HSV-1 infection in 2D dissociated cells from cerebral organoids
An ideal in-vitro complex system for neuroinflammation in AD is comprised of heterogeneous cell types, including neurons and glia cells, such as astrocytes and microglia (Fig. 1a). The resident innate immune cells in the central nervous system (microglia) are differentiated from the mesodermal lineage in-vitro62,63. As such, it was widely thought that cOrgs did not contain microglia. However, earlier studies discovered and characterized CD11b+ functional microglia that differentiated innately within these spontaneously differentiated cerebral organoids64,65,66,67,68. We differentiated 3D cOrgs for 2-4 months, followed by dissociation and culturing of the 2D dcOrgs for another month. We evaluated the percentages of cell types in 6-month dcOrgs by using flow cytometry and found that 7.1% of dcOrgs were positive for P2RY12, which is a marker for microglia (Fig. S1), and our observed fraction was consistent with reported fractions found in human and mouse brains69.
a Schematic of using dcOrgs for high-throughput assays. b Microscope images (20X magnification) of the mock, infected, treated or UV-HSV-1 dcOrgs, as well as mock, HSV-1+ or treated hiPSCs. c PCA plot using the human transcripts from the samples, using RNA-seq, color-coded by batches. d PCA plot of the HSV-1 viral transcripts from the samples, using RNA-seq, color-coded by batches.
We conducted HSV-1 infections in dcOrgs and consistently observed that >50% of cells were GFP+ 23 h post-inoculation (Fig. 1b and Fig. S2). ACV treatment on HSV-1-infected dcOrgs (that we termed as “Treated dcOrgs”) resulted in effective rescue and few GFP+ cells were observed under the microscope. As a negative control experiment, we infected dcOrgs with HSV-1 that had been UV-inactivated (that we termed as “UV-HSV-1 dcOrgs”) and observed no detectable GFP+ cells or plaques (Fig. S3). In parallel, we used influenza A (IAV) infected dcOrgs as another negative control. HSV-1 is a neurotropic DNA virus and IAV is a respiratory RNA virus. Unlike HSV-1, influenza was not associated with increased risk of AD70.
High inter-sample correlations observed in dcOrgs and hiPSCs using bulk RNA sequencing
To further evaluate the system with regard to the variability across biological and technical replicates of dcOrgs, we conducted bulk RNA-seq on dcOrgs and human induced pluripotent stem cells (hiPSCs) under different conditions (Fig. 1c, Data S1). Each set of conditions was performed in triplicate, followed by total RNA extraction, ribosomal RNA depletion and sequencing. We observed high correlations between replicate mock dcOrgs or HSV-1-infected dcOrgs (Pearson’s \(r\)>0.99; Fig. S4, Data S1). The correlations between mock versus HSV-1-infected dcOrg pairs were lower (\(r\)=0.75–0.82), indicating that there were many more transcriptome-wide perturbations than expected from the baseline technical variability. The correlations between HSV-1-infected versus treated dcOrg pairs ranged from \(r\)=0.81–0.89, indicating that ACV treatment had partially rescued some of the transcriptomic perturbations due to HSV-1 infection.
The correlations for biological replicates were high (\(r\)≥0.91 for mock or treated dcOrgs and \(r\)≥0.73 for HSV-1-infected dcOrgs, Data S1). The correlations between mock versus HSV-1-infected or treated hiPSC pairs were higher (\(r\)=0.95–1), because fewer human transcripts were differentially expressed due to HSV-1 infection in hiPSCs compared to dcOrgs. Previously, we found that RNA-seq data on cerebral organoids differentiated from 25 donors71 had similarly high inter-individual correlations \({r}^{2}\) > 0.7, with a mean \({r}^{2}\) of 0.94, indicating that dcOrgs differentiated from many individual donors can be a highly reproducible platform for HSV-1-transduced neuroinflammation.
Investigation of the top three principal components showed that most of the variation (57.2%) in the human transcripts was captured by the first principal component (PC1), as shown in the principal components analysis (PCA) plot in Fig. 1c and Fig. S5. The dcOrgs were also distinctively separated by their conditions into well-defined clusters on PC1. dcOrgs with UV-inactivated HSV-1 (UV-HSV-1) were clustered close to mock dcOrgs on PC1, whereas treated dcOrgs formed their own cluster between mock and HSV-1-infected dcOrgs on PC1. All hiPSCs clustered together into a single group regardless of the condition (mock, HSV-1-infected or treated). A fraction (0.96–1.4%) of all transcripts were viral transcripts in infected hiPSCs, and 0.028–0.035% of all transcripts were viral transcripts in treated hiPSCs. There were lower viral transcript counts in the hiPSCs than dcOrgs, despite using a higher multiplicity of infection (MOI) for the hiPSCs versus dcOrgs (MOI of 4 and 2, respectively), consistent with reports that stem cells are resistant to viral infection, unlike differentiated cells72,73.
To elucidate the viral transcriptome-wide expression patterns of HSV-1, we performed PCA using detected HSV-1 transcripts and found that most of the variation (62.4%) was captured by PC1 (Fig. 1d). All the HSV-1-infected dcOrgs clustered together on PC1 and PC2. However, both sets of treated dcOrgs showed distinctive clusters, primarily in PC1 and PC2. We found that 0.15-0.17% of all transcripts were viral transcripts in the first set of treated dcOrgs (dcOrgs1), 2.4-3% of total transcripts were viral transcripts in the second set of treated dcOrgs (dcOrgs2), whereas 60-81% of all transcripts were viral transcripts in HSV-1-infected dcOrgs, similar to our observation from scRNA-seq data from HSV-1-infected dcOrgs. There were 14-20 times more HSV-1 transcripts in the second set of treated dcOrgs compared to the first set of treated dcOrgs, suggesting that ACV treatment in dcOrgs2 was not as effective as in dcOrgs1 on suppressing viral transcript expression.
