Neurodegeneration risk factor, EIF2AK3 (PERK), influences tau protein aggregation

Abstract

Tauopathies are neurodegenerative diseases caused by pathologic misfolded tau protein aggregation in the nervous system. Population studies implicate EIF2AK3 (eukaryotic translation initiation factor 2 alpha kinase 3), better known as PERK (protein kinase R-like endoplasmic reticulum kinase), as a genetic risk factor in several tauopathies. PERK is a key regulator of intracellular proteostatic mechanisms-unfolded protein response and integrated stress response. Previous studies found that tauopathy-associated PERK variants encoded functional hypomorphs with reduced signaling in vitro. But, it remained unclear how altered PERK activity led to tauopathy. Here, we chemically or genetically modulated PERK signaling in cell culture models of tau aggregation and found that PERK pathway activation prevented tau aggregation, whereas inhibition exacerbated tau aggregation. In primary tauopathy patient brain tissues, we found that reduced PERK signaling correlated with increased tau neuropathology. We found that tauopathy-associated PERK variants targeted the endoplasmic reticulum luminal domain; and two of these variants damaged hydrogen bond formation. Our studies support that PERK activity protects against tau aggregation and pathology. This may explain why people carrying hypomorphic PERK variants have increased risk for developing tauopathies. Finally, our studies identify small-molecule augmentation of PERK signaling as an attractive therapeutic strategy to treat tauopathies by preventing tau pathology.

Keywords: EIF2AK3; ER stress; PERK; eIF2α phosphorylation; integrated stress response; neurodegeneration; tau aggregation; tauopathy; unfolded protein response.

Figure 1

Molecular function and population distribution of PERK variants in human diseases.A, pie chart shows the molecular functional classification of 1294 genetic human PERK variants reported in Genome Aggregation Database (gnomAD). B, pie chart shows the frequency of human PERK variants reported in gnomAD. About 1270 PERK variants are ultrarare (<0.1% frequency). About 14 PERK variants are rare (0.1% to 1% frequency). About 10 variants are common (>1% frequency). C, prevalence of the two most common PERK variants, haplotype A and haplotype B, across seven racial/ethnic groups found in gnomAD. The frequency of tauopathy-risk variant, haplotype B, ranges from ∼49% in East Asian populations to ∼5% in African population. Conversely, the frequency of tauopathy-protective variant, haplotype A, ranges from 94% in African population to ∼50% in East Asian populations. D and E, PERK protein cartoons show positions of missense variants linked to WRS (D) and tauopathies (E) as reported in gnomAD database and NCBI since 2000. Functional domains of the 1116 amino acid human PERK protein include IRE1-like luminal ER stress–sensing domain, ER transmembrane domain, and amino (N) and carboxyl (C) kinase lobes in the cytoplasmic domain. PERK missense variants linked to WRS are all ultrarare and target PERK’s N- and C- kinase lobes. PERK missense variants linked to tauopathies include common (green), rare (red), and ultrarare (blue) variants. Tauopathy PERK missense variants frequently map to PERK’s luminal domain; do not overlap with WRS variants; and do not target the kinase lobes. ER, endoplasmic reticulum; IRE1, inositol-requiring enzyme 1; NCBI, National Center for Biotechnology Information; PERK, protein kinase R-like endoplasmic reticulum kinase; WRS, Wolcott–Rallison syndrome.

Figure 2

Tauopathy-associated variants disrupt hydrogen bonds in the PERK luminal domain.A and B, amino acid sequence alignments show conservation of human tauopathy protective PERK variants, S136 and R240, with chimpanzee, monkey, rat, and mouse PERK proteins. Human tauopathy PERK risk variants, C136 and H240, are not reported in other mammalian PERK proteins. C–F, hydrogen bonds formed by the PERK 136 and 240 residues were modeled by PyMol using the mouse PERK luminal domain crystal structure (PDB ID: 4YZY; MMDB ID: 129295). C, the combination of the protective PERK S136 and R240 variants forms seven H-bonds (dashed lines) locally including one direct S136-R240 H-bond. D, the combination of the risk PERK C136 and protective R240 variants forms five H-bonds locally including one direct C136-R240 H-bond. E, the combination of the protective PERK S136 and risk H240 variants forms four H-bonds locally. S136-H240 cannot form direct H-bonds. F, the combination of risk PERK C136 and H240 variants forms two H-bonds. C136-H240 cannot form direct H-bonds. White = carbon; blue = nitrogen; red = oxygen; and yellow = sulfur. G, pathogenicity of PERK S136C and R240H missense changes was bioinformatically assessed by five algorithms: PolyPhen-2 (HumDiv), PROVEAN, MutationTaster, SIFT, and CADD. PERK S136C was pathogenic using SIFT and CADD. PERK R240H was pathogenic in all algorithms. CADD, combined annotation–dependent depletion; MMDB, Molecular Modeling Database; PDB, Protein Data Bank; PERK, protein kinase R-like endoplasmic reticulum kinase; PolyPhen-2, protein kinase R-like endoplasmic reticulum kinase; PROVEAN, protein variation effect analyser; SIFT, sorting intolerant from tolerant.

