF+C1q-induced inflammatory reactivity, i.e. that IL-1α+TNF+C1q may induce distinct inflammatory reactive signatures, one of which is marked by VCAM1 upregulation.

CROP-seq reveals two distinct inflammatory reactive signatures

To gain deeper insight into how regulators uncovered by CRISPRi screening and MRA may control distinct inflammatory reactive signatures, we turned to CROP-seq⁴⁹. By coupling CRISPRi perturbations to single-cell transcriptomics, CROP-seq enables the recovery of perturbation-associated changes in gene expression from a pooled experiment. We selected 30 regulators (Extended Data Fig. 5d; Supplementary Table 6) that were strong hits from CRISPRi screening and MRA for pooled knockdown in iAstrocytes treated with vehicle control or IL-1α+TNF+C1q, and then performed CROP-seq to characterize the effect of knockdown on IL-1α+TNF+C1q-induced gene expression (see Methods).

Before examining the effect of knocking down different regulators on inflammatory reactivity, we first focused on iAstrocytes transduced with non-targeting control (NTC) sgRNAs. In the absence of IL-1α+TNF+C1q, NTC iAstrocytes partitioned largely by cell cycle in uniform manifold approximation projection (UMAP), with cluster 3 corresponding to dividing cells. A small fraction (<10%) of endothelial-like and stromal-like cells were also present, corresponding to clusters 5 and 6, respectively (Fig. 4a, Extended Data Fig. 5a–c, Supplementary Table 4). Upon treatment with IL-1α+TNF+C1q, NTC iAstrocytes partitioned into two distinct clusters (clusters 1 and 2), with an additional small cluster (cluster 4) corresponding to cycling cells (Fig. 4a, Extended Data Fig. 5a–c, Supplementary Table 4).

Open in a separate window

Fig. 4 |

CROP-seq of iAstrocytes reveals two distinct inflammatory reactive signatures dependent on the canonical NF-kB pathway.

a, Uniform manifold approximation projection (UMAP) of single-cell transcriptomes of iAstrocytes transduced with a non-targeting control (NTC) sgRNA treated with vehicle control or IL-1α+TNF+C1q, colored by cytokine treatment or cluster assignment. b, Visualization of transcript levels of selected Cluster 1 and Cluster 2 markers in the iAstrocytes shown in panel a, overlaid onto the same UMAP embedding. c, Cellular pathway (MSigDB⁷⁹) and upstream transcription factor (TRRUST⁸⁰) enrichment analysis of Cluster 1 and Cluster 2 markers; TF – transcription factor. d, Aligned UMAP embedding (see Methods) of NTC sgRNA transduced iAstrocytes with iAstrocytes transduced with sgRNAs knocking down RELA, IKBKG, or MAP3K7. e, Heatmap of Cluster 1 and Cluster 2 marker transcript levels in IL-1α+TNF+C1q-treated NTC sgRNA iAstrocytes compared to IL-1α+TNF+C1q-treated RELA, IKBKG, or MAP3K7 sgRNA iAstrocytes. f, VCAM1/C3 levels (MFI: median fluorescence intensity) or percent positive cells measured by flow cytometry in NTC, RELA, IKBKG, or MAP3K7 sgRNA iAstrocytes treated with vehicle control or IL-1α+TNF+C1q (n = 3 wells). g, Representative images of NF-κB p65 immunostaining or Hoechst counterstain in iAstrocytes treated with vehicle control or IL-1α+TNF+C1q. Scale bar: 60 μm. h, Quantification of NF-κB p65 nuclear localization (integrated fluorescence intensity masked to Hoechst; n = 3 wells). In panel c, the adjusted P values were derived from Enrichr⁸¹. In panel f, P values were calculated using linear regression for MFI values or beta regression for percentages (two-sided Wald test; see Methods) and adjusted for multiple testing (P_adj; Holm’s method) per family of tests (all comparisons made within a plot). In panel c, P values were calculated using the two-sided Fisher’s exact test and adjusted for multiple testing using the Benjamini-Hochberg method. In panel h, the P value was calculated using the two-sided student’s t-test. In panels b and d, NTC sgRNA astrocytes in Cluster 1 or Cluster 2 are circled by colored dotted lines.

Upon further examination of the two major clusters induced by IL-1α+TNF+C1q, we found that iAstrocytes in cluster 1 expressed markers related to the acute phase response such as C3, IL6, and IL1B, whereas iAstrocytes in cluster 2 expressed markers related to interferon signaling such as VCAM1, CXCL10, and LTB⁵⁰ (Fig. 4b). Enrichment analysis of cluster 2 markers confirmed a strong, specific enrichment (i.e. unique to cluster 2 markers) for interferon-related pathways (e.g. “Interferon Gamma Response”, “Interferon Alpha Response”) and transcription factor regulons (e.g. STAT1, IRF1; Fig. 4c, Supplementary Table 5). On the other hand, cluster 1 markers showed a specific enrichment for pathways and transcription factor regulons related to the mTORC1-HIF1α/MYC axis of metabolic control⁵¹ (e.g. “mTORC1 Signaling”, “Hypoxia”, “Myc Targets V1”; HIF1A, CEBPD⁵²). For brevity, we will hereon provisionally refer to the inflammatory reactive astrocyte signature (IRAS) corresponding to cluster 1 as IRAS1 and the one corresponding to cluster 2 as IRAS2.

Focusing on the canonical NF-κB pathway, we confirmed that IL-1α+TNF+C1q induced nuclear localization of NF-κB p65 (Fig. 4d,​,e)e) and found that knockdown of RELA (NF-κB p65 subunit), IKBKG (NEMO), or MAP3K7 (TAK1) blocked the induction of both IRAS1 and IRAS2 by IL-1α+TNF+C1q, both in the CROP-seq experiment (Fig. 4f,​,g)g) as well as measured by flow cytometry of C3 or VCAM1 (Fig. 4h), markers of IRAS1 or IRAS2, respectively.

To gain a more global view, we examined how knockdown of each regulator altered the differential expression induced by IL-1α+TNF+C1q (Extended Data Fig. 5d, Supplementary Table 6; see Methods), and then performed hierarchical clustering on the log-fold-changes (weighted by statistical significance) of the union of all DEGs.

We found that regulators segregated into three major groups, with one group (G1) consisting largely of regulators involved in the canonical NF-κB pathway (RELA, IKBKG, NFKB1, REL, CHUK) and upstream signal transduction (IRAK4, RIPK1, MAP3K7), another group (G2) that contained acute phase response-related transcription factors (CEBPB, CEBPD, STAT3), and the last group (G3) consisting of regulators with weak knockdown phenotypes (Extended Data Fig. 5e). In terms of coordinately regulated genes, we recovered 10 gene modules that displayed distinct patterns across knockdown of different regulators (Extended Data Fig. 5e, Extended Data Fig. 6, Supplementary Table 7).

Gene modules M3, M4, and M9 all had reduced induction upon knockdown of G1 regulators and thus appeared to be downstream of the canonical NF-κB pathway. Interestingly, M4 genes contained and had similar enrichments as IRAS1 markers (e.g. IL6; see Supplementary Table 7 and Extended Data Fig. 6) and displayed reduced induction upon knockdown of G2 regulators (Extended Data Fig. 5e), whereas M3 and M9 genes contained and had similar enrichments as IRAS2 markers (e.g. VCAM1; see Supplementary Table 7 and Extended Data Fig. 6) and displayed increased induction upon knockdown of G2 regulators (Extended Data Fig. 5e). This led us to hypothesize that the acute phase response-related transcription factors CEBPB, CEBPD, and STAT3 promote IRAS1 while inhibiting IRAS2.

IRAS1 and IRAS2 are induced by autocrine-paracrine IL-6 and interferons

Given that IRAS1 iAstrocytes expressed markers related to the acute phase response, which is driven by IL-6 ⁵³, and that IRAS2 iAstrocytes expressed markers related to the response to interferons, we hypothesized that IL-1α+TNF+C1q induced iAstrocytes to secrete IL-6 and interferons, which then acted on the iAstrocytes in an autocrine-paracrine manner. We found that iAstrocytes derived from multiple hiPSC lines indeed secreted appreciable amounts of IL-6 and IFN-β in response to IL-1α+TNF+C1q (Fig. 5a, Extended Data Fig. 7d). iAstrocytes also secreted GM-CSF, which is known to act synergistically with IL-6 ⁵⁴, as well as CXCL10, which is produced in response to interferons⁵⁵.

