Introduction

Only about 2 % of the human genome are DNA sequences that will be translated into protein. The remaining 98 % are comprised of introns, regulatory elements, non-coding RNA, pseudogenes and repetitive elements including transposable elements. However, some sequences in what is generally considered “non-coding genome” do in fact contain sequences which encode proteins. This is true for specific lncRNAs which can encode peptides or functional proteins ¹ but also for a few copies of two transposable element families, Long INterspersed Element-1 (LINE-1) and Human Endogenous RetroViruses (HERV). Non-functional copies of retrotransposons, to which LINE-1 and HERV belong, cover about 44% ² of the human genome as remnants of an evolutionary ancient activity. Depending on the source, about 100 ³ to 146 ⁴ full-length LINE-1 elements with two open reading frames encoding ORF1 and ORF2 are present in the Human reference genome (GRCh38 Genome Assembly) and several incomplete HERV sequences encoding either or any combination of envelope (env), gag, pro or pol ⁵. The LINE-1 encoded protein ORF1p, an RNA binding protein with “cis” preference ^6,7, and ORF2p, an endonuclease and reverse transcriptase ^8,9 are required for the mobility of LINE-1 elements. As many other transposable elements (TEs) including HERVs, LINE-1 elements are repressed by multiple cellular pathways. It was thus generally thought that TEs are repressed in somatic cells with no expression at steady-state ^10–12. However, the aging process reduces the reliability of these repressive mechanisms ¹³. It is now, 31 years after the initial proposition of the “transposon theory of aging” by Driver and McKechnie ¹⁴, generally accepted that TE activation can be both, a cause and a consequence of aging ^15,16.

Sparse data has shown that the LINE-1 encoded protein ORF1p is expressed at steady-state in the mouse ventral midbrain ¹⁷, the mouse hippocampus ¹⁸ and in some regions of the human post-mortem brain ¹⁹ and recent data informed about the presence of full-length transcripts in cancer cells, human epithelial cells and mouse hippocampal neurons ²⁰. Repression of LINE-1 might thus be incomplete and if so, it remains unclear how cells then prevent cell toxicity associated with LINE-1 encoded protein activity. Indeed, LINE-1 encoded proteins have been demonstrated to induce genomic instability (ORF2p endonuclease-mediated ^17,21–26) and inflammation (ORF2p reverse transcriptase-mediated ^27–29) and these cellular activities might be causally related to organismal aging, cancer, autoimmune and neurological diseases ³⁰. For instance, LINE-1 activation can drive neurodegeneration of mouse dopaminergic neurons ¹⁷, of drosophila neurons ^31,32 and of mouse Purkinje neurons ³³ which can be at least partially rescued with nucleoside analogue reverse transcriptase inhibitors (NRTIs) or other anti-LINE-1 strategies. NRTIs are currently being tested in several clinical trials designed to target either the RT of HERVs or the RT encoded by the LINE-1 ORF2 protein. It is not known today, however, to which extend LINE-1 encoded proteins are expressed at steady-state throughout the mouse and human brain, whether there is cell-type specificity and whether activation of LINE-1 encoded proteins is associated with brain aging or human neurodegeneration. Here, using a deep-learning assisted cellular detection methodology applied to pyramidal large-scale images of the mouse brain mapped to the Allen mouse brain atlas combined with post-mortem human brain imaging, co-IP mass spectrometry and transcriptomic analysis of LINE-1 expression, we describe a brain-wide map of ORF1p expression and interacting proteins at steady-state and in the context of aging. We find a heterogenous but widespread expression of ORF1p in the mouse brain with predominant expression in neurons. In aged mice, neuronal ORF1p expression increases brain-wide and in some brain regions to up to 27%. In human dopaminergic neurons, young LINE-1 transcripts and specific full-length and coding LINE-1 copies are increased in aged individuals. We further describe endogenous mouse ORF1p interacting proteins revealing known interactors and unexpected interacting proteins belonging to GO categories related to RNA metabolism, chromatin remodeling, cytoskeleton and the synapse.

Results

Widespread and heterogenous expression of the LINE-1 encoded ORF1p protein in the wildtype mouse brain

To investigate the expression pattern and intensities of endogenous LINE-1 encoded ORF1p protein throughout the entire mouse brain, we devised a deep-learning assisted cellular detection methodology applied to pyramidal large-scale images using a comprehensive workflow complemented by an approach based on confocal imaging as schematized in Figure 1A. Briefly, starting from sagittal slide scanner images of the mouse brain, we defined anatomical brain regions by mapping the Allen Brain Atlas onto the slide scanner images using Aligning Big Brains & Atlases (ABBA). We then employed a deep-learning detection method to identify all cell nuclei (Hoechst) and categorize all detected cells into neuronal cells (NeuN+) or non-neuronal cells (NeuN-) and ORF1p-expressing cells (ORF1p+) or cells that do not express ORF1p (ORF1p-). This workflow allowed us then to characterize the cell identity of ORF1p+ cells and ORF1p intensity throughout the whole brain but also in specific anatomical regions. In parallel, we completed the approach using confocal microscopy on selected anatomical regions allowing for comparison with higher resolution. Importantly, the specificity of the ORF1p antibody, a widely used, commercially available antibody ^18,34–38, was confirmed by blocking the ORF1p antibody with purified mouse ORF1p protein resulting in the complete absence of immunofluorescence staining (Suppl Fig. 1A), by using an in-house antibody against mouse ORF1p¹⁷ which colocalized with the anti-ORF1p antibody used (Suppl Fig. 1B, quantified in Suppl Fig. 1C), and by immunoprecipitation and mass spectrometry used in this study (see below). Unexpectedly, we found a generalized and widespread expression of ORF1p throughout the brain of wildtype mice (Fig. 1B; Swiss OF1 mice, three months-old; whole brain except regions with particularly high cellular density (cerebellum, hippocampus, olfactory bulb) which impedes nuclei detection by deep-learning. ORF1p is detectable in all regions and subregions analyzed with heterogenous expression patterns (density and intensity) per region/subregions. The ten regions shown in Figure 1B exemplify visible different densities of ORF1p+ cells with varying levels of expression. Notably, the expression pattern of ORF1p in the hippocampus is similar to what has recently been published ¹⁸ (Fig. 1B, panel 2). Throughout the entire brain, the mean density of ORF1p+ cells per mm² was ≈ 305 ±18 (mean ±SEM), representing up to 20% of all detected cells (Fig. 1C). ORF1p+ cells in each mouse brain analyzed showed up to eight-fold disparities in intensity between low- and high-expressed cells (Fig. 1C). We then quantified nine anatomical regions according to the Allen Brain Atlas on four brains of three-month old mice (Fig. 1D) using the automated workflow (Fig. 1A) with regard to cell density (Fig. 1D), cell proportions (Fig. 1E) and fluorescent intensity of ORF1p+ cells (Fig. 1F). This approach permitted the analysis of about 10 000 ORF1p+ cells per animal highlighting the power of our large-scale analysis. Densities of ORF1p+ cells ranged from the lowest density in the hindbrain with 154 ±19 cells per mm² (mean ±SEM) to the highest density of ORF1p+ cells in the isocortex with 451 ±44 cells per mm² (mean ±SEM) and the thalamus with 446 ±50 cells per mm² (mean ±SEM). The proportion of ORF1p+ cells per anatomical region fluctuated between 10% ±2.1 (ventral striatum, mean ±SEM) and 31% ±1.6 (thalamus, mean ±SEM). The dorsal striatum (“striatal dorsal” in the Allen Brain Atlas denomination) exhibited the lowest ORF1p expression intensity (658 ±3 mean ±SEM) of all regions tested, the hindbrain the highest mean intensity of ORF1p per cell (mean ±SD 1221 ±548) as illustrated in Figure 1B and quantified in Figure 1D and 1F. Interestingly, cell density did not correlate with expression levels. Dorsal and ventral striatum for instance displayed similar ORF1p intensities per cell but exhibited significant differences in ORF1p cell density and proportion. The “midbrain motor” region as defined by the Allen Brain Atlas showed an intermediate cell density (mean ±SEM 265 ±16 cells per mm²) and a rather high ORF1p expression intensity (mean ±SEM 1006 ±533). Statistical analysis comparing mean density of ORF1p+cells per mm² or mean intensity per ORF1p+ cells among regions confirmed the heterogeneity concerning ORF1p expression throughout the mouse brain (Fig. 1D, 1F). Slide scanner and confocal images revealed an exceptionally high ORF1p expression intensity in the ventral region of the midbrain which we identified as the Substantia nigra pars compacta (SNpc). This region displayed an important density of ORF1p+ cells and a comparatively high level of ORF1p expression as illustrated by confocal imaging (Fig. 1B, panel 8), but could not be quantified independently with our brain-wide approach due to the geometrically-complex anatomy of this region and its small size (subregion-level in the Allen Brain Atlas hierarchy). Another region which could not be included in our brain-wide analysis was the cerebellum due to its extremely high density of cell nuclei. However, slide scanner and confocal imaging (Fig. 1B, panel 10) revealed that ORF1p is expressed in Purkinje cells, while not detectable in the molecular or granular layers.

