MEG3 activates necroptosis in human neuron xenografts modeling Alzheimer’s disease

1Mouse models have not allowed us to address the crucial question how the defining hallmarks of Alzheimer’s disease (AD)—amyloid-β (Aβ) plaques, neuronal tau tangles, granulovacuolar neurodegeneration (GVD), and neuronal cell loss—relate to each other. The major problem is that tau pathology in these models can only be induced artificially, by using frontotemporal dementia causing tau mutation or by injecting tau seeds isolated from AD patient brains. Thus, the fundamental unanswered questions remain as to whether Aβ can induce tau pathology and how neurons die in AD. Unknown human-specific features that escape modeling in rodents are likely underlying this conundrum.

2Two good models are available that replicate human features of AD: a three-dimensional cell culture based on different types of human cells and xenografted human neurons in mouse brains. In this study, we improved the previously Nod-SCID–based xenotransplantation model using the Rag2-/- (Rag2tm1.1Cgn) immunosuppressive genetic background and a single AppNL-G-F (Apptm3.1Tcs/Apptm3.1Tcs) knock-in gene to drive Aβ pathology. Human stem cell-derived neuronal progenitor cells (NPCs) transplanted into control Rag2-/- mice integrated well and developed dendritic spines (fig. S1, A and B). Two months after transplantation, the xenografted neurons already displayed some characteristic mature neuronal (NEUN and MAP2) and cortical markers (CTIP2, SATB2, TBR1, and CUX2). Unlike rodent neurons, human neurons expressed equal amounts of 3R and 4R tau splice forms at 6 months after transplantation (figs. S1C and S5H). Compared to the previously used model, the animals showed a considerably longer life span (>18 months longer), allowing for the study of healthy human neurons during brain aging.

3We xenografted 100,000 green fluorescent protein (GFP)–labeled H9-derived human cortical NPCs into Rag2-/-/AppNL-G-F (hereafter referred to as amyloid mice) or Rag2-/-/Appmm/mm (control mice). β sheet staining with dye X34 revealed robust plaque pathology in the amyloid mice (Fig. 1A and fig. S1, D and E). The number of cells (fig. S6H) and transcriptional profile of the human transplants at 2 months were very similar in control and amyloid mice (Fig. 2A). Thus, human neurons integrated and differentiated similarly in control and amyloid mice, in line with the fact that amyloid plaque pathology appeared only after 2 months in this model. Full-blown amyloid plaque pathology was seen at 18 months after transplantation. At this late point in time, human neurons in the control brain appeared generally healthy and displayed neuronal projections and dendritic spines intermingled with host microglial (IBA1) and astroglial (GFAP) cells (fig. S1D). In contrast, grafted neurons in the amyloid mice displayed severe dystrophic neurites associated with microgliosis (six to seven IBA1-positive microglia per Aβ plaque) and astrogliosis (two to three GFAP-positive astrocytes per plaque) (Fig. 1A and fig. S2, A and B). The glial cell recruitment to plaques was very similar to that seen in nongrafted control animals (fig. S3, A and B). Immunohistochemistry with anti–phosphorylated tau (p-tau) antibodies AT8 (p-tau Ser202 and Thr205), PHF1 (p-tau Ser396 and Ser404), and MC1 (pathological conformational tau epitope) demonstrated considerable deposition of neuritic plaque tau (NP-tau) in the human neurons of the grafted amyloid mice (Fig. 1, C and D). About 2% (AT8), 5% (PHF1), and 3% (MC1) of the surface was stained with the indicated antibodies in a 20-μm ring around the Aβ plaques (Fig. 1, E to G). Both NP-tau– and AT8-positive neurons appeared as early as 6 months after transplantation (fig. S1E), indicating that amyloid deposition drives tau phosphorylation early on in this model. Notable tau pathology was not observed at 18 months in mouse neurons in the same animal nor in human grafted control or mouse neurons grafted in amyloid mice (Fig. 1, C and D, and figs. S1E, S3, C and H, and S4, A to C).