Open in a separate window

Fig. 5 |

IL-6 and interferons act in an autocrine-paracrine manner downstream of IL-1α+TNF+C1q.

a, Cytokine concentrations in conditioned media (measured by multi-spot electrochemiluminescence) from iAstrocytes treated with vehicle control or IL-1α+TNF+C1q (n = 3 wells). b, VCAM1/C3 levels (MFI: median fluorescence intensity) or percent positive cells measured by flow cytometry from NTC sgRNA iAstrocytes compared to iAstrocytes transduced with sgRNAs knocking down genes or gene pairs involved in IL-6 or interferon (IFN) signaling as indicated (n = 3 wells). c, Levels of phosphorylated STAT3 (Y705; p-STAT3) or phosphorylated STAT1 (Y701; p-STAT1) measured by flow cytometry in iAstrocytes treated with vehicle control or IL-1α+TNF+C1q (n = 6 wells for p-STAT3 and n = 9 wells for p-STAT1). d, Transcript levels of CSF2 and GM-CSF overlaid onto the UMAP embedding from Fig. 3a. e, Concentration of GM-CSF or CXCL10 in conditioned media from iAstrocytes transduced with sgRNAs knocking down genes or gene pairs involved in IL-6 or IFN signaling as indicated (n = 4 wells). In panels a and c, P values were calculated using the two-sided Student’s t-test. In panels b and e, P values were calculated using linear regression for MFI values or beta regression for percentages (two-sided Wald test; see Methods), and only comparisons with statistically significant differences are marked. Where appropriate, P values were adjusted for multiple testing (P_adj; Holm’s method) per family of tests (all comparisons made within a plot).

To test whether the secreted IL-6 and interferons acted on the iAstrocytes in an autocrine-paracrine manner, we knocked down their receptors and downstream transcription factors and measured the induction of C3 and VCAM1, markers of IRAS1 and IRAS2 respectively, by flow cytometry. We found that knockdown of IL6R (IL-6 receptor) or IL6ST (gp130 – the signal transducing partner of IL-6R) decreased the induction of C3+ iAstrocytes; the induction of VCAM1 in iAstrocytes increased slightly with IL6ST knockdown but did not change with IL6R knockdown (Fig. 5b). Knockdown of STAT3, which is activated downstream of IL-6 ⁵⁶, similarly decreased the induction of C3+ iAstrocytes while increasing the induction of VCAM1+ iAstrocytes (Fig. 5b). We confirmed that IL-1α+TNF+C1q induced STAT3 activation as measured by increased STAT3 phosphorylation at Y705 (Fig. 5c).

On the other hand, concurrent knockdown of IFNAR1 and IFNAR2 (see Methods), which encode subunits of the type I interferon (IFN) receptor, or IFNGR1 and IFNGR2, which encode subunits of the type II interferon receptor, decreased the induction of both C3+ iAstrocytes and VCAM1+ iAstrocytes (Fig. 5b). Concurrent knockdown of STAT1 and STAT2, which are activated downstream of interferons⁵⁷, also decreased the induction of both C3+ iAstrocytes and VCAM1+ iAstrocytes (Fig. 5b). We confirmed that IL-1α+TNF+C1q induced STAT1 activation as measured by increased STAT1 phosphorylation at Y701 (Fig. 5c).

As an alternative approach to assay the induction of IRAS1 and IRAS2, we assayed the production of GM-CSF and CXCL10, which should be produced by IRAS1 and IRAS2 iAstrocytes respectively based on mRNA expression (Fig. 5d). Knockdown of IL6ST or STAT3 decreased the production of GM-CSF but increased the production of CXCL10, whereas knockdown of IFNAR1/2 or STAT1/2 decreased the production of both GM-CSF and CXCL10 (Fig. 5e).

A limitation of the above experiments is that only a single marker of IRAS1 (C3, GM-CSF) or IRAS2 (VCAM1, CXCL10) was examined at a time, even though these markers individually do not perfectly distinguish between IRAS1 and IRAS2 (Fig. 4b, Fig. 5c). To better resolve IRAS1 and IRAS2, we performed flow cytometry on iAstrocytes stained for VCAM1 and C3 simultaneously (see Methods). With IL-1α+TNF+C1q treatment, we were able to resolve VCAM1−/C3+, VCAM1+/C3−, and VCAM1+/C3+ iAstrocytes (Fig. 6a). Starting with the expression pattern of C3 and VCAM1 in our single-cell data (Fig. 4b), we mapped VCAM1−/C3+ iAstrocytes to IRAS1, and VCAM1+/C3− as well as VCAM1+/C3+ iAstrocytes to IRAS2 based on their response to additional IL-6 (in the form of an IL-6/IL-6R chimera) or IFN-β added concurrently with all possible combinations of IL-1α, TNF, and C1q (Fig. 6b; see Supplementary Text). Importantly, we found that IL-1α by itself induced VCAM1−/C3+ iAstrocytes and secretion of IL-6 and GM-CSF secretion, whereas TNF by itself induced VCAM1+/C3− and VCAM1+/C3+ iAstrocytes and higher secretion of CXCL10 compared to IL-1α (Fig. 6b, Extended Data Fig. 7d; see Supplementary Text).

Open in a separate window

Fig. 6 |

Differential regulation of distinct inflammatory reactive signatures by cytokines and cellular factors.

a, Representative contour plot of VCAM1 and C3 levels measured by flow cytometry in iAstrocytes treated with vehicle control or IL-1α+TNF+C1q. b, Percent VCAM1−/C3+, VCAM1+/C3−, or VCAM1+/C3+ cells measured by flow cytometry in iAstrocytes treated with vehicle control vs. all possible combinations of IL-1α, TNF, and C1q, in the absence or presence of additional IL-6/IL6R chimera (25 ng/mL) or IFN-β (5 ng/mL) added concurrently (n = 6 wells). c, Percent VCAM1−/C3+, VCAM1+/C3−, or VCAM1+/C3+ cells measured by flow cytometry in NTC sgRNA iAstrocytes compared to iAstrocytes transduced with sgRNAs knocking down genes involved in IL-6 or IFN signaling (n = 5 wells). d, IL-6 or IFN-β concentration in conditioned media (measured by multi-spot electrochemiluminescence) from NTC iAstrocytes vs. iAstrocytes with knockdown of CEBPB, STAT3, IRF1, or STAT1/2 treated with vehicle control or IL-1α+TNF+C1q (n = 4 wells). e, Model of transcription factors and signaling pathways controlling inflammatory reactivity induced by IL-1α+TNF+C1q. In panels b, c, and d, P values were calculated using linear regression for MFI values or beta regression for percentages (two-sided Wald test; see Methods), and only comparisons with statistically significant differences are marked. Where appropriate, P values were adjusted for multiple testing (P_adj; Holm’s method) per family of tests (all comparisons made within a plot).

With dual staining of VCAM1 and C3, we then revisited the effect of knocking down receptors and transcription factors involved in IL-6 or interferon signaling (Fig. 6c) and found that the results were consistent with those obtained with single staining of VCAM1 or C3 (Fig. 5b). In addition, we also tested the effect of knocking down CEBPB, CEBPD, NFKB2, or IRF1, which were strong hits from the VCAM1 CRISPRi screen. Similarly to knockdown of STAT3, knockdown of CEBPB, CEBPD, or NFKB2 reduced the induction of VCAM1−/C3+ iAstrocytes and increased the induction of VCAM1+/C3− or VCAM1+/C3+ iAstrocytes (Fig. 6c). Knockdown of IRF1 reduced the induction of VCAM1+/C3− and VCAM1+/C3+ iAstrocytes and also marginally reduced the induction of VCAM1− /C3+ iAstrocytes (Fig. 6c), similarly to STAT1/2 knockdown. We also tested the effect of STAT3, CEBPB, NFKB2, or IRF1 knockdown in iAstrocytes derived from two additional hiPSC lines and obtained consistent results (Extended Data Fig. 8a–b).

