Progression of pathology in PINK1-deficient mouse brain from splicing via ubiquitination, ER stress, and mitophagy changes to neuroinflammation

Pink1 ^−/− and wildtype (WT) control mice, which were derived from common ancestors and share the strain 129/SvEv genetic background, were bred and genotyped as previously reported [29]. Brain tissue from Pink1 ^−/−+A53T-SNCA double mutant mice was obtained as published [20].

Affymetrix oligonucleotide microarray profiling was performed with Genechip mouse genome 430 2.0 arrays as previously [30, 31], using cRNA from the brain cerebellar tissue as reported before [20] and from neuron-rich primary cultures from 3 Pink1 ^−/− versus 3 age and sex-matched WT control mice. Hybridization occurred on Affymetrix Genechip mouse genome 430 2.0 arrays, which represent 39,000 transcripts, of which more than half are anonymous or poorly understood, according to PubMed and GeneCards database searches. The biomathematical analysis was performed in the institute for medical genetics at Tuebingen University.

For protein-protein interaction (PPI) network analysis, the software tool String v.10 (https://​string-db.​org/​) with standard settings has been employed to visualize networks of significant dysregulations [32]. As recommended, gene symbols of factors with significant dysregulation were entered into the Multiple Proteins window with the Mus musculus option, the matching of the input with the correct factors was accepted, and the graphic interaction diagram was generated and archived. The Analysis button was used to generate automated network statistics; significant functional enrichments of GO (Gene Ontology) terms and KEGG pathways were exported into EXCEL files.

For an additional comprehensive transcriptome analysis, gene set enrichment analysis (GSEA, v2.2.3, http://​software.​broadinstitute.​org/​gsea/​index.​jsp) [28] was applied in order to see, if a priori defined sets of genes show statistically significant, concordant differences between mutant/WT in 18 months old cerebellum samples. For every gene, only the one entry with the lowest adjusted p value and the according log₂ transformed ratio was taken. GSEA default settings and Reactome v5.2 and KEGG v5.2 gene set database were used. Pathways with p value ≤0.05 and FDR q ≤ 0.25 were regarded as significant. Heat maps were produced with the Perseus software.

Neuron-rich primary cultures from the dissociated brain cerebral cortex of postnatal mice were prepared as previously described [33]. In short, 500,000 cells per well were seeded on a 0.01% (w/v) poly-d-lysine coated 6-well plate. In order to limit the growth of dividing cells, cytosine ß-D-Arabinoside was added on the second day of culture as before. The 3 plate pairs (mutant versus WT), where many singular neurons in homogeneous density with a dense network of processes contrasted with very few astrocytes and microglia present, were chosen out of a total of 12 plate pairs on culture day 12 for RNA extraction.

Mouse aging and dissection of mice was carried out as before, employing cerebellum from single mutant Pink1 ^−/− at three ages (10 mutants versus 10 WT) and midbrain from adult double mutant Pink1 ^−/−+A53T-SNCA mice at age 18 months (5 mutants versus 5 WT) for the extraction of global RNA and cDNA synthesis [20].

Immunohistochemistry was performed using an automated staining system (Leica Bond-III, Nussloch, Germany). The following antibodies were used: anti-GFAP, rabbit polyclonal antibody, dilution 1:14,000 (DakoCytomation, Glostrup, Denmark); anti-Iba1, rabbit antibody, dilution 1:1000 (Wako Pure Chemical Industries, Osaka, Japan).

Representative areas of striatum, substantia nigra, neocortex, and brainstem of Pink1 ^−/− (n = 1) and wild-type (n = 1) animals were analyzed using ImageJ software (Version 1.51 h; National Institutes of Health, Bethesda, Maryland, USA). We quantified positive cells in relation to all cells.

Matched pairs of male and female Pink1 ^−/− and Pink1 ^+/+ mice were used for analysis of bioactive lipids in brain tissue. Mice were 9–13 months (four each), 17 months (three each), and 21 months (four each) old at the time of tissue preparation (mean age 15.5 −/− and 17 +/+). Mice were sacrificed by carbon dioxide. Blood was drawn into K+ EDTA microvettes (Sarstedt) for plasma analysis by cardiac puncture. Subsequent intracardial perfusion with saline removed rests of blood. The lumbar spinal cord, olfactory bulb, and hippocampus were dissected, and tissue pieces of 3–5 mg were excised, rapidly frozen in liquid nitrogen, and stored at −80 °C until analysis. The precise tissue weight was determined on precision scales directly before tissue homogenization.

Sphingolipids were analyzed by liquid chromatography tandem mass spectrometry (LC-MS/MS) in different regions of the nervous system (olfactory bulb, hippocampus, spinal cord) at three different ages (8.5–12.5 weeks, 17.5 weeks, and 21 weeks). LC-MS/MS analyses were done on an API4000 triple quadrupole mass spectrometer equipped with an APCI (atmospheric pressure chemical ionization) ion source for the analysis of ceramides and with an ESI (Electrospray Ionization) ion source for the analysis of sphingosines, Glu-Cer/Lac-Cer (Sciex, Darmstadt, Germany) [34, 35]. All quadrupoles were working at unit resolution. Concentrations of the calibration standards, quality controls, and samples were evaluated by MultiQuant 3.0 (Sciex, Darmstadt, Germany) using the internal standard method (isotope-dilution mass spectrometry). Calibration curves were calculated by linear regression with 1/x weighting. The coefficient of correlation for all measured sequences was at least 0.99.

