Localized molecular chaperone synthesis maintains neuronal dendrite proteostasis

Proteostasis is maintained through regulated protein synthesis and degradation and chaperone-assisted protein folding. However, this is challenging in neuronal projections because of their polarized morphology and constant synaptic proteome remodeling. Using high-resolution fluorescence microscopy, we discovered that neurons localize a subset of chaperone mRNAs to their dendrites and use microtubule-based transport to increase this asymmetric localization following proteotoxic stress. The most abundant dendritic chaperone mRNA encodes a constitutive heat shock protein 70 family member (HSPA8). Proteotoxic stress also enhanced HSPA8 mRNA translation efficiency in dendrites. Stress-mediated HSPA8 mRNA localization to the dendrites was impaired by depleting fused in sarcoma—an amyotrophic lateral sclerosis-related protein—in cultured mouse motor neurons and expressing a pathogenic variant of heterogenous nuclear ribonucleoprotein A2/B1 in neurons derived from human induced pluripotent stem cells. These results reveal a crucial and unexpected neuronal stress response in which RNA-binding proteins increase the dendritic localization of HSPA8 mRNA to maintain proteostasis and prevent neurodegeneration.

Additional Declarations: Yes there is potential Competing Interest.GWY is a Scienti c Advisory Board member of Jumpcode Genomics and a co-founder, Board of Directors and Scienti c Advisory Board member, equity holder, and paid consultant for Locanabio and Eclipse BioInnovations.GWY is a visiting professor at the National University of Singapore.GWY's interests have been reviewed and approved by the University of California, San Diego in accordance with its con ict-of-interest policies.The authors declare no other competing nancial interests.
The mechanism underlying HSP subcellular distribution in neurons represents a major knowledge gap.
Most studies on induced HSP expression have investigated their upregulation by the transcription factor heat shock factor 1 (HSF1) 47,[70][71][72][73] .Recently, mRNAs encoding HSPA8 and HSP90AA were identi ed in the dendritic transcriptome under basal conditions 74 .Here, we report that neurons increase the transport and local translation of a subset of HSP mRNAs in the dendrites in response to proteotoxic stress.Combining high-resolution uorescence microscopy and molecular biology, we characterized changes in the subcellular localization of HSP mRNAs in primary hippocampal and motor neurons subjected to different proteotoxic insults.Fused in sarcoma (FUS) and heterogenous nuclear ribonucleoprotein A2/B1 (HNRNPA2B1), both implicated in amyotrophic lateral sclerosis (ALS), were identi ed as important regulators of the subcellular distribution of the constitutive HSP70 mRNA, HSPA8, and their actions were essential for dendritic proteostasis during stress.

Hippocampal neurons alter HSP mRNA distributions upon stress
To study how neurons tailor HSP expression to proteostatic demands, we subjected cultures of primary mouse hippocampal neurons to proteotoxic stress by inhibiting the proteasome.These cultures faithfully recapitulate the regulation of mRNA localization and translation in response to neuronal stimuli and the activation of HSP transcription upon stress 12,[75][76][77][78] .Dissociated cultures of hippocampi from postnatal day 0 mouse pups differentiate to express features of their mature in situ counterparts by day 17.Treatment with the proteasome inhibitor MG132 results in the accumulation of misfolded proteins and protein quality failure-prominent hallmarks of neurodegenerative disorders 79 .To study MG132-induced changes in subcellular mRNA localization, we isolated the total RNA from somas and projections separately harvested from neurons cultured in Transwell membrane lter inserts 80 (Fig. 1a).RNA sequencing (RNA-Seq) and differential expression analysis (DESeq2) revealed previously described transcript signatures speci c to somas and projections in steady-state (control (Ctrl)) neurons, e.g., dendritic localization of calcium/calmodulin-dependent protein kinase II alpha (CAMKII) and β-actin) 74,81 .Exposure to MG132 (10 µM for 7 h) signi cantly changed the expression of hundreds of RNAs between the soma and dendrites.In fact, in gene ontology analyses of the changed transcripts, the only biological function enriched in both compartments was "protein refolding" (Fig. 1c, S1a-S1c and Table S1).