In terms of cytokine secretion, we observed that production of IL-6 was decreased by CEBPB, STAT3, or STAT1/2 knockdown, whereas production of IFN-β was decreased by on STAT3, IRF1, or STAT1/2 knockdown (Fig. 6d).

Lastly, to complement our experiments employing CRIPSRi knockdown, we also tested the effect of small molecules known to modulate STAT3 or STAT1/2 activity and found that they modulated the induction of IRAS1 and IRAS2 as marked by dual VCAM1/C3 staining (Extended Data Fig. 9; see Supplementary Text).

Based on our findings so far and existing literature, we created a model (Fig. 6e) of how IL-1α+TNF+C1q treatment induces IRAS1 and IRAS2 through a primary NF-κB-dependent transcriptional response (Fig. 4d–h) followed by autocrine-paracrine IL-6 and interferon signaling feedback loops involving CEBPB/D, NFKB2, STAT3, STAT1/2, and IRF1 (Fig. 5, Fig. 6). Downstream of NF-κB, CEBPB/D, NFKB2, and STAT3 drive the induction of IRAS1 and inhibit the induction of IRAS2 through IL-6 (also see refs. ^{58, 59}), whereas STAT1/2 and IRF1 promote the induction of IRAS2 through interferons (also see ref. ⁶⁰), which inhibit the induction of IRAS1 (Fig. 6b–d). IL-1α polarizes astrocytes towards IRAS1 by driving IL-6 production (Extended Data Fig. 7d) and antagonizing interferon signaling (also see ref. ⁴⁶), whereas TNF polarizes astrocytes towards IRAS2 by driving interferon production (Extended Data Fig. 7d), which in turn antagonizes IL-1 signaling (also see ref. ⁴⁶). The discrepancy between the inhibitory effect of IFN-β on IRAS1 induction (Fig. 6b) and the weak requirement of STAT1/2 and IRF1 for IRAS1 induction (Fig. 6c) could be explained by the promiscuity of STAT1/2 in responding to and promoting the production of both IL-6 and IFN-β (Fig. 6d; also see refs. ^60,61–63). Likewise, STAT3 activation has also been reported in response to interferons^{64, 65}, and our data supports some role of STAT3 in the production of IFN-β (Fig. 6d); however, in our system STAT3 is much more strongly associated with IL-6 signaling given the consistent effect of IL-6 addition and STAT3 knockdown.

We do not intend for the terms IRAS1 and IRAS2 to be used outside of a provisional basis in this manuscript. Thus, considering the cellular pathways driving IRAS1 and IRAS2, we propose that IRAS1 be referred to as “IL-1/IL-6-responsive”, and IRAS2 as “TNF/IFN-responsive” reactivity.

To see if signatures of IL-1/IL-6-responsive and TNF/IFN-responsive reactive astrocytes are found across different experimental paradigms of neuroinflammation, we integrated the NTC sgRNA-transduced iAstrocytes from our single-cell RNA-seq data with previously published astrocyte single-cell RNA-seq datasets using the anchor-based data integration functionality of Seurat⁶⁶ and found substantial overlap (Fig. 7, Extended Data Fig. 10; see Supplementary Text for detailed exposition).

Open in a separate window

Fig. 7 |

Integration of iAstrocyte single-cell data with published single-cell datasets shows overlap of inflammatory reactive signatures across species in diverse disease contexts.

a, UMAP of integrated analysis (see Methods) of NTC sgRNA iAstrocytes with astrocytes from Barbar et al., colored by cluster assignment or astrocyte source. b-c, The same UMAP embedding as in a, showing only iAstrocytes colored by cytokine treatment or Fig. 4a cluster assignment (b) or Barbar et al. astrocytes colored by cytokine treatment (c). d, Cluster-averaged expression levels of selected cluster markers in Barbar et al. astrocytes (n = 2 cell lines). e, UMAP of integrated analysis of NTC sgRNA iAstrocytes with astrocytes from Wheeler et al., colored by cluster assignment or astrocyte source. f-g, The same UMAP embedding as in e, showing only iAstrocytes colored by cytokine treatment or Fig. 4a cluster assignment (f) or astrocytes from Wheeler et al. colored by EAE stage (g). h, Cluster-averaged expression levels of selected cluster markers in Wheeler et al. astrocytes (n = 8 mice). i, UMAP of integrated analysis of NTC sgRNA iAstrocytes with astrocytes from Hasel et al., colored by cluster assignment or astrocyte source. j-k, The same UMAP embedding as in i, showing only iAstrocytes colored by cytokine treatment or Fig. 4a cluster assignment (j) or astrocytes from Hasel et al. colored by LPS treatment (k). l, Cluster-averaged expression levels of selected cluster markers in Hasel et al. astrocytes (n = 12 mice).

Upregulation of reactive astrocyte signature markers in human disease

Given that we found signatures of IL-1/IL-6 responsive and TNF/IFN-responsive reactive astrocytes in various experimental models of neuroinflammation, we sought to test whether they can be found in human disease. In Alzheimer’s disease (AD), neuroinflammation mediated by microglia and astrocytes contributes to disease progression⁶⁷ potentially through IL-6⁶⁸ and interferons⁶⁹. We performed immunostaining of the IL-1/IL-6-responsive marker C3 and the TNF/IFN-responsive marker VCAM1 in post-mortem brain tissue derived from AD cases (61–92 years old, Braak stage IV-VI) vs. age-matched controls (62–102 years old, Braak stage 0-III; see Supplementary Table 10 for clinical metadata). We found a significant increase in the abundance of C3+ astrocytes in AD (Fig. 8a–b), consistent with previous studies⁹. Notably, C3+ astrocytes tended to be localized near C3+ plaques (Fig. 8a), which we deduced to be amyloid plaques given the fact that they are known to accumulate complement proteins such as C3 ⁷⁰. Amyloid plaques are also known to accumulate SERPINA3 ⁷¹, which is a IL-1/IL-6 responsive marker and also a marker of disease-associated reactive astrocytes found in mouse models of AD⁷². As for VCAM1, we did not observe a statistically significant difference in the abundance of VCAM1+ astrocytes between AD cases vs. controls (Fig. 8c–d). Lastly, C3+ astrocytes tended to be VCAM1− (416 VCAM1− vs. 31 VCAM1+ astrocytes among C3+ astrocytes aggregated across all individuals; Fig. 8a), and VCAM1+ astrocytes tended to be C3− (135 C3− vs. 31 C3+ astrocytes among VCAM1+ astrocytes aggregated across all individuals; Fig. 8c); there were 31 VCAM1+/C3+ astrocytes among 12,885 total astrocytes.

Open in a separate window

Fig. 8 |

Markers of distinct inflammatory reactive signatures are upregulated in astrocytes in human Alzheimer’s disease (AD) and hypoxic-ischemic encephalopathy (HIE) and are regulated by Stat3 in a mouse model of neuroinflammation.

a-h, Post-mortem human brain sections. a,c, Representative immunofluorescence images of C3+ (a) or VCAM1+ (c) astrocytes (marked by arrowheads) in an AD case compared to an age-matched control. Staining for VCAM1 and C3 was performed simultaneously (the same image is shown twice with different marker combinations). Scale bars correspond to 20 μm. b,d, Quantification of the percentage of C3+ (b) or VCAM1+ (d) astrocytes in AD cases (n = 8 individuals) vs. age-matched controls (n = 8 individuals). e,g, Representative immunofluorescence images of C3+ (e) or VCAM1+ (g) astrocytes (marked by arrowheads) in a HIE case compared to an age-matched control. Staining for VCAM1 or C3 was performed separately. Scale bars correspond to 10 μm. f,h, Quantification of the percentage of C3+ (f) or VCAM1+ (h) astrocytes in HIE cases (n = 4 individuals) vs. age-matched controls (n = 4 individuals). i-n, Brain sections from mice injected with vehicle or LPS. i,l, Representative immunofluorescence images of GFAP in the corpus callosum (i) or hippocampus (l). Scale bars correspond to 100 μm. j,m, Representative images of C3, Isg15, and Gfap staining in the corpus callosum (j) or hippocampus (m) in wild-type (WT; n = 6 for untreated or LPS-treated) or Stat3 astrocyte-specific conditional knockout (Stat3-cKO; n = 4 for untreated or 6 for LPS-treated) mice. Staining for C3 and Isg15 was performed simultaneously (the same image is shown twice with different marker combinations). Representative Isg15−/C3+ (solid arrowheads), Isg15+/C3− (empty arrowheads), and Isg15+/C3+ astrocytes (striped arrowheads) are labeled. k,n, Quantification of the percentage of Isg15−/C3+, Isg15+/C3−, and Isg15+/C3+ astrocytes in the corpus callosum (k) or hippocampus (n). P values were calculated using the two-sided Mann-Whitney U test in b, d, f, and h and beta regression (two-sided Wald test; see Methods) in k and n. For beta regression in k and n, sex was included as a covariate.