Figure 3

Tau aggregation in cell culture does not induce ER stress.A and B, protein lysates were prepared from wildtype and PS19 mouse brains. Soluble and insoluble protein lysate fractions were immunoblotted for total human Tau (HT7) and phospho-human Tau (AT8). Arrowheads mark positions of tau protein. C, Biosensor cells were transfected with wildtype or PS19 soluble brain lysate. After 24 h, Tau-YFP aggregates (puncta) were imaged by fluorescent microscopy. The white box outlines magnified image of Biosensor cells with puncta. The scale bar represents 50 μm. D, fluorescent puncta were quantified after wildtype or PS19 mouse brain lysate transfection, and the puncta number was normalized to cell number (∗∗p ≤ 0.01, one-tailed Student’s t test, n = 6 independent transfections, mean ± SD). E–I, the mRNA levels of PERK-, IRE1-, ATF6-, and ERAD-regulated genes were examined by RNA-Seq of Biosensor cells transfected with wildtype or PS19 mouse brain protein lysate for 24 h. E and F, gene expression levels of 31 PERK-regulated genes in PS19 brain lysate–treated cells relative to wildtype brain lysate–treated cells are shown as log₂ fold change. Graph I shows levels of the five PERK-regulated genes most significantly reduced between wildtype and PS19-treated cells and expression levels of the PERK gene itself. The violin plot (F) shows levels of the entire PERK-regulated gene set. G and H, gene expression changes of 32 IRE1-regulated genes in PS19-treated cells relative to wildtype treated are shown. The graph (G) shows levels of the five IRE1-regulated genes most differentially expressed between wildtype and PS19-treated cells and expression levels of IRE1 gene itself. The violin plot (H) shows levels of the entire IRE1-regulated gene set. I and J, gene expression changes of 74 ATF6-regulated genes in PS19 brain lysate–treated cells relative to wildtype treated are shown. The graph (I) shows levels of the five ATF6-regulated genes most differentially expressed between wildtype and PS19-treated cells and expression levels of ATF6 gene itself. The violin plot (J) shows levels of the entire ATF6-regulated gene set. K, the violin plot shows levels of 74 ERAD-regulated genes. L, GO analysis identifies significantly decreased ER stress term (GO:0034976) in PS19-treated Biosensor cells. ERAD pathway (GO:0036593) shows no significant (ns) change after PS19 brain treatment. Error bars in E, G, and I represent mean ± SD. Black circles and squares represent five independent experimental replicates. (∗p ≤ 0.05, ∗∗p ≤ 0.01, ns, two-tailed Student’s t test). The red horizontal line in figures F, H, J, and K marks the median level of expression of the gene set, and the thin horizontal blue lines delimit upper and lower gene expression quartiles in the violin plots. (∗∗∗∗p ≤ 0.0001, one-sample t test and two-tailed Wilcoxon signed rank test, n = 5 experimental replicates). Detailed gene expression RNA-Seq information is shown in Data S3. ATF6, activating transcription factor 6; ER, endoplasmic reticulum; ERAD, ER stress–associated degradation; GO, Gene Ontology; IRE1, inositol-requiring enzyme 1; PERK, protein kinase R-like endoplasmic reticulum kinase.