Fig. 1:

In order to confirm ORF1p expression by an independent method, we performed Western blot analysis on six micro-dissected regions from the mouse brain (Swiss/OF1 mouse, three-month old). As shown in Suppl Fig. 1D, ORF1p is expressed in all six regions with varying expression levels confirming the overall presence of ORF1p throughout the brain. We then chose two regions with significantly divergent ORF1p expression intensities as identified and quantified on pyramidal large-scale images: the frontal cortex (low) and the ventral midbrain (intermediate to high). We confirmed a significant higher expression of ORF1p in the ventral midbrain compared to the frontal cortex using an approach based on the unbiased, automated quantification of multiple z-stacks using a confocal microscope (Fig. 1G) and by Western-blotting on micro-dissected regions (Fig. 1H,I). In concordance with the findings stemming from the large-scale image quantification pipeline (Fig. 1F), the ventral midbrain showed ≈ 2-times higher expression of ORF1p than the frontal cortex as quantified in Figure 1G (1.8-fold) and Figure I (2.3-fold) validating our cellular detection methodology for pyramidal large-scale imaging and underscoring the heterogeneity of ORF1p expression levels in the mouse brain. To investigate intra-individual expression patterns of ORF1p in the post-mortem human brain, we analyzed three brain regions of a neurologically-healthy individual (Fig. 1J) by Western blotting. ORF1p was expressed at different levels in the cingulate gyrus, the frontal cortex and the cerebellum underscoring a widespread expression of human ORF1p across the human brain. In summary, our findings reveal the consistent presence of ORF1p expression throughout the mouse brain in all anatomical regions analyzed with high regional variability in terms of density of ORF1p+ cells per mm2 per region and ORF1p+ cell intensity per region. This finding raises several questions concerning cell-type identity of ORF1p expressing cells and potential functions or consequences of ORF1p expression in the mouse and human brain at steady-state.

ORF1p is predominantly expressed in neurons

Following our observation of a wide-spread expression of endogenous ORF1p throughout the brain, we first addressed the question of the cellular identity of ORF1p+ cells. To this end, we used the neuron-specific marker NeuN, commonly used to identify post-mitotic neurons in the central nervous system ³⁹. This allowed us to determine the proportion of neuronal (NeuN+) or non-neuronal cells (NeuN-) expressing ORF1p (ORF1p+) or not (ORF1p-). Making use of our large-scale imaging approach (Fig. 1A), we observed drastic dissimilarities in detected cellular proportions between the white and grey brain matter. As expected, we observed only 1% of NeuN+ cells in the white matter (corpus callosum; Fig. 2A) validating both, the neuronal marker NeuN as such and the ABBA superposition of the Allen Brain Atlas onto the sagittal brain slices. In the grey matter, our approach detected 30.5% NeuN+ cells (dark red and yellow bars in Fig. 2A) which, according to the literature, should include all post-mitotic neurons with only minor exceptions ^39–42 and corresponds to the reported proportion of neurons present in the mouse brain ⁴³. The nine identified grey matter regions in Fig. 2A display the proportions of the different cell types per region. The proportion of all cells in a given region which are positive for ORF1p (dark red bars) differed between regions (lowest proportion: hindbrain: 7%; highest proportion dorsal striatum: 26.6%). In the isocortex and the midbrain motor-related regions, the majority of neurons detected express ORF1p (54% and 59% by large-scale analysis, Suppl Fig. 2A; 68.7% and 68.8% by confocal imaging (Fig. 2B, quantified in C and Suppl Fig. 2B), respectively), while in the midbrain sensory related regions the proportion dropped to 25% whereas it reached 82% in the thalamus (Suppl Fig. 2A). Altogether, nearly half of all NeuN+ cells throughout the mouse brain expressed ORF1p (mean of all regions: 48.2%; Suppl Fig. 2A). Regarding the cell identity of ORF1p+ cells brain-wide, more than 70% were identified as neuronal by the large-scale approach (Suppl Fig. 2C). This contrasted somewhat with results obtained by the second approach using confocal imaging on multiple z-stacks which indicated that 91.3% (frontal cortex) and 88.5% (ventral midbrain) of ORF1p+ cells were neuronal (Fig. 2D). This difference in percentages of ORF1p+ expressing neurons among all neurons between the large-scale image cell detection methodology and the confocal workflow is most probably due to technical limitations inherent to our large-scale pipeline. Indeed, with the latter approach, region-dependent differences in cell density and signal intensity levels might be the cause for an underestimation of the proportion of ORF1p+ cells being neuronal due to difficulties in cell detection by StarDist/Cellpose (high cell density) on a single focal plan, technical difficulties which are widely reduced by the multiple z-stack based approach when using a confocal microscope. Notably, frontal cortex and ventral midbrain present similar proportion of neurons expressing ORF1p and ORF1p although the percentage of NeuN+ cells between these two regions is significantly different (Suppl Fig. 2D). As we could not rule out that ORF1p might also be expressed in non-neuronal cells, we turned to non-neuronal markers specific for different glial cell populations using two different astrocytic markers (GFAP, Sox9), the astro- and oligodendrocytic marker S100β and the microglial marker Iba1^43,44 and performed co-staining with ORF1p followed by confocal imaging as illustrated in Figure 2E. We screened multiple images of frontal cortex, ventral midbrain, hippocampus and striatum and did not find a single ORF1p+ cell, which could unambiguously be defined as non-neuronal. This indicated that ORF1p is not or only very rarely expressed in non-neuronal cells. To further confirm the predominant presence of expression of ORF1p in neurons and the absence of ORF1p expression in non-neuronal cells, we used fluorescence-activated cell sorting (FACS) to isolate neurons (using a NeuN antibody) and non-neuronal cells (NeuN-) from the adult mouse brain followed by Western blotting with an antibody against ORF1p (Fig. 2F, 2G). We detected ORF1p exclusively in the neuronal population, confirming the results based on two different imaging approaches. Finally, to assess whether predominant, if not exclusive ORF1p expression in neurons is mouse brain specific or a pattern also applicable to the human brain, we investigated the identity of ORF1p expressing cells in the cingulate gyrus of a healthy human brain. Similar to what we found in the mouse brain, we observed sparse NeuN expression in the white (Suppl Fig. 2E) and extensive NeuN staining in the grey matter corresponding to the cortical layers (right, separated by a dashed white line) with ORF1p+ cells predominantly located in the grey matter (images shown in Figure 2H located in the grey matter). All cells stained by ORF1p were co-stained with NeuN indicating that ORF1p was expressed in neuronal cells in the human brain (Fig. 2H). However, due to the lower signal quality inherent to human post-mortem sections compared to mouse sections, the identity of ORF1p+ cells was estimated to be 80% neuronal by the automated image analysis pipeline, although no ORF1p+ / NeuN-cells could be clearly identified (Fig. 2I). Of all neurons identified, 37.2% were ORF1p+ (Fig. 2J), indicating that, similar to the mouse brain, only a fraction of neurons express ORF1p (Fig. 2H, right).

Fig. 2:

In summary, ORF1p expression in the mouse and human brain is widely restricted to neurons of which a proportion express ORF1p. This raises the question of the function and consequences of ORF1p expression specifically in neurons but also on the dynamic regulation of this expression upon exogenous (exposome) or endogenous (aging) challenges.