High co-abundance correlations in intracellular Aβ and HSV-1 were predominantly driven by Aβ aggregates versus Aβ monomers
Several prior studies have shown that accumulation of intracellular Aβ (aggregated Aβ42 in particular) precedes the accumulation of extracellular amyloid plaques and may be an early event in the pathogenesis of AD74,75,76. In the mammalian brain, amyloid plaques are primarily produced by neurons and are also produced by other cell types such as astrocytes and oligodendrocytes77,78,79,80,81. We next evaluated the impact of viral infections in dcOrgs on intracellular AD-associated molecular readouts, such as the abundance of intracellular Aβ or pTau. We performed 3-channel flow cytometry to quantify intracellular protein abundance from 20,000-30,000 single cells for each condition, followed by machine learning based gating (Fig. 2a). A Zombie Violet dye was used to label dead cells, Alexa Fluor-647 or Allophycocyanin (APC) fluorophore-conjugated antibodies were used to detect the abundance of Aβ, pTau or cell-type specific markers, and the HSV-1 virus was tagged with GFP on VP26.
a Schematic of the flow cytometry experiments and computational analyses. Replicate experiments are biological replicate experiments that were conducted at different times. The results from the conditions (mock, infected, treated) within a replicate experiment were conducted at the same time on the same batch of dcOrgs and viral/ACV stocks to minimize technical differences across the conditions. b UMAP plots showing the normalized fluorophore intensities for Aβ aggregates using aducanumab (red), HSV-1 (green) and Zombie dye (blue) across mock, HSV-1-infected and ACV-treated samples. c UMAP plots showing the normalized fluorophore intensities for Aβ aggregates using solanezumab (red), HSV-1 (green) and Zombie dye (blue) across mock, HSV-1-infected and ACV-treated samples. d Pairwise correlation heatmaps using the intensities for HSV-1, Zombie dye and aducanumab (Aβ aggregates) or solanezumab (Aβ monomers) across mock, HSV-1-infected and ACV-treated samples, calculated by taking the differences in correlations between infected dcOrgs and mock dcOrgs, or treated dcOrgs and mock dcOrgs. e Boxen plots showing the aducanumab or solanezumab fluorophore intensities for GFP– cells (without HSV-1) versus GFP+ cells (with HSV-1) within the same HSV-1-infected sample, as well as GFP– cells versus GFP+ cells within the same ACV-treated sample.
We evaluated the protein abundance of intracellular Aβ species, as well as co-detection of Aβ species and HSV-1, by using monoclonal antibodies for Aβ. The aducanumab antibody, which is selective for aggregated Aβ, was the first AD treatment to have received approval from the U.S. Food and Drug Administration (FDA)82,83,84. In parallel, we used the solanezumab antibody that selectively binds to monomeric Aβ and had failed to slow cognitive decline in people with preclinical AD or mild dementia85,86.
Uniform Manifold Approximation and Projection (UMAP) plots and co-detection analyses on the normalized intensities from the flow cytometry results revealed that there were higher intracellular aggregated Aβ in HSV-1-infected dcOrgs compared to mock dcOrgs or ACV-treated dcOrgs (Fig. 2b-c). High co-detection of HSV-1 and intracellular Aβ42 abundance was found primarily observed in dead cells within dcOrgs (Figure S6), similar to our prior observations from HCMV-infected dcOrgs33. We found strong correlations in the co-detection of intracellular Aβ aggregates and HSV-1 using aducanumab (Δ\(r\)=0.77 and 0.78 for two replicates; Fig. 2d). There was a positive but weaker correlation in the co-abundance of intracellular Aβ monomers and HSV-1 using solanezumab (Δ\(r\)=0.65 and 0.64 across two replicates; Fig. 2d). ACV treatment on HSV-1-infected dcOrgs reduced the correlations in co-abundance between Aβ aggregates and monomers with HSV-1 (Δ\(r\)=0.26-0.35; Fig. 2d). These results indicated that HSV-1 abundance in dcOrgs is highly correlated with the intracellular abundance of aggregated Aβ in dead cells, and that the abundance of aggregated Aβ could be reduced by ACV treatment.
Direct comparisons in HSV-1– cells versus HSV-1+ cells within the same samples indicate cell intrinsic effects on increased abundance of Aβ aggregates and monomers
Because not all cells within the HSV-1-infected dcOrgs or ACV-treated dcOrgs were positive for HSV-1 abundance, we sought to understand if the increased abundance of Aβ aggregates and monomers were driven by cell intrinsic or extrinsic factors. If the increased abundance in these Aβ readouts were driven by cell intrinsic factors, we would expect to observe differences in the intensity distributions of Aβ from cells that were positive for HSV-1 protein abundance (HSV-1+ cells) versus cells that did not have HSV-1 protein abundance (HSV-1– cells) within the same dcOrg sample. Alternatively, if the increased abundance in the Aβ was driven by cell extrinsic factors, we would not expect to observe differences in the intensity distributions from HSV-1– cells versus HSV-1+ cells. We found that the aducanumab intensities in live HSV-1+ cells were significantly higher than the intensities in live HSV-1– cells, indicating that increased intracellular aggregated Aβ abundance was driven by cell intrinsic factors (Fig. 2d). Similar results were obtained for the solanezumab intensities (Fig. 2d).