Mammals contain over 400 genes encoding molecular chaperones and co-chaperones 82 ; of these, only 16 were upregulated in both fractions with increased enrichment in either the soma (e.g., the inducible HSP70 HSPA1A1) or the projections (e.g., the constitutive HSP70 HSPA8).Interestingly, while mRNAs for 11 chaperones were speci cally increased in neuronal projections, only the mRNA encoding the sHSP CRYAB was enriched in the soma (Fig. 1d-1f).Importantly, co-chaperones colocalized with their chaperone partners and HSP mRNA distributions matched the subcellular locations of their known folding clients.For instance, HSP90s and their co-chaperone P23 were enriched in projections upon stress.In dendrites, HSP90 supports the delivery of α-amino-3-hydroxy-5-methyl4-isoxazolepropionic acid receptors to the spine membrane, which is critical for synaptic transmission in the hippocampus 83 .
Likewise, DNAJs localized with their refolding partners, constitutive (HSPA8) and inducible (HSPA1A) HSP70s.In contrast with the signi cant upregulation of HSP mRNAs in projections, the levels of CAMKII and β-actin mRNAs, which are well-known to localize to dendrites, were unchanged 74,81 (Fig. 1f).These data suggest that neurons identify the need for HSP and co-chaperone mRNAs and distribute them to the same compartments as their client proteins.

A subset of HSP mRNAs speci cally localize to the dendrites upon stress
To de ne the principles of selective neuronal HSP mRNA localization and to identify proteotoxicityinduced changes in their subcellular distributions, we combined single-molecule uorescence in situ hybridization (smFISH) with immuno uorescence (IF) to localize single mRNAs in primary hippocampal neurons using established markers of dendrites (microtubule-associated protein 2 (MAP2)), axons (microtubule-associated protein tau (TAU)), and spines (postsynaptic density 95 (PSD95); Fig. 2a) 84,85 .
Single mRNAs were identi ed and quanti ed using FISH-quant 86,87 .To collect statistics on mRNA localization across the neuronal morphology, we updated the computational pipeline Analysis of RNA Localization In Neurons (ARLIN ) 88 .This pipeline was used to validate the RNA-seq data by studying the main HSPs implicated in proteostasis loss during neurodegeneration: HSPA1A, HSPA8, HSP90AA, HSP90AB, and HSP110 50,58,66,89,90 .We rst analyzed changes in the soma and quanti ed the signi cant transcriptional induction and increased concentrations of these constitutive and inducible HSP mRNAs, while the induction of DNAJB1 and DNAJB5 (used as controls) was lower; the same pattern was detected by RNA-seq (Fig. 2b-d).
We next ascribed the mRNAs enriched in neuronal projections to either the axons or dendrites by combining smFISH with IF of TAU and MAP2.Remarkably, we did not detect any HSP mRNAs in the axons of Ctrl or MG132-stressed neurons (Fig. 2e).Instead, the mRNAs of all HSP of interest localized to the dendrites and were signi cantly enriched by MG132 exposure (Fig. 2f and 2g).However, their concentrations varied greatly; from an average of 100 Hspa8 mRNAs to only ve Hspa1a mRNAs per dendrite.Hspa1a mRNA was retained in the soma, whereas Hspa8, Hsp90aa, and Hsp110 mRNAs were enriched in the dendrites after MG132 stress, con rming the RNA-seq data (Fig. 2h).These results, together with the unchanging distribution of CamkII mRNA in the dendrites in response to MG132, strongly suggest that neurons selectively target speci c HSP mRNAs to the dendrites upon stress (Fig. 2g  and 2h).
Since Hspa8 was the most abundant dendritic HSP mRNA measured, we veri ed that its dendritic localization occurs under disease-related stress conditions (Fig. 2i and 2j).We used hypoxia and hypoxiareoxygenation injury (1% O 2 for 3 h and 4 h reoxygenation), which generates reactive oxygen species resulting in protein misfolding 91 and brain damage 92,93 or neuronal exposure to 500 nM of oligomeric amyloid-β peptides for 24 h (oAβ 1-42 ) 94,95 , which accumulate in the hippocampus in Alzheimer's disease and cause dendritic attrition 94,95 .Dendritic Hspa8 mRNA localization signi cantly increased after reoxygenation or oAβ 1-42 exposure (Fig. 2e).Notably, although the level of transcriptional induction was much lower than after MG132 exposure, these stresses relocated somatic Hspa8 mRNAs to the dendrites (Fig. 2j).Therefore, the subcellular localization of Hspa8 mRNAs to the dendrites is promoted by diverse proteotoxic stresses.These results suggest a common stress-induced regulatory mechanism that directs speci c HSP mRNAs to the dendrites.