Tissue pieces of approximately 4 mg were homogenized in 200-μl extraction buffer (50 μl per mg). For analysis of sphingolipids, 20 μl plasma or homogenized tissue samples consisting in 0.4 mg tissue were extracted with methanol:chloroform:HCl (15:83:2) after spiking with the respective internal standards, which was Cer17:0 for ceramides and sphingosine-D7 and sphingosine-1-phosphate-D7 for sphingosines. A Luna C18 column (150 mm × 2 mm ID, 5 μm particle size, 100 Å pore size; Phenomenex, Aschaffenburg, Germany) was used for chromatographic separation. The HPLC mobile phases consisted of water-formic acid (100:0.1, v/v) (A) and acetonitrile-tetrahydrofuran-formic acid (50:50:0.1, v/v/v) (B). For separation, a gradient program was used at a flow rate of 0.3 ml/min. The initial buffer composition was 60% (A)/40% (B). It was maintained for 0.6 min, then linearly changed to 0% (A)/100% (B) over 3.9 min, and held for 6.5 min. Subsequently, the ratio was linearly changed back within 0.5 min to 60% (A)/40% (B) and then held for another 4.5 min. The running time for every sample was 16 min. The injection volumes were 15 μl for ceramides and 10 μl for sphingosines. The analyses were done in multiple reaction monitoring (MRM) mode. For every analyte, two transitions were recorded, one for quantification and the other for identification. The peak area of the analyte was normalized for the peak area of the internal standard. Precursor to product ion transitions for quantification were: m/z 539 → 264 for C_16:0-Cer, m/z 567 → 264 for C_18:0-Cer, m/z 595 → 264 for C_20:0-Cer, m/z 651 → 264 for C_24:0-Cer, m/z 649 → 264 for C_24:1-Cer, m/z 700 → 264 for C_16:0-GluCer, m/z 729 → 264 for C_18:0-GluCer, m/z 862 → 264 for C_16:0-LacCer, m/z 891 → 264 for C18:0-LacCer, m/z 973 → 264 for C24:1-LacCer, m/z 553 → 264 for C17:0-Cer, m/z 300 → 282 for Sph and m/z 380 → 264 for S1P. The dwell times were 15 or 50 ms.

To assess genotype-dependent differences on individual ceramides in specific regions, mice of different ages were summarized. To assess progression over age, ceramides across regions were pooled, log2-transformed to linearize the data, and subsequently summed to get a global readout for all ceramides. Total ceramides were then plotted over time and genotype-dependent differences at different ages were analyzed using two-way ANOVA for “genotype x age”.

The human SH-SY5Y neuroblastoma cell line with dopaminergic properties was stably transduced by lentivirus either with a control (NT for Non-Target knock-down) shRNA or a shRNA directed against PINK1 and maintained under puromycin (1 μg/ml) selection in RPMI medium containing 10% Fetal Calf Serum (FCS), as published already [6]. These PINK1-KD and NT control cell lines had the stability of their KD controlled repeatedly over many months. They were switched to HBSS medium without FCS, to subject them to a starvation time course as previously described [13].

RNA was isolated with the RNeasy mini kit (Qiagen) and then treated with DNase I. cDNA was synthesized with SuperScript III reverse transcriptase using oligo(dT)₂₀ and random primers (Invitrogen). cDNA from 20 to 25 ng RNA were utilized in a 20 μl reaction volume using the StepOnePlus Real-Time PCR System and the appropriate murine (lowercase) or human (uppercase) TaqMan gene expression assays (Applied Biosystems): for mouse Pink1 (Mm00550827_m1), Creb3 (Mm00501607_m1), Ddx58 (Mm01216853_m1), Hebp1 (Mm00469161_m1), Ifit1 (Mm00515153_m1), Ifit3 (Mm01704846_s1), Irf3 (Mm00516784_m1), Mapk8 (Mm01218957_m1, Mm01218946_m1, Mm00489514_m1), Mapk9 (Mm00444239_m1), Mapk14 (Mm01301009_m1), Mavs (Mm00523170_m1), Mfn1 (Mm00612599_m1), Nfkbia (Mm00477798_m1), Rsad2 (Mm00491265_m1), Srsf10 (Mm01193320_m1), Tbk1 (Mm00451150_m1), Tnf (Mm00443258_m1), for human PINK1 (Hs00260868_m1), DDX58 (Hs01061436_m1), HEBP1 (Hs00211123_m1), IFIT1 (Hs03027069_s1), IFIT3 (Hs00155468_m1), IRF3 (Hs01547283_m1), LRRK2 (Hs00411197_m1), MAVS (Hs00920075_m1), MFN1 (Hs00966851_m1), RSAD2 (Hs00369813_m1), SQSTM1 (Hs00177654_m1), TBK1 (Hs00179410_m1). mRNA expression was normalized to the TATA binding protein gene expression or the Hypoxanthine Phosphoribosyltransferase 1 gene expression (Tbp: Mm00446973_m1, TBP: Hs99999910_m1, HPRT1: Hs99999909_m1). Relative expression changes were calculated with the 2^−ΔΔCt method [36].

The drug FCCP, which is known to uncouple the mitochondrial membrane gradient and trigger mitophagy [37], was administered over 24 h at 10 μM concentration to human SH-H5Y neuroblastoma cells or murine embryonal fibroblasts, which had been cultured in DMEM plus 10% FCS and grown to confluency (approximately 4 × 10⁶ cells) in T25 flasks as previously described [38]; then, the cells were collected, and the RNA was extracted with TRIzol methodology.

The synthetic dsRNA polymer poly(I:C), which induces the RNA sensors that activate innate immunity [39, 40], was purchased from InvivoGen in the low molecular weight variant with the transfection agent LyoVec and used at a concentration of 1 μg/ml (for SH-SY5Y cells) or 2 μg/ml (for MEFs) as recommended by the manufacturer during 16 h before harvesting the cells and extracting RNA/protein.

Previously established primary skin fibroblast cultures from 3 homozygous PARK6 patients (passages 12–14) were employed as published [12, 19, 41‐43], in addition to one sex-/age-matched control (the principal investigator G.A., passage 14) and four matched control fibroblast lines from Coriell depository (catalog number AG02261/passage 6/age 61, AG06103/passage 16/age 29, AG06858/passage 5, age 47, AG12207/passage 14/age 68).