Co-abundance of HSV-1 and aducanumab were higher in dead cells versus live cells in HSV-1-infected dcOrgs (Δ\(r\)=0.75 and 0.84 in dead cells versus Δ\(r\)=0.38 and 0.62 in live cells; Figure S6a-d). Similarly, co-abundance of HSV-1 and solanezumab were higher in dead cells versus live cells in HSV-1-infected dcOrgs (Δ\(r\)=0.81 and 0.79 in dead cells versus Δ\(r\)=0.63 and 0.67 in live cells; Figure S6e-h). We evaluated IAV-infected dcOrgs and observed that IAV and aducanumab were more highly co-detected in live cells in IAV-infected dcOrgs (Δ\(r\)=0.44 and 0.42; Fig. 3a-d, Figure S7a-d), versus dead cells in IAV-infected dcOrgs (Δ\(r\)=0.12 and 0.16). Similarly, co-abundance of IAV and solanezumab were higher in live cells in IAV-infected dcOrgs (Δ\(r\)=0.22 and 0.25; Figure S7e-h), compared to dead cells in IAV-infected dcOrgs (Δ\(r\)=-0.01 and -0.02).
a UMAP plots showing the normalized fluorophore intensities for Aβ aggregates using aducanumab (red), IAV (green) and Zombie dye (blue) across mock and IAV-infected samples. b UMAP plots showing the normalized fluorophore intensities for Aβ aggregates using solanezumab (red), IAV (green) and Zombie dye (blue) across mock and IAV-infected samples. A cell with a darker blue color had a higher intensity from the Zombie dye, compared to a cell with a lighter blue, white or black color; however, gating for live/dead cells was conducted by using the raw intensities and not by using the normalized intensities (that are represented in the UMAP plots). c Pairwise correlation heatmaps using the intensities for IAV, Zombie dye and aducanumab (Aβ aggregates) or solanezumab (Aβ monomers) across mock and IAV-infected samples. The change in correlations (Δcorrelations) were calculated by taking the differences in correlations between infected dcOrgs and mock dcOrgs. d Boxen plots showing the aducanumab or solanezumab fluorophore intensities for GFP– cells (without IAV) versus GFP+ cells (with HSV-1) within the same IAV-infected sample in live-gated cells.
These results supported a hypothesis that cell intrinsic factors may be contributing to HSV-1 infection on aducanumab intensities and support the observations that HSV-1 infection led to increased intracellular abundance of Aβ aggregates and monomers, but preferentially of aggregated Aβ.
Intracellular abundance of HSV-1 in dcOrgs is most strongly correlated with aggregated Aβ, total tau, pTau-181 and pTau-217
To further define the effects of HSV-1-induced neuroinflammation on Aβ aggregation and tau phosphorylation in dcOrgs, we compared the co-abundance of HSV-1 in dcOrgs with several antibodies targeting the precursor APP, Aβ, Aβ1-40, total tau, pTau-181, pTau-217 and pTau-202,205. In terms of the amyloid antibodies, we observed strong correlation in intensities of HSV-1 with Aβ using the 6E10 antibody (Δ\(r\)=0.62 and 0.51 across 2 replicates; Fig. 4a).
a Pairwise Δ\(r\) heatmaps using the intensities for HSV-1, Zombie dye and APP, Aβ, tau and different pTau residues across HSV-1-infected dcOrgs and ACV-treated dcOrgs. The Δ\(r\) values were calculated by taking the difference in correlations from infected dcOrgs and mock dcOrgs or treated dcOrgs and mock dcOrgs. b Top panel shows the concentrations of Aβ42/40/38 in pg/mL, Aβ42/40 ratios and Aβ42/38 ratios detected from conditioned media that had undergone heat inactivation for mock, HSV-1 + , or treated dcOrgs, and dcOrgs with UV-inactivated HSV-1 (UV-HSV-1). Bottom panel shows the concentrations of Aβ42/40/38 in pg/mL, Aβ42/40 ratios and Aβ42/38 ratios detected from conditioned media that had undergone heat inactivation for mock or IAV-infected (IAV + ) dcOrgs. P-values shown were calculated using 2-sided t-tests with comparison to mock dcOrgs.
Total tau and hyperphosphorylated tau are additional hallmark biomarkers of AD. Site-specific tau phosphorylation changes had been associated with clinical biomarkers87,88,89. pTau-Thr217 (pTau-217) and pTau-Thr181 (pTau-181) had been reported be closely linked to amyloid and tau pathology90; pTau-Thr205 (pTau-205) was associated with a decline in synaptic homeostasis and later stages of disease progression; whereas there were no significant changes in pTau-Ser202 (pTau-202) throughout disease progression87,88,89,91,92. We found strong correlations in intensities of HSV-1 with total tau using the HT7 antibody (Δ\(r\)=0.77 and 0.63; Fig. 4a), pTau-181 (Δ\(r\)=0.85 and 0.76), pTau-217 (Δ\(r\)=0.73 and 0.72), pTau Ser202 and Thr205 (AT8 antibody; Δ\(r\)=0.38 and 0.72).
Conditioned media from HSV-1-infected dcOrgs or HSV-1-infected and ACV-treated dcOrgs had consistently lower concentrations of secreted Aβ42, Aβ40 and Aβ38 compared to conditioned media from mock dcOrgs (2-tailed t-test P≤3.2×10-5; Fig. 4b, Data S2). There were significantly lower secreted Aβ42/40 or Aβ42/38 ratios from conditioned media with HSV-1-infected dcOrgs (2-tailed t-test P = 0.0012; Fig. 4b), that were primarily driven by lower levels of secreted Aβ42. ACV treatment restored secreted Aβ42/40 ratios to similar ratios as detected in conditioned media with mock dcOrgs but did not completely restore secreted Aβ42/38 ratios.
