Proteomic profiling of interferon-responsive reactive astrocytes in rodent and human

Abstract

Astrocytes are a heterogeneous population of central nervous system glial cells that respond to pathological insults and injury by undergoing a transformation called ‘reactivity’. Reactive astrocytes exhibit distinct and context-dependent cellular, molecular, and functional state changes that can either support or disturb tissue homeostasis. We recently identified a reactive astrocyte sub-state defined by interferon-responsive genes like Igtp, Ifit3, Mx1, and others, called interferon-responsive reactive astrocytes (IRRAs). To further this transcriptomic definition of IRRAs, we wanted to define the proteomic changes that occur in this reactive sub-state. We induced IRRAs in immunopanned rodent astrocytes and human iPSC-differentiated astrocytes using TNF, IL1α, C1Q, and IFNβ and characterized their proteomic profile (both cellular and secreted) using unbiased quantitative proteomics. We identified 2335 unique cellular proteins, including IFIT2/3, IFITM3, OASL1/2, MX1/2/3, and STAT1. We also report that rodent and human IRRAs secrete PAI1, a serine protease inhibitor which may influence reactive states and functions of nearby cells. Finally, we evaluated how IRRAs are distinct from neurotoxic reactive astrocytes (NRAs). While NRAs are described by expression of the complement protein C3, it was not upregulated in IRRAs. Instead, we found ~90 proteins unique to IRRAs not identified in NRAs, including OAS1A, IFIT3, and MX1. Interferon signaling in astrocytes is critical for the antiviral immune response and for regulating synaptic plasticity and glutamate transport mechanisms. How IRRAs contribute to these functions is unknown. This study provides the basis for future experiments to define the functional roles of IRRAs in the context of neurodegenerative disorders.

Graphical Abstract

INTRODUCTION

Astrocytes, the most abundant glial cells in the mammalian central nervous system (CNS) are a heterogeneous population of cells with critical roles in health, development, and disease (Han, Kim, Molofsky, & Liddelow, 2021). Under normal physiological conditions, astrocytes provide trophic support to neurons, participate in the formation, function, and elimination of synapses (Allen, 2013; Allen et al., 2012; Ç. Eroglu et al., 2009), help maintain the integrity of the blood-brain barrier (Alvarez, Katayama, & Prat, 2013), phagocytose protein aggregates (Prakash et al., 2021) and synapses (Chung et al., 2013; Tasdemir-Yilmaz & Freeman, 2014) reuptake and recycle glutamate and other neurotransmitters (Li, Schaffner, & Barker, 1999; Q.-Y. Liu, Schaffner, Li, Dunlap, & Barker, 1996; Rusakov & Lehre, 2002), and perform other functions to regulate CNS homeostasis. Most of these functions are performed by regulation of proteins in the cell soma or by secreting proteins into the extracellular environment to affect the health and function of neighboring cells. In response to pathogenic insults, disease, or following infection and injury, homeostatic astrocytes undergo a transformation called ‘reactivity’ that causes profound changes to their molecular signature and function (Han et al., 2021). In addition to robust changes in gene expression, morphology, and function (Hasel, Aisenberg, Bennett, & Liddelow, 2023), these heterogeneous populations of reactive astrocytes are now beginning to be defined at the protein and lipid level (Guttenplan et al., 2021; Soto et al., 2023). These additional layers of reactive astrocyte complexity are helping to expand our understanding of the changes associated with reactive astrocytes and whether they hinder or support the CNS environment.

Recently we used single cell RNA-seq and spatial transcriptomics to define a novel sub-state of reactive astrocytes called ‘interferon-responsive reactive astrocytes’ (IRRAs) in the adult mouse brain following acute systemic inflammation (Hasel et al., 2021). Like many sub-states of reactive astrocytes, IRRAs represent a lowly abundant population - with less than 5% of all cortical astrocytes undergoing this transcriptomic change. IRRAs are also closely associated with blood vessels on the surface of the brain and around the ventricles, suggesting a strategic position that enables interactions with both central and peripheral immune cells. We took advantage of published scRNA-seq datasets to define IRRAs in several mouse models of disease including the 5xFAD model of Alzheimer’s disease (AD), the demyelinating mouse model experimental autoimmune encephalomyelitis (EAE), and acute stab wound injury (Hasel et al., 2021). Not restricted only to rodent models of disease, IRRA marker genes and proteins have been reported in human post mortem tissue from patients with AD (Xia, Bacskai, Knowles, Qin, & Hyman, 2000) and multiple sclerosis (Sørensen et al., 1999; Stüve & Ransohoff, 2009), as well as in the mouse EAE demyelination model (Tani et al., 1996). More recently we report that IRRAs also localize to amyloid plaques in the APP/PS1 mouse (Castranio et al., 2022). Responding to inflammation as quickly as 3 hours (Hasel et al., 2021), IRRAs can be defined at the transcriptomic level by a suite of stereotypical reactive astrocyte genes (e.g. Serpina3n, Lcn2, etc.) as well as a number of IRRA-specific transcripts like Igtp, Ifit1, Ifit3, Oasl, Mx1, Cxcl10, and Gbp2 (Hasel et al., 2021). Notably, the gene expression profile of IRRAs is distinct from that of neurotoxic reactive astrocytes (NRAs) - a previously identified sub-state of reactive astrocytes that are induced by three microglial inflammatory factors (tumor necrosis factor (TNF), interleukin alpha (IL1α), and complement component 1Q (C1Q) (Liddelow et al., 2017). While NRAs are highly inflammatory and toxic to damaged neurons and oligodendrocytes in vitro and in vivo, the exact function of IRRAs including their influence the health of their neighboring CNS cells is not known.

In addition to describing their anatomical location we have previously shown that the transcriptomic signature of IRRAs in vivo can be recapitulated in vitro by treating serum-free cultured astrocytes with a cocktail of TNF, IL1α, C1Q, and IFNβ (Hasel et al., 2021). IFNβ is a type I interferon secreted by many different cell types including innate immune cells like macrophages and dendritic cells, as well as non-immune cells such as fibroblasts and epithelial cells. Traditionally, IFNβ is associated with a wide array of functions including anti-inflammatory, antiviral, and pro-inflammatory responses. In astrocytes specifically, type I interferon signaling regulates synaptic plasticity, glutamate transport mechanisms, and hippocampal function (Hosseini et al., 2020). IFNβ can also directly alter the expression of an astrocyte-specific glutamate/aspartate transporter (GLAST/Slc1a3) (Costello & Lynch, 2013) further highlighting the importance of type 1 interferon signaling in astrocytes.

While gene expression profiles generated at the single cell/nuclei level is a powerful way to define new sub-states of reactive astrocytes, multimodal characterization of the changes that delineate them from physiologically normal astrocytes remains key to understanding their roles in pathological contexts. Defining the protein profiles of reactive astrocyte sub-states like IRRAs is therefore beneficial as proteins dictate many aspects of cell function and are also the primary immunoregulatory signaling molecules that affect the states of neighboring cells. Further, the formation of mRNA is only the first step in a long sequence of events resulting in the synthesis of a protein and many different protein isoforms can be generated from a single gene during the processing of mRNA (as a result of alternative splicing, mRNA editing, and polyadenylation) (Franks, Airoldi, & Slavov, 2017; Raj & van Oudenaarden, 2008; Wilkins, Sanchez, Williams, & Hochstrasser, 1996). In this study, we sought to answer three questions to better understand the molecular underpinnings of IRRAs: What is the proteomic and secretomic signature of IRRAs? Does the proteome of IRRAs correlate with their transcriptome? Finally, how are IRRAs different from NRAs? We used Tandem Mass Tag (TMT)-based relative quantitative proteomics to answer these questions and identify and validate new molecular markers for IRRAs. Understanding the proteomic changes that occur in different sub-states of reactive astrocytes will bring us one step closer to understanding their function in the context of different pathological contexts.

1. METHODS

1.1. Animals

Sprague Dawley rats from Charles River were used to obtain postnatal day (P)5/6 pups. All procedures were conducted in accordance with the animal care standards of the National Institute of Health and approved by NYU Grossman School of Medicine’s Institutional Animal Care and Use Committee.