Figure 4

PERK signaling prevents Tau-YFP aggregation. A, Biosensor cells were transfected with wildtype or PS19 brain lysate and coincubated with GSK2656157 (5 μM), GSK2606414 (5 μM), Salubrinal (2.5 μM), or ISRIB (5 μM). After 24 h, Tau-YFP fluorescent puncta were imaged by fluorescent microscopy. The scale bar represents 30 μm. B, quantification of puncta number from (A) normalized by cell number. P value was calculated by two-way ANOVA Tukey’s multiple comparisons test, mean ± SD. ∗p ≤ 0.05, ∗∗p ≤ 0.01, and ∗∗∗∗p ≤ 0.0001 (n ≥ 5 experimental replicates). C, Biosensor cells were transfected with PS19 brain lysate and coincubated with ISRIB for 24 h. Media were replaced with/without Salubrinal for another 24 h. Tau-YFP fluorescent puncta were imaged by microscopy after these drug treatments. The scale bar represents 25 μm. D, quantification of puncta number from (C) normalized by cell number. E and F, PERK^−/− or PERK^+/+ MEFs were transduced with TauRD(P301S)-YFP. After 48 h, cells were imaged by fluorescence microscopy (E, the scale bar represents 25 μm), and protein lysates were prepared. F, soluble protein fractions were immunoblotted for Tau-YFP and HSP90 (loading control). Insoluble fractions were immunoblotted for Tau-YFP and Ponceau stained (loading control). G and H, eIF2α^A/A or eIF2α^S/S MEFs were transduced with TauRD(P301S)-YFP. After 48 h, cells were imaged by fluorescence microscopy (G, the scale bar represents 25 μm), and protein lysates were prepared. H, soluble protein fractions were immunoblotted for Tau-YFP or HSP90 (loading control). Insoluble fractions were immunoblotted for Tau-YFP and lamin A/C (loading control). eIF2α, eukaryotic initiation factor 2 alpha; MEF, mouse embryonic fibroblast; PERK, protein kinase R-like endoplasmic reticulum kinase.

Figure 5

IRE1 and ATF6 pathway inhibitors cause Tau-YFP aggregation in cell culture. Biosensor cells were transfected with PS19 brain lysate and coincubated with ATF6 inhibitor Ceapin-A7 (10 μM), IRE1 inhibitor, 4u8c (10 μM), or ATF6 pathway activator, AA147 (10 μM). After 24 h, Tau-YFP fluorescent puncta were imaged by fluorescent microscopy. The scale bar represents 30 μm. B, quantification of puncta number from (A) was normalized by cell number. p Value was calculated by two-way ANOVA Tukey’s multiple comparisons test, mean ± SD, not significant (ns), ∗∗∗∗p ≤ 0.0001 (n ≥ 5 experimental replicates). ATF6, activating transcription factor 6; IRE1, inositol-requiring enzyme 1.

Figure 6

PERK pathway activity is reduced in Alzheimer’s disease (AD) patient brains.A and B, protein lysates were prepared from 1 mg of five Braak I (normal) and five Braak VI (AD) patient hippocampi. Soluble (A) and insoluble (B) protein lysate fractions were immunoblotted for total human Tau (HT7), phospho-Tau (AT8), and GAPDH (loading control). Individual brain identification numbers are listed above the blots, and associated clinicopathology information are provided in Table 1. C, Biosensor cells were transfected with normal (Braak I) or AD (Braak VI) soluble brain lysate. After 24 h, Tau-YFP aggregates (fluorescent puncta) were imaged by fluorescent microscopy. The scale bar represents 50 μm. D–F, soluble brain lysates from (A) were immunoblotted for phospho-PERK and total PERK; protein levels were quantified by densitometry and normalized by loading controls, PERK (D) and GAPDH (A). PERK was not detected in insoluble fraction (Fig. S3). ∗∗p ≤ 0.01, not significant (ns), two-tailed Student’s t test. Mean ± SD. G and H, gene expression levels of PERK-regulated genes in hippocampi of AD brains (n = 8) relative to non-AD brains (n = 10) from GSE173955 (56) are shown as log₂ fold change. The graph (G) shows levels of the five PERK-regulated genes most significantly reduced between AD brains and non-AD brains and expression levels of PERK gene itself. The violin plot (H) shows levels of the entire PERK-regulated gene set. Detailed clinicopathology information are available from Ref. (56) and summarized in Table 1. I and J, gene expression levels of PERK-regulated genes from hippocampi of AD brains (n = 12) relative to non-AD brains (n = 10) from Ref. (55) (GSE159699) are shown as log₂ fold change. The graph (I) shows levels of the five PERK-regulated genes most significantly reduced between AD and non-AD brains and PERK gene expression levels itself. The violin plot (J) shows levels of the entire PERK-regulated gene set. Detailed clinicopathology information of these AD and non-AD brains are available from Ref. (55) and summarized in Table 1. Error bars in G and I represent mean ± SD. Black circles and squares represent individual AD or non-AD brains. (∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, not significant [ns], two-tailed Student’s t test). The red horizontal line in figures H and J marks the median level of gene expression, and the thin horizontal blue lines delimit upper and lower gene expression quartiles in the violin plots. (∗∗p ≤ 0.01, one-sample t test and two-tailed Wilcoxon signed rank test). PERK, protein kinase R-like endoplasmic reticulum kinase.