Stress promotes HSP mRNA localization to the proximal dendrites and spines
We next investigated whether stress induces changes in the distribution of HSP mRNAs within the dendritic compartment, and particularly in the dendritic spines, which receive synaptic inputs.First, we considered that stress might drive HSP mRNAs into distal dendrites, broadening their distribution.To test this hypothesis, we measured the distance that each mRNA traveled from the soma and grouped them into bins of 25 µm to analyze their dendritic distributions (Fig. 3a).Stress induced signi cant increases in the number of Hspa8, Hsp90aa, Hspa90ab, and Hsp110 mRNAs in the bins proximal to the soma.However, no signi cant increases in any of the mRNAs were observed at distances ≥ 100 µm from the soma.Thus, although the dendritic concentration of HSP mRNAs increased upon MG132-stress, their distribution over the dendritic shaft was comparable to that in non-stressed neurons, with more mRNAs in the proximal dendrites (Fig. 3a).
We examined whether mouse primary motor neurons, which have more extensive and thicker dendrites than hippocampal neurons, similarly regulate Hspa8 mRNA.Dissociated cultures of embryonic day 13 mouse spinal cords were matured for at least 3 weeks, when cultured motor neurons express properties of their in situ counterparts 96 .Upon MG132 exposure, cultured motor neurons Hspa8 mRNA concentrations in the soma and dendrites, promoting dendritic localization (Fig. 3b-3d); however, Hspa8 mRNA was more concentrated in the proximal dendrites in both stressed and non-stressed neurons (Fig. 3e).These results indicate that the regulation of HSP mRNA localization is shared by different neuron types and that their movement towards distal dendrites is constrained or subjected to retrograde transport.
Considering the concentrated localized translation that occurs in the dendritic spines, we hypothesized that they would require relatively more HSPs to prevent aberrant interactions among unfolded proteins.Thus, we used ARLIN to analyzed stress-induced changes in the contiguity between HSP mRNAs (detected by smFISH) and the postsynaptic density (detected by PSD95 IF) 88 (Fig. 3f).Quantifying mRNAs within 600 nm of the center of the PSD95 signal showed that MG132 stress increased the number of spines containing at least one Hspa8, Hsp90aa, Hsp90ab, or Hsp110 mRNA (Fig. 3f and 3g and Fig. S2a).We used ARLIN to bin the dendritic shafts into 25 µm segments and organized them by distance from the soma to examine changes in the number of HSP mRNAs.The frequency of spines containing HSP mRNAs was higher in the proximal dendrites than in the distal, with up to ~ 20% and 40% of spines localizing Hspa8 mRNAs in control and MG132-stressed neurons, respectively (Fig. 3g).To determine whether localization near a dendritic spine was regulated or due to increased HSP mRNA density, we used the ARLIN segmentation module to run random simulations of HSP mRNA localizations while maintaining the position of the PSD95 signal.The contingency between all HSP mRNAs and dendritic spines was similar to that in random simulations over the dendritic shafts of control and MG132-stressed neurons, with no signi cant differences (P > 0.05 (unpaired t-test)) observed between the experimental and simulated data in each bin (Fig. 3i and Fig. S2a).Given the density of the cultures, we were unable to quantify mRNAs in the distal dendrites (> 150 µm from the soma) without introducing errors in their assignment to speci c dendrites.We conclude that the increased localization of HSP mRNAs in dendrites upon stress increases the number of spines containing them.

Active transport of individual HSP mRNAs in dendrites
We envisioned two non-exclusive mechanisms to promote dendritic Hspa8 mRNA localization upon stress: active mRNA transport from the soma to the dendrites or enhanced mRNA stability in the dendrites.To distinguish them, we disrupted microtubule polymerization with nocodazole to prevent intracellular transport.The signi cant reduction in the number of dendritic Hspa8 mRNAs upon combined MG132/nocodazole treatment con rmed the importance of active transport from the soma to increase dendritic Hspa8 mRNA levels upon stress (Fig. 4a).As such, longer exposure to MG132 favored dendritic increases in HSPA8 mRNA over a somatic increase, and this subcellular distribution was maintained after MG132 was withdrawn (Fig. S3a and S3b).Thus, stress triggered an initial transport of Hspa8 mRNAs to dendrites that remained stable during recovery, suggesting RBP-mediated transport.mRNAs are transported in dendrites bidirectionally by motor proteins: dynein for minus-end and kinesin for plus-end movement along microtubules.To identify the motor protein directing Hspa8 mRNAs to dendrites, we considered they harbor microtubules oriented in both plus and minus directions, while axons feature only plus-end-out oriented microtubules (Fig. 4b) 97 .Since dynein is the only motor exiting the soma using dendrite-speci c (minus-end out) microtubules, we propose it directs HSP mRNAs to dendrites.To test this hypothesis, motor neurons were microinjected in the nucleus with a plasmid expressing a dominant negative dynein inhibitor CC1 fused to GFP (a gift of Dr. Adam Hendricks) 98 or GFP alone.CC1 blocks the interaction between Dynein and Dinectin important for dynein-mediated cargo movement in cells.Neurons expressing CC1 had a signi cant decrease in dendritic Hspa8 mRNA levels upon MG132 exposure (Figs.4b and 4c).Under permissive conditions, Hspa8 mRNA dendritic decreased by CC1 expression was only signi cant in the most proximal 25 µm of dendrite (Fig. 4d).This result suggests a dynein-dependent targeting of Hspa8 mRNA to dendrites upon stress.