1.2. Cell Culture

1.2.1. Rat astrocytes

Astrocytes were isolated via immunopanning from postnatal day (P)5/6 rat forebrain and cultured using a previously described well-established serum-free protocol (Foo et al., 2011). Briefly, 6–8 rat pups were used in each individual experiment (biological replicate). The cortices from the 6–8 rat pups were pooled, and enzymatically digested using papain followed by mechanical dissociation to obtain a single-cell suspension. The single cells were then incubated on successive negative immunopanning plates to remove endothelial cells (BSL1), microglia and macrophages (secondary antibody only), immune cells (lymphocytes, neutrophils, macrophages), and oligodendrocyte lineage cells (O4) before positively selecting for astrocytes using ITGB5. Isolated astrocytes were cultured in a defined, serum-free medium containing 50% neurobasal, 50% DMEM, 100 U/ml penicillin, 100 μg/ml streptomycin, 1 mM sodium pyruvate, 292 μg/ml L-glutamine, 1× SATO, 5 μg/ml of N-acetyl cysteine and 5 ng/ml Heparin Binding EGF-like growth factor (HBEGF; Peprotech #100–47). Purified astrocytes were maintained in serum-free media to retain a non-reactive gene expression profile (Foo et al., 2011, Liddelow et al., 2017).

1.2.2. Generation of interferon-responsive reactive astrocytes

To induce physiologically normal astrocyte to adopt an IRRA phenotype, the purified astrocytes were maintained in serum-free media for 6 days and then treated with rat IL1α (3 ng/ml, Sigma #I3901), human TNF (30 ng/ml, Cell Signaling Technology #8902SF), human C1Q (400 ng/ml, MyBioSource #MBS143105) and rat IFNβ (1000 U/ml, Sigma #I8907) for 24 hours. Cell lysate and conditioned media samples were collected and stored at −80 °C until proteomic profiling (see below).

1.2.3. Human Induced Pluripotent Stem Cell Lines

We used six hiPSC lines that were reprogrammed from skin fibroblasts of healthy donors using the fully automated New York Stem Cell Foundation Global Stem Cell Array^® (Paull et al., 2015). Demographic information for each line can be found in Table 1. All lines passed stringent quality controls, including karyotyping, mycoplasma testing, pluripotency efficiency, and identity match. Astrocytes derived from these lines have been extensively characterized in our previous studies (Barbar, Jain, et al., 2020; Barbar, Rusielewicz, Zimmer, Kalpana, & Fossati, 2020).

Table 1.

Demographic information for each hiPSC line.

Line ID	Age at biopsy	Sex
051064-01-MR-007	53	M
051104-01-MR-040	56	F
051121-01-MR-017	52	F
050659-01-MR-013	64	F
050743-01-MR-023	50	M
051106-01-MR-045	57	F

1.2.4. Human Induced Pluripotent Stem Cell differentiation into astrocytes and induction of interferon-responsive reactive astrocytes

All cultures were maintained in a 37 °C incubator with 5% CO₂. For astrocyte differentiation, hiPSCs were plated at 1.5×10⁵ cells per well in a 6-well geltrex-coated plate and fed with mTeSR1 maintenance medium with 10 μM Y27632 for the first 24 hours after seeding (Marotta, Rao, & Fossati, 2022). Differentiation was performed using a serum-free protocol that we have previously established; differentiation media and patterning reagents are fully detailed in our previous publications (Barbar, Jain, et al., 2020; Barbar, Rusielewicz, et al., 2020; Labib et al., 2022). On Day 79, neural cultures were enzymatically dissociated using Accutase (ThermoFisher, Waltham, MA, United States; A1110501) and fluorescence-activated cell sorting was used to collect CD49f+ astrocytes as previously described (Barbar, Jain, et al., 2020). To model an interferon-responsive reactive astrocyte cell state, purified CD49f+ astrocytes were maintained in PDGF medium until day 93, when they were treated with the following recombinant proteins: human TNF (30 ng/ml, R&D Systems; 210-TA-020), rat IL1α (3 ng/ml, Sigma #I3901), human C1Q (400 ng/ml, MyBioSource; MBS143105), and rat IFNβ (1000 U/ml, Sigma #I8907) for 24 hours. PDGF medium used for culturing iPSC-derived astrocytes contains DMEM/F-12, PenStrep (100x), 2-Mercaptoethanol (1000x), MEM non-essential amino acids (1x), N2 supplement (100x), B27 supplement without vitA (50x), human insulin solution (25 μg/ml), hrPDGRaa (10 ng/ml), hrIGF1 (10ng/ml), hrHGF (5 ng/ml), NT3 (10 ng/ml), T3 (60 ng/ml), Biotin 100ng/ml, cAMP (1 μM). The conditioned media from all 6 lines were collected, snap froze using Corning CoolRack in dry ice, and stored at −80 °C until proteomic profiling (see below).

1.3. Tandem Mass Tag Proteomics

1.3.1. Preparation of cell samples for proteomics

We performed five biological replicates (n=5) to obtain five IRRA and five PBS-treat control samples for mass spectrometry. Collected astrocyte cell pellets were processed as follows. Briefly, 8 M urea containing protease inhibitors (Complete Mini, Roche Diagnostics, GmbH Mannheim, Germany) was used to solubilize the cells and processed as previously described (D’Acunzo et al., 2023; Villén & Gygi, 2008), with a few modifications: After solubilization, cysteines were reduced with 5 mM final concentration of Dithiothreitol (Pierce DTT, Thermo Scientific, Rockford, IL, USA) followed by alkylation using 14 mM Iodoacetamide (Sigma-Aldrich, St.Louis, MO, USA). Finally, digestion was carried out “in-solution” in the presence of 1.6 M urea, supplemented with protease inhibitors and 5 ng/nL endoproteinase Lys-C, (Mass Sectrometry Grade, Promega, Madison, WI, USA) which was added to each sample for 2 hours, followed by Trypsin (Gold, Mass Spectrometry Grade (Promega, Madison, WI, USA), at 5 ng/μl, overnight. The next day, the digestion was stopped by acidification to 0.1% trifluoroacetic acid (v/v) (Thermo Scientific, Rockford, IL, USA). After brief centrifugation, the peptides were isolated using solid phase extraction (SPE) columns (SepPak Vac 1ccm, tC18 Cartridges, Waters Corporation, Milford, MA). TMT labelling of purified peptides and the remaining proteomics procedures were performed as previously described (Erdjument-Bromage, Huang, & Neubert, 2018; Huang et al., 2017; Lawlor, Nazarian, Lacomis, Tempst, & Villanueva, 2009; Xu et al., 2021) with minor adjustments, as follows: TMT labels 126, 127N, 127C, 128N, 128C, 129N, 129C, 130N, 130C, and 131 (TMT10 plex Mass Tag Labeling Kit, Thermo Scientific, Rockford, IL, USA) were added to each sample at a w/w label/peptide ratio of 12:1 and mixed briefly by vortexing. The mixture was incubated at room temperature for 1 hour, quenched with 10 μl 5% w/v hydroxylamine (Sigma-Aldrich, St. Louis, MO, USA) for 15 minutes, and finally acidified with 10 μl 10% v/v formic acid (Thermo Scientific, Rockford, IL, USA). An aliquot from each reaction was desalted (Rappsilber, Mann, & Ishihama, 2007) with Empore C18 High Performance Extraction Disks (3M, St. Paul, MN, US). The eluted peptides were partially dried under vacuum and analyzed by liquid Chromatography with tandem mass spectrometry (LC-MS/MS) with a Thermo Easy nLC 1000 system coupled online to a Q Exactive High Field Orbitrap mass spectrometer with a NanoFlex source (Thermo Fisher Scientific) (Huang et al., 2017) to determine labelling efficiencies, which were found to be 97%–98%. To ensure equal amounts of labelled peptides, samples were mixed and analyzed in test runs by LC-MS/MS. The final sample mixture containing mixed TMT channels was prepared by readjusting the volume of each sample so that they contained equal amounts of labelled peptides. The mixture was desalted by using SEP cartridges. Eluted peptides were analyzed in replicates by LC-MS/MS.

1.3.2. Preparation of conditioned media samples for proteomics

Conditioned media from rodent and human astrocytes that were treated with TICI or PBS (n=5 biological replicates) were collected for TMT proteomics as previously described (Lawlor et al., 2009). The 2 ml initial volume was concentrated to 50 μl using a Microcon-10kDa Centrifugal Filter Unit with Ultracel-10 membrane. The final concentrated conditioned media were subjected to SDS-PAGE from which 1 cm gel bands containing all the proteins were cut out for tryptic digestion and peptide extraction followed by TMT-labeling and mass spectrometry as described above.