IAV infection in dcOrgs similarly led to decreased concentrations of secreted Aβ42, Aβ40 and Aβ38 detected in the conditioned media, compared to conditioned media with mock dcOrgs (2-tailed t-test P≤0.0098; Fig. 4b). However, secreted Aβ42/40 and Aβ42/38 ratios in conditioned media with IAV-infected dcOrgs did not differ significantly from ratios detected in conditioned media with mock dcOrgs (1-tailed Wilcoxon P = 0.72 and 0.095 respectively; Fig. 4b). We made similar observations using conditioned media that were UV-irradiated to inactivate residual viruses (Figure S8).
These results showed that HSV-1 infection in dcOrgs resulted in significantly lower ratios of secreted Aβ42/40, but IAV infection in dcOrgs did not affect the ratios of secreted Aβ42/40. In addition, ACV treatment of HSV-1-infected dcOrgs rescued the ratios of secreted Aβ42/40 to similar ratios as detected from mock dcOrgs. The results indicated that HSV-1 infection in dcOrgs led to a lower proportion of Aβ42 peptides that were secreted into the media, consistent with prior reports from AD patients’ CSF samples93,94.
Cell type enrichment analyses indicate a depletion of neurons and enrichment of glia in HSV-1-infected dcOrgs
Glia-neuronal morphological and functional changes that are important in AD include astrogliosis (marked by increased proportions of GFAP+ reactive astrocytes) and neuronal loss95,96. We sought to evaluate differences in the HSV-1-infected dcOrgs by performing flow cytometry with 15 cell-type markers to evaluate the enrichment or depletion of cell types. Using the cell counts from the flow cytometry results, we found that there were decreased proportions of NeuN+ neurons, VMAT2+ dopaminergic neurons, TuJ1+ immature neurons and EOMES+ intermediate progenitor cells in HSV-1-infected dcOrgs compared to mock dcOrgs, which indicated that there was likely to be neuronal loss in dcOrgs due to HSV-1 infection (Fig. 5, Figure S9-11). Live glia was mostly increased in proportions, such as GFAP+ astrocytes, GLAST+ astrocytes, Iba1+ microglia, P2RY12+ microglia, and O4+ oligodendrocytes. ACV treatment of HSV-1-infected dcOrgs reduced the depletion in proportions of neurons and neural progenitor cells and reduced the increased proportions of reactive glia cells (Fig. 5). CD4+ or CD45+ monocytes/macrophages did not show a huge increase in proportions among infected live cells and the proportions were similar after ACV treatment, which showed that the increased proportions of Iba1+ or P2RY12+ cells were likely to be driven by microglia and not monocytes nor macrophages. Our results were consistent with reactive astrogliosis, activation of microglia and neuronal death previously observed in AD patients4,97,98,99.
Odds ratios calculated for HSV-1-infected live dcOrgs versus mock live dcOrgs, or ACV-treated live dcOrgs versus mock live dcOrgs, using 15 cell-type markers. Multiple sets of replicate experiments were performed for each marker (depicted by the colored points). Shaded points represent significant odds ratios with P≤0.05 and open points represent non-significant odds ratios with P > 0.05. The tan bars represent average odds ratios<1 across both replicates and the blue bars represent average odds ratios>1 across both replicates.
Abundance of Iba1+ microglia and GFAP+ astrocytes were highly positively correlated with HSV-1
We next asked about the correlations in co-abundance of HSV-1 with cell-type marker abundance using our flow cytometry approach and conducted two sets of independent experiments for each marker (Figs. S12-14). We observed positive correlations in the co-abundance of HSV-1 and NeuN+ neurons (Δ\(r\)=0.26 and 0.24; Figure S12-14), as well as positive correlations in the co-abundance of HSV-1 and EOMES+ intermediate progenitor cells (Δ\(r\)=0.3 and 0.44). Correlation analyses of glia cell-type marker abundance with HSV-1 protein abundance pointed to Iba1+ microglia with the highest correlations (Δ\(r\)=0.66 and 0.39). There were negative correlations observed in the abundance of HSV-1 and P2RY12+ microglia (Δ\(r\)=-0.28). We also observed positive correlations in the abundance of GFAP+ astrocytes with HSV-1 (Δ\(r\)=0.32 and 0.49; Figure S13), and GLAST+ astrocytes with HSV-1 (Δ\(r\)=0.49 and 0.19). To ensure the reproducibility of the Δ\(r\) values, we conducted extensive replication experiments to establish the distributions of Δ\(r\) for the antibody markers used in our study and found that this approach can enable us to detect experimental failures that led to outlier Δ\(r\) values in an unbiased manner (Figure S15-19). These results showed that there were high correlations in the co-abundance of HSV-1 with neuronal and glia markers (NeuN, EOMES, Iba1, GFAP and GLAST), indicating that there were complex interactions between HSV-1 with neuronal cells, microglia and astrocytes to recapitulate the observed neuroinflammatory molecular and transcriptomic signatures.
Gene set enrichment analyses identified that differentially expressed transcripts due to HSV-1-infection in dcOrgs were enriched for AD-associated genes
After characterizing HSV-1-infected 2D dcOrgs and observing that many of the previously reported molecular signatures associated with AD that were observed in HSV-1-infected 3D brain organoids27,55,56,58, could be recapitulated in HSV-1-infected 2D dcOrgs, we further dissected the transcriptomic signatures of HSV-1-infected 2D dcOrgs. The identification of AD-associated transcriptomic readouts will enable us to use the massive RNA-seq and scRNA-seq datasets available from AD human post-mortem brains to identify if there may be subsets of individuals whose AD pathology may have arisen from chronic neuroinflammation due to HSV-1 reactivation, or more broadly, neurotropic pathogens. Identifying AD-associated transcriptomic readouts will also enable us to use emerging scRNA-seq technologies to identify cell types and cell type specific networks associated with neuroinflammation-induced AD molecular pathology such as aggregated Aβ, pTau, secreted Aβ42/40 ratios and cell-type proportional changes that we had observed.