ORF1p expression is increased in the aged mouse brain

ORF1p is expressed at steady-state throughout the brain, but if this expression is dynamically regulated is not known. Aging has unequivocally been linked to LINE-1 regulation ^16,45 both as a trigger and as a consequence of LINE-1 activation but whether this is true for the brain has not been thoroughly investigated with ORF1p as a read-out. We therefore addressed the question whether advanced age (16-month-old mice) was paralleled by an enhanced presence of ORF1p expression in the brain compared to young, three-month old mice. In the context of aging, we globally observed a reduction in the proportion of ORF1p+/NeuN+ cells using the cell detection workflow applied to large-scale images described in Figure 1A, phenomenon mainly driven by the midbrain motor, the dorsal striatum, the pallidum and the thalamus regions (Fig. 3A, dark red bars, Suppl_Table1). The confocal approach applied to two regions, the frontal cortex and the ventral midbrain (Fig. 3B) confirmed a loss of ORF1p+/NeuN+ cells in the ventral midbrain with no change in cell proportions in the frontal cortex in accordance with the large-scale imaging approach (Fig. 3A). The predominantly neuronal identity of ORF1p+ cells, however, was unchanged (Suppl Fig. 3A) just as the proportion of neurons expressing ORF1p (Suppl Fig. 3B). We observed a significant decrease of NeuN+ cells in the aged ventral midbrain (Fig. 3B, Suppl Fig. 3C). We next analyzed ORF1p expression levels in the context of brain aging. Interestingly, the mean intensity of ORF1p expression increased significantly with age throughout the brain (13% increase brain-wide; Fig. 3C). Frequency distribution analysis unveiled a shift in ORF1p mean expression per cell in aged mice (Fig. 3D). Importantly, the Hoechst mean intensity within nuclei of ORF1p+ cells, serving as an internal control, showed no significant change (Fig. 3E). Among nine analyzed regions, five demonstrated a general increase in ORF1p mean intensity per cell in aged mice (p≤0.05), a change independent from inter-individual variations in both young and aged mice (Fig. 3F). An increase of ORF1p expression was also observed in three others regions albeit not significant. The only exception was the isocortex which remained unchanged with aging. The general increase of ORF1p expression (fold change intensity) in the whole brain, reaching nearly a 30% increase in some regions, is represented on the heatmap in Figure 1G. These results were confirmed by the confocal imaging approach: ORF1p expression in the frontal cortex remained unchanged and the ventral midbrain region increased significantly in aged mice as quantified in Figure 3H and shown with a representative image in Figure 3I. Overall, these results highlight an age-dependent increase in ORF1p expression in neurons throughout the brain with some regions showing an increase of up to 27 % in ORF1p intensities.

Fig. 3:

Coding LINE-1 transcripts are increased in aged human dopaminergic neurons

Following the observation of increased ORF1p expression in the aged mouse brain, among which the ventral midbrain, and given the age-related susceptibility of dopaminergic neurons in the SNpc to cell death and to degeneration in PD ⁴⁶, we turned to a RNA-seq dataset of laser-captured micro-dissected post-mortem human dopaminergic neurons of brain-healthy individuals ⁴⁷, in order to interrogate full-length LINE-1 mRNA expression profiles as a function of age. To avoid read-length bias to which TE analysis is particularly sensitive, we analyzed only the data derived from 50bp paired-end reads of linearly amplified total RNA as this dataset represented all age categories (n=41; with ages ranging from 38 to 97; mean age: 79.88 (SD ±12.07); n=6 ≤65y; n=35 >65y; mean PMI: 7.07 (SD ±7.84), mean RIN: 7.09 (±0.94)). As age-related dysregulation of TEs might not be linear, we considered individuals with ages-at-death younger or equal to 65 years as “young” (n=6, 38-65 years, mean age 57.5 years (SD ±9.9)) and individuals older than 65 years as “aged” (n=35, 65-97 years, mean age 83 years (SD±7.8). The expression of the dopaminergic markers tyrosine hydroxylase (TH) and LMX1B were similar in both populations indicating no apparent change of dopaminergic identity of analyzed melanin-positive dopaminergic neurons (Suppl Fig. 4A). Next, we compared the expression of repeat elements at the class, family and name level based on the repeat masker annotation implemented in the UCSC genome browser using a commonly used mapping strategy for repeats ⁴⁸. No overt dysregulation of repeat elements at either level of repeat element hierarchy was observed (Suppl Fig. 4C-F), however there was a significant increase in several younger LINE-1 elements including L1HS and L1PA2 at the “name” level (Fig. 4A,B). This was not observed for HERVK-int, a human endogenous retrovirus family with some copies having retained coding potential (Fig. 4B) or other potentially active TEs like HERVH-int, HERV-Fc1 or SVA-F with the exception of a trend for an increase in AluYa5 transcripts in the >65y group, a young Alu family mobilized by the LINE-1 retrotransposition machinery (Suppl Fig. 4G). Interestingly, L1HS expression was highly correlated with L1PA2 expression and this correlation extended to almost all younger LINE-1 subfamilies weaning down with evolutionary distance (Fig. 4C). This was not true for other active TEs as L1HS was negatively correlated with HERVK-int expression (Fig. 4C). Several regulators of LINE-1 activity have been identified ^17,49 and correlation of their expression with L1HS might allow to infer their relevance of interaction (activation or repression) with L1HS in human dopaminergic neurons. Spearman correlation analysis revealed three known repressors of LINE-1 activity; EN1 (Engrailed 1 ¹⁷, Suppl Fig. 5A) with important functions for dopaminergic neuron homeostasis ⁵⁰, CBX5/HP1a, a heterochromatin binding protein binding to the histone mark H3K9me3, thereby mediating epigenetic repression ⁵¹ (Suppl Fig. 5B) and XRCC5/6, also known as Ku86/Ku70, which are essential for DNA double-stranded break repair through the nonhomologous end joining (NHEJ) pathway and limit LINE-1 full-length insertions ⁵² (Suppl Fig. 5C). The transcripts of these genes showed, although not statistically significant, a trend for decreased expression in the elderly (Suppl Fig. 5D-G). Based on the increase of young LINE-1 families L1HS and L1PA2 in aged human dopaminergic neurons and the finding that ORF1p was increased in the aged mouse brain, we focused our attention on LINE-1 elements with coding potential for ORF1 and ORF2 according to the L1Basev2 annotation which are specific elements comprised in the L1HS and L1PA2 annotation at the “name” level. Most of the 146 full-length and coding LINE-1 termed UIDs (= Unique Identifier) in the L1Base are L1HS elements (76.03%), whereas the remaining 35 UIDs belong to the evolutionary older L1PA2 family (Suppl Fig. 6B). The L1Base annotation is based on the human reference genome (GRCh38) and annotates 146 human full-length (>6kB), intact LINE-1 elements (ORF1 and ORF2 intact) with a unique identifier from 1 to 146 ⁴. Attribution of sequencing reads to a specific, individual TE copy is problematic ⁵³ and several approaches have been proposed to circumvent this problem including the mapping of unique reads ⁴⁸.While several tools using expectation maximization algorithms in assigning multi-mapping reads have been developed and successfully tested in simulations ^48,54, we used a different approach in mapping unique reads to the L1Base annotation of full-length LINE-1. Specific “hot” LINE-1 loci in a given cellular context have been identified ³, but usage of the L1Base annotation enabled an unbiased approach albeit ignoring polymorphic LINE-1 sequences. Unique read mapping strategies for repeat elements, especially young LINE-1 elements, will unavoidably underestimate LINE-1 locus-specific expression levels ⁴⁸, but will be most accurate in assigning reads while allowing the comparison of two different conditions analyzed in parallel. Assuming that expression of UIDs was correlated with mappability, we plotted a mappability count of each UID (see methods) against its mean normalized read count expression of the six individuals ≤65y. Non-parametric Spearman correlation revealed no correlation between UID mappability and expression (Suppl Fig. 6A) indicating no apparent bias between the two parameters. However, individual UID dependency of mappability on expression cannot be excluded, especially for high expressing UIDs like UID-16 for example (Suppl Fig. 6A,E). Expression of LINE-1 at the locus-level has been attributed to artefacts not representing autonomous transcription including differential high intronic read counts ⁵⁵, pervasive transcription or reads attributable to passive co-transcription with genes when the LINE-1 element is intronic ⁵⁶. To evaluate the latter, we determined the number of intronic (46.58%) and intergenic UIDs (78/146; Suppl Fig. 6C) and identified the corresponding genes for intronic UIDs (Suppl Fig. 6D). Of the 146 UIDs, 140 passed the threshold of >3 reads in at least 6 individuals. Differential expression of UID between “young” and “aged” dopaminergic neurons revealed several significantly deregulated full-length LINE-1 loci (Fig. 5A). However, while no single locus stood out, paired analysis of the expression of all UIDs indicated a general increase (Fig. 5B, especially of low expressed UIDs. The comparative analysis of the sum expression of UIDs per individual comparing young (≤65y) with elderly human dopaminergic neurons, however, did not reach statistical significance (Fig. 5C). Several specific loci were dysregulated in particular, for instance UID-68 (Fig. 5A), a L1HS element located on chromosome 7 (chr7: 141920659-141926712) in between two genes, CLEC5A (C-type lectin domain containing 5A) and OR9A4 (olfactory receptor family 9 subfamily A member 4). To rule out any influence of “hosting” gene transcription interference on measurable UID-68 expression differences (Fig. 5D, left), we performed Spearman correlation which did not indicate any correlation between CLEC5A or OR9A4 expression with UID-68 (Fig. 5D). Further, UID-68 had a high mappability count of 16 (range of all UIDs: 1-30, mean 9.0 (SD ±6.05), Suppl Fig. 6A) indicating that UID-68 might be a candidate for an age-dependent gain of activity. L1HS UID-129, located on chromosome 15 (chr15: 54926081-54932099) is intergenic, at ≈ 2Mb distance upstream and ≈ 200Mb downstream from the next genes. However, the mappability count of UID-129 was only 1, indicating possible mapping biases inherent to the unique read-based mapping strategy employed (Fig. 5E and Suppl Fig. 6A). Another UID increased in individuals >65y, the L1HS UID-37 (chr10:98,782,942-98,788,971, minus strand), located in intron 3 of the HPSE2 gene (mappability count of 13, minus strand), showed no correlation with its “hosting” gene indicating potential autonomous transcription of this LINE-1 element and suggesting its contribution to the increase of full-length LINE-1 transcripts (Suppl Fig. 7A). We also inspected UID-127 (chr13:40,356,291-40,362,321, mappability count: 14), a L1PA2 element which slightly decreased in elder individuals (Fig. 5A). We found a positive association with LINC00598 (intron 6 of 6; ≤65y: r=0.2, p=0.71; >65y: r=0.37, p=0.03), a non-coding RNA which hosts UID-127 in its 6^th intron, indicating transcriptional co-regulation potentially indicative of non-autonomous transcription of UID-127 (Suppl Fig. 7B). The decrease in the expression of UID-137 (Fig. 5A) was mostly driven by one young individual with high expression (Suppl Fig. 7C) and thus not reflecting an overall decrease (Suppl Fig. 7C, p=0.23). In conclusion, TE expression analysis of this human dataset covering an age-span of 59 years indicates an increase in the expression of young LINE-1 elements including those which have coding potential in elderly dopaminergic neurons. A slight net sum increase of UID transcripts/cell might be sufficient for the production of “above steady-state” levels of ORF1p and ORF2p. Other TEs with coding potential, namely members of the HERV family, were not increased. Further, correlation analyses suggest that L1HS expression might possibly be controlled by the homeoprotein EN1, a protein specifically expressed in dopaminergic neurons in the ventral midbrain ⁵⁰, the heterochromatin binding protein HP1, two known regulators of LINE-1, and the DNA repair proteins XRCC5/6.