1.3.3. Data analysis

All data were analyzed by MaxQuant proteomics software (version 1.5.5.1) with the Andromeda search engine (Cox et al., 2011) using mouse, human, and rat databases (Supplementary Figure 1a). The mouse (Mus musculus) Uniprot database had 21,989 entries (downloaded on April 10, 2022), human (Homo sapiens) Uniprot database had 20,950 entries (downloaded on May 3, 2022) and rat (Rattus norvegicus) Uniprot database had 22,825 entries (downloaded on April 9, 2022). Reporter ion mass tolerance was set to 0.01 Da, the activated precursor intensity fraction value was set to 0.75, and the false discovery rate was set to 1% for protein, peptide-spectrum match, and site decoy fraction levels. Peptides were required to have a minimum length of seven amino acids and a mass no greater than 4,600 D. The reporter ion intensities were defined as intensities multiplied by injection time (to obtain the total signal) for each isobaric labeling channel summed over all MS/MS spectra as previously validated (Tyanova, Temu, & Cox, 2016) (Supplementary Figure 2a). Reporter ion intensities plotted from each sample of each replicate followed a normal distribution (Supplementary Figure 1b), and the protein reporter ion intensity plots between 9 samples showed very good (>99%) correlations between samples (Supplementary Figure 1c). Since the reporter ion intensities for the PBS2 sample was very low, we omitted this sample from all our analysis moving forward and only considered 9 samples in total between TICI and PBS-treated rat astrocytes. The histograms of reporter ion intensities of proteins detected in each TMT channel showed normal distribution of intensities (Supplementary Figure 2b). Mass spectra were subjected to label-free quantitation by using MaxQuant proteomics data analysis workflow (version 1.5.5.1) with the Andromeda search engine (Cox et al., 2011; Tyanova, Temu, & Cox, 2016) and P values, log₂ fold change values, etc. were analyzed on the Perseus platform available with MaxQuant (Tyanova, Temu, Sinitcyn, et al., 2016). We performed t-tests using Benjamini-Hochberg correction which inherently controls the false discovery rate using sequential modified Bonferroni correction for multiple hypothesis testing (Tyanova, Temu, Sinitcyn, et al., 2016). The complete lists generated from analyzing the TICI vs. PBS rodent cellular proteomics data are available as Supplementary Table 1, TICI vs. PBS rodent secretome data as Supplementary Table 2, and TICI vs. PBS human secretome as Supplementary Table 3.

1.4. Pathway enrichment analysis

We performed pathway analysis using the STRING platform (Szklarczyk et al., 2023; von Mering et al., 2003) which extracts curated data from KEGG, Gene Ontology, Reactome, Biocarta, and BioCyc database to predict pathways to be affected given the list of proteins. All the proteins from the proteomics analysis of rodent cells were considered for pathway enrichment analysis. Functional enrichment analysis was performed on the STRING database (version 11.5), where each protein was ranked by its log₂ fold change value (TICI vs. PBS). The inputted proteins were significantly enriched in a total of 23 pathways in the “biological processes” category.

1.5. Western Blot

For western blot analysis, we used three biological replicates (n=3) for downstream processing as follows. TICI and PBS-treated immunopanned astrocytes were collected with cold RIPA buffer containing 1X protease inhibitor cocktail (Cell Signaling Technology), agitated at 4 °C for 45 minutes, briefly sonicated, and stored at −80 °C. The total protein concentration of each sample was evaluated using the microplate procedure of the Pierce bicinchoninic acid assay (Thermo Fisher Scientific) per the manufacturer’s instructions. 20 μg of total protein from each sample was diluted in 4X Laemmli buffer and incubated at 95 °C for 5 minutes. The protein samples were run on a 12% gel (Bio-Rad Mini-PROTEAN TGX Precast Gel) at 90 V for 80 minutes. The proteins from the gel were then transferred onto a PVDF membrane using the Trans-Blot Turbo Transfer System (Bio-Rad). The membranes were incubated with primary antibodies overnight at 4 °C followed by secondary antibodies (LI-COR) for 1 hour at room temperature. The resulting protein bands were developed using the LI-COR kit per the manufacturer’s instructions. Beta-actin or GAPDH antibodies were used as loading controls. Information about the primary and secondary antibodies along with their catalog numbers and dilutions are provided in Supplementary Table 4. The final bands were visualized on the Odyssey Imager (LI-COR) and pixel density quantified using ImageJ2 version 2.9.0/1.53t.

1.6. Cytokine Array

Cytokines present in the TICI-treated rodent astrocyte secretome were quantified using the Proteome Profiler Rat XL Cytokine Array kit (R&D Systems #ARY030) according to the manufacturer’s instructions. Briefly, the conditioned media samples from TICI and PBS-treated astrocytes were collected from two independent experiments and applied on to the PVDF membranes that were precoated with the captured antibodies. The membranes were then washed with the washing buffer and incubated with the detection antibodies. The presence or absence of the cytokines were visualized with streptavidin-HRP mixture and the chemi reagent mix. The immunoblot images were captured using the Bio-Rad Gel Doc XR imaging system and the pixel density of the cytokine spots on the membranes were quantified using ImageJ2 version 2.9.0/1.53t.

1.7. Comparative analysis of astrocyte transcriptome and proteome

The abundances of proteins detected by mass spectrometry were compared to RNA sequencing data from Hasel et al. 2021 in which primary rat astrocytes were treated in vitro with PBS (control) or TNF, IL1α, C1Q, and IFNβ, (Hasel et al., 2021). RNA sequencing FASTQ files were obtained from the NCBI Gene Expression Omnibus Sequence Read Archive (accession GSE165069). FASTQ files were adaptor trimmed using TrimGalore v0.6.7 and mapped to the Ensembl mRatBN7.2 (release 104) transcriptome using Salmon v1.9.0 (Patro, Duggal, Love, Irizarry, & Kingsford, 2017). The Surrogate Variable Analysis (sva) package v3.42.0 (Leek, Johnson, Parker, Jaffe, & Storey, 2012) was used to model and remove unwanted noise, and DESeq2 v1.34.0 (Love, Huber, & Anders, 2014) was used for differential expression testing with log-fold change shrinkage (Zhu, Ibrahim, & Love, 2019). Proteins detected by mass spectrometry were next mapped to their corresponding genes using MaxQuant annotations. Downstream comparison was performed using only genes for which a corresponding protein was detected, and which were detected in the Hasel et al. 2021 RNA-seq data and not determined by DESeq2 to be an extreme count outlier (those with a differential expression p-value of ‘NA’).

Genes corresponding to multiple proteins were compared in an all-to-all manner, with all possible protein-gene pairs included in downstream comparison. This was done as follows: After filtering out the genes for which there was no corresponding protein detected in the proteomics data, and filtering out the proteins for which no corresponding gene transcript was detected in the RNA sequencing data, 2,107 genes remained in the RNA-seq dataset and 2,076 proteins remained in the proteomics dataset. Among those, 47 genes from the RNA-seq dataset corresponded to two different proteins in the proteomics data, as annotated by MaxQuant. (2,107 unique genes + 47 duplicate genes = 2,154 protein-gene pairs). Among the 2,076 unique proteins in the filtered proteomics dataset, 62 proteins correspond to more than one gene as annotated by MaxQuant (52 proteins map to two genes, 7 proteins map to three genes, 1 protein maps to 4 genes, 1 protein maps to 5 genes, and 1 protein maps to 6 genes; 2,076 unique proteins + 52 duplicate entries + 7 triplicate entries (x2) + 1 quadruplicate entry (x3) + 1 quintuplicate entry (x4) + 1 sextuplicate entry (x5)). This resulted in 2,154 protein-gene pairs.

Pearson correlations were calculated and plotted using ggpubr v0.5.0 (Kassambara, 2023) comparing log₁₀ TPM (transcripts per million) from the RNA-seq data and log₁₀ TMT intensity from the proteomics data, as well as comparing the log₂-fold change in gene expression estimated by DESeq2 to the log₂-fold change in protein abundance estimated by Perseus (Tyanova, Temu, Sinitcyn, et al., 2016), for each protein-gene pair. For the latter correlation plots, protein-gene pairs were categorized as differentially abundant at the protein-level with a Perseus p-value less than 0.05, and categorized as differentially expressed at the RNA-level with a DESeq2-adjusted p-value less than 0.05. Protein-gene pairs labeled as significant at both the protein and RNA level were defined as “concordant” if both protein and gene expression were changed in the same direction (i.e., up-up or down-down) or “discordant” if the two differed in direction. The lists of these common markers are provided in Supplementary Table 5.