The isolation of total proteins from the primary skin fibroblasts was carried out as described [42]. Samples of 20 μg were heated at 90 °C for 5 min and then separated in 10% tris–glycine polyacrylamide gels, using Precision Plus Protein™ All Blue Standards as size marker. Transfer to nitrocellulose membranes (Protran, GE Healthcare) was done at 50 V for 90 min, with blocking in 5% BSA solution in 1X TBS-T for 1 h at room temperature (RT). Primary antibodies against LRRK2 (1:1000, NBP1–49954, Novus), IFIT3 (1:500, 15,201–1-AP, Proteintech), IFIT1 (1:500, 23,247–1-AP, Proteintech), DDX58 (1:700, 3743, Cell Signaling Tech), RSAD2 (1:500, 11,833–1-AP, Proteintech), and β-Actin (1:5000, A5441, Sigma-Aldrich) occurred in 1X TBS-T solutions overnight at 4 °C. Fluorescent-labeled α-mouse (1:15,000, IRDye 800CW, Li-Cor) and α-rabbit (1:15,000, IRDye 680RD, Li-Cor) were the secondary antibodies. Fluorescence detection occurred on the Li-Cor Odyssey Classic Instrument.

Previously, we documented the effects of PINK1 deficiency on brain regions such as midbrain, striatum, and cerebellum at different ages in the absence of stress or in the presence of A53T-alpha-synuclein overexpression as stressor, identifying marked effects of PINK1-deficiency on mitochondrial biology and excitability via global transcriptome profiles [20‐22, 29, 44]. Irrespective of the size of fold-changes, we now document all significant expression dysregulations in Pink1 ^−/− brain, which were consistent (1) across the lifespan and (2) across diverse brain areas (Additional file 1: Table S1). As expected, the loss of Pink1 transcript constituted the most significant effect, being detected by two microarray probesets. An additional downregulation was observed for the splicing repressor Srsf10 (also called Fusip1) (to 63% on average), a stress response factor which regulates the levels of the inflammasome component Caspase 1 [45]. A downregulation was found also for the vesicle endocytosis factor Clta (to 32%), which controls antibody isotype switching [46]. A converse upregulation was documented by two independent probesets for the splicing activator Srrm1 (to 187%), which is responsible for splicing of the lymphocyte surface protein CD44 [47]. Another upregulation detected by two independent oligonucleotide probesets appeared for the precursor RNA processing factor Hnrnpr, which is known to regulate c-Fos and thus neuronal excitability, but also the expression of classical and non-classical MHC class I proteins (probeset 2610528B01Rik showing transcript elevation to 172% and probeset Gm17388 to 131%) [48, 49]. Currently, there is almost no experimental evidence implicating PINK1 in splicing [22] or in endocytosis [50]. Against our expectations, these expression dysregulations did not increase with age, so the fold-changes at age 18 months were very similar to those at 6 weeks. Thus, the focus on the expression of individual transcripts failed to identify markers of pathology progression.

To further elucidate the expression effects of PINK1-deficiency in nervous tissue, we now considered the enrichment of functional pathways and of protein interaction networks, employing automated bioinformatics tools. This approach was focused on the cerebellum, in view of several advantages: (1) The cerebellum is relatively big and easy to dissect with negligible anatomical variance; (2) its expression changes will not get diluted as in other brain regions where substantial neuron population heterogeneity renders many effects minimal upon mixed tissue analysis; and (3) it does not suffer from the neuron loss and astrogliosis, which are expected to occur in the vulnerable brain regions of PD models and which would distort expression profiles.

To illustrate the impact of PINK1 deficiency on specific gene networks within the global cerebellar transcriptome and to visualize the progression from age 6 weeks via 6 months until age 18 months, we produced a STRING diagram of protein-protein interaction clusters for each age (Additional file 2: Fig. S1A–S1C). In these figures, it is evident that the progression of pathology involves progressively increasing numbers of individual factors and pathways, so that at age 18 months, only the most severe effects could be visualized.

In mice at age 6 weeks, around 250 factors showed significant expression changes (with false discovery rate, FDR, adjusted p value below 0.05) in Affymetrix oligonucleotide microarray studies (Additional file 3: Table S2A, first datasheet). Their analysis regarding the enrichment of specific pathways and protein-protein interactions with the automated bioinformatics tools at the STRING web-server detected significant dysregulation for 158 molecules within the Gene Ontology (GO) Cellular Component term “intracellular membrane-bounded organelle” (FDR q value = 6.39e−08) (Additional file 3: Table S2B), in good consistency with the known mitochondrial localization of PINK1 protein during stress. Interestingly, an enrichment was notable for the KEGG pathway “Spliceosome” (q value = 0.04 only, in view of the few dysregulations at this initial stage of pathology) (a blown-up interaction diagram detail around the splicing factor Sfrs10 is shown in Fig. 1a). The STRING diagram (Fig. 1a, Additional file 2: Fig. S1A), which represents this age, shows the intracellular membrane-bounded organelle factors highlighted in red color.

In mice at age 6 months, over 1300 factors showed expression changes at this significance threshold (Additional file 3: Table S2A, second datasheet). Their bioinformatics analysis detected significant dysregulation for 826 molecules within the GO Cellular Component term "intracellular membrane-bounded organelle" (q value = 7.61e−16) (Additional file 3: Table S2C). Significant enrichment in the KEGG pathways “Ubiquitin-mediated proteolysis” (q = 1.01e−05) and “Protein processing in endoplasmic reticulum” (q = 3.34e−04) were notable. The corresponding STRING protein interaction diagram (Additional file 2: Fig. S1B) for this age again shows the intracellular membrane-bounded organelle factors highlighted in red color.