Fig. 4:

Fig. 5:

Endogenous ORF1p interactors in the mouse brain

To go further in our understanding of steady-state neuronal ORF1p expression across the mouse brain, we immunoprecipitated ORF1p and performed quantitative label-free LC-MS/MS to identify potential protein partners of ORF1p in the mouse brain. We successfully immunoprecipitated endogenous ORF1p from whole brain lysates (Fig. 6A) and identified a total of 424 potential protein interactors associated with ORF1p (Suppl_Table2) in 5 independent experiments (n=5 mice). Using Gene Ontology (GO) analysis, we identified several interacting proteins belonging to GO terms related to known functions of the ORF1p protein in RNA binding, preferentially ⁵⁷ but not exclusively in cis ⁵⁸, for instance RNA decapping and mRNA catabolic process, or related to the known presence of ORF1p in ribonucleoprotein particles ^59,60 (GO: cytoplasmic ribonucleotide granule) or the presence of ORF1p in p-bodies ⁵⁸ as shown in Figure 6B and listed in Suppl_Table3. Other GO terms that emerged, to our knowledge not previously associated with ORF1p, were related to cGMP-mediated signaling (GO: cGMP-mediated signaling and 3’-5’phosphodiesterase activity: i.e. PDE4A, PDE4B, PDE4DIP) and the cytoskeleton (GO: microtubule depolymerization, cytoskeleton organization, microtubule and tubulin binding, cytoskeletal motor activity and protein binding). cGMP signaling is regulated by 3’-5’ phosphodiesterases (PDEs) which degrade 3’,5’-cyclic guanosine monophosphate (cGMP) and 3’,5’-cyclic adenosine monophosphate (cAMP), an activity essential for cell physiology for the integration of extra- and intracellular signals including neuronal excitability, synaptic transmission and neuroplasticity ^61,62. Further, several ORF1p interacting proteins were constituents of the mating-type switching/sucrose nonfermenting complex (SWI/SNF complex), i.e. ARID1A, ARID1B, SMARCA2, SMARCB1, SMARCC2), an ATP-dependent chromatin remodeler complex disrupting nucleosome/DNA contacts to facilitate DNA/chromatin accessibility by shifting, removing or exchanging nucleosomes along DNA ^63,64. Finally, we also observed proteins belonging to the GO term “neuronal cell body”, corroborating with the neuron-specific presence of ORF1p in the brain. A comparative analysis with previous mass spectrometry studies ^60,65–70 aimed at identifying ORF1p interacting proteins unveiled significantly more common proteins than randomly expected (overrepresentation test; representation factor 2.6, p< 5.4e-08; Fig. 6C), including LARP1, STAU2, ATXN2, RALY, TARBP2 or DDX21 (for a full list see Suppl_Table4). The presence of a significant number of overlapping ORF1p interactors in different non-neuronal human cells (HEK^60,65,66, HeLa⁶⁷, human breast and ovarian tumors⁷⁰ and hESCs⁶⁸) and mouse brain cells (our study), suggest conserved key interactors between both species and between cell types, with a subset of these proteins regulating RNA degradation and translation potentially relevant for the LINE-1 lifecycle itself. ORF1p interactors found in mouse spermatocytes ⁶⁹ were also present in our analysis including CNOT10, CNOT11, PRKRA and FXR2 among others (Suppl_Table4). To unravel the physical interactions between the identified interactors of endogenous ORF1p within the mouse brain, we used the STRING database (Search Tool for Recurring Instances of Neighboring Genes, https://string-db.org/). This analysis generated a network representation, where physical interactions are represented by edges (Fig. 6D). In analogy with the GO term analysis, ORF1p displayed interactions with various clusters, including well-known RNA decapping complexes directed against LINE-1 RNA, which also encompassed DCP2 and DCP1A which had not previously been identified as interacting with ORF1p ⁷¹. Furthermore, ORF1p exhibited interactions with the SWI/SNF complex (highlighted in red) as well as subunits of the RNA polymerase II complex suggesting a direct or indirect association with accessible chromatin, a hitherto unknown interaction of ORF1p with chromatin compartments within the nucleus. Notably, a multitude of novel interactors belonged to the “neuronal cell body” and “neuron projections” clusters, proposing potential neuron-specific partners of ORF1p such as Grm2/5, Bai1, Epha4, Kcnn2, Grik2 and Dmd among others. A last cluster, formed by Ncoa5 (Nuclear Receptor Coactivator 5), Nxf1 (Nuclear RNA Export Factor 1), Ranbp2 and Nup133 (both nucleoporins), might imply a role for these interactions in L1-RNA nuclear export and/or a mechanism for the LINE-1 RNP to gain access to the nucleus in post-mitotic neurons. Altogether, the identification of known and novel interactors of ORF1p in the mouse brain suggests roles of ORF1p in the LINE-1 life cycle (RNA binding and metabolism, RNP formation, nuclear access) but also suggests potential novel physiological roles of ORF1p in the brain related to cytoskeleton organization, cGMP signaling, neuron-specific functions (i.e. synaptic signaling, Suppl_Table3) and chromatin organization and/or transcription regulation.