1.8. Comparative analysis of neurotoxic and interferon-responsive reactive astrocytes

To compare the common and unique proteins between IRRAs and NRAs, we used data from a previously published proteomic analysis of NRAs (Guttenplan et al., 2021) which is also provided in Supplementary Table 6. We first selected all the proteins from this dataset that differed significantly between NRAs and control astrocytes (203 total proteins, based on FDR<0.1) and from the dataset from our current study comparing IRRAs and control astrocytes (104 total proteins, based on P<0.05). Then, we identified the common proteins between the two groups (based on gene names) and also the proteins unique to either of the two groups.

1.9. Immunolabeling of proteins in human post-mortem tissues of Alzheimer’s and healthy patient brains

Immunohistochemistry was performed on post-mortem brain samples from Alzheimer’s disease (AD) and age-matched non-symptomatic (NS) donors (demographic and clinical information is listed in Table 2). The formalin-fixed paraffin-embedded human AD and NS tissues were sectioned at 5 μm and mounted onto glass microscope slides. The slides were placed on a 60°C hotplate for 30 minutes and moved through HistoChoice clearing aged (2 washes, 5 minutes each) to dewax the paraffin. Next, the sections were rehydrated through decrease concentrations of ethanol: 100% (2 washes, 5 minutes each), 95% (1 wash, 5 minutes), 70% (1 wash, 5 minutes), 1x PBS (1 wash, 5 minutes). The sections were then treated with a blocking buffer (10% normal goat serum and 0.4% Triton X-100 prepared in 1x PBS) for 1 hour at room temperature in a humid chamber. Following blocking, they were treated with the primary antibodies (Supplementary Table 4) and incubated overnight (16–20 hours) at 4 °C in a humid chamber. The following day, sections were washed with 1×PBS (3 washes, 10 minutes each) and treated with the secondary antibodies (Supplementary Table 4) for 1 hour at room temperature in a darkened humid chamber. To remove unbound secondary antibodies, three 10-minutes washes in 1×PBS was performed followed by a nuclear co-stain with DAPI (0.5 μg/ml prepared in PBS) for 1 minute. The sections were then treated with TrueBlack Lipofuscin Autofluorescence Quencher (Biotium) for 1 minute to reduce background fluorescence. Finally, sections were mounted with Fluoromount-G, and a 22 × 50 mm cover glass was placed onto the sections. Images were acquired on a Keyence BZ-X710 using a 60× oil objective. For quantification, tissue sections from 3 different AD and NS individuals and 3 different ROI per section were used. Unpaired t-tests were performed using the means from each tissue.

Table 2. Clinical characterization of post-mortem brain patient donors.

Sex, age, and clinical diagnosis (amyloid, Braak, CERAD scores) of human patient post-mortem brain samples used for immunostaining in this study.

Sample ID	Sex	Age (yrs)	A (Amyloid)	B (Braak)	C (CERAD)	Clinical diagnosis
TN13-11-A	Male	88	A2	B2	C2	Non-symptomatic
TN13-21-A	Female	49	A3	B3	C3	Non-symptomatic
TN13-29-A	Female	89	A1	B1	C1	Non-symptomatic
TN13-66-A	Female	95	A0	B0	C0	Non-symptomatic
TN11-45-A	Male	69	A4	B4	C4	Alzheimer’s disease
TN12-60-A	Male	92	A6	B6	C6	Alzheimer’s disease
TN16-37-A	Female	92	A7	B7	C7	Alzheimer’s disease
TN17-49-A	Male	94	A8	B8	C8	Alzheimer’s disease

1.10. Statistics

Statistical significance was determined by unpaired 2-tailed Student t tests for comparison of TICI and PBS groups based on their log₁₀ reporter ion intensities, pixel densities, or antibody intensities using GraphPad Prism 9 version 9.5.1. The number of independent experiments (n) is noted in figure legends. All data were from biological triplicates or quadruplicates (indicated in the figure legends). We used R to make the heatmap, PCA, and scatter plots. The code for comparing the proteomics and transcriptomics datasets is uploaded on our lab GitHub repository (https://github.com/liddelowlab/irra_proteomics).

2. RESULTS

2.1. Proteomic profiling of rodent control and interferon-responsive reactive astrocytes identifies thousands of proteins

To generate IRRAs in vitro, we isolated and cultured rodent astrocytes and treated them with a cocktail of TNF, IL1α, C1Q, and IFNβ (TICI) for 24 hours (Figure 1a). Astrocytes treated with PBS (vehicle) were used as control cells. After treatment we isolated total proteins from cell lysates and processed them for quantitative tandem mass-tag (TMT) proteomics. We identified a total of 2,335 proteins based on the intensities of the reporter ions obtained for each sample, out of which 102 proteins were significantly different in amount between TICI- and PBS-treated astrocytes based on the P value<0.05, out of which 76 were abundant and 26 were depleted in the TICI astrocytes, respectfully (based on positive and negative log₂ fold change values) (Supplementary Table 1; Figure 1b). The principal component analysis (PCA) of the top 10% highly variable proteins showed a clear separation of the TICI-treated and PBS-treated astrocytes (Figure 1c). Some of the most relatively abundant and depleted proteins can be clearly visualized in both scatter plot (Figure 1d) and heatmap (Figure 1e) which highlight a clear shift in the proteome of IRRAs compared to control astrocytes. Proteins specific to interferon response, like interferon induced protein with tetratricopeptide repeats 2 and 3 (IFIT2, IFIT3), interferon-induced transmembrane protein 3 (IFITM3), and antiviral response proteins like 2’-5’-oligoadenylate synthetase like (OASL), and MX dynamin like GTPase 1 and 2 (MX1, MX2) were among the most relatively abundant proteins in IRRAs. The transcription factor STAT1 (signal transducer and activator of transcription 1)—a critical regulator of interferon responses (Platanias, 2005)—was also upregulated in IRRAs, as was arginosuccinate synthase 1 (ASS1), a protein involved in urea cycle and energy metabolism. Another upregulated protein, CD44 is a cell surface glycoprotein reported to suppress immune responses and exert protective roles in inflammation (Neal, Boyle, Budge, Safadi, & Richardson, 2018) and may be important in regulating reactive astrocyte interferon response. Of the depleted proteins, the urea transporter protein solute carrier family 14 member 1 (SLC14A1), gap junction protein alpha 1 (GJA1; also known as connexin 43), complexin 2 (CPLX2), and alpha-2-macroglobulin (A2M) were the most significantly changed in amount in the TICI-treated astrocytes. Slc14a1 is expressed by astrocytes both in mice and human and is not expressed by other CNS cell types (Zhang et al., 2014, 2016) - suggesting a sole role of astrocytes in CNS urea trafficking in response to infection (Gropman, Summar, & Leonard, 2007). GJA1 is also an astrocyte-specific protein that is critical for formation of astrocyte networks and transfer of metabolic resources between astrocytes. Depletion of this molecule occurs during stress and injury where astrocytes putatively mitigate cytotoxicity (Almad et al., 2016). We validated some of the top proteins identified by TMT proteomics using immunoblotting (Supplementary Figure 4a,d) and show that ASS1 and CD44 are more abundantly present in the cellular lysates of TICI-treated astrocytes compared to the PBS-treated cells.(Supplementary Figures 4b–c,e–f, 5a–b). We also validate that IRRAs upregulate phosphorylated STAT1 (pSTAT1) which was not seen in the control astrocytes (Supplementary Figures 3d, 4h–i), confirming the astrocytic immune activation in response to interferon (Platanias, 2005). Notably, we did not specifically detect pSTAT1 in our mass spectrometry data, however, an increase in both STAT1 (mass spectrometry; Supplementary Figure 4g) and the phosphorylated protein (western blotting) suggests activation of STAT1 may mediate conversion to an IRRA phenotype. Overall, these data indicates that the astrocyte interferon response is directed by a combination of several proteins that are likely involved in downstream functions.

Figure 1. Proteomic analysis of interferon-responsive reactive rodent astrocytes.