4Pathological, β sheet X34– positive tau reactivity was seen in the xenografted human neurons (fig. S2C). Furthermore, Gallyas silver staining and tau immunogold labeling of PHF fibril-like structures extracted with sarkosyl from grafted neurons in the amyloid mice (Fig. 1, H and I) confirmed the progression of p-tau into pathological states (>20 tau fibrils per electron microscopy grid (control n = 4, amyloid n = 4) (fig. S6F). Such pathology was largely absent in grafted control or nongrafted amyloid mice (Fig. 1H and figs. S6F and S3, I and J). Finally, and of clinical relevance, we found that the increase in p-tau181 and p-tau231 was statistically significant in the plasma of the grafted amyloid mice but not in control grafted or nongrafted mice (Fig. 1, J and K), reflecting increased secretion of soluble p-tau biomarkers into the bloodstream in response to amyloid pathology, as in human AD.

5It is possible to quantify accurately up to 0.1 ng of human DNA in a mixture with 100 ng mouse genomic DNA using quantitative PCR (qPCR) (fig. S6D). Using this assay, we estimated that up to 50% fewer human neurons were present in the amyloid animals compared with controls (Fig. 1L and fig. S6G) at 6 and 18 months after transplantation. Of note, no significant increases in human neurons in the control situation were observed (fig. S6E), and the number of mouse neurons was also not significantly affected in nongrafted control or amyloid mice (fig. S3, E to G). Thus, human, but not mouse, neurons developed AD-like tau pathology upon exposure to amyloid plaques in vivo and degenerated much like their counter-parts in the human brain by an unknown death process.

6Previous work has found opposite effects of Rag2-/- immunodeficiency on amyloid plaque pathology and cellular responses. Comparison of soluble Aβ (Aβ40 and Aβ42) and guanidine-extractable Aβ (Aβ40 and Aβ42) revealed no significant differences between Rag2-/-/AppNL-G-F and AppNL-G-F mice. Bulk RNA sequencing indicated that the main aspects of the innate microglia and astroglial cellular transcriptomic responses to amyloid plaques are maintained in the Rag2-/-/AppNL-G-F immunodeficient mouse compared with a previously characterized nonimmunosuppressed amyloid mouse (APP/PS1) (Fig. 1B) [coefficient of determination (R2) = 0.81 for differentially expressed genes]. Thus, although we acknowledge the lack of an adaptive immune system as a weakness of the model, we note that all key features of the neuropathology of AD were reproduced.

7We isolated the transplanted neuron-positive brain regions from mice at 2, 6, and 18 months after transplantation and extracted total RNA for sequencing (mean human reads: ~13.3 million) (fig. S5, A and B). Mapping to mouse and human reference databases generated speciesspecific datasets. As expected, up-regulation of the mouse genes Axl, B2m, Cst7, Ctss, Itgax, Trem2, Tyrobp, Lyz2, Ccl3, and Ccl4 confirmed microglial activation when exposed to amyloid in Rag2-/-/AppNL-G-F mice (Fig. 1B; fig. S7, A to C; and table S2). Human samples clustered according to genotype and age (fig. S5C). Differential expression (DE) analysis of the human grafts at 2 months showed only a few significantly down-regulated DE genes of unknown relevance (Fig. 2A and table S1). In contrast, DE of grafts in amyloid and control conditions at 6 and 18 months revealed 916 up-regulated and 73 down-regulated genes, and 533 up-regulated and 34 down-regulated genes, respectively (Fig. 2, B and C, and table S1). Gene alterations of interest included CD74, a marker for tau tangle-containing cells; HBB, a cortical pyramidal neuronal marker; and A2M, which is associated with neuritic plaques in AD human brains (fig. S5E). In addition to the neuronal signature, in line with previous findings, we also observed astroglial and oligodendrocytic genes, including MOBP, MBP, OPALIN, S100A6, and APOE, among others, indicating that the grafted neural stem cells also generated glia in vivo (fig. S5F). A limitation of the current study is the lack of an accurate estimate of the cellular composition of the graft that might influence the transcriptional profile of the grafted cells. Nevertheless, the overall expression profiles of the grafts exposed to amyloid plaques at 6 and 18 months showed strong correlations (R2 = 0.88), suggesting that the pathological cell states of neurons at 6 and 18 months are similar (fig. S5G). We found a prominent enrichment using GSEA (gene set enrichment analysis) between previously published AD datasets (table S3), including the ROSMAP (Religious Orders Study and Rush Memory and Aging Project) cohort, and our data in transplanted neurons at 6 and 18 months after transplantation [adjusted P value (Padj) < 0.05] but not at 2 months (fig. S8, A and B). Furthermore, GSEA also revealed a clear enrichment in up-regulated genes from transplanted neurons and neurons directly reprogrammed from fibroblasts of AD patients (6 months: Padj = 1.64 × 10-5; 18 months: Padj = 2.19 × 10-4). These data confirmed that our transplanted neurons capture AD-relevant transcriptional signatures.