Fig. 6:

Discussion

While LINE-1 derepression in aging has been extensively explored in peripheral tissues and various pathologies, including cancer, our understanding of LINE-1, particularly ORF1p expression, in the central nervous system remains limited ^20,72,73. A recent search of ORF1p peptides in mass spectrometry data spanning 29 different healthy tissues did not reveal the presence of ORF1p in the brain, suggesting that its presence might lie below detection limits ²⁰. Only a few studies explored ORF1p encoded LINE-1 protein expression in the brain, most focusing on a specific region (in mice ^17,33 in rats ⁷⁴ and in human post-mortem brain ¹⁹), but it remained unclear if ORF1p is expressed throughout the entire brain, exhibits cell-type specificity, and most intriguingly, if its expression is influenced by the aging process. Here, we demonstrate that ORF1p is expressed throughout the entire mouse brain and in at least three regions of the human post-mortem brain at steady-state. Leveraging a comprehensive workflow that incorporates brain atlas registration and machine learning algorithms (as described in the methods section), we quantified tens of thousands of brain cells, enabling a profound analysis of cell proportions, cell identities, densities and ORF1p expression levels across the entire brain. Surprisingly, more than one-fifth of detected cells expressed ORF1p. Strikingly, regional variations in ORF1p expression levels were observed, with each region exhibiting distinct proportions, cell density, and signal intensity of ORF1p+ cells. In a non-neurologically diseased human brain, ORF1p is expressed in all three regions examined, that is the cingulate gyrus, the frontal cortex and the cerebellum. This is in accordance to an earlier study using histological staining, which found ORF1p expression in the human frontal cortex, the hippocampus, in basal ganglia, thalamus, midbrain and the spinal cord ¹⁹. This suggests, similarly to the mouse brain, a generalized expression across the human brain. On the transcriptomic level using long-read sequencing of GTEx tissues, brain and liver were highlighted as the organs displaying the highest expression of putatively active, full-length LINE-1 elements ⁷⁵. However, when the authors looked at sub-regions, they found transcript expression in cerebellar hemispheres and the putamen, but not in the caudate and the anterior cingulate gyrus and frontal cortex ⁷⁵. This is in contrast to our data and the data from Sur et al, where ORF1p was found to be expressed in the latter two regions using two different antibodies. We used the anti-human LINE-1 ORF1p antibody clone 4H1, a well characterized antibody ^73,76. While the sample size for the staining of human post-mortem tissues certainly needs to be increased in order to draw quantitative conclusions, the presence of the protein in two independent studies does point to a steady-state expression of ORF1p in the human brain.

In the mouse brain, we find ORF1p to be expressed predominantly if not exclusively in neurons using immunofluorescence and fluorescence-activated cell sorting (FACS). This result is consistent with previous studies, such as the identification of ORF1p in excitatory neurons within the mouse frontal cortex ⁷⁷, in parvalbumin neurons in the hippocampus ¹⁸, its presence in neurons in the ventral midbrain ¹⁷ and the recognition of morphological similarities between stained neurons and ORF1p+ cells in a post-mortem hippocampus sample of a healthy individual ¹⁹. We also detected ORF1p in Purkinje cells in the mouse and in bulk human cerebellum. Neuronal specificity or preference of LINE-1 expression was also shown on the transcriptomic level in recent studies investigating LINE-1 expression in the mouse hippocampus, where neuronal LINE-1 expression exceeded that of astrocytes and microglia by approximately twofold ²⁰, is abundant in parvalbumin interneurons ¹⁸ and single-nuclei RNA-seq data from the mouse hippocampus and frontal cortex which confirmed globally that repetitive elements including LINE-1 are more active in neurons than in glial cells ⁷⁷. In the human brain, LINE-1 transcripts are found in greater quantities in neurons compared to non-neuronal cells by single-nucleus sequencing ⁷⁸. Furthermore, retrotransposition-competent LINE-1 elements (similar to UID) are found expressed exclusively in neurons ⁷⁹. While ORF1p expression is suggested to be expressed in microglia under experimental autoimmune encephalomyelitis conditions in the spinal cord ⁸⁰, no evidence of such expression was observed in non-neuronal cells under non-pathological condition.

On average, throughout the mouse brain, the majority of neurons was positive for ORF1p and in some regions (i.e. the thalamus) around 80% of neurons expressed ORF1p. Comparing the results of both imaging approaches, the percentages of neurons expressing ORF1p in the ventral midbrain and frontal cortex were roughly similar (around 70% of neurons expressed ORF1p as quantified by confocal imaging and about 60% of neurons were identified as ORF1p+ using the slide scanner approach). In the human cingulate gyrus, we found that 37.2% of neurons express ORF1p and that 80% of cells expressing ORF1p were neurons, which are proportions similar to some regions of the mouse brain. It is however possible that these percentages are underestimated due to technical issues inherent to the machine-learning based algorithm for cell detection as our observations often indicated a positive signal in neurons which were classified as negative due to a particular shape or our stringent intensity threshold. A question which arises based on these findings is whether specific features distinguish ORF1p+ and ORF1p-neurons. One hint comes from a recent study suggesting that in the mouse hippocampus, it is the parvalbumin positive neurons that predominantly express ORF1p ¹⁸. We have made a similar observation in the mouse ventral midbrain, where TH-positive dopaminergic neurons express higher levels of ORF1p compared to surrounding, non-dopaminergic neurons ¹⁷ (Fig. 1B, panel 8). In the cerebellum, we observed ORF1p staining in Purkinje cells but not in the surrounding granular and molecular layer neurons (Fig. 1B, panel 10). Parvalbumin positive neurons are inhibitory neurons, so are Purkinje cells. However, dopaminergic neurons are modulatory neurons exerting excitatory and inhibitory effects depending on the brain region they act on. Specific neurons in the granular layer (i.e. Golgi and unipolar brush cells) of the cerebellum are inhibitory, but ORF1p negative, indicating that the decisive feature might not be the excitatory or inhibitory nature of a neuron. Another possibility is a cell-type specific chromatin organization permissive for the expression of LINE-1 and future single-cell studies in the mouse and human brain might reveal those differences.

Because transposable elements are known to become active in somatic tissues during aging ^{15,16,28,81,82}, we aimed to investigate whether there was a corresponding increase at the protein level. In aged mice, ORF1p expression significantly increased throughout the mouse brain consistent with a previously documented increase in ORF1p outside the central nervous system in aged rats ^74,83 and aged mice ⁸¹. By quantifying the mean intensity of ORF1p in over 70 000 cells identified as ORF1p+, we were able to characterize the extent of this increase in each anatomical sub-region. Remarkably, apart from the isocortex which did not show any change, ORF1p expression increased in all other brain regions by 7% to 27%, indicating a generalized increase of ORF1p expression in neurons throughout the brain (13%). We did not detect any change in cell identity of ORF1p expressing cells, that is, ORF1p expression remained predominantly if not exclusively neuronal. Intriguingly, we observed a loss of NeuN+ cells, particularly those expressing ORF1p, throughout the brain which was more pronounced in specific regions (ventral midbrain) than in others (isocortex). The loss of NeuN+ cells could either be due to a loss of neuronal identity, as described recently in the context of neuronal aging ⁸⁴ and in the context of Alzheimer disease and related tauopathies ⁸⁵, to a neurodegenerative process per se or to gliosis related to the aging process ⁸⁶ (as we observe a slight increase of ORF1p-/NeuN-cells per mm²).

Interestingly, a region with a strong increase in ORF1p expression with aging (ventral midbrain) also had a significant loss of NeuN+ cells while a region with no change in ORF1p expression with aging (isocortex, frontal cortex) did not lose NeuN+ cells. However, further investigations are necessary to validate a correlation and to investigate an underlying mechanism. An increase of ORF1p might have several direct or indirect consequences on a cell or here, on a neuron. As ORF1p is translated from a polycistronic LINE-1 RNA together with ORF2p, albeit in much higher amounts (the estimated ratio ORF1p to ORF2p is 240:1) ⁸⁷, it can be expected that a LINE-1 ribonucleotide particles are formed and ORF2-dependent cell toxicity in form of genomic instability ^17,21 and single-stranded cytoplasmic DNA triggered inflammation ^27,28,81 might result. This has been shown in mouse dopaminergic neurons where oxidative stress induced LINE-1 causally contributed to neurodegeneration ¹⁷.

Neurodegeneration was partially prevented by anti-LINE-1 strategies among which NRTIs ¹⁷ and similar LINE-1 protein-dependent neuronal toxicity has been shown in drosophila ^31,32 and the mouse cerebellum ³³.

In order to test whether an increase in LINE-1 is a feature of human brain aging, we turned to a unique RNA-seq dataset of human laser-captured dopaminergic neurons of 41 individuals ranging from 38 to 99 years ⁴⁷. In accordance with our focus on LINE-1 sequences which are full-length and coding, we developed a rationale to interrogate LINE-1 families with representatives that are coding (L1HS, L1PA2, multimappers; RepeatMasker) and to specifically investigate full-length LINE-1 elements that have intact open reading frames for ORF1p and ORF2p (unique reads; L1Basev2 ⁴). Indeed, we find an increase in L1HS and L1PA2 elements in individuals ≥65y as well as an increase in specific full-length LINE-1 elements but only a trend for increase of all full-length LINE-1 in sum in the elderly. This analysis has technical limitations inherent to transcriptomic analysis of repeat elements especially as it is based on short-read sequences. Nevertheless, we tried to rule out several biases by demonstrating that mappability did not correlate with expression and that the expression of intronic full-length LINE-1 elements is not correlated with the expression of their “hosting” gene. Interestingly, dysregulated full-length LINE-1 elements in aged dopaminergic neurons did not correspond to those identified in bladder cancer ⁸⁸ indicating the intricate nature of this expression across tissues and pathological conditions. Overall, a slight net sum increase of UID transcripts/cell might be sufficient for the production of “above steady-state” levels ORF1p and ORF2p. Further, a dissociation of LINE-1 transcript and protein levels in aging has been observed recently in excitatory neurons of the mouse cortex. In the absence of transcriptional changes of LINE-1, protein levels of ORF1p were increased ⁷⁷.