In addition to AD, we also examined neonatal brain tissue derived from age-matched controls (38–41 gestational weeks) vs. cases of hypoxic-ischemic encephalopathy (HIE; 37–41 gestational weeks), in which neuroinflammation involving IL-1, TNF, IL-6, and interferons^73–75 contributes significantly to neuronal injury and negative neurologic sequela⁷⁶. Compared to AD, HIE may be better modeled by our experimental paradigm (hiPSC-derived astrocytes treated with cytokines) given its acute course and the immature status of astrocytes in neonates^{77, 78}. We observed increased abundance of both C3+ and VCAM1+ astrocytes in HIE (Fig. 8e–h). However, experimental limitations prevented us from staining for C3 and VCAM1 simultaneously.

Cloning of NFIA and SOX9 cDNA into dox-inducible cassette

To obtain NFIA cDNA, we designed PCR primers to amplify cDNA corresponding to transcript ENST00000403491(NFIA isoform 1) from astrocyte cDNA (Forward primer complementary sequence: ATGTATTCTCCGCTCTGTCTCAC; reverse primer complementary sequence: TCCCAGGTACCAGGACTGTG). We chose to amplify cDNA corresponding to NFIA isoform 1 because the cDNA clone (BC022264) used in Li et al.¹⁷ corresponds to transcript ENST00000371187 (NFIA isoform 2), which we found was not expressed highly in human astrocytes. To obtain SOX9 cDNA, we ordered cDNA clone OHu19789 (which corresponds to MGC clone BC056420) from GeneScript and then amplified SOX9 cDNA from the plasmid (Forward primer complementary sequence: ATGAATCTCCTGGACCCCTTCA; reverse primer complementary sequence: TCAAGGTCGAGTGAGCTGTGT). We then inserted NFIA and SOX9 cDNA joined by a T2A sequence into an AAVS1 safe-harbor plasmid containing a dox-inducible cassette (Addgene plasmid no. 105840, gift from Michael Ward; digested with AflIIand ClaI) using Gibson assembly (New England Biolabs; cat. no. E2611L), resulting in pKL100 (available on Addgene under ID no. 185708).

In addition to pKL100, an additional AAVS1 safe-harbor donor plasmid with dox-inducible NFIA and SOX9 was created by replacing the puromycin resistance and EF1α:mCherry portion of pKL100 with hygromycin resistance, resulting in pKL123. This plasmid was not used to generate any data in this manuscript but is also available on Addgene under ID no. 187889).

Generation of hiPSC lines with stable integration of NFIA-SOX9 cassette and CRISPRi cassette

WTC11, TCW-1E44, and 162D hiPSCs were transfected with pC13N-dCas9-BFP-KRAB⁸⁵ (Addgene plasmid no. 127968) to stably integrate constitutive CRISPRi machinery into the CLYBL locus using TALEN-based editing as previously described⁸⁵. CRISPRi WTC11, TCW-1E44, and 162D hiPSCs were then transfected with pKL100 to stably integrate the dox-inducible NFIA-SOX9 cassette into the AAVS1 locus using the same TALEN-based editing approach. Briefly, CRISPRi WTC11, TCW-1E44, and 162D hiPSCs were dissociated with Accutase to a single-cell suspension and plated at 500,000 cells per well in a Matrigel-coated 6-well plate in E8 + 10 μM Y-27632. The next day, a media change with E8 + Y-27632 was performed and then the hiPSCs were transfected with 2.5 ug of pKL100, 1.25 ug of left and right AAVS1 TALEN plasmids (Addgene plasmid no. 59025 and 59026, gift from Danwei Huangfu), and 0.5 ug of Bcl-XL plasmid (pEF1-BCL-XL-wpre-polyA P1102, gift from Xiaobing Zhang, described in ref ⁸⁶) using Lipofectamine Stem Transfection Reagent (ThermoFisher Scientific cat. no. STEM00015) following the manufacture’s protocol. After hiPSCs reached confluence, they were dissociated with Accutase and passaged to a 10 cm dish in E8 + 10 uM Y-27632 + 0.1 ug/mL puromycin (ThermoFisher Scientific cat. no. A1113803) to select for clones with stable integration. Y-27632 was maintained until stable colonies formed, and puromycin was maintained for 5–7 days. The hiPSCs were then sorted for mCherry+ cells, which were then plated at 500,000 cells per well in a Matrigel-coated 6-well plate for transfection with 1.2 ug of Cre recombinase mRNA (TriLink Biotechnologies cat. no. L-7211) using Lipofectamine Stem to remove the puromycin resistance gene and mCherry. The Cre-transfected hiPSCs were then expanded and sorted for mCherry- cells, which were then plated for colony picking to generate monoclonal hiPSC lines. hiPSC clones were tested for integration of the NFIA-SOX9 cassette and removal of the puromycin resistance and mCherry by genomic PCR with the following pairs of primers:

AAVS1 FWD: CTGCCGTCTCTCTCCTGAGT

bGHpolyA REV: GCTGGCAACTAGAAGGCACAG

AAVS1 FWD: CTGCCGTCTCTCTCCTGAGT

Puro REV: GTGGGCTTGTACTCGGTCAT

TRE3G FWD: GTGTTGTGGAATTGCTCCAG

AAVS1 REV: AAGAGTGAGTTTGCCAAGCAGT

T2A-SOX9 FWD: AATCCCGGCCCTATGAATCTCCTG

T2A-SOX9 REV: TGGGCGATCGATTCAAGGTCGAGT

Neural induction of hiPSCs

Embryoid body (EB)-based neural induction of WTC11, TCW-1E44, and 162D hiPSCs stably integrated with CRISPRi machinery and dox-inducible NFIA-SOX9 was performed as previously described⁸⁷ with some modifications. Briefly, on day 0, hiPSCs were dissociated to a near single-cell suspension with Accutase, resuspended in neural induction media (NIM; see recipe below) + 10 nM Y-27632, and then transferred to an Aggrewell 800 plate (StemCell Technologies cat. no. 34815) pre-coated with Anti-adherence rinsing solution (StemCell Technologies cat. no. 07010) at 3 million cells per well for EB formation. The next day (day 1), a half media change with NIM was performed and LDN193189 (LDN; Tocris cat. no. 6053) and SB431542 (SB; Tocris cat. no. 1614) were added to final concentrations of 0.1 μM and 10 μM, respectively. A half media change of NIM + LDN + SB was performed every day until day 7, when EBs were transferred to a Matrigel-coated 6-well plate after performing a half media change (one well from the Aggrewell plate would be transferred to one well in the 6-well plate). The next day (day 8), a full media change with NIM + LDN + SB was performed, and then every other day afterwards until day 14, during which time neural rosettes would appear in the attached EBs. On day 14, neural rosettes were detached non-enzymatically with the following method: the attached EBs were incubated Neural Rosette Selection Reagent (NRSR; StemCell Technologies cat. no. 05832) for 1 hour at 37 °C to weaken the attachment of the neural rosettes, the NRSR was aspirated and replaced with DMEM/F12, and the rosettes were detached with targeted jetting of DMEM/F12 using a wide-orifice P1000 pipette tip and then collected into a conical tube; targeted jetting with DMEM/F12 and collection of released rosettes was repeated until the majority of rosettes had detached. Rosettes were then spun down at 100 g for 3 min, resuspended in NPC media (see recipe below), and transferred to a Matrigel-coated 6-well plate (rosettes collected from 1–2 wells would be replated into 1 well depending on the yield). From day 15 to 21, a full media change with NPC media was performed every other day, during which time neural progenitor cells (NPCs) would spread out from the attached rosettes and cover the well completely. Once NPCs reached confluency, they were dissociated with Accutase and replated at high density (at least 1 million cells per well of a 6-well plate) in Matrigel-coated plates for expansion.