In mice at age 18 months, about 3500 factors showed expression changes at this significance threshold, too many to be processed by STRING bioinformatics (its limit is at 2000 nodes). Thus, only the about 1600 dysregulated factors with an adjusted p value below 0.01 were analyzed regarding interactions and enrichments (Additional file 3: Table S2A, third datasheet). Significant enrichment for 1031 molecules within the GO Cellular Component term intracellular membrane-bounded organelle (q value 6.34e−67) was detected (Additional file 3: Table S2D). At this stage of progression, significant enrichment was observed in several GO biological processes (Additional file 3: Table S2E), particularly 537 factors in “cellular response to stimulus” (q value 2.82e−14), 165 factors in “cellular response to stress” (q value 1.02e−09), and 23 factors in “regulation of mitophagy” (q value 0.038). These data are in agreement with the notion that the well-documented role of PINK1 in mitophagy is a small part of a much broader role of PINK1 in stimulus-dependent signaling and in stress responses. Significant enrichment was also observed in multiple KEGG pathways (Additional file 3: Table S2F), prominently for 47 factors in the “MAPK signaling pathway” (q value 2.64e−06), for 32 factors in the "ubiquitin-mediated proteolysis pathway" (q value 2.64e−06), for 34 factors in the "Protein processing in endoplasmic reticulum" pathway (q value 1.68e−05), and for 20 factors in the “Bacterial invasion of epithelial cells” pathway (q value 6.37e−05). Mildly significant was the enrichment of 21 factors of the “Dopaminergic synapse” pathway (q value 0.005), 16 factors of the “GABAergic synapse” pathway (q value 0.008), and 18 factors of the “Glutamatergic pathway” pathway (q value 0.015). These data are in excellent consistency with previously established roles of PINK1 as a component of MAPK signaling [51, 52], as an activator of PARKIN which prevents bacterial invasions [18], and as a modifier of dopaminergic, GABAergic, and glutamatergic signaling in the nigrostriatal pathway [44, 53‐56]. The agreement of our bioinformatics analysis of the global transcriptome with previously published cell biology data provides credibility to this automated approach.

In the STRING interaction diagram of cerebellar transcriptome changes at age 18 months, the appearance of neuroinflammatory factors was highlighted, manually placing the components of the significantly enriched KEGG pathways “HTLV-I infection”, “Toll-like receptor signaling”, “TNF signaling pathway”, “Fc epsilon RI signaling”, “T cell receptor signaling pathway”, “RIG-1-like receptor signaling”, “Hepatitis C”, and “Influenza A” in the lower left corner and highlighting the Toll-like receptor signaling pathway factors in red color (Additional file 2: Fig. S1C). A detail of this interaction diagram with focus on the inflammatory factors Creb3, Irf3, Nfkbia, Mapk8, Mapk9, and Stat1 is shown in Fig. 1b.

Heat maps were then used to focus on prominent examples among the significantly altered pathways, representing each dysregulated component at the three ages, not only for the Parkinson-resistant cerebellar tissue but also for comparing the Parkinson-vulnerable midbrain/brainstem and striatum (Table 1). This approach visualizes the temporal and spatial appearance of pathology as well as the effect sizes in a color code, with deep red mirroring strong upregulations and deep blue shades for strong downregulations. It also illustrates the number of factors and possibly the relevance of each pathway. Although the effect of PINK1 on ubiquitin-mediated proteolysis and on mitophagy are excellently studied, providing proof of principle that these moderate changes of many pathway components are relevant, it was novel to detect a selective dysregulation in particular for Ube3a that has a known role in the degradation of cytoplasmic misfolded proteins like alpha-synuclein [57‐59], for Xiap with an established function in ubiquitination, mitochondrial apoptosis, and innate immunity [60], for Dnm1l (encoding DRP1) with previously defined effects on PINK1-dependent mitochondrial fission [7, 61], as well as for Slc6a1 as a GABA transporter known to be induced by neuroinflammation [62]. The regional comparison in Table 1 made it evident that PINK1-deficiency-mediated downregulations of stress-response factors dominate in the PD-resistant cerebellar tissue, while the same factors show converse progressive upregulation in the PD-affected midbrain/brainstem tissue, e.g., Keap1, Mbtps1, Rad23a, Sec61a1, Hif1a, Pi4k2a, the autophagy factors Map1lc3b and Sqstm1. Further opposing regulations occurred for the ubiquitin-binding factor Tollip, which regulates Toll-like receptor signaling as well as the autophagic clearance of ubiquitin conjugates and protein aggregates [63]. This supersensitive upregulation in tissues, which are known to suffer PD-specific stress, might represent a cellular effort to compensate PINK1 deficiency further downstream.

If the STRING analysis is conducted with high stringency only for those cerebellar factors, which are significantly downregulated with >1.5-fold change, then much information is lost and the KEGG pathway of MAPK signaling becomes prominent at the different ages (Additional file 4: Table S3). This marked effect is not unexpected, since PINK1 (short for PTEN induced kinase 1) is downstream of PTEN, a phosphatase that antagonizes the kinase PI3K, which is an upstream modulator of the MAPK phosphorylation cascade at various levels. Furthermore, the interaction of PINK1 with the MAPK pathway has already been demonstrated experimentally in neuroblastoma cells, astrocytes, HeLa, and HepG2 liver cells [51, 64‐66].