8Functional Gene Ontology (GO) enrichment analysis of significantly up-regulated genes (Padj < 0.05) at 6 months using DAVID (Database for Annotation, Visualization and Integrated Discovery) and REVIGO (Reduce and Visualize Gene Ontology) covered semantic space around positive regulation of transcription, protein phosphorylation, positive regulation of mitogen-activated protein kinase cascade, inflammatory responses, including tumor necrosis factor (TNF) and interferon signaling, cell proliferation, aging, tissue regeneration, and myelination (Fig. 2D and table S1). Some reports have suggested that neurons in AD display signatures of hypomaturity, dedifferentiation, and cell cycle reentry. Using previously published datasets (table S1) and GSEA, we found that cellular signatures of hypomaturity, dedifferentiation, P53, hypoxia-inducible factor α signaling, nuclear factor κB (NF-κB), transforming growth factor-β (TGFβ) signaling, (positive and negative) regulation of cell cycle, and cell cycle reentry were indeed enriched in the 6 and 18 month grafts but not at 2 months (Fig. 2E). Conversely, mature neuronal marker pathways, such as synaptic plasticity, synaptic transmission, long-term potentiation, and axon guidance were not significantly enriched (Fig. 2E).

9How neurons die in AD remains controversial. As shown in Fig. 1L and fig. S6G, about 50% of xenografted human neurons were lost in the amyloid mice. We did not observe any significant alterations in the expression of genes associated with cell death mechanisms, such as apoptosis or ferroptosis, but found significant up-regulation of MLKL, the gene encoding the executor protein of necroptosis (Fig. 2, B and C, and figs. S2E and S6, A to C). This was not observed in the transcriptome of the host tissue (fig. S7, A to C). We used pRIPK1- (Ser166), pRIPK3- (Ser227), and pMLKL-specific (Ser358) antibodies to stain brain tissue from 18-month-old grafted and nongrafted mice. We noticed intense punctuate staining with a vesicular pattern in the soma of the neurons in grafted amyloid mice. These vesicular structures co-stained with casein kinase 1 delta (CK1δ), a marker for granulovacuolar degeneration (Fig. 2L and fig. S6I). Nongrafted control and amyloid animals or mouse neurons derived from NPCs and similarly transplanted into mice did not display these pathologies (figs. S3D and S4D).

10We validated this observation by qPCR on RNA extracted from the temporal gyrus of AD and age-matched control brain samples, revealing significant up-regulation of MLKL and RIPK3 (Fig. 2, F to H). We have, on the basis of our initial observations in the xenograft model, extensively confirmed the presence of necroptosis markers in GVD in AD patients in a previous publication. Thus, our model has predictive value for the neuropathology of AD.