(a) Experimental protocol for the isolation and culture of astrocytes from rat brains, treatment of TICI (TNF, IL1α, C1Q, and IFNβ) or PBS factors, collection and processing of cells and conditioned media samples for LC-MS/MS. (b) Total number of significant up and down proteins in TICI-treated versus PBS-treated rodent astrocytes based on P<0.05. (c) Principal component analysis (PCA) plot of the top 10% of the most differentially regulated cellular proteins between TICI (red) and PBS (blue) rodent astrocytes. (d) Scatter plot illustrating the abundant (red) and depleted (down) cellular rodent proteins, with a few key ones highlighted. (e) Heat map of the top 10% most variable cellular rodent proteins (represented by gene symbols) plotted using the Z-scored TMT intensity values shows a clear separation of the TICI and PBS groups indicating a dramatic proteome change in IRRAs compared to control astrocytes. n=5 biological replicates.

To evaluate the pathways related to the identified proteins, we performed pathway enrichment analysis (Supplementary Figure 3a). The top pathways predicted from our proteomics data were associated with the cellular response to interferon alpha and interferon beta – both type I interferons that share some overlapping signaling mechanisms (Ivashkiv & Donlin, 2014). In general, type I interferons trigger antiviral responses by binding to the interferon-α/β receptor (IFNAR), leading to STAT1 activation (Supplementary Figure 3b). This activation ultimately results in the transcription of interferon-stimulated genes (ISGs) (Schoggins, 2019) (Supplementary figure 3b). Our unbiased proteomics detected the direct functional products of ISGs, and the resulting interferon-stimulated proteins may act alone or in synergy to achieve cellular outcomes, such as antiviral defense (negative regulation of viral genome replication and viral processes, highlighted in our pathway analysis), antigen presentation, cytokine stimulation, antiproliferative activities, and the stimulation of adaptive immunity (Supplementary Figures 3a,b). Finally, we also found pathways related to actin cytoskeleton remodeling and the reorganization of cellular anatomical structures, which possibly indicate the morphological changes that astrocytes undergo in response to interferons.

2.2. Rodent and human interferon-responsive reactive astrocytes secrete PAI1 and other proteins into their environment

We next asked what proteins IRRAs secrete that may influence neighboring CNS cells. We collected conditioned media from TICI- and PBS-treated astrocytes and subjected them to TMT proteomics analysis. We identified 73 unique proteins in the rodent astrocyte secretome, out of which 10 were significantly different in abundance (P<0.05) (Supplementary Table 2). Of these, the plasminogen activator inhibitor-1 (PAI1; encoded by Serpine1), which is involved in cell migration and tissue remodeling (Baker & Strickland, 2020; Wilhelm et al., 2018), was the topmost relatively abundant protein in the TICI-treated astrocyte secretome followed by complement factor B (CFB), desmoplakin (DSP), and mitotic arrest deficient 1 like 1 (MAD1L1) (Figure 2a–d). In vitro studies have shown that Serpine1 mRNA and PAI1 protein are exclusively restricted to astrocytes and not detected in cultured cortical neurons (Docagne et al., 1999). Furthermore, PAI1 in the CNS is required for the maintenance of neuronal cellular structure (Vivien & Buisson, 2000) and therefore its secretion by IRRAs could indicate a putative neurotrophic function in the face of infection/inflammation. In vivo, PAI1 can also recruit other immune cells like microglia and influence their motility and phagocytic function (Jeon et al., 2012). Among other proteins secreted by rodent astrocytes, SPARC-like protein 1 (SPARCL1) was the most significantly depleted followed by alpha-2-macroglobulin (A2M), neurocan (NCAN), and CAMP Responsive Element Binding Protein 5 (CREB5) (Figure 2e–h). We also used cytokine arrays to assess additional secretory proteins released by IRRAs. We confirmed PAI1 to be increased in the secretome of TICI-treated astrocytes compared to PBS-treated control astrocytes, which served as a good positive control for the experiment and validated our finding from TMT proteomics. Additionally, a few chemokines including C-C Motif Chemokine Ligands 2, 5, and 20 (CCL2, CCL5, and CCL20) were abundant in the secretome of IRRAs (Supplementary Figure 5c,d).

Figure 2. Proteins in the secretome of rodent and human interferon-responsive reactive astrocytes.

(a-d) Proteins that were significantly more abundant in the conditioned media of TNF, IL1α, C1Q, and IFNβ (TICI)-treated compared to PBS-treated control rodent astrocytes. (e-h) Proteins that were significantly depleted in the conditioned media of TICI-treated astrocytes compared to PBS-treated control astrocytes. (i-l) Some of the significant (P<0.05) proteins that were abundant (PSAP, PTX3, PAI1) and depleted (KRT1) in the conditioned media of TICI-treated compared to PBS-treated control human IRRAs. n=5 biological replicates.

To relate the secretome alterations to the proteomic changes in IRRAs, we compared the entire rodent cellular proteome and the secretome datasets and identified 38 proteins that were common to both (Supplemental Figure 3c). Proteins detected in both the cell and secretome, including class I major histocompatibility complex molecules (such as beta-2 microglobulin (B2M), A2M, CFB, and complement C1q B chain (C1QB)), play roles in the regulation of proteolysis and the defense response.

We next wanted to determine if IRRAs are evolutionarily conserved from rodents to humans. We differentiated astrocytes from human induced pluripotent stem cells (hiPSCs) (Barbar, Jain, et al., 2020) and treated them with the TICI factors. Similar to rodent IRRAs, we performed TMT proteomics on the conditioned media and compared the proteins identified in the TICI- versus PBS-treated secretome. We identified a total of 81 total proteins in the human IRRA secretome out of which 7 proteins were significantly more abundant and 2 were depleted with TICI treatment (P value<0.05) (Supplementary Table 3). Of the more abundant proteins, prosaposin (PSAP), pentraxin 3 (PTX3), PAI1, and keratin 1 (KRT1) were the most significant. Other upregulated proteins in the human IRRA secretome were NLR family CARD domain-containing protein 4 (NLRC4), myotubularin-related protein 13 (SBF2), and MARCKS-related protein (MARCKSL1; MARCKS). Astrocyte-derived pentraxin 3, also known as TNF-inducible gene 14 protein, has been reported as upregulated in inflammatory stress and exhibits protective roles in ischemic stroke (Bonsack, Borlongan, Lo, & Arai, 2019; Shindo et al., 2016) by supporting blood brain barrier (BBB) integrity. Additionally, PSAP can also serve a protective role by rescuing astrocytes (and neurons) against oxidative damage (B. Liu et al., 2018).

2.3. The proteome of interferon-responsive reactive astrocytes positively correlates with their transcriptome

Previous study from our lab using bulk-RNA sequencing reported IRRAs express genes like Igtp, Ifit1, Ifit3, Oasl, Mx1, Cxcl10, and Gbp2 (Hasel et al., 2021). We therefore asked if the proteomic profile of IRRAs correlates with this previously published transcriptomic signature. 2,076 of the proteins we observed via proteomics had an annotated gene or multiple genes with transcripts detected by RNA-seq, resulting in a total of 2,154 protein-gene pairs. (Figure 3a). For these 2,154 mRNA-protein pairs, we found a positive correlation between the protein abundance (measured by TMT reporter ion intensity) and mRNA expression (transcripts per million) for each protein-mRNA pair for both the TICI-treated as well as the PBS-treated, unstimulated control astrocytes (Figure 3b,c). Given the correspondence between protein abundance and mRNA expression, we next examined whether changes in protein abundance between IRRAs and control non-reactive astrocytes similarly corresponds to changes in gene expression in IRRAs compared to control astrocytes. Sixty-seven mRNA-protein pairs were significantly regulated in the same direction i.e., they were both either abundant or depleted both at the RNA or protein level (concordant markers) Supplemental Table 5. Ifit2/3, Mx1/2, Gbp2/5, Oas1a/L/L2, Psmb9/10, Bst2, Ass1, Serpine1, etc., and their corresponding proteins were some of the concordant molecules abundant in IRRAs both at the RNA and protein level (Figure 3d, f). These gene/protein pairs are robust responders to interferons and regulate cellular innate responses in response to viral pathogenic triggers. The oligoadenylate synthase (OAS) family of proteins detect exogenous nucleic acid and initiate antiviral responses, making them useful biomarkers in various neurological diseases. They also play a role in other cellular functions, including the regulation of apoptosis (Mullan et al., 2005) and nuclear events such as pre-mRNA splicing (Sperling et al., 1991). Similarly, in addition to their antiviral response, the family of IFIT proteins have immunomodulatory functions in response to inflammation. For example, the production of IFIT2 in macrophages has been shown to inhibit lipopolysaccharide-induced expression of inflammatory cytokines like tumor necrosis factor (TNF), interleukin-6 (IL6), and CXC-chemokine ligand 2 (CXCL2) (Berchtold et al., 2008). Whether IFIT proteins mount a similar immunomodulatory response in astrocytes remains to be seen.