We can only speculate about the reason for an increase in ORF1p in the aged brain. A recent single-cell epigenome analysis of the mouse brain suggested a specific decay of heterochromatin in excitatory neurons of the mouse brain with age which was paralleled by an increase in ORF1p, albeit equally in excitatory and inhibitory neurons, again not indicating any dependency of ORF1p regulation on the excitatory or inhibitory nature of neurons ⁷⁷. Chromatin and particularly heterochromatin disorganization are a primary hallmark of aging ⁸² but other repressive cellular pathways which control the LINE-1 life cycle might also fail with aging ¹³. Another possibility is a loss of accessibility of repressive factors to the LINE-1 promoter or an age-dependent decrease in their expression. Matrix correlation analysis of several known LINE-1 regulators, both positive and negative, revealed possible regulators of young LINE-1 sequences in human dopaminergic neurons. Despite known and most probable cell-type unspecific repressive factors like the heterochromatin binding protein CBX5/HP1 ⁵¹ or the DNA repair proteins XRCC5 and XRCC6 ⁴⁹, we identified the homeoprotein EN1 as negatively correlated with young LINE-1 elements including L1HS and L1PA2. EN1 is an essential protein for mouse dopaminergic neuronal survival ⁵⁰ and binds, in its properties as a transcription factor, to the promoter of LINE-1 ¹⁷. As EN1 is specifically expressed in dopaminergic neurons in the ventral midbrain, our findings suggests that EN1 controls LINE-1 expression in human dopaminergic neurons as well and serves as an example for a neuronal sub-type specific regulation of LINE-1.

The heterogenous, brain-wide presence of ORF1p expression at steady-state is intriguing. In cancer cell lines or mouse spermatocytes, ORF1p interacts with several “host” proteins, some if not most of which are related to the LINE-1 life cycle. However, a profile of endogenous ORF1p interactors in the mouse brain might inform on possible other and organ-specific functions besides its binding to the LINE-1 RNA in “cis” ⁶⁹. Among the total 424 potential interactors of endogenous ORF1p in the mouse brain, 38 partners had been previously identified by mass spectrometry in human cancers, cancerous cell lines and mouse spermatocytes ^60,65–70 (Suppl_Table4). Further, GO term analysis contained expected categories like “P-body”, mRNA metabolism related categories and “ribonucleoprotein granule”. We also identified NXF1 as a protein partner of ORF1p, a protein found to interact with LINE-1 RNA related to its nuclear export ⁸⁹. This suggests the conservation of key interactors probably essential for completing or repressing the LINE-1 life cycle in both species, despite the divergence of mouse and human ORF1p protein sequences ⁹⁰. Along these lines, several ORF1p protein partners we identified might complete the list of post-transcriptional regulators implicated in LINE-1 silencing. Recent work conducted on human cancerous cell lines has demonstrated that MOV10 orchestrates the recruitment of DCP2 for LINE-1 RNA decapping ⁷¹. In our analysis, we identified DCP2 along with DCP1A, known to enhance the decapping activity of DCP2 ⁹¹, and DCP1b, a pivotal component of the mRNA decapping complex ⁹². Intriguingly, MOV10 was not detected in our mass spectrometry analysis, despite its established role in recruiting DCP2 and forming a complex with L1-RNP to mediate LINE-1 RNA decapping, as reported by Liu et al ⁷¹. However, we found two enhancers of mRNA decapping, EDC3 and EDC4, both core components of P-bodies, a membrane-less organelle known to contain L1-RNP ⁵⁸. Multiple ubiquitin-ligase proteins were found although not appearing as a significantly enriched GO term. These results complete the picture of the post-transcriptional and translational control of ORF1p and suggest that these mechanisms, despite a steady-state expression, are operational in neurons. Further, several neuron-specific interactors were identified belonging to GO term categories “neuron projection” (75 proteins) and “neuronal cell body” (5 proteins), again pointing to the neuron-predominant expression of ORF1p in the mouse brain. Other interesting aspects were raised from this analysis. Among significantly enriched GO terms, several were related to the cytoskeleton, the functional consequences of which need to be determined in future studies. Our screen also identified PDE10A as an interactor of ORF1p in the mouse brain, a PDE almost exclusively expressed in medium spiny neurons of the striatum and a target for treatment of neurological diseases related to basal ganglia function like Huntington’s disease, schizophrenia and Tourette syndrome ⁹³. Interestingly, PDE10A inhibition is related to beta-catenin signaling, another GO term which emerged from our screen ⁹⁴. Finally, we found components of RNA polymerase II and the SWI/SNF complex as partners of ORF1p. This further indicates that ORF1p has access to the nucleus in mouse brain neurons as described for other cells ^95,96, implying that ORF1p potentially has access to chromatin. These findings give rise to intriguing questions regarding the potential function of ORF1p in neuron in health and disease as (i) ORF1p is widely distributed throughout the brain under normal physiological conditions, (ii) ORF1p shows a wide range expression levels within and inbetween regions, (iii) ORF1p is expressed predominantly if not exclusively in neurons, (iv) but not in all neurons and (v) interacts with proteins that might not directly relate to the LINE-1 life cycle, some of which are neuron-specific. In addition, physicochemical properties of ORF1p to form compacted nucleic acid - bound complexes with sequestration potential were shown ^90,97. Future loss-of-function studies should help to shed light on the necessity of ORF1p for neuronal functions if they exist. Our data spurs the idea of a possible “physiological” function of ORF1p as an integrative protein with exapted function in neuronal homeostasis and a loss of restriction in the aged brain limiting LINE-1 expression to steady-state levels.

Materials & Methods

Animals

Swiss OF1 wild-type mice (Janvier) were housed on a 12h light/dark cycle with free access to water and food. Mice were sacrificed at 3-month or 16-month. Animal experiments were performed according to the EU directive 2010/63/EU.

Mouse tissue dissection and protein extraction

Tissues were extracted from 3-month-old and 16-month-old Swiss/OF1 mice. Briefly, the two hemispheres were separated in ice cold PBS -/-. For each mouse, one hemisphere was rinsed and fixed in 4% PFA for 1h followed by 24h of incubation in 30% sucrose. Hemispheres were kept at −20°C until being sliced on a freezing microtome (Epredia, HM 450) with a 20µm thickness. The other hemisphere was dissected in ice cold PBS -/- 1X and 6 brain regions were rinsed, cut in small pieces and dissociated separately using a large (21G) to small gauge (27G) needle in RIPA lysis buffer for 5 min. Lysates were kept on ice for 25 min, were sonicated for 15 min and supernatants were collected after a 30 min centrifugation at 4°C at 14 000 rpm.

Proteins were quantified and Laemmli buffer was added before boiling for 10min at 95°C to be used for Western Blot.

Human Samples

Cerebellum, frontal cortex and cingulate gyrus human samples were provided by Neuro-CEB biobank from a 78-year-old healthy male individual and conserved at −80°C.

Human Samples pulverization and protein extraction

We used the dry pulverizer Cryoprep (Covaris) for pulverization of tissue blocs. Each sample was disposed in a liquid-nitrogen precooled Tissue-tube bag and dry cryo-pulverized with one impact at the maximum level. The pulverized brain sample was then weighed and resuspended in lysis buffer (mg/v) (0.32M sucrose, 5mM CaCl₂, 3mM Mg(CH3COOH)₂, 0.1mM EDTA, 10mM Tris-HCL pH8, 1mM DTT, 0.1% TritonX-100 and Protease Inhibitors), kept on ice for 30 min with gentle up-and-down pipetting until homogenization. We added 2X RIPA buffer (v/v) to totals fractions for 30 min on ice. We then sonicated samples 2 times for 15 min. AtlasSupernatants were collected after a 30 min centrifugation at 14 000 rpm at 4°C, proteins were quantified and Laemmli buffer was added to be used for Western Blot. All samples were boiled 10 min at 95°C to be used for Western Blot.