We performed four sets of analyses to identify differentially expressed genes (DEGs) from HSV-1-infected dcOrgs versus mock dcOrgs (HSV-1-Inf-vs-Mock dcOrgs), HSV-1-infected dcOrgs versus HSV-1-infected and ACV-treated dcOrgs (HSV-1-Inf-vs-Treated dcOrgs), and HSV-1-infected dcOrgs versus dcOrgs with UV-inactivated HSV-1 (HSV-1-Inf-vs-UV-HSV-1 dcOrgs), shown in Data S3 and volcano plots in Figure S20. There were no viral transcripts detected in RNA-seq data from UV-HSV-1 dcOrgs, confirming the efficiency of viral inactivation (Data S4, Figure S3). The first two sets of experiments (HSV-1-Inf-vs-Mock dcOrgs and HSV-1-Inf-vs-Treated dcOrgs) had two sets of replication experiments (dcOrgs1 and dcOrgs2). In addition, we performed DEG analyses to compare HSV-1-infected hiPSCs from the same donor versus mock hiPSCs (Inf-vs-Mock hiPSCs) and HSV-1-infected hiPSCs versus HSV-1-infected and ACV-treated hiPSCs (Inf-vs-ACV hiPSCs).
Gene ontology analyses showed that biological processes such as regulation of apoptosis, autophagy and mitochondrion organization were enriched among DEGs in both sets of Inf-vs-Mock dcOrgs and Inf-vs-ACV dcOrgs1 but not in Inf-vs-ACV dcOrgs2 (Data S5). Pathways such as integrin signaling pathway, gonadotropin-releasing hormone receptor pathway and angiogenesis were enriched among DEGs in both sets of Inf-vs-Mock dcOrgs and Inf-vs-ACV dcOrgs1 but angiogenesis was not an enriched pathway in Inf-vs-ACV dcOrgs2 (Data S5).
Next, we obtained lists of genes that were associated to 21 common neurodegenerative, neuropsychiatric or autoimmune diseases, as well as human traits and related diseases identified from GWAS studies and reported in the GWAS Catalog100. To evaluate if the DEGs in infected versus mock dcOrgs were enriched for genes associated to any of the 21 common diseases or traits, we conducted gene set enrichment analyses (GSEA)101,102 with the gene lists in Data S6.
We observed that DEGs in both sets of HSV-1-infected versus mock dcOrgs (Inf-vs-Mock dcOrgs1 and Inf-vs-Mock dcOrgs2) were enriched for AD-associated genes (P = 0.039 and P = 5.1×10-3 respectively; Fig. 6a, Data S6). Similarly, the DEGs in HSV-1-infected dcOrgs versus dcOrgs with UV-inactivated HSV-1 (Inf-vs-UV dcOrgs2) were enriched for AD-associated genes (P = 0.015), providing further support that the transcriptomic profiles of dcOrgs with UV-inactivated HSV-1 were similar to mock dcOrgs. As control experiments, we evaluated whether the DEGs in HSV-1-infected versus mock hiPSCs might be enriched for AD-associated genes, but we did not observe any enrichment (P = 0.67; Fig. 6a). We conducted GSEA using a published list of DEGs identified from HSV-1-infected versus mock glioblastoma-derived (SH-SY5Y) neurons103, but we did not observe any enrichment for the neuronal DEGs with AD-associated genes (P = 0.38). We wondered whether inflammation induced by an RNA virus such as IAV, might similarly perturb AD-associated genes and therefore, we performed GSEA on DEGs identified from IAV-infected versus mock dcOrgs. However, we did not observe an enrichment for AD-associated genes (P = 0.79; Fig. 6a). Collectively, these results indicate that HSV-1-induced neuroinflammation in dcOrgs led to transcriptomic perturbations in AD-associated genes that may be driven by cell types other than neurons within the dcOrgs.
a Figure showing the GSEA results for HSV-1-infected cells or IAV-infected cells versus other conditions (mock, ACV or UV). -log10(Nominal P-value) from GSEA (y-axis) of the differentially expressed genes from our datasets with GWAS-associated genes for common diseases and traits (x-axis): Alzheimer’s disease (AD), amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), Parkinson’s disease (PD), Attention-deficit hyperactivity disorder (ADHD), autism spectrum disorders (ASD), bipolar disorder (BD), major depressive disorder (MDD), obsessive-compulsive disorder (OCD), schizophrenia (SCZ), Tourette syndrome (TS), Crohn’s disease (CD), inflammatory bowel disease (IBD), psoriasis (PSO), rheumatoid arthritis (RA), type 1 diabetes (T1D), type 2 diabetes (T2D), body mass index (BMI), brain volume (BRAIN), height (HEIGHT), waist-to-hip ratio (WHR). b Figure showing the GSEA results with a focus on the ACV-treated dcOrgs. -log10 (Nominal P-value) from GSEA (y-axis) of the differentially expressed genes from our datasets with GWAS-associated genes for common diseases and traits (x-axis). c Schematic figure of the scRNA-seq experiments where the cells from the mock samples were defined as “true uninfected” (TU) in tan colors, and the cells from the HSV-1-infected dcOrgs that did not contain many viral transcripts were defined as “abortively infected” (AI) in tan colors, and the cells from the HSV-1-infected dcOrgs that contained viral transcripts were defined as “true infected” (TI) in green colors. The TI and AI cells are collectively defined as “infected” (I). d -log10(Nominal P-value) from GSEA (y-axis) of the pseudobulk differential expression of various comparisons (TI versus TU, PU versus TU, I versus TU or TI versus PU). e Scatter plot for our discovery dataset showing the log2 fold change in infected versus ACV-treated dcOrgs, Inf-vs-ACV dcOrgs1 (y-axis) and infected versus mock dcOrgs, Inf-vs-Mock dcOrgs1 (x-axis), with the Pearson’s \(r\) correlations of the log2 fold changes of transcripts in each category shown on the bottom right. f Scatter plot for our replication dataset showing the log2 fold change in infected versus ACV-treated dcOrgs, Inf-vs-ACV dcOrgs2 (y-axis) and infected versus mock dcOrgs, Inf-vs-Mock dcOrgs2 (x-axis), with the Pearson’s correlations (\(r\)) of the log2 fold changes of transcripts in each category shown on the bottom right. g Venn diagram showing the numbers and percentages of genes that were rescued or not rescued in both our discovery and replication datasets (dcOrgs1 and dcOrgs2). h Venn diagram showing the numbers and percentages of genes that were in the exacerbated 1 or exacerbated 2 categories in both our discovery and replication datasets (dcOrgs1 and dcOrgs2).