RNAs, proteins, and components of the translation machinery phase separate into neuronal granules of ~ 700 nm in diameter with distinctive compositions 34,37,99 .The concomitant localization and similar dendritic distribution of all HSP mRNAs suggested that they can be assembled and transported in the same neuronal granules.To identify mRNAs traveling with the most abundant dendritic mRNA, Hspa8, we performed two-color smFISH in hippocampal neurons to detect Hsp90aa, Hsp90ab, or Hsp110 localizing within 700 nm of, and thus coexisting with, each Hspa8 molecule (Fig. 4b).We observed higher frequencies of Hspa8 coexisted with any of the other HSP mRNAs in MG132-stressed neurons than in non-stressed neurons, and in proximal dendrites than in distal dendrites (Fig. 4c).To differentiate regulated and random mRNA colocalization, we created a new module in ARLIN 88 that randomly simulated the positions of Hsp90aa, Hsp90ab, or Hsp110 mRNAs over the dendritic shaft.After binning the shaft into 25 µm segments, ARLIN averaged the shortest distances between Hspa8 and the closest HSP obtained in a hundred100 random simulations.Although the coexistence between HSP mRNAs signi cantly increased upon MG132 exposure, the levels were similar between the experimental and simulated data in control and MG132-stressed conditions (Fig. 4c).Similar results were obtained when quantifying the coexistence of Hsp90aa, Hsp90ab, or Hsp110 mRNAs with Hspa8 mRNA (Fig. S3c).These results suggest that HSP mRNAs are packaged and transported in individual neuronal granules, as previously reported for other dendritic mRNAs 100 .
HSPs concentrate in the same neuronal compartments as their mRNAsvialocalized translation Transcript shuttling and local translation is the most e cient way for neurons to target proteins to the dendrites and their spines 21,22 .We tested whether this regulatory mechanism supports the subcellular distribution of the inducible and constitutive HSP70s (Fig. 5).To quantify the increases in inducible HSP70 (HSPA1A) in the somas and dendrites upon MG132 stress, we cultured hippocampal neurons in Transwell membrane lter inserts 80 .Proteins were isolated from each compartment separately and analyzed by Simple Western.Exposure to MG132 for 7 h only increased HSPA1A in the soma compartment (Fig. 5a), matching the somatic retention of its mRNA (Fig. 1e and 1h).The mRNA level of the constitutive HSP HSPA8 increased in both compartments and matched the increased HSPA8 IF signal observed in MG132-stressed hippocampal neurons (Fig. S4a).These results suggest that localized synthesis determines an HSP's subcellular distribution.
To investigate whether the mRNAs in each compartment are translated there, we generated a single Hspa8 mRNA translation reporter using the SunTag translation reporter system [101][102][103][104] (Fig. 5b).This plasmid reporter contains Hspa8's 5′ and 3′ untranslated regions (UTRs), coding sequence (CDS), and promoter sequences.We placed 24×GCN4 epitopes at the 5′ end of the CDS to detect nascent peptides as soon as they exit the ribosome tunnel.Translating mRNAs are indicated by colocalization of the smFISH and IF signals detecting the GCN4 nucleotide and amino acid sequences, respectively, while individual signals indicate untranslated mRNAs (green) or fully synthesized proteins that have diffused away from their mRNA (magenta; Fig. 5b).We used the IF signal intensity to quantify the number of peptides being synthesized per mRNA as a read out of the number of ribosomes reading an mRNA.The HSPA8 translation reporter resulted in the expression of the expected 125 kDa protein (Fig. S4b) and enabled the detection and quanti cation of single mRNA translation in individual mouse embryonic broblasts (MEFs; Fig. S4c and S4d).Stressing MEFs with MG132 signi cantly increased the number of translated mRNAs by an average of 20-45%.To investigate HSPA8 translation in neurons, this plasmid was intranuclearly microinjected into mature cultured motor neurons derived from dissociated murine spinal cords.Like endogenous Hspa8 mRNA, the dendritic localization of GCN4-HSPA8 mRNA increased upon MG132 exposure (Fig. S4e); however, HSPA8 accumulation in the dendrites resulted in a high uorescent background that obscured the visualization and quanti cation of individual translation events (Fig. S4f).Thus, we added a C-terminal SET binding protein 1 (SETB) degron to decrease the reporter's half-life and reduce the background under control conditions (Fig. 5b).