Going even further, instead of considering only the expression dysregulations beyond a significance threshold, we employed a tool that takes all genes with their expression ratios into account to compare them with known gene sets of functional relevance. This approach even recognizes compound effects, when most components of a pathway are subtly dysregulated in the same direction. This Gene Set Enrichment Analysis (GSEA) software thus performs a comprehensive assessment of the complete transcriptome data and is useful as an independent complement of the STRING approach. In the cerebellum, the most salient result was a significant effect for the KEGG pathway “Parkinson’s disease”, but this biomathematical finding was mainly due to the loss of Pink1 transcript. Again, at the age of 6 weeks among the KEGG pathways, spliceosome downregulation appeared prominently (Additional file 5: Table S4). Downregulations of “Antigen processing and presentation” as well as "Ubiquitin-mediated proteolysis" were already significant. Among the Reactome pathways at this age, downregulations in the “innate immune system” already had nominal significance. At the age of 6 months, again an impairment of "endoplasmic reticulum stress responses" appeared with a significant downregulation of the Reactome pathway “Activation of chaperone genes by Xbp1s”. At the age of 18 months, among the Reactome pathways, “antiviral mechanism by IFN stimulated genes” became prominent, and in particular the “Negative regulators of RIG I and MDA5 signaling” just touched significance in FDR. For the maximal pathology at the age of 18 months in the cerebellum, the top dozen dysregulated pathways from the KEGG database and the top dozen dysregulated pathways from the Reactome database were documented with individual components and their expression scores (Additional file 6: Table S5). Moreover, four Reactome pathways with special relevance for this manuscript were analyzed in their progressive dysregulation by heat maps (Additional file 7: Table S6). Thus, the temporal order of pathway involvement and the late-stage prominence of neuroinflammation were reproducible in an alternative approach.

Experimental validation was performed for several dysregulated components of these pathways. As shown in Additional file 8: Fig. S2A, a significant downregulation was confirmed for the stress responsive splicing factor Srsf10 [67], while significant upregulations were observed for the cerebellar mRNA levels of Creb3 (human LZIP, synonymous LUMAN), which influences endoplasmic reticulum protein processing [68, 69], and for Nfkbia (NF-Kappa-B inhibitor alpha), which cooperates with Creb3 to modulate the nuclear control of stress and inflammation responses [70‐73], but not for other candidates within the Mapk (MAP kinase) gene family (Additional file 8: Fig. S2A). Of particular interest was the dysregulation of Ube3a within the ubiquitination pathway, because only one splice isoform changed its expression, which might be target of the spliceosome adaptation observed above, and because UBE3A is involved in the degradation of alpha-synuclein as the main driver of Parkinson pathogenesis [59].

We then focused on JNK1 (Mapk8 mRNA) expression, a crucial component of the stress-dependent phosphorylation cascades that control endoplasmic reticulum protein processing, neuroinflammation, and apoptotic or autophagic cell death [74‐77]. JNK1 activity is also regulated by alternative splicing [78, 79]; therefore, three representative exon-exon junctions were assessed. Quantitative reverse transcriptase PCR (qPCR) in Pink1 ^−/− cerebellum at age 18 months in the absence of additional stressors demonstrated a significant but mild upregulation, 1.15-fold ±0.04 SEM with p = 0.01 (Additional file 8: Fig. S2B). This finding is consistent with a previous report of increased p38 MAPK activation in PINK1-deficient mouse astrocytes [51]. However, the expression change of JNK1 was too subtle to be validated by quantitative immunoblots. This technique has inherent variability and limited linearity, so it requires an impractically high number of samples to detect 1.1-fold changes. Thus, a subtle dysregulation of the spliceosome pathway, the endoplasmic reticulum protein processing pathway, and neuroinflammatory response pathways was observable in global microarray transcriptome profiles upon unbiased bioinformatics analyses in old Pink1 ^−/− mouse cerebellum, and the alteration of these pathways could be validated experimentally by qPCR for several key factors.

Overall, PINK1 deficiency triggered a dysregulated expression of many membrane-associated factors already in the first weeks of life. The progression of pathology occurred mainly through involvement of more pathways and protein interaction networks, rather than through stronger expression dysregulation. Pathological mitophagy and neuroinflammation profiles were apparent by the age of 18 months in Pink1 ^−/− cerebellum.

To test if the molecular profile of immune activation is reflected by cellular responses in the brain tissue, immunohistochemistry of GFAP as marker of astroglia and of Iba1 as marker of microglia was performed. Parallel processing of the sections was necessary to visualize their subtly increased immunoreactivity in diverse areas of the Pink1 ^−/− brain, which was prominent at the myelinated fiber tracts, e.g., the corticospinal projections through the striatum, the brainstem, and also in the cerebellar white matter (Fig. 2a, b). This increase was not detectable in nuclei like the striatal matrix or the substantia nigra and was not consistent in the cerebral cortex upon counting reactive cells in sections at even intervals throughout the brain (Fig. 2c). Thus, the dysregulation of membrane-associated factors in old animals leads to a cellular immune response most likely at myelinated axons.

To further assess the relevance of the subtle transcriptional changes at the effector molecule level, we decided to analyze ceramides of different chain lengths and saturation and their sugar-modified metabolites because ceramides trigger lethal mitophagy [80], lactosyl-ceramides maintain neuroinflammation [81], and ceramides accumulate upon lysosomal dysfunctions and LRRK2-deficiency [82] and are expelled by activated astrocytes via exosomes [83]. Hence, ceramide pathology may be a key factor for the progression of Parkinson’s disease.

Ceramides, glucosyl-ceramides, and lactosyl-ceramides were increased in the Pink1 ^−/− olfactory bulb starting from 9 months on (Fig. 2d). Analysis of variance for repeated measurements revealed significant differences between genotypes (F (1:20) = 7297; P = 0.0137) and for the interaction “ceramide by genotype” (F (10:200) = 4.558; P < 0.0001). Subsequent post-hoc analyses for each ceramide and sphingolipid separately were significant for Cer18:0 (P < 0.0001), LacCer24:1 (P = 0.0285), and GluCer18:0 (P = 0.0042). For individual ceramides, there was no site-specific significant progression with increasing age, but analysis of log-2 transformed total ceramides revealed age-dependent differences. In wildtype animals, log-2 Cers linearly increased over time, but in Pink1^−/−, levels were already increased at 9 months and remained at this elevated level, so that genotype-dependent differences were strongest in young mice (2-way ANOVA “genotype x age” for genotype F (1, 54) = 16.98, P = 0.0001; for the interaction F (2, 54) = 2.531, P = 0.089) (Fig. 2e). Hence, the observed increase of ceramide levels is strongly suggestive of ongoing neuro-inflammation in the brain starting in young adulthood, particularly in the olfactory bulb—a brain area which is among the first to be affected by Parkinson’s disease.