HSV-1 and IAV infections in dcOrgs perturbed the expression of transcripts in several innate immune pathways
The expression of transcripts in several human immune pathways (cGAS/STING, IFN, IRF3, JAK/STAT, RIG-I and TLR) were significantly differentially perturbed in both sets of HSV-1-infected versus mock dcOrgs (Data S3-4, Figure S21a). ACV treatment rescued the expression for some of these transcripts (Figure S21b). As expected, HSV-1-infected dcOrgs versus dcOrgs with UV-inactivated HSV-1 showed similar differential expression as HSV-1-infected versus mock dcOrg samples (Figure S21c). IAV-infected versus mock dcOrgs did not result in much significant differential expression of these human immune transcripts, except for transcripts in the RIG-I and TLR pathways (Figure S21d). Similarly, the expressions of these transcripts were not significantly perturbed in hiPSCs (Figure S22). These results from HSV-1-infected dcOrgs are consistent with reports that the cGAS/STING pathway is a major pathway associated with aging-related inflammation in neurodegenerative diseases such as AD58,104.
A strong transcriptional activation of the innate immune response was observed in our RNA-seq datasets from HSV-1 infection in dcOrgs and reported previously from HSV-1 infection in 2D dcOrgs and 3D cerebral organoids28,57. This innate immune induction is most likely produced by microglia or astrocytes within the cerebral organoids, although a small fraction of neurons had been reported to produce type I interferon response105. Neurons had also been reported to be activated directly by microbes to release neuropeptides and modulate the innate immune response106,107. As such, the interferon induction observed in our data is likely to be primarily driven by glia cells and to a lesser extent, by neurons within these dcOrgs.
ACV treatment in HSV-1-infected dcOrgs rescued the expression in AD-associated genes
To determine the effect of ACV, we evaluated if ACV treatment of HSV-1-infected dcOrgs could prevent the transcriptomic perturbations in AD-associated genes that we had observed. GSEA analyses of our discovery dataset (Inf-vs-ACV dcOrgs1) showed an enrichment for AD-associated transcripts (P = 0.023; Fig. 6b), indicating that ACV treatment could rescue HSV-1-induced transcriptomic perturbations in AD-associated genes. However, our replication dataset (Inf-vs-ACV dcOrgs2) did not show an enrichment for AD-associated transcripts, indicating that ACV treatment in our replication experiment did not rescue HSV-1-induced transcriptomic perturbations in AD-associated genes (P = 0.53; Fig. 6b). As there were 14-20 times more HSV-1 viral transcripts detected in our replication dataset compared to our discovery dataset (Fig. 1d), we hypothesized that there might be different extents of inhibition of HSV-1 DNA synthesis in the two experiments.
Using a previously reported annotation of viral transcripts108, we observed that there was a high correlation in the P-value ranked differential expression for the leaky late viral transcripts (γ1) across both datasets (Spearman’s ρ = 0.63, P = 1.6×10-3), but not for the true late viral transcripts (γ2) across both datasets (Spearman’s ρ = 0.2, P = 0.47; Figure S23, Data S7). These results were consistent with prior reports that ACV treatment can perturb the expression of true late (γ2) viral genes, but not the expression of immediate early, early or leaky late (γ1) viral genes108,109. The results also indicated that the observed enrichment of AD-associated genes was due to dosage-dependent expression of γ2 viral genes, indicating that the expression of one or more γ2 genes led to differential expression in many of the AD-associated genes.
We performed GSEA using the DEGs from ACV-vs-Mock dcOrgs2 and observed an enrichment for AD-associated genes (P = 0.028; Fig. 6b), which showed that the transcriptomic profile of ACV-treated dcOrgs2 was similar to HSV-1-infected dcOrgs; thus, the results from ACV-vs-Mock dcOrgs2 were similar to Inf-vs-Mock dcOrgs2 and showed a similar enrichment for AD-associated genes. These results indicated that ACV treatment can rescue transcriptomic perturbations in AD-associated DEGs due to HSV-1 infection by a mechanism that is dependent on inhibition of viral DNA synthesis. The results are different from our prior results that ACV treatment cannot rescue transcriptomic perturbations in type 1 diabetes (T1D)-associated DEGs due to HSV-1 infection in stem cell derived pancreatic islets (sc-islets)110.