Figure 3. Comparison of rodent interferon-responsive reactive astrocyte (IRRA) transcriptome and proteome.

(a) Method illustrating the comparison of IRRA RNA-seq data from Hasel et al., 2021, and proteomic data from current study. Out of the 19,447 genes in the RNA-seq dataset, 2,107 genes were selected that corresponded to the 2,076 proteins in our proteomics dataset. (b,c) Both PBS and TNF, IL1α, C1Q, and IFNβ (TICI)-treated rodent astrocytes showed positive correlation between their transcriptome and proteome (R=0.4, P<2.2e-16). (d,e) Tables showing the top 10 concordant upregulated and downregulated gene/protein pairs (represented by gene symbols) along with their corresponding log₂ fold change (FC) values from the proteomics and transcriptomics experiments. (f) Plot showing the concordant (green) and discordant (red) expression changes (represented by gene symbols) between the two datasets. The changes occurring in the proteomics data only and not in the RNA-seq data is also shown (orange; protein only) and so are the changes occurring in the RNA-seq data only and not in the proteomics data (violet; RNA only). The number of significantly different protein-gene pairs is shown as ‘n’ values for each group. The RNA-seq data from Hasel et al., 2021 was collected from n=3 biological replicates and compared to the current proteomics data from n=5 biological replicates.

Similarly, some markers like complexin 2 (Cplx2), cell adhesion molecule 4 (Cadm4), transferrin receptor protein 1 (Tfrc), and cystatin C (Cst3), were depleted both at the gene expression and protein level (Figure 3e,f). Cplx2 in astrocytes is a key modulator of exocytosis and synaptic activity via glutamate release (Hazell & Wang, 2011). Another downregulated gene/protein pair is glutathione S-transferase mu 1 (Gstm1). It has been demonstrated that GSTM1 is required for the activation of nuclear factor-κB (NFKB) and the production of proinflammatory mediators (Kano et al., 2019). The downregulation of Gstm1/GSTM1 indicates the cellular reprogramming against a proinflammatory phenotype.

We found eight significant mRNA-protein pairs that exhibited a discordant abundance shifts i.e., upregulated at the RNA level but downregulated at the protein level and vice-versa. PDZ and LIM domain protein 4 (Pdlim4), galectin (Lgals3), proteasome subunit beta type-6 (Psmb6), and adenylosuccinate lyase (Adsl), and A2m were upregulated at the mRNA level but found to be depleted at the protein level; while disheveled segment polarity protein 3 (Dvl3), sorbitol dehydrogenase (Sord), and dynein axonemal heavy chain 12 (Dnah12) were abundant at the protein level but were downregulated genes in our RNA-seq data. Finally, fourteen mRNA-protein pairs were significantly changed only in the proteomics dataset, 966 were significant only in the transcriptomics dataset, and 1,099 were not significant in either of the datasets (complete lists provided in Supplemental Table 5). By determining the relationship between the astrocytic transcriptome and proteome, we have identified the mRNA-protein pairs with concordant and discordant abundance shifts in the comparison between IRRA and “homeostatic” astrocytes. There is a strong correlation between the fold changes observed at the protein level and the RNA level, thus providing evidence that the significant proteins in the IRRA proteome mostly positively correlates with their transcriptome with many genes likely translated into proteins. These differentially expressed genes/proteins represent high-confidence targets for future studies involving IRRA mechanisms, functions, and biomarker characterization in various neuroinflammatory disorders.

2.4. Interferon responsive reactive astrocyte proteins are enriched in human Alzheimer’s disease post-mortem brain

Next, we wanted to determine if IRRA-specific proteins were also present in astrocytes in human post-mortem brain samples. We have previously shown that IRRA-specific transcripts are present in the 5xFAD mouse model of AD (Hasel et al., 2021), so we chose a cohort of AD patient and age-matched non-symptomatic (NS) controls (Sadick et al., 2022) and stained formalin-fixed paraffin embedded frontal cortex samples for some of the top IRRA proteins. In the AD brain, we observed that IFIT3 was mostly restricted to select GFAP⁺ astrocyte cell bodies in the upper cortical layers (Figure 4a, Supplementary Figure 6a). Around 35% of the GFAP⁺ astrocytes in this region were also IFIT3⁺ (compared to less than 5% of the GFAP+ cells in the NS tissue) and around 20% of all cells were IFIT3⁺GFAP⁺ (compared to 1% in NS tissue) indicating that the IRRAs comprise a small percentage of all cells in the AD cortex and are regionally restricted to certain regions in the AD brain (Figure 4b,c). Furthermore, we also observed IFIT3⁺ puncta outside the GFAP⁺ cell bodies (around 20% in the AD tissue and less than 1% in the NS tissue; Figure 4d) which could be present in other cell types in the CNS including microglia or even GFAP- astrocytes. Next, we stained for IFITM3, another top IRRA marker and found very high abundance of IFITM3⁺ puncta in the GFAP⁺ cell bodies in the AD tissue compared to the NS tissue (Figure 4e, Supplementary Figure 6b). Specifically, around 40% of the GFAP⁺ astrocytes in AD contained IFITM3⁺ puncta within their cell bodies (compared to 10% in NS tissue) (Figure 4f). Compared to the NS tissue, the AD tissue did not have any significant increase in the IFITM3⁺GFAP⁺ cells (Figure 4g). Although we observed some IFITM3⁺ puncta outside the GFAP⁺ cell bodies in the AD tissue (white arrows, Figure 4h), there was no significant difference between the AD and NS groups at this brain region (Figure 4h). Finally, we wanted to validate if an increased protein detected in human iPSC-derived IRRAs are also present in human AD. We stained both AD and NS brains for PAI1, a protein upregulated in rodent IRRAs cellular proteome and secretome as well as human IRRA secretome. Although we did not find a statistically significant overall increase in PAI1 puncta in human AD versus NS groups (Figure 4j–l), in one AD patient tissue (male, 94 years), we observed several PAI1⁺ puncta to be highly concentrated within GFAP⁺ cell bodies (magnified image; Figure 4i). Overall, these experiments suggests that IRRA proteins identified in the rodent astrocyte cultures can be used to detect populations of astrocytes in the human post-mortem brain, with enrichment in the AD brain.

Figure 4. Immunostaining for IRRA markers in human post-mortem brain.

(a,e) Non-symptomatic (left) and Alzheimer’s disease (right) cortical sections stained for (a) Interferon induced protein with tetratricopeptide repeats 3 (IFIT3) and (e) interferon induced transmembrane protein 3 (IFITM3) (green) and co-stained for the canonical astrocyte marker GFAP (magenta), and nuclear stain DAPI (blue). Yellow arrowheads mark astrocytes positive for GFAP and IFIT3 or IFITM3. (e) White arrowheads mark GFAP negative cells that are positive for IFIMT3. (b-d and f-h) Quantification of %IFIT3⁺ and %IFIMT3⁺ cells in human tissue. (i,j) Non-symptomatic (left) and Alzheimer’s disease (right) cortical sections stained for Plasminogen activator inhibitor-1 (PAI1) and co-stained for GFAP (magenta) and DAPI (blue). Yellow arrowheads mark astrocytes positive for GFAP and PAI1. White arrowheads mark GFAP negative cells that are positive for PAI1. (k-l) Quantification of %PAI1⁺ cells in human tissues. For all plots, each dot represents an average value calculated from 3 ROIs from three independent human tissue. %Marker⁺ astrocytes = (Marker⁺GFAP⁺ cells/GFAP+ cells)x100; %Marker⁺GFAP⁺ cells = (Marker+GFAP+ cells/DAPI+ cells)x100; %Marker+ other cells = (Marker⁺GFAP⁻ cells/DAPI⁺ cells)x100; Scale bar = 40 μm. Data is from 3 different human brain samples. *P< 0.05, **P< 0.01.