11We identified 36 up-regulated and 21 down-regulated long noncoding RNAs (lncRNAs) in the grafts of 6-month-old animals (table S1). Some of these noncoding RNAs (e.g., NEAT1) have been implicated in AD, but overall, it is unclear whether or how they contribute to pathogenesis. In our data, the lncRNA MEG3 (Maternally Expressed 3) was the most strongly (~10-fold) up-regulated gene in human neurons exposed to amyloid pathology (Fig. 2, B to C, and fig. S2D) but not in the host mouse neurons (fig. S7, D to F). MEG3 has been linked to cell death pathways through p53, is involved in the TGFβ pathway, and has been associated with Huntington’s disease.

12While MEG3 has not been implicated in AD, we found up-regulation of MEG3 in AD patient brains in a single-nucleus transcriptomic database. We confirmed two- to threefold up-regulation of MEG3 in RNA extracted from the temporal gyrus of AD patients (table S5) using qPCR (Fig. 2I). MEG3 in situ hybridization combined with cell-specific immunohistochemical markers NeuN (neurons), IBA1 (microglia), and GFAP (astrocytes) demonstrated its exclusive expression in the nucleus of neurons (fig. S9, A and B) and its strong enrichment in AD brains [2 to 3 puncta per nucleus in control and 8 to 10 puncta per nucleus in AD brain (P = <0.0001)] (Fig. 2, J and K). In AD, MEG3-expressing neuronal nuclei appeared blotched, with reduced 4′,6-diamidino-2-phenylindole (DAPI) intensity, and MEG3-positive neurons also displayed high levels of the necroptosis marker pMLKL (fig. S9C).

13We performed Sanger sequencing on brain cDNA and found that from 15 known transcriptional variants, MEG3 transcript variant 1 (accession number NR_002766.2) was most abundantly expressed in human adult brain. A lentivirus was used to express MEG3 V1 in H9-derived mature cortical neurons (Fig. 3, A and B), resulting in strong reduction of cell viability at 9 days compared with control (Fig. 3C). Immunohistochemical analysis revealed the presence of activated necroptotic markers pRIPK1 (Ser166), pRIPK3 (Ser227), and pMLKL (Ser358) (Fig. 3, D to F). Necroptosis-positive neurons displayed a reduced amount of cyto-skeletal filament neurofilament-H (NF-H). Immunoblot analysis confirmed increased pRIPK1 (Ser166) levels (Fig. 3G). Notably, MEG3-induced cell loss was rescued by necroptosis inhibitors ponatinib, dabrafenib, or necrosulfonamide and by CRISPR-mediated deletion of RIPK1, RIPK3, and MLKL in vitro (Fig. 3, I and J, and fig. S10, H to J).

14We performed RNA sequencing of MEG3-transduced neurons 7 days after transduction and compared the transcriptomic signatures with those from the xenografted neurons using GSEA. We ranked the genes from the amyloid and control xenografts according to their log fold change differential expression (at 6 months; see Fig. 2B) and compared those to the top up-regulated genes in response to MEG3 transduction in neurons in vitro. We found a significant enrichment of the up-regulated genes (normalized enrichment score = 1.5, Padj = 1.2 × 10-5) (Fig. 3H and table S4), suggesting that some transcriptional changes in the xenografted neurons in vivo might be the consequence of up-regulated MEG3. DE analysis of MEG3 transduced neurons revealed no changes in necroptosis genes (fig. S10, A to C). DAVID GO analysis of leading-edge genes from GSEA on up-regulated genes from 6-month transplanted human neurons and in vitro MEG3-expressing neurons revealed signatures of NF-κB and TNF signaling, which are related to necroptosis induction. In addition, signatures of lipid transport, interferon-γ signaling, positive regulation of peptidyl-tyrosine phosphorylation, and apolipoprotein L signaling were observed (fig. S10D and table S4).