Neural induction media (NIM) formulation:

DMEM/F12 (+HEPES, +Glutamine) basal media (ThermoFisher Scientific cat. no. 11330032)

2% (v/v) B27 supplement minus Vit. A (ThermoFisher Scientific cat. no. 12587010)

1% (v/v) N2 supplement (ThermoFisher Scientific cat. no. 17502048)

Neural progenitor cell (NPC) media formulation:

DMEM/F12 (+HEPES, +Glutamine) basal media (ThermoFisher Scientific cat. no. 11330032)

2% (v/v) B27 supplement minus Vit. A (ThermoFisher Scientific cat. no. 12587010)

1% (v/v) N2 supplement (ThermoFisher Scientific cat. no. 17502048)

20 ng/mL bFGF (PeproTech cat. no. 100–18B)

Purification of NPCs

To remove contaminating neural crest cells from the NPC cultures, we used fluorescence activated cell sorting (FACS) to select for CD133+/CD271− cells following the protocol described in Cheng et al.⁸⁷ with modifications. Briefly, NPCs were dissociated to a single-cell suspension with Accutase, resuspended in FACS buffer (see recipe below), and then incubated with PE-conjugated CD133 antibody (1:50 dilution; Miltenyi Biotec cat. no. 130-113-108) and PerCP-Cy5.5-conjugated CD271 antibody (1:50 dilution; BD Biosciences cat. no. 560834), including single antibody-stained and unstained controls. After antibody incubation, the cell suspension was diluted 10x with FACS buffer, spun down, and resuspended in FACS buffer for sorting. CD133+/CD271− cells were sorted using a BD FACSAria Fusion cell sorter at 5,000–10,000 events per second, and then plated at 100,000 cells per cm² onto Matrigel-coated plates. Media was then changed every other day until the NPCs reached confluency, at which point the NPCs were passaged for expansion. After 2–3 additional expansion passages after sorting, NPCs were characterized by immunostaining and qPCR for NPC markers such as PAX6 and Nestin, and then cryopreserved in NPC media + 10% DMSO.

FACS buffer formulation:

DPBS (Milipore Sigma cat. no. D8537)

1% (w/v) BSA (Milipore Sigma cat. no. A9647)

2 mM EDTA (Milipore Sigma cat. no. 324506)

Generation of iAstrocytes from NPCs

For iAstrocyte differentiation, CD133+/CD271− sorted NPCs generated from WTC11, TCW-1E44, and 162D iPCS with stably integrated CRISPRi machinery and dox-inducible NFIA-SOX9 were dissociated to a single-cell suspension with Accutase and then replated at 7,500–15,000 cells per cm² in NPC media onto a 10-cm or 15-cm dish coated with Matrigel diluted at 1:200 in DMEM/F12. The next day, media was changed to ScienCell Astrocyte Media (ScienCell Research Laboratories cat. no. 1801) + 2 μg/mL doxycycline (Millipore Sigma cat. no. D9891) to initiate iAstrocyte differentiation. A full media was then changed every other day, with doxycycline maintained at 2 μg/mL throughout the differentiation process. When the differentiating NPCs reached confluency within 3–4 days, the culture was dissociated with Accutase and split 1:4–1:10 onto new Matrigel coated dishes for expansion, with some cells saved for cryopreservation. After confluency was reached within 5–6 days, the cultures were dissociated and split 1:4–1:8, saving some cells for cryopreservation. Expansion of the cultures at 1:4–1:8 split with cryopreservation of cells after each split was continued until day 20 of differentiation, yielding iAstrocytes. To maintain the post-split cell density between 10,000–20,000 cells per cm², the split ratio was adjusted based on the proliferative tendency of the differentiating astrocyte precursors, which varied those generated from WTC11, TCW-1E44, and 162D hiPSCs. In terms of yield, starting from ~400,000 NPCs plated into a 10 cm dish, in theory ~100 million iAstrocytes could be obtained by around day 20 assuming that the cells are split 1:4 for a total of four times. However, the yield may vary depending on the proliferative tendency of the astrocyte precursors.

For cryopreservation, iAstrocytes were resuspended in ScienCell Astrocyte media + 10% DMSO, transferred into Corning CoolCell alcohol-free freezing containers (Corning cat. no. 432001) in cryogenic vials, and then transferred to liquid nitrogen. For experiments, cryopreserved iAstrocytes were thawed by warming cryovials in a 37 °C water bath until no ice was left, transferring the contents to a 15 mL conical tube, adding 4 volumes of DPBS, equilibrating at RT for 3 min, spinning down cells at 300 g for 3 min, and then resuspending the pellet in ScienCell Astrocyte Media.

Experiments were performed using iAstrocytes derived from WTC11 hiPSCs unless otherwise indicated.

Generation of hiPSC-derived astrocytes using alternative protocols

In parallel to the generation of iAstrocytes, WTC11 hiPSC-derived astrocytes were also generated from CD133+/CD271− sorted NPCs according to the protocol published in TCW et al.¹⁶, which mirrored the process described above for iAstrocyte generation with the exception of adding doxycycline. In addition to the protocol published in TCW et al., hiPSC-derived astrocytes were also generated from WTC11 iPSCs with stably integrated CRISPRi machinery and dox-inducible NFIA-SOX9 according to Li et al.¹⁷, or from WTC11 iPSCs with stably integrated CRISPRi machinery according to Krencik et al.⁸⁸.

Compared to Li et al.¹⁷, our iAstrocyte protocol differs in the following aspects (see Methods): 1) using a splice isoform of NFIA that is more highly expressed in astrocytes (ENST00000403491 in this study vs. ENST00000371187 in Li et al.¹⁷; Extended Data Fig. 1c, 2) ensuring the purity of neural progenitor cells through sorting for CD133+/CD271− cells⁸⁹, 3) culturing astrocyte precursors during the differentiation process in monolayer, which is more straightforward in terms of media changes and cell dissociation compared to spheroid culture in Li et al.¹⁷, and 4) using a commercially available astrocyte media (ScienCell Astrocyte Media) adapted from TCWet al.

Primary mouse astrocyte culture

P0 C57BL/6J pups were decapitated and the brains were removed and placed into ice cold HBSS. Cortices were removed from the rest of the brain in HBSS and dissociated using the Pierce Primary Neuron Isolation Kit (ThermoFisher Scientific cat. no. 88280). Instead of using the Neuronal Culture Medium from the kit, cells from the dissociated cortices were plated onto PDL-coated 96-well plates in primary astrocyte medium (see recipe below) at the recommended density. Media was changed every other day for 6 days, at which point the culture consisted mainly of astrocytes.

Primary astrocyte medium:

DMEM (ThermoFisher Scientific cat. no. 11965084)

10% FBS (VWR cat. no. 89510–186; lot no. 184B19)

1% Penicillin-streptomycin (ThermoFisher Scientific cat no. 15140122)

Induction of inflammatory reactivity in hiPSC-derived astrocytes or primary mouse astrocytes

iAstrocytes or hiPSC-derived astrocytes generated according to TCW et al.¹⁶ were plated onto Matrigel-coated (1:200 diluted Matrigel) 96-well plates, 24-well plates or 6-well plates at 20,000 cells per cm² in ScienCell Astrocyte Media without addition of doxycycline; hiPSC-derived astrocytes generated according to Li et al.¹⁷ or Krencik et al.⁸⁸ were plated onto Matrigel-coated (1:200 diluted Matrigel) 96-well plates at ~62,500 cells per cm² in Astrocyte Maturation Media (AMM; see recipe below) without addition of doxycycline. A full media change was performed the next day, and then every other day afterwards. Five days after plating, hiPSC-derived astrocytes were treated with vehicle control or IL-1α (3 ng/mL; Peprotech cat. no. AF-200-01A), TNF (30 ng/mL; Peprotech cat. no. AF-300-01A), and C1q (400 ng/mL; Complement Technology cat. no. A099) with a full media change in the appropriate media to induce inflammatory reactivity according to Liddelow et al.⁹. For experiments involving all possible combinations of IL-1α, TNF, and C1q, the same concentrations as above were used. For experiments involving treatment with additional IL-6 or IFN-β, a IL-6/IL-6R chimeric protein (25 ng/mL; R&D systems cat. no. 8954-SR-025) or IFN-β (5 ng/mL; Peprotech cat. no. AF-300-02B) was added concurrently with all possible combinations of IL-1α, TNF, and C1q, which were added at the same concentration as above. We found that using a IL-6/IL-6R chimeric protein was necessary given our observation that addition of IL-6 itself had no additional effect on iAstrocytes, which is likely due to the fact that IL-6R is not expressed at high levels constitutively in most cell types⁹⁰. For experiments involving napabucasin (Selleckchem cat. no. S7977) or RGFP966 (Selleckchem cat. no. S7229), the drugs were added concurrently with cytokines, with the final concentration of DMSO kept less than or equal to 0.1% in all conditions.