Cell type enrichment from bulk RNA-seq data aligns with results from flow cytometry
We ran our previously published Orgo-Seq method71 to predict for cell type enrichment or depletion by using our bulk RNA-seq datasets and a previously published scRNA-seq reference dataset111. We observed that astrocytes were increased in proportions across all our datasets, whereas excitatory neurons and inhibitory neurons were decreased in proportions across most of our datasets (Figure S24), which were consistent with our results from the flow cytometry experiments. However, microglia, oligodendrocytes and oligodendrocyte progenitor cells were predicted to be depleted across most of our RNA-seq datasets. To validate the gene expression for some of the markers, we conducted quantitative real-time PCR (qRT-PCR) and found that the expression of astrocytes (GFAP) and oligodendrocytes (SOX10) were increased in HSV-1-infected dcOrgs versus mock dcOrgs, whereas the expression of neural progenitor cells (Nestin) and neurons (RBFOX3 and MAP2) were decreased (Figure S25).
Single-cell RNA-seq analyses indicated that AD-associated DEGs from HSV-1-infected dcOrgs were driven by cells that were exposed to HSV-1 but did not contain viral transcripts
To identify transcriptomic changes at individual single cells, we generated scRNA-seq data using Parse Evercode WT Mini V2 with 2D HSV-1-infected versus mock dcOrgs (Figs. S26-29). We conducted analyses using pseudobulk data from subsets of the cells from our scRNA-seq data. Individual cells within the mock dcOrgs were defined as “true uninfected” (TU) cells. Individual cells in HSV-1-infected dcOrgs but did not contain many viral transcripts were collectively defined as “pseudo-uninfected” (PU) cells that may be comprised of cells that were uninfected, or “abortively infected” (AI) cells (Fig. 6c). Individual cells in HSV-1-infected dcOrgs that contained viral transcripts were defined as “true infected” (TI) cells. We could also combine the PU and TI cells to collectively analyze these cells as “infected” (I) cells.
There were multiple cell types that had preferentially higher levels of viral transcripts (in the TI cells compared to the TU cells), as shown in Data S9, such as cortical neurons, glia progenitor cells, interneurons, mesoderm-derived cells, oligodendrocytes, BMP-responsible cells, cilia-bearing cells, proteoglycan-expressing cells and unfolded protein responsible cells (mapped to the Tanaka reference dataset). By mapping the cell types to the Allen reference dataset, there were similar cell types that had preferentially higher levels of viral transcripts in the TI cells compared to the TU cells, such as microglia-PVM, endothelial cells, oligodendrocytes, oligodendrocyte progenitor cells (OPCs), and neuronal subtypes such as chandelier, L2/3 IT, L4 IT, L5 ET, L5 IT, L6 CT, L6 IT, L6 IT Car3, L6b, Lamp5, Lamp5 Lhx6, Pvalb, Sncg, Sst, Sst Chodl and VIP neurons.
Unlike the GSEA results from the bulk RNA-seq data, we found that there was no enrichment for AD-associated genes in the I/TU analyses from the 2D scRNA-seq data (P = 0.14; Fig. 6d; Data S6). Instead, we did observe an enrichment for AD-associated genes in the PU/TU analyses (P = 3.7×10-3). These PU cells did not harbor high abundance of viral transcripts but had been exposed to other cells that had high abundance of viral transcripts (TI cells). These PU cells may be comprised of abortively infected (AI) cells and prior literature reported that astrocytes were the main AI cells that produce interferon beta in HSV-1-infected mammalian brains112,113,114. Using the scRNA-seq data, we identified changes in cell type proportions among the TU, TI, PU and I cells and found that there were multiple cell types that increased or decreased in proportions (Data S9). However, these results remain to be replicated with additional scRNA-seq experiments.
ACV treatment exacerbated the expression for 19-23% of human transcripts that were expressed in dcOrgs
To explore the contribution of unexpected transcriptomic perturbations due to ACV treatment, we defined two categories of unexpectedly perturbed transcripts. First, we defined “Exacerbated 1” group as genes whose dysregulation were exacerbated by ACV treatment. For instance, if HSV-1 infection in dcOrgs down-regulated the human gene expression compared to mock dcOrgs, but ACV treatment on HSV-1-infected dcOrgs further down-regulated the gene expression compared to HSV-1-infected dcOrgs, or the reciprocal scenario with up-regulation, the gene would be classified into the Exacerbated 1 group. Next, we defined “Exacerbated 2” group as genes where HSV-1 infection in dcOrgs did not perturb the human gene expression but ACV treatment in HSV-1-infected dcOrgs significantly perturbed the expression of the gene.
Collectively, both groups of exacerbated genes comprised of 19% or 23% among all genes that were expressed in dcOrgs (Fig. 6e-f). We observed that exacerbated genes comprised of similarly high percentages of 22% or 23% among AD-associated GWAS genes (Data S8). On the other hand, the expression for most of the genes (22% or 40%) were rescued by ACV treatment, and similarly 26% or 41% of AD-associated genes were rescued by ACV treatment. These results reaffirmed that ACV treatment can rescue most AD-associated transcriptomic perturbations due to HSV-1 infection, despite resulting in high percentages of exacerbated gene expression.
ACV treatment rescued AD-associated gene expression in dcOrgs
We further explored the overlaps between transcripts that were rescued or not rescued across the dcOrg replicates. The highest overlap was among transcripts that were rescued in dcOrgs1 but were not rescued in dcOrgs2 (Fig. 6g), which further supported our observations that ACV treatment was more effective in the discovery experiment (Inf-vs-ACV dcOrgs1) than the replication experiment (Inf-vs-ACV dcOrgs2). As expected, the second highest overlap in transcripts were rescued in both dcOrgs1 and dcOrgs2, demonstrating that despite the dosage-dependent differences of ACV treatment in both sets of experiments, there were several transcripts in common that were rescued across both datasets.