2.5. Interferon-responsive reactive astrocytes are distinct from neurotoxic reactive astrocytes

Finally, we asked whether IRRAs have different proteomic signatures from our previous-defined NRAs. Given that NRAs are induced via TNF, IL1α, and C1Q (TIC) treatment and also exhibit a unique proteome compared to homeostatic control astrocytes (Guttenplan et al., 2021; Labib et al., 2022), we hypothesized that the presence of IFNβ induces production of several unique proteins that may modify the state and function of IRRAs. We used the previously published dataset (Guttenplan et al., 2021) to compare the common and unique proteins between IRRAs and NRAs as well as identify the proteins that are unique to each of the individual cell subtypes (Figure 5, Supplementary Table 6). Only fourteen proteins were common to the upregulated proteomes of IRRAs and NRAs. These included: ASS1, NFKB2, Intercellular adhesion molecule 1 (ICAM1), A2M, Guanylate binding protein 5 (GBP5), etc. Out of the 203 significantly regulated proteins in NRAs when compared to control astrocytes, 189 were unique to the NRA proteome which included C3, a previously identified NRA marker (Liddelow et al., 2017). Of the 104 significantly changed proteins in IRRAs, 90 were unique to the IRRA proteome, including the interferon-stimulated proteins OAS1A, IFIT3, IFIT2, OASL, IFIH1, STAT1, USP18, etc. It should be noted that these comparisons are solely dependent on the proteomics methods used in these two studies and therefore we are limited by the detected proteins in these experiments and datasets. However, these data highlight that multiple modalities can be used to identify specific sub-states of reactive astrocytes. IRRAs, like NRAs have measurable and reproducible transcriptomic and proteomic signatures that are both evolutionarily conserved and distinct from one another.

Figure 5. Similarities and differences between rodent interferon-responsive reactive astrocytes and neurotoxic reactive astrocytes.

The list of significantly different proteins from the neurotoxic reactive astrocyte proteome from Guttenplan et al., 2021, was used to perform comparisons with the significantly different proteins from our current study (104 total proteins that were P<0.05). The proteins are represented by their gene symbols; 14 were found to be common to both groups, 189 were unique to NRAs while 90 were unique to IRRAs. The significant proteins list from both the studies contained both abundant and depleted proteins. The proteomics dataset from Guttenplan et al., 2022 was from n=5 biological replicates and compared to the current proteomics data from n=5 biological replicates.

3. DISCUSSION

Here we characterized the proteome and secretome profile of interferon-induced reactive astrocytes (IRRAs). While activation with TIC is necessary and sufficient to induce a neurotoxic reactive astrocyte (NRA) phenotype, the presence of IFNβ profoundly impacts the astrocyte cell state and response by regulating a core group of proteins that are canonically known to be antiviral and neuroprotective in nature. Many IRRA-specific upregulated proteins (e.g. IFIT1/2/3, MX1/2/3, OASL1/2, GBP2/5, etc.) are implicated in antiviral responses, and/or have been identified in different neurological and neurodegenerative disorders (Ju et al., 2022; Lai et al., 2021; Zhong et al., 2018). In addition to the protein changes in rodent astrocytes, we also identified proteins in the secretome both rodent and human IRRAs, noting that these molecules are released into the cellular microenvironment and can modulate the activity of neighboring cells by functioning as extracellular signaling molecules. For example, the secretion of CFB can affect the proliferation and degradation of blood cellular components during inflammation (Noris & Remuzzi, 2013). PAI1 in the local environment can recruit other immune cells like microglia and influence their motility and phagocytic function (Jeon et al., 2012). Interestingly, PAI1 was not only abundant in the cellular proteome of rodent IRRAs but was also abundant in both the rodent and human IRRA secretome. The gene Serpine1 has been previously reported as enriched in astrocytes, and the increase in the resulting protein PAI1 in AD brains is associated with increased plaques. PAI1 is also speculated to be neuroprotective in acute injury (Flavin, Zhao, & Ho, 2000; Higgins, 2006); however, we do not know how astrocytic PAI1 release may regulate inflammatory events and this topic warrants further investigation.

IRRAs have been reported in various human and non-human primate models of injury, infection, and disease, including SARS-CoV-2 infection of primary human astrocytes (Kong et al., 2022). Specifically, several IRRA markers were found to be significantly upregulated in astrocytes in response to the virus, including IFIT1/2/3, MX1/2, OAS1/2, and OASL. While the literature on interferon signaling is mostly focused on its roles in antiviral defenses, some interferon responsive proteins also play a role in modulating tissue homeostasis in other pathologies. For example, reactive astrocytes upregulate IRRA-specific markers like IFIT1 and IFIT3 in the primary visual cortex of marmosets seven days after an ischemic stroke (Boghdadi et al., 2020). We have previously identified IRRA marker genes to be enriched in astrocytes in the experimental autoimmune encephalomyelitis mouse model of Multiple Sclerosis (MS) and acute stab wound injury model, in addition to the 5xFAD model of AD. In fact, astrocytic response to type I interferons has also been reported in CNS lesions from MS patients (Rothhammer et al., 2016). GFAP⁺ astrocytes in these brains were enriched for pSTAT1 and MX1, suggesting a role for IRRAs in regulating inflammation in MS. Even though IRRAs have been detected in different pathological environments, their functional roles in these contexts remain elusive. Given that astrocyte subsets are highly dynamic, their functions should be evaluated in the context of their context-dependent, local triggers.

The proteomic profiling of astrocyte subtypes is particularly important because many neurological and neurodegenerative disorders are manifested at the protein level. Proteins regulate almost all aspects of astrocyte state and functions, including the support and elimination of synapses via the release of thrombospondins (TSPs) 1–4, and SPARC family proteins (e.g. SPARCL1) into their microenvironment (Christopherson et al., 2005; C. Eroglu, 2009; Kucukdereli et al., 2011), regulation of BBB via PTX3 (Shindo et al., 2016), and receptor-mediated phagocytosis of cellular debris via receptor protein-tyrosine kinase (AXL) and tyrosine-protein kinase Mer (MERTK) (Konishi et al., 2020), among many other functions. Recent advances in scRNA-seq technologies have revealed the heterogeneity of astrocytes with the identification of several new cell subpopulations. However, the analysis of mRNA is not a direct reflection of the protein content in a cell since the formation of mRNA is only the first step in a long sequence of events resulting in the synthesis of a protein (Buccitelli & Selbach, 2020). Thus, RNA-seq cannot accurately identify mRNA transcripts successfully translated into functional proteins. Several biological checkpoint mechanisms exist in the generation of a final functionally-relevant protein molecule, such as posttranslational modifications. It is estimated that many different protein isoforms can be generated from a single RNA molecule as a result of alternative splicing, mRNA editing, polyadenylation, etc. (Buccitelli & Selbach, 2020). Characterizing the proteomic signatures of glial cell subtypes is therefore crucial to more comprehensively evaluate their state and function in a given pathological microenvironment.

Even though we profiled the proteome of IRRAs using TMT proteomics, a major limitation of this study is the missing of lowly abundant proteins in our cell and conditioned media samples. While we were able to validate several IRRA proteins in vivo in human post-mortem tissue sections, our in vitro model of IRRAs may not fully recapitulate the complete in vivo environment in which astrocytic proteins are constantly regulating intracellular signaling events like energy production, participating in extracellular communication neighboring cells, insults, or toxic pathogens, all of which occur in a complicated yet collaborative manner. Thus, the findings from this study raise several key questions: 1) How do the IRRA core proteins regulate intracellular and extracellular signaling events? 2) How do the IRRA proteins influence homeostatic astrocytic functions? 3) What new gain-of-functions do IRRA proteins confer to astrocytes? Answering these questions will be key to understanding the functional roles of IRRAs.