Motor neurons were microinjected with the optimized reporter plasmid, cultures were treated with dimethyl sulfoxide (DMSO; Ctrl) or MG132 11 h later, and translation was analyzed by smFISH-IF 7 h later (18 h after microinjection).The short window between injection and detection was critical to quantify the e ciency of Hspa8 mRNA translation in dendrites while avoiding GCN4-HSPA8-SETB accumulation (Fig. 5c and 5d).In control motor neurons, a few reporter mRNAs localized to the dendrites and were translated by monosomes or polyribosomes throughout the shaft.Likewise, somatic reporter mRNAs were translated at different e ciencies (Fig. 5c, upper panel).MG132 exposure increased the transcription and somatic and dendritic localization of reporter mRNAs (Fig. 5c).Its accumulation impaired accurate quanti cation of the somatic mRNA translation e ciency, although bright magenta spots depicting polyribosomes colocalized with the mRNA signal in control and MG132-stressed neurons (Fig. 5c, magni cation in bottom left panel).In dendrites, colocalization between the peptide and mRNA signals revealed mRNAs translated by monosomes and polysomes at varying distances from the soma (Fig. 5c, magni cation in bottom right panel).The percentage of mRNAs being translated per dendrite signi cantly increased upon MG132 exposure (Fig. 5d).Similarly, dendritic mRNA translation was more e cient in MG132-stressed neurons, as measured by the number of ribosomes reading a single mRNA (Fig. 5e).Thus, transcripts of the constitutive Hspa8 mRNAs escaped the translational repression associated with MG132 105 and instead its translation e ciency was increased in response to proteostasis demands.These results strongly suggest that combining increased mRNA localization and maintained translation e ciency during stress provides an on-demand source of dendritic HSPA8.

The active transport of Hspa8 mRNAs by FUS ensures dendrite proteostasis
Neurons contain many RBPs relevant to dendritic RNA transport, and several, including TAR DNA binding protein/TARDBP, FUS, and fragile X messenger ribonucleoprotein 1, tightly couple mRNA transport to translation, which is vital for neuronal function [41][42][43] .Accordingly, mutations in these RBPs lead to diverse neurodegenerative disorders 106 .We next identi ed RBPs that can mediate the dendritic transport of Hspa8 mRNA by binding to linear or structural zip codes in its 3′ UTR (Fig. 6a).We in vitro transcribed the Hspa8 3′ UTR and LacZ (as a negative control) sequences fused to 2×PP7 stem loops and attached them to amylose magnetic beads using the PP7 capsid protein fused to the maltose binding protein (PCP-MBP) 85 .Mass spectrometry (MS) identi ed six RBPs speci cally bound to the Hspa8 3′ UTR in crude protein extracts obtained from control or MG132-stressed mouse Neuro-2A (N2A) cells (Fig. 6a).Among them, we validated the binding of staufen double-stranded RNA binding protein 2 (STAU2) because of its well-known function in stabilizing and transporting speci c mRNAs to dendrites 31,107,108 (Fig. 6b).We also validated the binding of FUS to the Hspa8 3′ UTR.Although FUS was consistently present in the MS pro les of both the Hspa8 3′ UTR and the LacZ control, it showed relatively higher binding to the 3′ UTR in pulldowns (Fig. 6b).FUS was of particular interest because it regulates several steps in mRNA maturation, including transport to dendrites, and FUS mutations lead to dendritic retraction in motor neurons, leading to ALS and frontotemporal dementia 27,[109][110][111][112][113][114] .