The highest overlap in AD-associated transcripts was driven by transcripts that were rescued in both dcOrgs1 and dcOrgs2 (Figure S30). There was an enrichment in the ratios of AD-associated transcripts that were rescued by ACV treatment, compared to the overall ratios of all transcripts that were rescued by ACV treatment (OR = 1.33, P = 0.062), indicating that ACV treatment may have a preferential rescue for AD-associated genes in HSV-1-infected dcOrgs. These results are consistent with our prior results that ACV treatment did not rescue the expression of T1D-associated genes in HSV-1-infected sc-islets110.
Similarly, there were modest enrichment of RA-associated or T1D-associated transcripts that were rescued in both dcOrgs1 and dcOrgs2, compared to the overall set of all transcripts (OR = 1.33 and 1.47, P = 0.068 and 0.074 respectively, Figure S30). There were 7 other diseases or traits with modest enrichments of transcripts that were rescued in both dcOrgs1 and dcOrgs2, such as Parkinson’s disease (PD; OR = 1.55, P = 0.032), ADHD (OR = 1.24, P = 0.071), obsessive compulsive disorder (OCD; OR = 1.96, P = 0.0043), schizophrenia (SCZ; OR = 1.21, P = 0.069), inflammatory bowel disease (IBD; OR = 1.29, P = 0.088), height (OR = 1.14, P = 0.03) and waist-hip-ratio (WHR; OR = 1.34, P = 7.3×10-4). These results indicate that ACV treatment may rescue the expression of AD-associated genes and genes associated with several other common diseases or traits.
On the other hand, the 4-way Venn diagrams depicting both groups of exacerbated genes did not show high overlaps in any subsets that were common between both dcOrg replicates (Fig. 6h, Figure S31). This observation indicated that different sets of genes were transcriptionally exacerbated by each ACV treatment, unlike the rescued genes.
Analyses on human post-mortem brain RNA-seq revealed that 25-31% of patients with late-onset AD have transcriptomic signatures similar to HSV-1 infected dcOrgs
A previous study used RNA-seq data generated using human post-mortem brain samples from patients with late-onset AD (LOAD) and performed molecular subtyping of the LOAD patients into 5 subtypes (A, B1, B2, C1 and C2)115. Globally, there were weaker fold change differences observed in RNA-seq data from human post-mortem brains versus dcOrgs (Figure S32a). For better visualization, we rescaled the fold change differences observed in human post-mortem brains to illustrate that there were some pathways with high degrees of similarities in fold changes between the human post-mortem brains and dcOrgs, such as Amyloid, Blalock, Immune, Synapse/Myelin and Tau (Figure S32b).
We conducted a modified gene set enrichment analysis test using the DEGs from the dcOrgs with the most informative DEGs from the post-mortem brains and identified a significant enrichment of the DEGs in dcOrgs with subtype A (Figure S32c). Inf-vs-Mock dcOrgs1, Inf-vs-Mock dcOrgs2, Inf-vs-ACV dcOrgs1 and ACV-vs-Mock dcOrgs2 had significant positive associations with subtype A (FWER = 1×10-8, 9.6×10-9, 6.4×10-10 and 4×10-7 respectively). Subtype A comprises of 25% and 31% of the LOAD patients from two cohorts115, indicating that the transcriptomic signatures from HSV-1-infected dcOrgs can be used to model transcriptomics from a subset of AD human post-mortem brains.
Replication studies by using additional donor-derived dcOrgs showed high co-abundance of HSV-1 and aducanumab, solanezumab and pTau-181, and transcriptomic enrichment for AD-associated genes
We sought to replicate our results by using dcOrgs that were differentiated from another set of donors. We had reprogrammed 9 donor-derived hiPSC lines from lymphoblastoid cell lines with homozygous APOE ε3/ε3 alleles by using our previously established protocols116,117. We pooled hiPSCs in equal proportions from the 9 donors (and we termed the pool as A3P) and extracted DNA from the A3P hiPSC pool across 4 weeks for sequencing116,118. We used our PoolSeq algorithm to deconvolute donor proportions from the sequencing data118 and observed that at least 5 out of the 9 donors had proportions above the baseline after 4 weeks (Figure S33). We differentiated the A3P hiPSCs into dcOrgs and used 8-month dcOrgs to conduct replication experiments.
We found that there is a strong correlation in co-abundance of HSV-1 and aducanumab in HSV-1-infected A3P dcOrgs, compared to mock dcOrgs (Δ\(r\)=0.73, Figure S34), similar to our prior results from the PGP1 dcOrgs. ACV-treatment reduced the correlation in co-abundance of HSV-1 and aducanumab in treated A3P dcOrgs (Δ\(r\)=0.38). There was a weaker correlation in co-abundance of HSV-1 and solanezumab in HSV-1-infected A3P dcOrgs (Δ\(r\)=0.59). We conducted HSV-1 infection and ACV-treatment on HSV-1-infected A3P dcOrgs for RNA-seq and found that the DEGs in HSV-1-infected A3P dcOrgs versus mock A3P dcOrgs, and HSV-1-infected A3P dcOrgs versus ACV-treated A3P dcOrgs, were enriched for AD-associated genes (Figure S35). These results indicate that the AD-associated Aβ molecular and transcriptomic readouts induced by HSV-1 infection in dcOrgs are likely to replicate across donor lines.
In summary, our HSV-1-induced neuroinflammatory 2D dcOrgs system can recapitulate AD-associated molecular and transcriptomic signatures that can be used as a novel human NAM for neuroinflammation associated with AD. Collectively, the HSV-1-induced neuroinflammatory 2D dcOrgs can complement ongoing research development in the use of 3D cOrgs and animal models in recapitulating certain molecular and transcriptomic signatures associated with subsets of a common, complex disease such as AD.