Given that PINK1 is important for stress responses [5, 6, 13], and in view of the enhanced release of inflammatory cytokines from acutely prepared Pink1 ^−/− brain slices [84], we decided to study very young Pink1 ^−/− neuron-rich primary culture in vitro—where the previous mechanical dissection, the lack of glial support, and the presence of the toxin cytosine arabinoside (cytarabine) exert combined stress [85, 86]. Hence, murine neuron-rich primary cultures were established at postnatal age and maintained for 12 days to assess early effects of PINK1 deficiency on global transcriptome readouts (Additional file 9: Table S7). Given that genomic insertion events may influence the expression of neighboring genes across a distance of 3 MegaBases [87], it was reassuring that only one dysregulated transcript, Nipal3, was encoded in the vicinity of Pink1.

Two findings had obvious credibility, but limited novelty, namely the dysregulation of Dram1 and Ret mRNA levels. Ordered by significance, the Dram1 (the abbreviation stands for DNA-damage regulated autophagy modulator 1) transcript 1.7-fold upregulation was the strongest early change after Pink1 loss (p value ≈ 0.0001). Similar to PINK1, DRAM1 is connected to the pathway of selective autophagy, since it promotes the targeting of mycobacteria to degradation [88], and is thus contributing to innate immunity responses. DRAM1 mRNA is selectively downregulated in brain tissue of PD patients [89]. A similarly meaningful observation was the 1.7-fold upregulation of Ret (proto-oncogene RET receptor tyrosine kinase) (p value ≈ 0.001). RET trophic receptor signaling was already shown to rescue histological and biochemical phenotypes of Pink1 deletion in flies [15]. In a human neuroblastoma cell line, the administration of its receptor ligand GDNF compensates morphological and bioenergetic deficits of PINK1-deficient cells without affecting mitophagy. Furthermore, GDNF stimulation rescued mitochondrial defects in PARKIN-deficient cells, while the loss of dopaminergic midbrain neurons in aged RET-deficient animals was rescued by PARKIN overexpression and exacerbated by PARKIN deficiency [16]. Intriguingly, GDNF transfected macrophages significantly ameliorated neuroinflammation and neurodegeneration in a PD mouse model [90]. Thus, the Ret mRNA upregulation in Pink1-deficient primary neuron cultures after dissociation stress may also form part of the neuroinflammatory response pathway.

In addition to the known candidates, we observed a remarkable and novel enrichment of additional factors involved in antiviral state, innate immunity, and neuroinflammation (Additional file 9: Table S7) including Gin1 (gypsy retrotransposon integrase 1) [91], Hmox1 (Heme oxygenase 1) [92, 93], Hebp1 (Heme-binding protein 1) [94], Snx16 (Sorting nexin 16) [95], Timp1 (Tissue inhibitor of metalloproteinase 1) [96, 97], Dnmt2 (DNA methyltransferase 2) [98‐101], Rsad2 (Radical S-adenosyl methionine domain containing 2 = Viperin) [102‐106], Adamts4 (a disintegrin-like and metallopeptidase—reprolysin type—with thrombospondin type 1 motif, 4) [107‐109], Csde1 (Cold shock domain containing E1, RNA binding = UNR) [110], and Ifit3 (interferon-induced protein with tetratricopeptide repeats 3) [111, 112].

Further enrichments in the transcriptome profile were apparent for DNA/RNA-quality control and for antioxidant/hypoxia factors. The DNA/RNA quality control factors included Zfp148, Cdk7, Tia1, Mbd1, Mus81, Mat2a, and Hipk2 [113‐127]. The hypoxia/antioxidant factors, which are credible in view of the well-documented oxidative stress and bioenergetics deficit in PINK1-deficient cells [21, 41], included Srxn1, Hlf, Steap1, Tsga10, and Hipk2 [126‐132].

Intriguingly, Foxp1 (Forkhead Box P1) showed a 1.7-fold downregulation (p value ≈ 0.03). This differentiation factor of midbrain dopaminergic neurons was previously found downregulated also in mice with knock-out of alpha-synuclein, while its upregulation was documented in mice with transgenic overexpression of A53T-alpha-synuclein, which model the PARK1 and PARK4 variants of PD [30, 31, 133]. Thus, the dysregulation of Foxp1 transcripts levels appears to respond both to acute stress and to constant mutations of at least two PD genes.

Overall, already at neonatal stages, a significant expression dysregulation of innate immunity factors in Pink1 ^−/− primary neurons became apparent under in vitro stress conditions.

For independent validation of these observations, and to exclude that a microbial infection of the cultures had produced artifacts, we next studied the transcript levels of candidate genes by qPCR in brain homogenate extracts from independent animals. Given that the stress-evoked observations in primary Pink1 ^−/− neurons were not evident in global transcriptome screenings of single-mutant Pink1 ^−/− mouse brains even at old age [20], we reasoned that additional challenges may be necessary to manifest PINK1-dependent stress response dysregulations. Thus, we studied double-mutant mice, where the Pink1 ^−/− is combined with 1.5-fold overexpression of A53T-alpha-synuclein (SNCA) selectively in neurons as a trigger of Parkinsonian pathology and as a genetic interactor of PINK1 [19, 20]. These double-mutant mice show a potentiated phenotype with appearance of Lewy-like inclusion bodies and with lethality from the age of 14 months onward [20]. The global transcriptome in several brain regions throughout their lifespan was previously compared to single mutant Pink1 ^−/− mice, and the data are publically available in a database [20]. It has always been our experience that the heterogeneity of neuron populations in midbrain/brainstem and the dissection variance make it very difficult to demonstrate subtle expression changes.