For primary mouse astrocytes, IL-1α+TNF+C1q treatment (same concentration as above) occurred after 6 days of culture. All assays were performed at 24 hours after cytokine treatment (see Fig. 1g) unless otherwise stated.

Astrocyte Maturation Media (AMM) formulation:

DMEM/F12 (+HEPES, +Glutamine) basal media (ThermoFisher Scientific cat. no. 11330032)

1% (v/v) B27 supplement minus Vit. A (ThermoFisher Scientific cat. no. 12587010)

0.5% (v/v) N2 supplement (ThermoFisher Scientific cat. no. 17502048)

1% (v/v) Antibiotic-Antimycotic (ThermoFisher Scientific cat. no. 15240062)

10 ng/mL CNTF (R&D Systems cat. no. 257-NT)

10 ng/mL BMP-4 (R&D Systems cat. no. 314-BP)

CRISPRi-mediated gene knockdown in hiPSC-derived astrocytes using lentiviral sgRNA delivery

For experiments involving CRISPRi-mediated gene knockdown, iAstrocytes were transduced with lentivirus containing single-guide RNAs (sgRNAs) at the time of plating (see Fig. 2b). CRISPRi sgRNAs were cloned into pMK1334 ⁸⁵ (Addgene cat. no. 127965) digested with BstXI and BlpI as previously described in Gilbert et al.²⁴. Lentivirus containing sgRNAs was produced by transfecting HEK293T cells with pMK1334 and 3^rd generation lentiviral packaging plasmids with TransIT-Lenti Transfection Reagent (Mirus cat. no. MIR6606) according to the manufacturer’s instructions. The lentivirus was then precipitated using Lentivirus Precipitation Solution (ALSTEM cat. no. VC150) according to the manufacturer’s instructions, resuspended in DPBS at 1/10 of the original volume, and then aliquoted and stored at −80 °C. The functional titer of the lentivirus was then tested on iAstrocytes by serial dilution followed by measurement of BFP+ cells 48 hours after transduction. For all experiments involving CRISPRi-mediated gene knockdown in iAstrocytes, sufficient sgRNA lentivirus was added to transduced >70% of iAstrocytes.

Immunostaining of hiPSC-derived astrocytes

iAstrocytes and hiPSC-derived astrocytes generated according to TCW et al.¹⁶ from WTC11, TCW-1E44, and 162D hiPSCs were plated at 20,000 cells per cm² onto a Greiner μClear 96-well plate (Greiner Bio-One cat. no. 655087) coated with 1:200 diluted Matrigel, and then treated with vehicle control or IL-1α+TNF+C1q as described above. 24 hours after cytokine treatment, the astrocytes were washed with DPBS and then fixed with 4% paraformaldehyde (diluted from a 16% solution; Electron Microscopy Sciences cat. no. 15710) for 15 min at room temperature (RT). After washing three times with DPBS, blocking and permeabilization was performed with DPBS + 3% BSA + 0.1% Triton X-100 (Millipore Sigma cat. no. X100) for 30 min at RT. Primary antibodies against GFAP (1:500, rabbit polyclonal; ThermoFisher Scientific cat. no. PA1-10019), S100β (1:500, mouse monoclonal; Millipore Sigma cat. no. S2532), GLAST (1:500, mouse monoclonal; Miltenyi Biotec cat. no. 130-095-822), NFIA (1:200, rabbit polyclonal; Atlas Antibodies cat. no. HPA008884), SOX9 (1:500, mouse monoclonal; Atlast Antibodies cat. no. AMAb90795), Cx43 (1:250, rabbit polyclonal; ThermoFisher Scientific cat. no. 71–0700), glutamine synthetase (1:500, mouse monoclonal; ThermoFisher Scientific cat. no. MA5-27749), or vimentin (1:200, rabbit monoclonal; Cell Signaling Technologies cat. no. 5741) were then added in blocking buffer and incubated overnight at 4 °C. Afterwards, the samples were washed with DPBS + 0.1% Triton X-100 three times, incubated with pre-adsorbed secondary antibodies (1:500 Goat anti-mouse IgG AF647, 1:500 Goat anti-rabbit IgG AF555; Abcam cat. no. ab150119 and ab150086) for 1 hour at RT, washed two times with DPBS + 0.1% Triton X-100, incubated with 1 μg/mL Hoechst (ThermoFisher Scientific cat. no. H3570) and 1:10 ActinGreen 488 (ThermoFisher Scientific cat. no. R37110) for 20 min at RT, and then washed two additional times before imaging.

For immunostaining of C3 and IFIT3, iAstrocytes derived from WTC11, TCW-1E44, and 162D hiPSCs washed, fixed, and blocked as above and then incubated with primary antibodies against C3 (1:250, mouse monoclonal; BioLegend cat. no. 846302), IFIT3 (1:500, rabbit polyclonal; ThermoFisher Scientific cat. no. PA5-22230), or Na⁺/K⁺ ATPase (1:500, chicken polyclonal; VWR cat. no. 89162–626) overnight at 4 °C in blocking buffer. Afterwards, washing and secondary antibody incubation was performed as above (1:500 Goat anti-mouse IgG AF647, 1:500 Goat anti-rabbit IgG AF555, 1:500 Goat anti-chicken IgY AF647; Abcam cat. no. ab150119, ab150086, and ab150175) with 1 μg/mL Hoechst as a counterstain.

Samples were imaged with an IN Cell Analyzer 6000, using a 20X 0.45 NA objective, 2×2 binning, 100–400 ms exposure, an aperture width of ~1 Airy unit, and 16 fields per well.

Glutamate uptake assay

WTC11 iAstrocytes or TCW astrocytes were plated onto a 96-well plate (with some wells left devoid of cells) and maintained as described above. At 5 days after plating, the media was replaced with Krebs-Ringer Phosphate Glucose buffer (KRPG: 145 mM NaCl, 5.7 mM sodium phosphate, 4.86 mM KCl, 0.54 mM CaCl₂, 1.22 mM MgSO₄, 5.5 mM glucose, pH 7.35) with 100 μM glutamate for 3 hours, after which the conditioned media was collected for measurement of glutamate concentration using a colorimetric assay (Millipore Sigma cat. no. MAK004) following the manufacturer’s instructions. The background signal values from wells with no cells and no glutamate (n = 3) were averaged and subtracted from raw signal values, the background-corrected assay signal values from wells containing no astrocytes with 100 μM glutamate added (n = 3) were averaged, and the percent glutamate uptake by astrocytes in a given well was calculated as follows: (average assay signal from wells containing no astrocytes with 100 μM glutamate added − assay signal from astrocyte well treated with 100 μM glutamate) / (average assay signal from wells containing no astrocytes with 100 μM glutamate added) * 100.