Western-Blot

We used 1.5mm NuPAGE 4-12% Bis-Tris-Gel (Invitrogen™). Proteins samples were loaded and gel migration was performed with NuPAGE™ MES SDS Running Buffer (Invitrogen™) for 45 minutes at 200mV. Gels were transferred onto a methanol activated PVDF membrane (Immobilon) in a buffer containing: Tris 25 mM, pH=8,3 and Glycine 192 mM, during 1h30 at 400 mA. Membranes were blocked 30 min with 5% milk in TBST (0,2% Tween 20, 150 mM NaCl, 10 mM Tris pH:8). Primary antibodies were diluted in 5% milk in TBST, and membranes were incubated o/n at 4 C°. After 3 x 10 min washing in TBST, membranes were incubated for 1h30 with the respective secondary antibodies diluted at a concentration of 1/2000 in 5% milk TBST. Membranes were washed 3 x 10 min in TBST and were revealed by the LAS-4000 Fujifilm system using Clarity Western ECL Substrate (Bio Rad) or Maxi Clarity Western ECL Substrate (Bio Rad).

Immunostaining

Sagittal mouse brains slices were fixed for 10min in PFA 4% and rinsed 3 times for 10min in PBS -/-. Slices were then incubated 20 min in glycine 100mM, washed 3 times for 5min in PBS and immersed in 10mM citrate pH 6 at 62°C during 45min for antigen retrieval. Slices were then immersed 3 times in PBS with Triton X-100 0.2% and incubated in blocking buffer for 1,5 h (PBS with Triton X-100 0.2% and FBS (10%) previously inactivated 20min at 56°C (Gibeco, 16141061). Primary antibodies (ORF1p antibody: abcam ab216324; NeuN antibody: GeneTex GTX00837) were diluted (1/200 and 1/500 respectively) in blocking buffer and incubated with slices overnight at 4°C and then washed 3 times for 10 min with PBS. For validation, an in-house ORF1p antibody was used (09) (guinea pig, 1/200). For non-neuronal marker (GFAP antibody: Millipore AB5541; Iba1 antibody: GeneTex GTX101495; Sox9 antibody: RnDsystems AF3075; S100β antibody: Sigma S2532), antibodies were diluted at 1/500.

Suitable secondary antibodies (Invitrogen) and Hoechst (Invitrogen, 15586276) were incubated for 1,5h at 1/2000 in PBS with inactivated FBS (10%) and washed 3 times 10 min in PBS. To quench tissue autofluorescence, especially lipofuscin, TrueBlack Plus (Biotium) in PBS was used during 10min. Slices were rinsed 3 times in PBS and mounted with Fluoromount (Invitrogen™). For human cingulate gyrus stainings, the same protocol was performed, with the difference that a human ORF1p antibody (Abcam 245249) was used. Mouse and human brain slices were imaged by the Axioscan 7 Digital Slide Scanner (Zeiss) or a Spinning Disk W1 confocal microscope (Yogogawa).

Blocked peptide

The ORF1p antibody (abcam ab216324) was incubated 2h on a turning wheel with excess (4:1) of mouse ORF1p recombinant protein as in ¹⁷ before the blocked antibody was used in the above-described immunofluorescence protocol.

Quantification of confocal acquisitions

Analysis was performed with a custom-written plugin developed for the Fiji software, using Bio-Formats ⁹⁸ and 3D Image Suite ⁹⁹ libraries. Code is freely available online at https://github.com/orion-cirb/DAPI_NEUN_ORF1P. Nuclei were detected using Hoechst channel downscaled by a factor of 2 with the 2D-stitched version of Cellpose ¹⁰⁰ (percentile normalization = [1-99], model = ‘cyto’, diameter = 30, flow threshold = 0.4, cell probability threshold = 0.0, stitching threshold = 0.75). The segmented image is rescaled to its original size, and the obtained 3D nuclei are filtered by volume to avoid false positive detections. NeuN and ORF1p+ cells were detected in their respective channel in the same manner as nuclei, but with different Cellpose settings (model = ‘cyto2’, diameter = 40, flow threshold = 0.4, cell probability threshold = 0.0, stitching threshold = 0.75). Then, each cell was associated with a nucleus having at least half of its volume in contact with it. Cells without any associated nucleus were filtered out. Each nucleus was thus labeled according to NeuN and/or ORF1p positivity.

ABBA Registration and Qupath analysis

We registered each sagittal brain section with the Allen Mouse Brain Atlas (CCFv3 ¹⁰¹), using the Aligning Big Brains & Atlases plugin ¹⁰² in Fiji. Registration results were imported into QuPath software ¹⁰³ for downstream processing.

In each brain subregion, cell analysis was done with custom Groovy scripts developed for QuPath. Code is freely available online at https://github.com/orion-cirb/QuPath_ORF1P. In brief, Hoechst nuclei were detected with StarDist 2D ¹⁰⁴, applying the DSB 2018 pretrained model with the following parameters: percentile normalization = [1-99], probability threshold = 0.75, overlap threshold = 0.25. Cells in NeuN and ORF1p channels were detected with Cellpose 2D (percentile normalization = [1-99], model = ‘cyto2’, diameter = 30, flow threshold = 0.4, cell probability threshold = 0.0). Nuclei and cells were then filtered by area and intensity in their respective channel, once again to avoid false positive detections. Minimum intensity threshold was based on the channel background noise. This noise was estimated for each subregion as the mean intensity of pixels not belonging to any nucleus or cell detected in the corresponding channel. Finally, each cell was associated with a nucleus having its centroid located inside the cell mask. Cells without any associated nucleus were filtered out. Each cell containing a nucleus was thus identified as NeuN+ or NeuN- and ORF1p+ or ORF1p-. Intensity values were normalized by subtracting the background noise computed in the corresponding channel and subregion. As a last step, subregional results were merged into regional ones and data were analyzed using the Pandas Python library ¹⁰⁵.

FACS

Mouse brains were dissociated with Adult Brain Dissociation kit (Miltenyi Biotec, 130-107-677) and incubated with the coupled antibody NeuN Alexa 647 (Abcam, ab190565) or the control isotype IgG Alexa 647 (Abcam, ab199093). Stained cells were filtered a last time with a 40µm filter before FACS sorting (FACS ARIA II). Neuronal and non-neuronal cells were separately collected in PBS -/- 2m EDTA and then centrifugate (5min at 700rpm). Pellets were resuspended in RIPA for protein extraction in an appropriate volume in order to achieve equal cell concentrations (10 000 cells/µl).

RNA-seq analysis

The RNA-seq dataset from Dong et al ⁴⁷ was downloaded from dbGAP (phs001556.v1.p1) and contains unstranded paired-end 50bp and 75bp reads from pooled laser-capture micro-dissected dopaminergic neurons from human post-mortem brain (107 samples) from 93 individuals w/o brain disease. RNA-seq had been done on total and linearly amplified RNA. We focused our analysis on data obtained with 50bp reads, in order to avoid mappability bias, while still regrouping all age categories (n=41; with ages ranging from 38 to 97 (mean age: 79.88 (SD ±12.07); n=6 ≤65y; n=35 >65y; mean PMI: 7.07 (SD ±7.84), mean RIN: 7.09 (±0.94)). Sequencing reads were aligned on the Human reference genome (hg38) using the STAR mapper (v2.7.0a) 3 and two different sets of parameters. Genome-wide individual repeat quantification was performed using uniquely mapped reads and the following STAR parameters: -- outFilterMultimapNmax 1–-alignEndsType EndToEnd–-outFilterMismatchNmax 999–- outFilterMismatchNoverLmax 0.06. Repeats class, family and name quantification was performed using a random mapping procedure and the following parameters : -- outFilterMultimapNmax 5000–-outSAMmultNmax 1–-alignEndsType EndToEnd–- outFilterMismatchNmax 999–-outFilterMismatchNoverLmax 0.06–-outSAMprimaryFlag OneBestScore–-outMultimapperOrder Random. Repeats annotations were downloaded from the UCSC Table Browser (repeatMasker database: https://genome.ucsc.edu/cgi-bin/hgTables) and coordinates of LINE-1 full length and coding elements were downloaded from the L1base database 2 (http://l1base.charite.de/l1base.php; ⁴) selecting LINE-1 full length elements containing two predicted complete open reading frames for ORF1 and ORF2 (UID= Unique IDentifier) from the LINE-1 database (http://l1base.charite.de/l1base.php) and corrected genomic intervals with the repeat masker annotation of the corresponding genomic locus. Repeat quantification from the aligned data was done using a gtf file composed of all genes (Gencode v29) and all individual repeat elements. This strategy was used to avoid overestimation of repeat elements due to overlaps with expressed genes. For individual repeat quantification of the full length L1 elements (L1base), we therefore used a gtf of all genes and all L1base entries, and ran the FeatureCounts tool ¹⁰⁶ with the following parameters: -g gene_id -s 0 -p. In the context of the family-based analysis, we used a gtf with all genes and all annotated repeats elements and ran FeatureCounts with -g gene_family -s 0 -p -M. Before DeSeq2 analysis, we remove all genes and repeat elements with less than 10 reads in a minimum of n individuals, n being the number of individuals in the condition containing the fewest individuals (“young” condition: n=6, 38-65y, mean 57.5 years). We use the same conditions with genes and UIDs with less than 3 reads in a minimum of n individuals. Finally, we calculated the scaling factors using DeSeq2 on all genes + all repeat elements or all gene + UID according to the quantification method and then applied these scaling factors to the corresponding counts tables.