In midbrain from adult double-mutant Pink1 ^−/−+A53T-SNCA mice at the age of 18 months, a significant 1.16-fold upregulation for Mapk9 (p value = 0.0007), a significant 2.15-fold upregulation for Rsad2 (p value = 0.0007), a trend towards significance (p value = 0.08) for the 0.87-fold downregulation (a dysregulation so subtle that it would usually be deemed irrelevant) of Hebp1, and a significant 1.92-fold upregulation for Tnf alpha were observed (p value = 0.03) (see Additional file 10: Fig. S3).

Thus, also in midbrain, the PINK1 deficiency led to dysregulated expression levels of key factors in the anti-microbial defense, when an additional challenge represented by mutant A53T-alpha-synuclein exceeded the allostatic threshold.

We decided to test in vitro whether (I) similar observations indeed occur in neural cells or have to be attributed to microglia, (II) the mouse data can be translated to human, (III) transient changes can be defined after acute stress, and (IV) trophic and nutrient deprivation stress with concomitant autophagy can trigger these stress response changes, always in dependence on PINK1 deficiency. The starvation stress was initially chosen rather than established Parkinsonian toxins like MPTP or CCCP [6, 134], since PINK1 was recently shown to respond to mutations in ATXN2, which is a known starvation response factor [135‐138], and in view of the known transcriptional induction of PINK1 and PARKIN during trophic deprivation [13]. The human neuroblastoma cell line SH-SY5Y with stable lentiviral knockdown (KD) of PINK1 versus a non-target (NT) control sequence were subjected to acute starvation by changing the culture conditions from RPMI medium with 10% Fetal Calf Serum (FCS) to HBSS medium (low glucose, no amino acids) without FCS. The transcriptional regulation was documented over 2 days for several key factors of the antiviral defense by qPCR. The overview on this systematic survey of the mitochondria-associated antiviral defense pathway is represented in Fig. 3a, with blue color illustrating deficiency and PINK1-dependent downregulation, while red color illustrates PINK1-dependent upregulation. The KD was stable, since PINK1 mRNA in this cell line was reduced to 30% before stress and to 9% after stress at the 16-h time point (Fig. 3b). As described, the starvation induced PINK1 mRNA in a phasic manner with a peak at 12 h [13]. Similar to the previous findings in Pink1 ^−/− mice, an upregulation of RSAD2 and a downregulation of HEBP1 were observed in the starving PINK1-deficient neuroblastoma cells. RSAD2 was induced up to 1.6-fold in control cells, and this response was potentiated by 46% in PINK1-KD cells, with strong significance at 2 and 4 h (Fig. 3c). The up to 3.3-fold induction of HEBP1 levels in control cells was significantly diminished at 16 h in PINK1-KD cells (Fig. 3d).

Systematic further assessment of the innate immunity pathway revealed further regulations, with PINK1-deficiency modulating the alterations caused by starvation. A supersensitive transcriptional upregulation was observed not only for RSAD2 but also for TBK1 and IRF3. TBK1 was induced up to 1.9-fold in control cells, and this induction was further increased to 2.2-fold in PINK1-KD cells, with strong significance at 4, 8, 24, and 48 h (Fig. 3e). Similarly, IRF3 levels were elevated up to 2.2-fold in control cells, and this elevation was exacerbated to 2.4-fold in PINK1-KD cells with strong significance at 24 and 48 h (Fig. 3f).

In contrast, a downregulation was observed for a known posttranslational modification target of PINK1/PARKIN signaling, mitochondrial MFN1. The 2.0-fold starvation-triggered induction of MFN1 transcript was blunted in PINK1-KD cells by 32%, with strong significance at 12 and 16 h (Fig. 3g). Starvation triggered an initial short upregulation and later consistent downregulation of IFIT3 levels to 15% in control cells. In PINK1-KD cells, a strong downregulation of IFIT3 levels appeared already at 2 h and was stably diminished further to 6% of control at 48 h, with significant differences between NT and PINK1-KD cells from 2 to 16 h (Fig. 3h). The starvation-triggered up to 1.7-fold induction of IFIT1 levels was also diminished in PINK1-KD cells by 41% with strong significance at 8, 12, and 16 h (Fig. 3i). Finally, the starvation-triggered almost 3.7-fold induction of the mitochondria-associated innate immunity factor MAVS was blunted in PINK1-KD cells by 40%, with strong significance at 12 and 16 h (Fig. 3j).

No significant dysregulation was documented for the transcript levels of TMEM173 (STING) as an endoplasmic reticulum-associated antiviral factor (data not shown). Given that experimental testing of multiple mRNAs at multiple time-points followed by statistical evaluation via Student’s t tests will exaggerate the significance, an adjustment of the significance threshold according to Bonferroni is useful, by dividing the nominal p values indicated in each panel of Fig. 3 through the number of analyses done. After this correction, the downregulation of PINK1, MFN1, MAVS, IFIT1, and IFIT3 and the upregulation of RSAD2 and TBK1 remain significant as particularly robust findings.

In order to provide a time-correlation and comparison with autophagy factors that are also regulated by starvation, the curves for LRRK2 and SQSTM1 are shown as well (Fig. 3k–m), given that LRRK2 is a positive regulator of inflammation and autophagy, with mutations that trigger autosomal dominant Parkinson’s disease; SQSTM1 (sequestosome 1 = p62) was studied as an adaptor between the autophagy machinery and ubiquitinated cargo, whose mutation cause neurodegenerative disorders such as motor neuron disease. A significant indirect correlation was observed again between the PINK1-deficiency and an exacerbated induction of LRRK2 transcript levels, as previously reported for fibroblasts and neuronal cells derived from PARK6 patients [139].