Generation of hiPSC-derived brain endothelial-like cells and measurement of barrier integrity

hiPSCs were differentiated to brain endothelial-like cells as previously described⁹¹. Briefly, hiPSCs were dissociated with Accutase and seeded on Matrigel-coated plates in E8 medium containing 10 μM Y27632 at a density of 15,000 cells/cm². Differentiation was initiated 24 hours after seeding by changing to E6 medium, with daily medium changes for 4 days. Next, cells were expanded with serum-free basal endothelial cell medium (EC medium) supplemented with 50x diluted B27 (Thermo Fisher Scientific), 1x GlutaMAX (Thermo Fisher Scientific), 10 μM retinoic acid (Sigma Aldrich), and 20 μg/ml FGF2 for 2 days without a media change. Following this treatment, cells were collected by a 20-minute incubation in Accutase and seeded onto Transwell filters (1.1 cm² polyethylene terephthalate membranes with 0.4 μm pores; Fisher Scientific) coated with a mixture of 400 μg/ml collagen IV (Sigma Aldrich) and 100 μg/ml fibronectin (Sigma Aldrich). The following day, cells were switched to EC medium lacking FGF2 and RA. For co-culture with iAstrocytes, filters were transferred to 12-well plates containing iAstrocytes and the same medium was utilized. Starting at this time, transendothelial electrical resistance (TEER) was measured using STX2 chopstick electrodes and an EVOM2 voltameter (World precision Instruments) approximately every 24 hours. TEER readings on empty Transwell filters were subtracted from all measurements to reflect the resistance of only the cultured cells.

Generation of hiPSC-derived neurons (iNeurons) and GCaMP iNeurons

hiPSC-derived neurons (iNeurons) were generated from WTC11 hiPSCs with stably integrated dox-inducible NGN2 (NGN2 iPSCs) according to Fernandopulle et al.⁹². To generate GCaMP iNeurons, NGN2 iPSCs were transduced with a lentivirus delivering GCaMP6m (gift from Dr. Michael Ward). To facilitate segmentation of neuron soma, an additional lentivirus transduction was performed to deliver pMK1334 containing a non-targeting sgRNA, which confers BFP expression localized to the nucleus. Clonal lines were then isolated by colony picking, differentiated to neurons according to Fernandopulle et al.⁹², and evaluated for homogeneity of GCaMP6m expression and the presence of spontaneous calcium oscillations. A clonal line satisfying the above criteria was selected for GCaMP imaging experiments.

Measurement of calcium activity in GCaMP iNeurons

Briefly, GCaMP iNeurons at day 3 of differentiation (see Fernandopulle et al.⁹²) were replated onto poly-D-lysine-coated 96-well plates (Corning cat. no. 354640) at 62,5000 cells per cm² in neuron media + 2 μg/mL doxycycline. For co-culture experiments, iAstrocytes were added on day 3 at 10,000 cells per cm² in an equivalent volume of ScienCell Astrocyte Media + 2 μg/mL doxycycline. For mono-culture experiments, an equivalent volume of neuron media + 2 μg/mL doxycycline was added. On day 6, half of the media was replaced with fresh neuron media + 2 μg/mL doxycycline; on day 10, half of the media was replaced with fresh neuron media without doxycycline. On day 17, the cultures were treated with vehicle control or IL-1α+TNF+C1q (using the same final concentration as described above for astrocyte experiments) by performing a half media change with fresh neuron media. Calcium activity in GCaMP iNeurons was recorded on day 18 and day 20 with an IN Cell Analyzer 6000, using a 20X 0.45 NA objective, 2×2 binning, environmental control set to 37 °C and 5% CO₂, an aperture width of ~1 Airy unit, and 1 frame (800 ms exposure) collected per second for 40 seconds per field (1 field per well).

Isolation of synaptosomes and labeling with pHrodo

Synaptosomes were isolated from fresh Innovative Grade US Origin Rat Sprague Dawley Brains (Innovative Research, Inc.; Cat. No. IGRTSDBR) with the Syn-PER™ Synaptic Protein Extraction Reagent (ThermoFisher Scientific cat. no. 87793) according to the manufacture’s protocol with minor changes. Briefly, 10 mL of Syn-PER Reagent supplemented with 1x protease inhibitor cOmplete Mini, EDTA free (Roche cat. no. 11836170001) and 1x phosphatase inhibitor PhosSTOP (Roche cat. no. 4906845001) were added per gram of brain tissue. Dounce homogenization was performed on ice and homogenate was transferred to a conical tube and centrifuged at 1200 × g for 10 minutes at 4°C. The pellet was discarded, the supernatant was transferred to a new tube, and the centrifugation step was repeated. The supernatant was then centrifuged at 15,000 × g for 20 minutes at 4°C. The supernatant was removed and the wet pellet was weighed. The synaptosome fractions were resuspended at a concentration of 50 mg/ml. 3 μM of pHrodo Red, succinimidyl ester (ThermoFisher Scientific cat. no. P36600) was added to the synaptosome fraction and incubated for 45 min at room temperature in the dark. After diluting the solution 1:10 in DPBS, the synaptosomes were spun down at 2500 × g for 5 min. The supernatant was removed and then the synaptosomes were washed two times with DPBS. The pHrodo-labelled synaptosomes were resuspended in DMEM/F12 + 5% DMSO at a stock concentration of 50 mg/ml, aliquoted, and then frozen in liquid nitrogen for later use.

Measurement of synaptosome phagocytosis

For synaptosome phagocytosis experiments, pHrodo-labeled rat synaptosomes were used for iAstrocytes, and pHrodo-labeled iNeuron synaptosomes were used for Li et al.¹⁷ and Krencik et al.⁸⁸ hiPSC-derived astrocytes. Briefly, astrocytes were incubated with pHrodo-labeled synaptosomes resuspended in the appropriate astrocyte media (ScienCell Astrocyte media for iAstrocytes, AMM for Li et al. and Krencik et al. astrocytes) at 1 mg/mL for 3 hours at 37 °C; for negative controls, some samples were pre-treated with 10 uM cytochalasin D (Millipore Sigma cat. no. C8273) for 15 min and also incubated with synaptosomes in the presence of 10 uM cytochalasin D to inhibit phagocytosis. After incubation with pHrodo-labeled synaptosomes, astrocytes were washed with DPBS, dissociated with Accutase, and pHrodo fluorescence was measured by flow cytometry. The gating strategy to determine the percent of phagocytic cells was based on the separation between the fluorescence histograms of samples treated or not treated with cytochalasin D.

Measurement of iNeuron viability in the presence of astrocyte conditioned media

Conditioned media was collected from iAstrocytes treated with vehicle control or IL-1α+TNF+C1q for 24 hours, spun down at 300 g for 10 min to remove dead cells, and transferred to day 17 iNeurons after removing the original iNeuron media. Unconditioned ScienCell Astrocyte Media was used as a negative control. After 72 hours, iNeuron viability was assessed by adding 10 ug/mL Hoechst and 1 μΜ TO-PRO-3 (ThermoFisher Scientific cat. no. T3605) in DPBS, incubating for 10 min at 37 °C, and then imaging on an IN Cell Analyzer 6000, using a 10X 0.45 NA objective, 2×2 binning, environmental control set to 37 °C and 5% CO₂, an aperture width of ~1 Airy unit, 200 ms exposure, and 4–9 fields per well. The percent of dead neurons (stained by TO-PRO-3) was calculated after image processing and segmentation with CellProfiler (see Data Analysis section).

Bulk RNA-seq library prep

hiPSC-derived astrocytes (iAstrocytes, TCW et al., Li et al., and Krencik et al. astrocytes), NPCs, and iNeurons were cultured in their respective media and treated with vehicle control or IL-1α+TNF+C1q for 24 hours. RNA extraction was then performed with Zymo Quick-RNA Microprep kit (Zymo Research cat. no. R1051). 50–100 ng of RNA was then used to construct bulk RNA-seq libraries using the QuantSeq 3’ mRNA-Seq Library Prep Kit FWD for Illumina (Lexogen cat. no. 015.96) following the manufacturer’s instructions. The concentration of QuantSeq libraries mRNA-seq library was quantified using the Qubit dsDNA HS Assay Kit (ThermoFisher Scientific cat. no. Q32851) on a Qubit 2.0 Fluorometer. Library fragment-length distributions were quantified with High Sensitivity D5000 Reagents (Agilent Technologies cat. no. 5067–5593) on the 4200 TapeStation System. The libraries were sequenced on an Illumina NextSeq 2000 instrument with single-end reads.