In conclusion, the proteome of IRRAs positively correlates with their transcriptome, with several genes that are translated into quantifiable proteins. We thus propose that IRRAs are defined by a core group of interferon-stimulated proteins. These proteins include IFIT1/2/3, IFITM3, MX1/2/3, OASL1, IFIH1, PAI1, STAT1, among many others. Several of these proteins are already described as interferon-responsive proteins in the peripheral immune system or in other pathophysiological contexts in the CNS. We report that IRRAs are distinct from NRAs at the protein level, and the presence of interferon signaling likely modulates reactive astrocyte response to their environment, that may include neurons and oligodendrocytes susceptible to NRA-induced lipid toxicity. Can interferons counteract the toxic effects of NRAs, or will they exacerbate this effect? This remains unknown, but interferons (especially IFNβ) are broadly associated with antiviral properties (Kasper & Reder, 2014), and previous studies have shown that cells activated by IFNβ can upregulate anti-inflammatory cytokines (Filipi & Jack, 2020). However, some studies have also indicated their proinflammatory properties (Bolívar et al., 2018; Galligan et al., 2010) depending on their activation state in a disease context (Axtell & Steinman, 2008). Thus, further investigations are needed to better evaluate if reactive astrocytes that encounter IFNβ released by neighboring cells like microglia and macrophages, endothelial cells, etc. (Axtell & Steinman, 2008) undergo a shift in their proteomic profile to carry out immunomodulatory functions.

Supplementary Material

Fig S2

Supplementary figure 2. Key to TMT reporter ions and their intensities. (a) Table showing the medians of protein intensities for rat cells, key to TMT reporter ion in each channel. (b) Histograms of reporter ion intensities of proteins detected in each TMT channel. Abbreviations: TICI, TNF, IL1α, C1Q, and IFNβ. Data shown is for n=5 biological replicates.

Fig S3

Supplementary figure 3. Pathway analysis. (a) Pathway enrichment analysis using all the rodent cellular proteins highlights biological processes involved in interferon response and antiviral defense, among other immune responses. (b) Schematic depicting proteins from our dataset related to a known interferon signaling pathway involving STAT1/2 activation. Proteins in black are detected in our dataset and previously characterized as an interferon-stimulated molecular marker. Proteins in grey have been previously identified in literature but not detected in our dataset. (c) Comparison of all the detected proteins from the rodent cellular proteome and secretome datasets resulted in 38 proteins that were common to both. The overlap of the detected proteins are shown (Proteins are represented by gene symbols).

Fig S4

Supplementary figure 4. Validation of select interferon-responsive reactive astrocyte (IRRA) proteins with immunoblotting. Select abundant IRRA proteins that were significantly more abundant than in control astrocytes were validated in immunopanned rat astrocytes using western blotting. (a,d,g) Log₁₀ TMT intensity values for arginosuccinate synthase 1 (ASS1), CD44 antigen (CD44), and signal transducer and activator of transcription 1 (STAT1) plotted from the TMT proteomics dataset. n=5 biological replicates. (b,e,h) Western blot of ASS1, CD44, and phosphorylated STAT1 (pSTAT1) showing it to be more abundant in TICI-treated astrocytes compared to PBS-treated control astrocytes. n=3 biological replicates. (c,f,i) Quantification of ASS1, CD44, and pSTAT1 in PBS vs. TICI-treated astrocytes. All data are from 3 biological replicates. P values were calculated from unpaired Student’s t-test. *P < 0.05; **P < 0.01, ****P<0.0001. Abbreviations: TICI, TNF, IL1α, C1Q, and IFNβ.

Fig S1

Supplementary figure 1. TMT proteomics method and assessment of data. (a) The protocol for TMT-based proteomics beginning samples preparation, labeling, and LC-MS/MS. (b) Distribution of ion intensity in each group (TNF, IL1α, C1Q, and IFNβ (TICI), PBS) for each replicate (1–5) for rat astrocyte cells. (c) TMT reporter ion intensities correlate linearly among the samples used for the LC-MS/MS analysis. Each dot represents a single protein labelled by the respective TMT channel. The log₂ transformed TMT reporter ion intensities (estimation of the relative protein abundances) for each TMT channel are plotted against the log₂ transformed intensities of the other nine TMT channels. The Spearman rank correlation value of the fitted regression line is labelled in blue, on the top left corner of each plot.

Fig S5

Supplementary figure 5. Original blots. (a,b) Original (raw) western blot images indicating lanes, ladders, and the actin loading controls. n=3 biological replicates. (c,d) Original (raw) cytokine array membranes from the 1^st and 2^nd biological replicates. Symbols (protein names): CCL2, 5, and 20 (C-C motif chemokine ligand 2, 5, and 20), cystatin C (CST3), insulin like growth factor binding protein 2 (IGFBP2), interleukin 1 alpha (IL1A), lipocalin 2 (LCN2), C-X-C motif chemokine ligand 5 (CXCL5), and matrix metallopeptidase 2 (MMP2). Abbreviations: TICI, TNF, IL1α, C1Q, and IFNβ.

Tab S1

Supplementary Table 1. List of all proteins identified in TNF, IL1α, C1Q, and IFNβ (TICI)-treated and unstimulated rat astrocytes (cells). Proteins are sorted according to the level of significance in the difference between TICI and PBS treatments (P value; highlighted by a black box), from the highest to the lowest. Significant (P<0.05) changes were colored as follows: red lines indicate proteins more abundant in TICI-treated astrocytes when compared to PBS-treated controls, while blue lines show proteins less abundant in TICI-treated astrocytes when compared to PBS-treated controls.

Tab S2

Supplementary Table 2. List of all proteins identified in TNF, IL1α, C1Q, and IFNβ (TICI)-treated and unstimulated rat astrocyte secretome (conditioned media). Proteins are sorted according to the level of significance in the difference between TICI and PBS treatments (P value; highlighted by a black box), from the highest to the lowest. Significant (P<0.05) changes were colored as follows: red lines indicate proteins more abundant in TICI-treated astrocyte conditioned media when compared to PBS-treated controls, while blue lines show proteins less abundant in TICI-treated astrocyte conditioned media when compared to PBS-treated controls.

Tab S4

Supplementary Table 4. List of antibodies used for western blot analysis and immunofluorescence staining of human brain tissue. Antibodies are listed in alphabetic order, according to the name of the antigen they recognize. Also provided is the catalog number, molecular weight, species that the antibody is raised in, the dilution factor used, and link to the webpage.

Tab S3

Supplementary Table 3. List of all proteins identified in TNF, IL1α, C1Q, and IFNβ (TICI)-treated and unstimulated human astrocyte secretome (conditioned media). List of all proteins identified in TICI-treated and unstimulated human astrocyte secretome (conditioned media). Proteins are sorted according to the level of significance in the difference between TICI and PBS treatments (P value; highlighted by a black box), from the highest to the lowest. Significant (P<0.05) changes were colored as follows: red lines indicate proteins more abundant in TICI-treated astrocyte conditioned media when compared to PBS-treated controls, while blue lines show proteins less abundant in TICI-treated astrocyte conditioned media when compared to PBS-treated controls.

Tab S5

Supplementary Table 5. List of significant concordant and discordant markers in IRRAs proteome and transcriptome. This file lists all the significant markers that were concordant and discordant between the IRRA proteomics and RNA-seq datasets. Also provided are lists of markers that were significant at the protein level only but not at the mRNA level per the RNA-seq data.

Fig S6

Supplementary figure 6. (a) Interferon induced protein with tetratricopeptide repeats 3 (IFIT3⁺) puncta in the GFAP⁺ cell bodies in AD tissue. Magnifications of select cells shown at the bottom. (b) Interferon induced transmembrane protein 3 (IFITM3⁺) puncta in the GFAP⁺ cell bodies in AD tissue. Magnifications of cell bodies shown at the bottom.

Tab S6

Supplementary Table 6. List of all proteins identified in neurotoxic reactive astrocytes. List of all proteins identified in (TNF, IL1α, C1Q)TIC-treated and unstimulated rat astrocytes from Guttenplan KA, et al., 2021. Proteins are sorted according to the level of significance in the difference between TICI and PBS treatments (False Discovery Rate (FDR); highlighted by a black box), from the highest to the lowest. Significant (FDR<0.1) changes were colored as follows: red lines indicate proteins more abundant in TIC-treated astrocyte conditioned media when compared to PBS-treated controls, while blue lines show proteins less abundant in TIC-treated astrocyte conditioned media when compared to PBS-treated controls.

Main Points.

Characterizing the molecular underpinnings of interferon-responsive reactive astrocytes:

What is their proteomic signature?

Does their proteome correlate with their transcriptome?

How are they different from neurotoxic reactive astrocytes?

DATA AVAILABILITY STATEMENT

The raw mass spectrometry data generated during this study are available at MassIVE (UCSD, https://massive.ucsd.edu/ProteoSAFe/static/massive.jsp) with accession number MSV000091426.