Overall, the data confirm that key factors of the anti-microbial response are modulated in their transcriptional regulation by PINK1 in human neural cells after acute starvation stress.

To further assess if indeed mitochondrial dysfunction underlies changes in innate immunity, we used the uncoupling drug FCCP to trigger mitophagy and then study the transcriptional response of the above key factors in PINK1-knockdown and control SH-SY5Y neuroblastoma cells. Mitophagy needs at least 12 h to occur and subsequent transcriptional responses require additional time. Therefore, the RNA extraction and qPCR analyses were performed later than in the previous experiment, at 24 h after drug administration. As shown in Additional file 11: Fig. S4 above, FCCP in SH-SY5Y cells elicited a transcriptional induction of the mitochondria associated factors PINK1, DDX58, and MAVS and of TBK1, accompanied by a downregulation of the RNA sensor IFIT1, in control cells. A significant difference between control cells and PINK1 knockdown cells after FCCP treatment was only observed for MAVS, which failed to be induced in the absence of PINK1. These data confirm that key inflammatory factors respond to mitochondrial dysfunction and that particularly the levels of the mitochondrial antiviral signaling factor MAVS depend on PINK1 levels.

Pink1 ^−/− murine embryonal fibroblasts (MEFs) are a useful tool to test if the presence of lentiviral knock-down double-strand RNA in the human neuroblastoma cells distorts the results and to test whether these effects apply only to neuroinflammatory processes, or represent responses of the innate immunity system of any cell also outside the nervous system. Therefore, MEFs were subjected to FCCP treatment and the transcriptional response of the key antiviral factors was studied in the presence or absence of PINK1. As shown in Additional file 11: Fig. S4 below, FCCP treatment of MEF cells after 24 h led to transcriptional upregulation of murine Mavs, accompanied by downregulation of Rsad2, Ddx58, Tbk1, Irf3, Mfn1, Ifit3, and Ifit1, both in control WT cells and Pink1 ^−/− cells. Thus, again the mitochondrial antiviral signaling factor Mavs showed a selective dependence on mitochondrial dysfunction, whereas the levels of all other antiviral key factors decreased in parallel to the FCCP-triggered autophagic elimination of mitochondria. Fibroblasts tolerate a reduction of mitochondrial mass by prominent glycolysis, whereas neuronal cells depend on mitochondria and will compensate mitophagy with mitochondrial biogenesis [37]. Therefore, fibroblasts and neurons may respond to FCCP with opposing regulations of mitochondria-dependent innate immunity factor levels even in control cells.

In view of the strong PINK1-dependent dysregulation of several sensors of pathogenic RNA and of an RNA-virus budding suppressor, namely RSAD2, IFIT3, and IFIT1 in the starvation experiment (Fig. 3c, h, i), we also stressed both cell types (neuroblastoma and MEF cells) with poly(I:C) as a pathogenic RNA-analogue and determined the transcriptional response after 16 h (Fig. 4a). Neuroblastoma cells responded by massive inductions of RSAD2, DDX58, IFIT3, and IFIT1, moderate induction of IRF3, and a downregulation of PINK1. The additional knockdown of PINK1 significantly blunted the upregulations of RSAD2 and DDX58 in SH-SY5Y cells. In MEF cells, again a downregulation of Pink1 was found, now paralleled by moderate downregulation of the other two mitochondrial outer membrane factors, Mfn1 and Mavs. Massive inductions of Rsad2, Ddx58, Ifit3, and Ifit1 reoccurred, now with a moderate increase of Tbk1. The Pink1 deletion significantly blunted the upregulations of Rsad2 and Ifit3 in MEFs.

The substantial upregulations of RNA sensors and their modulation by PINK1 encouraged us to attempt validating these effects also at the protein level, so quantitative immunoblots were performed (Fig. 4b). In neuroblastoma cells, the poly(I:C) triggered inductions were significantly blunted by PINK1 knockdown for RSAD2, DDX58, IFIT3, and IFIT1. In comparison, MEF cells also responded by strong inductions to poly(I:C), but the PINK1-dependent reductions of RSAD2, DDX58, and IFIT3 were not significant. IFIT1 could not be analyzed, since the available antibodies in our hands recognized only the human, but not the murine variant. Overall, these data clearly establish PINK1 as enhancer of anti-microbial responses and particularly of RNA sensors.

Particularly strong and early PINK1-dependent effects were further assessed by qPCR in unstressed primary skin fibroblasts from PARK6 patients versus control individuals, which were previously characterized regarding mitochondrial dysfunction and expression profiles and were found to constitute a useful model of PD [1, 5, 12, 19, 41‐43]. In the fibroblasts of homozygous PARK6 patients, IFIT3 transcript was reduced to 31% (p = 0.0059), while RSAD2 transcript was increased to 307% (p = 0.044) (Fig. 5a). These data in fibroblasts from manifest PARK6 patients at advanced age confirmed the previous findings in starving neuroblastoma cells, verifying that the loss of function of PINK1 leads to a deficit of the mitochondria-associated antiviral RNA sensor IFIT3, in parallel to a supersensitive induction of downstream RSAD2 (viperin) as an inhibitor of many RNA and DNA viruses. In quantitative immunoblots, IFIT3 protein was reduced to 0.27-fold (±0.10, p = 0.03) (Fig. 5b). The RSAD2 protein levels were not significantly changed within this analysis of three patients versus three control samples. These data confirm that the IFIT3 mRNA downregulation is not offset by compensatory molecular efforts, but instead is translated into a deficit of this key antiviral factor at the protein level, thus altering innate immunity responses at the advanced age of manifest PARK6 patients.