Antibody staining for flow cytometry

For antibody staining of cell-surface proteins (VCAM1, TFRC), hiPSC-derived or primary mouse astrocytes were dissociated with Accutase, washed with DPBS, incubated with conjugated primary antibodies for 20 min on ice in DPBS + 1% BSA, washed with DPBS, and then resuspended in DPBS + 1% BSA for flow cytometry. For antibody staining of intracellular proteins (C3), hiPSC-derived astrocytes were dissociated with Accutase, washed with DPBS, fixed with 2% paraformaldehyde for 10 min at RT, washed twice with DPBS + 0.5% Tween 20 (Millipore Sigma cat. no. P9461), incubated with unconjugated primary antibody for 20 min at RT, washed with DPBS + 0.5% Tween 20, incubated with conjugated secondary antibody for 20 min at RT, washed with DPBS + 0.5% Tween 20, and then resuspended in DPBS + 0.5% Tween 20 for flow cytometry. For dual staining of VCAM1 and C3, the protocol for intracellular protein staining was used.

Conjugated primary antibodies:

PE-Cy7 mouse anti-VCAM1 (1:80 dilution; BioLegend cat. no. 305818)

PE-Cy7 mouse anti-TFRC (1:80 dilution; BioLegend cat. no. 334112)

Unconjugated primary antibodies:

Mouse anti-C3 (1:250 dilution; BioLegend cat. no. 846302)

Rabbit anti-VCAM1 (1:250 dilution; Abcam cat. no. ab134047)

Conjugated secondary antibodies:

AF488 goat anti-mouse IgG (1:1000 dilution; ThermoFisher Scientific cat. no. A-11029)

AF568 goat anti-rabbit IgG (1:1000 dilution; ThermoFisher Scientific cat. no. A-11036)

Pooled CRISPRi screening

To identify transcriptional regulators of inflammatory reactivity, we created a custom sgRNA library targeting the human transcription factors²⁹, using 5 sgRNAs with the highest predicted activity scores from Horlbeck et al.³² per gene. The library was created by cloning a pool of sgRNA-containing oligonucleotides custom-synthesized by Agilent Technologies into our optimized sgRNA expression vector as previously described²⁴. To screen against the druggable genome, we used the H1 sub-library from Horlbeck et al.³². The transcription factor and druggable genome libraries were packaged into lentivirus as previously described⁸⁵. For each experimental replicate, ~10 million iAstrocytes were plated onto 4 Matrigel-coated 15-cm dishes, transduced with the lentiviral transcription factor or H1 sgRNA library so that 70–80% of cells were transduced, treated with vehicle control or IL-1α+TNF+C1q for 24 hours, and then incubated with pHrodo-labeled rat synaptosomes or stained for cell-surface VCAM1 (using the PE-Cy7 mouse anti-VCAM1 antibody) as shown in Fig. 3a. iAstrocytes were sorted into pHrodo high vs. low or VCAM1 high vs. low (top and bottom 30% of cells) using a BD FACSAria Fusion cell sorter at 5,000–10,000 events per second, and then pelleted for genomic DNA extraction. sgRNA abundances were then measured using next-generation sequencing as previously described⁸⁵. The screens were performed with two experimental replicates per condition.

We chose to transduce iAstrocytes with sgRNA libraries at 70–80% transduction because we found that we could not enrich for sgRNA-transduced iAstrocytes with puromycin selection, which changed the morphology and behavior of iAstrocytes. Although a transduction rate of 70–80% is higher than the usual target of 30%, it is not problematic for the following reasons: (i) in this regime, most cells will only be transduced with 1 or 2 sgRNAs, (ii) the majority of sgRNAs do not have a phenotype because most genes don’t have a phenotype in the screen; thus, the rate of cells with 2 sgRNAs that both cause phenotypes is very low, (iii) the screens are carried out at very high representation, i.e. each sgRNA is represented in hundreds of independent cells. Since pairings of sgRNAs during double infections can be assumed to be random, each sgRNA will mostly occur by itself or in combination with sgRNAs that don’t cause a phenotype, and the few instances where it is paired with other sgRNAs that do have a phenotype constitute a negligibly small fraction that will not substantially affect the average phenotype measured in a pooled screen.

CROP-seq

sgRNAs targeting the top hits from the CRISPRi screens and also candidate regulators selected based on literature were cloned into pMK1334. The concentration of each sgRNA plasmid was measured using the Qubit dsDNA HS Assay Kit on a Qubit 2.0 Fluorometer, and then the plasmids were pooled. iAstrocytes were transduced with lentivirus generated from the sgRNA pool at low multiplicity of infection so that <30% of cells were transduced, treated with vehicle control or IL-1α+TNF+C1q for 24 hours, and then sorted for sgRNA-transduced cells via FACS. Sorted iAstrocytes were then used as input for single-cell RNA-seq using Chromium Next GEM Single Cell 3’ v3.1 reagents (10X Genomics cat. no. PN-1000121, PN-1000127, and PN-1000213), loading ~45,000 cells per reaction into four reactions, with two reactions for vehicle control-treated iAstrocytes and two reactions for IL-1α+TNF+C1q-treated iAstrocytes. To facilitate association of single-cell transcriptomes with sgRNAs, sgRNA-containing transcripts were amplified as described in Tian et al.⁸⁵. The sgRNA-enrichment libraries were separately indexed and sequenced as spike-ins alongside the whole-transcriptome single-cell RNA-seq libraries on a NovaSeq 6000, recovering on average ~29,000 transcriptome reads and ~5,000 sgRNA-containing transcript reads per cell.

Measurement of cytokine concentrations

Conditioned media was collected from iAstrocytes, spun down at 300 g for 10 min to remove debris, and then frozen at −80 °C until analysis. The concentration of selected cytokines (IFN-α2a, IFN- β, IFN-γ, IL-6, CXCL10, GM-CSF) was measured by multi-array electrochemiluminescence using custom U-PLEX plates from Meso Scale Discovery following the manufacturer’s instructions.

Immunostaining of human neuropathological samples

Immunostaining of mouse tissue

Data analysis

DATA AVAILABILITY STATEMENT

Bulk RNA-seq data of hiPSC-derived astrocytes generated in this study shown in Extended Data Fig. 3 are available on the Gene Expression Omnibus (GEO) under accession code GSE182307. The raw single-cell RNA-seq data and UMI count matrices from the CROP-seq experiment are available on GEO under accession code GSE182308. Processed data from the CRISPRi screens can also be interactively explored on CRISPRbrain (https://www.crisprbrain.org/screens/) by selecting “Astrocyte” as the screening cell type. The public databases used in this study include Enrichr (https://maayanlab.cloud/Enrichr/), GENCODE (https://www.gencodegenes.org/human/), GEO (https://www.ncbi.nlm.nih.gov/geo/), and Synapse (https://www.synapse.org/). Previously published datasets available on GEO reanalyzed in this study include GSE143598, GSE120411, GSE76097, GSE148611, and GSE130119; previously published datasets available on Synapse reanalyzed in this study include syn21861229.

We thank Brandon Desousa, Vukasin Jovanovic, Zuzana Krejciova, and Nawei Sun for contributions to preliminary studies and discussions. We thank Professors Anna Molofsky, Aimee Kao, and Michael Oldham for serving on KL’s thesis committee. We thank members of the Kampmann lab (Greg Mohl, Sydney Sattler, Olivia Teter) for discussions and feedback on the manuscript. We thank Brian Woo for cloning the transcription factors sgRNA library, and Biswa Ramani for help with obtaining primary mouse astrocytes. We thank the Conklin lab for the gift of the WTC11 hiPSC line. This research was supported by NIH grant F30 AG066418 to KL, CIRM grant EDUC4-12812 & NIH grant T32 NS115706 to IVLR, Chan Zuckerberg Initiative Ben Barres Early Career Acceleration Awards to ESL and M. Kampmann, NIH/NINDs grants (R01NS097551, P01NS083513, R21NS119954) to SF, and NIH grants P30 EY02162-39 and R03AG063157 to EMU. SF is a Harry Weaver Neuroscience Scholar of the National Multiple Sclerosis Society. The TCW-1E44 iPSC line was generated with the support of NIH NIA K01AG062683 (J. TCW) and the Druckenmiller Fellowship from the New York Stem Cell Foundation (J. TCW). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.