In order to test for the mappability of each UID (= full-length and coding LINE-1), we extracted the bed track « main on human:umap50 (genome hg38) from the UCSC genome browser (≈ 7Mio regions) directly into Galaxy (usegalaxy.org) and joined genomic intervals with a minimum overlap of 45bp of this dataset with a dataset containing the annotation of UIDs extracted from L1Basev2 ⁴ corrected in length with repeat masker and completed with information on whether the UID is intra- or intergenic and, if intragenic, in which gene (NM_ID, chr, strand, start, end, gene length, number of exons, gene symbol) the fl-LINE-1 is located, which resulted in 1266 regions. We then used the “group on data and group by” function in Galaxy and counted the number of overlapping 50kmers with all 146 UIDs (=mappability score). Correlation analysis (non-parametric Spearman) was then done between the mappability score and the normalized read counts.

Immpunoprecipitation of ORF1p from the mouse brain

For immunoprecipitation, we used ORF1p (abcam, ab245122) and IgG rabbit (abcam, ab172730) antibodies. The antibodies were coupled to magnetic beads using the Dynabeads® Antibody Coupling Kit (Invitogen, 14311D) according to the manufacture’’s recommendations. We used 5µg of antibody for 1 mg of beads and used 1.5mg of beads for IP. Individual mouse brain lysates (n=5), homogenized using dounce and sonicated, were incubated with ORF1p or IgG-control coupled beads and a small fraction was kept as input. Each of these two tubes containing coupled beads and brain lysates were diluted in 5 ml buffer (10 mM Tris HCl, 150 mM NaCl, protease inhibitor). The samples were then incubated overnight on a wheel at 4°C. Samples were then washed 3 times with 1 ml buffer (10 mM Tris HCl pH 8, 200 mM NaCl) using a magnet and then resuspended in the same buffer. The samples were boiled in Laemmli buffer (95°C, 10 min) and 20 µl of each sample were loaded on a 4-12% Nupage gel (Invitrogen, NP0336) to be revealed by WB. For samples used in Mass Spectrometry study, beads were washed with buffer (10 mM Tris HCl pH 8, 200 mM NaCl) using a magnet. After 3 washes with 1ml buffer the beads were washed twice with 100 µL of 25 mM NH4HCO3 (ABC buffer). Finally, beads were resuspended in 100 μl of 25mM ABC buffer and digested by adding 0.20 μg of trypsine/LysC (Promega) for 1 hour at 37 °C. A second round of digestion was applied simultaneously on the beads by adding 100 µL of 25 mM ABC buffer and to the previous digest by adding 0.20 µg of trypsin/LysC for 1 hour at 37 °C. Samples were then loaded into homemade C18 StageTips packed by stacking three AttractSPE® disk (#SPE-Disks-Bio-C18-100.47.20 Affinisep) into a 200 µL micropipette tip for desalting. Peptides were eluted using a ratio of 40:60 CH3CN:H2O + 0.1% formic acid and vacuum concentrated to dryness with a SpeedVac device. Peptides were reconstituted in 10 µL of injection buffer in 0.3% trifluoroacetic acid (TFA) before liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis.

Mass Spectrometry

Online chromatography was performed with an RSLCnano system (Ultimate 3000, Thermo Scientific) coupled to a Q Exactive HF-X with a Nanospay Flex ion source (Thermo Scientific). Peptides were first trapped on a C18 column (75 μm inner diameter × 2 cm; nanoViper Acclaim PepMapTM 100, Thermo Scientific) with buffer A (2/98 MeCN/H2O in 0.1% formic acid) at a flow rate of 2.5 µL/min over 4 min. Separation was then performed on a 50 cm x 75 μm C18 column (nanoViper Acclaim PepMapTM RSLC, 2 μm, 100Å, Thermo Scientific) regulated to a temperature of 50°C with a linear gradient of 2% to 30% buffer B (100% MeCN in 0.1% formic acid) at a flow rate of 300 nL/min over 91 min. MS full scans were performed in the ultrahigh-field Orbitrap mass analyzer in ranges m/z 375–1500 with a resolution of 120 000 at m/z 200. The top 20 intense ions were subjected to Orbitrap for further fragmentation via high energy collision dissociation (HCD) activation and a resolution of 15 000 with the intensity threshold kept at 1.3 × 105. We selected ions with charge state from 2+ to 6+ for screening. Normalized collision energy (NCE) was set at 27 and the dynamic exclusion of 40s. For identification, the data were searched against the Mus Musculus (UP000000589_10090 012019) Uniprot database using Sequest HT through proteome discoverer (version 2.4). Enzyme specificity was set to trypsin and a maximum of two-missed cleavage sites were allowed. Oxidized methionine, Met-loss, Met-loss-Acetyl and N-terminal acetylation were set as variable modifications. Maximum allowed mass deviation was set to 10 ppm for monoisotopic precursor ions and 0.02 Da for MS/MS peaks. The resulting files were further processed using myProMS97 v3.10.0. FDR calculation used Percolator and was set to 1% at the peptide level for the whole study. The label free quantification was performed by peptide Extracted Ion Chromatograms (XICs), reextracted by conditions and computed with MassChroQ version 2.2.21 ¹⁰⁷. For protein quantification, XICs from proteotypic peptides shared between compared conditions (TopN matching) with missed cleavages were used. Median and scale normalization at peptide level was applied on the total signal to correct the XICs for each biological replicate (n=5). To estimate the significance of the change in protein abundance, a linear model (adjusted on peptides and biological replicates) was performed, and p-values were adjusted using the Benjamini–Hochberg FDR procedure. Proteins with at least three peptides, identified in each biological replicates of ORF1p condition, a 10-fold enrichment and an adjusted p-value ≤ 0.05 were considered significantly enriched in sample comparisons. Unique proteins were considered with at least three peptides in all replicates. Protein selected with these criteria were used for Gene Ontology enrichment analysis and string network analysis.

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository ¹⁰⁸ with the dataset identifier PXD047160.

GO term and STRING network analysis

Gene Ontology analysis was performed using GO PANTHER (PANTHER: Making genome-scale phylogenetics accessible to allPaul D. Thomas, Dustin Albou and Huaiyu MiProtein Society. 2022;31(1):8-22. doi:10.1002/pro.4218) and String network physical interactions were retrieved using the STRING database v11.5 (https://string-db.org/) and then implemented in Cytoscape software ¹⁰⁹.

Statistical analyses

In column comparisons, data in each column were tested for normality using two normality and lognormality tests (Shapiro-Wilk test and Kolmogorov-Smirnov test). Data which passed the normality tests were analyzed subsequently by a parametric test, data which did not pass the normality tests were analyzed by a non-parametric statistical test as indicated in the figure legends. The significance threshold was defined as p<0.05. Statistical analyses were done with PRISM software (v10).

Acknowledgements

This work was supported by the Fondation de France (00086320, to J.F.), the Fondation du Collège de France (to J.F. and T.B.), the Fondation NRJ/Institut de France (to J.F.) and the National French Agency for Research (ANR-20-CE16-0022 NEURAGE). We gratefully acknowledge the Orion Technological Core (IMACHEM-IBiSA) of CIRB, member of the France-BioImaging research infrastructure, especially Estelle Anceaume and Julien Dumont for assistance with slide scanner and spinning disk acquisition and Magalie Fradet for assistance with FACS analysis. We also thank the Fondation Bettencourt Schueller for their support.

Contributions

T.B. carried out most of the experimental work and analyzed the data. T.B., S.S. and J.F. analyzed the transcriptomic data. O.M.B. and S.S contributed to the experimental work and performed the FACS analysis. T.B. and J.F. wrote the manuscript. T.B., P.M. and H.M. developed the image analysis pipeline. B.L. carried out the MS experimental work and D.L. supervised MS and performed data analysis with T.B.. Read alignment, quality control and mapping were done by N.S.. J.F. and R.L.J. conceived and supervised the project.

Competing interests

The authors declare no competing interests.

Supplementary Figures

Suppl Fig. 1

Suppl Fig. 2

Suppl Fig. 3

Suppl Fig. 4

Suppl Fig. 5

Suppl Fig. 6

Suppl Fig. 7