15Blocking MEG3 up-regulation in NPCs using a previously characterized short hairpin RNA (shRNA) antisense construct significantly improved neuronal survival and was associated with down-regulated expression of necrosome proteins (Fig. 3, K and L, and fig. S10, E to G). It is possible that necroptosis markers are limited to the neurons that are remaining at the stage of pathological analysis. We thus treated xenografted mice (three groups of n = 5) from 2 until 6 months after transplantation with the orally available necroptosis kinase inhibitors ponatinib (30 mg/kg) and dabrafenib (50 mg/kg). Ponatinib inhibits RIPK1 and RIPK3 and is a US Food and Drug Administration-approved drug for the treatment of acute lymphoid leukemia and chronic myeloid leukemia. Dabrafenib is a more specific inhibitor of RIPK3. Immunostaining at 6 months revealed a large reduction in levels of pRIPK1, pRIPK3, and pMLKL in both treatment groups compared with untreated mice (Fig. 4, A to D), without alteration of the glial response to amyloid (fig. S9D). qPCR analysis of neuronal cell numbers revealed a significant increase in neurons in the dabrafenib-treated group compared with the control (Fig. 4E). Transplanting RIPK1 knockout NPCs generated using CRISPR-Cas9, as discussed earlier, demonstrated that neurons were not viable, in agreement with the lethal phenotype of RIPK1 in mouse. In contrast, deletion of the RIPK3 gene significantly improved neuronal survival (Fig. 4, F to H). Immunofluorescence staining confirmed the significant decrease of necrosome signals in RIPK3-deficient human neurons.

16Our results demonstrate that amyloid pathology is sufficient to induce hallmark neuropathology of AD, including neuronal tangles and secretion of p-tau181 and p-tau231 biomarkers in blood, by simply exposing human neurons to amyloid plaques. We asked the crucial question of how neurons in AD die. Reports have proposed apoptotic mechanisms, but no convincing evidence has demonstrated that this drives neuronal cell loss in AD. Previous publications, one instigated by the observations in the xenograft mice, have confirmed that necroptosis occurs in the AD brain. We demonstrate here that neuronal loss can be rescued by treatment with clinically relevant necroptosis inhibitors or by ablating the RIPK3 gene in transplanted neurons. Thus, we suggest that neuronal death in AD is largely driven by necroptosis, linking the loss of neurons to inflammatory processes that are upstream of this well-studied death pathway. Therapies that prevent neuronal cell loss, in combination with more-mainstream Aβ- and tau-targeted interventions, might be useful additions to the current efforts to develop disease-modifying strategies for AD. Necroptosis is an active area of drug development in cancer and ALS.

17Our data suggest that necroptosis is downstream of the accumulation of pathological tau and is induced by the up-regulation of the non-coding RNA MEG3, possibly via TNF inflammatory pathway signaling. It needs to be mentioned that the imprinting status of MEG3 and its expression may differ among human embryonic stem cell lines. lncRNAs are important regulators of gene expression and influence a variety of biological processes, including brain aging and neurodegenerative disease. The large number of noncoding RNAs that are differentially expressed in our AD model warrants further investigation.

18It is intriguing that human transplanted neurons display an AD phenotype, whereas mouse neurons interspersed within the graft or neurons derived from mouse NPCs and transplanted in a similar way as their human counterparts do not display such a phenotype. This suggests that unknown human-specific features define the sensitivity of the neurons to amyloid pathology. We noticed that the transplanted human neurons display signatures of down-regulation of mature neuronal properties and up-regulation of immature signaling pathways, which aligns with a recent study analyzing the transcriptional profiles in neurons directly derived from fibroblasts of AD patients (induced neurons). Those neurons did not, however, show tangles, amyloid, or necroptosis markers, and it was therefore not clear how those observations relate to the classical hallmarks of AD. The contribution of these immature signatures to the disease process needs further investigation, but our data indicate that necroptosis is an important contributor to cell death in AD.

19Transplanted human neurons, in contrast to transplanted mouse neurons, become diseased when exposed to amyloid pathology. Understanding the molecular basis of the resilience of mouse neurons to amyloid pathology will not only help to model the disease better but might also stimulate research into pathways that protect against neurodegeneration.