To investigate the roles of STAU2 and FUS in the stress-induced dendritic localization of Hspa8 mRNA, we knocked down their expression in cultured motor neurons by co-microinjecting two speci c short hairpin (sh)RNAs for each along with a green uorescent protein (GFP)-expressing plasmid to identify the injected neurons (Fig. 6c-6h and Fig. S5a).Knocking down STAU2 signi cantly decreased the somatic and dendritic density of Hspa8 mRNA (quanti ed as mRNAs per pixel of soma or dendrite area) in control and MG132-stressed neurons (Fig. 6e); however, MG132 still signi cantly increased the density of Hspa8 mRNA in dendrites (p < 0.001 by unpaired t-test), demonstrating that STAU2 knockdown did not prevent Hspa8 mRNA dendritic transport.To assess speci city, we analyzed the constitutive non-HSP eEf1a1 mRNA in parallel.STAU2 knockdown had a milder effect on eEf1a1 mRNA density in the soma and dendrites (Fig. 6f).On the contrary, knocking down FUS did not change the somatic concentration of Hspa8 mRNA but signi cantly decreased its dendritic density upon MG132 exposure (Fig. 6d and 6g), and led to an overall decrease in eEf1a1 mRNA density in the soma and dendrites (Fig. 5h).The ratio of dendritic to somatic eEf1a1 mRNA density did not change upon stress, whereas the FUS depleted neurons displayed decreases in these ratios (Fig. S5b).Thus, FUS plays an essential role in targeting Hspa8 mRNAs to dendrites during stress.
We next evaluated whether FUS knockdown could weaken dendritic proteostasis using the proteostasis reporter FLUC-GFP 115 .Impaired proteostasis leads to GFP aggregation, which was quanti ed by the granularity of the GFP signal.We injected the FLUC-GFP plasmid into motor neurons either alone or with the FUS shRNA plasmids.Knocking down FUS increased GFP granularity in dendrites even under control conditions, suggesting that normal FUS levels are essential for neuronal proteostasis under physiological conditions.The loss of dendritic proteostasis upon FUS knockdown was further increased upon MG132 treatment, leading to signi cant increases in GFP granularity and in the sizes of the GFP aggregates (Fig. 6i and 6j).These results reveal regulated FUS expression as a determinant of neuronal proteostasis.
The D290V mutation in HNRNPA2B1 impairs HSPA8 mRNA localization in human-derived motor neurons Mutations in FUS and other RBPs, including HNRNPA2B1, lead to ALS 116 .Through a recent collaboration, we found that mouse motor neurons microinjected with a plasmid expressing the familiar FUS R521G mutation had signi cant lower Hspa8 mRNAs levels in soma and dendrites (Fernandez, M. et al, submitted to Cell Stress and Chaperones).As the HSPA8 3′ UTR sequence and length differs between mice and humans, we next investigated whether human motor neurons localize HSPA8 mRNA to dendrites upon MG132 stress and which RBPs mediate their localization.Interestingly, the human HSPA8 3′ UTR sequence contains ve putative HNRNPA2B1 binding sites that are not present in mice (Fig. 7a).In patients with ALS, the D290V mutation in HNRNPA2B1 is rare.However, it promotes the accumulation of the detergent-insoluble HNRNPA2B1 protein in the nucleus and changes the subcellular distribution of mRNAs during stress 117 .Because of this, human motor neurons differentiated from patient broblastderived induced pluripotent stem cells (iPSCs) do not recover from stress as well as neurons differentiated from healthy donors 118,119 .Thus, we compared the ability of healthy (control) and HNRNPA2B1 D290V -expressing motor neurons (from two patients each) to localize HSPA8 mRNAs to the dendrites upon MG132 exposure (Fig. 7a and 7b).
Importantly, human motor neurons behaved like mouse neurons, increasing the HSPA8 mRNA level and its dendritic localization upon MG132 stress (Fig. 7b and 7c).The D290V mutation impaired the somatic accumulation of HSPA8 mRNAs in HNRNPA2B1 D290V -derived motor neurons, and remarkably, both sets of HNRNPA2B1 D290V -expressing motor neurons had signi cantly less dendritic HSPA8 mRNA than healthy motor neurons and this decrease was more signi cant for patient 2 than patient 1 (Fig. 7c).As a result, the distribution ratio of HSPA8 mRNA in the dendrites relative to the soma was lower in HNRNPA2B1 D290V motor neurons than in controls upon MG132 stress (Fig. 7d).Therefore, HNRNPA2B1's role in localizing HSPA8 mRNAs to dendrites is compromised by the D290V mutation, contributing to impaired proteostasis and proteotoxic damage.
Overall, decreased FUS expression and HNRNPA2B1 mutation have similar consequences in stressed motor neurons: impaired Hspa8 mRNA dendritic localization and decreased stress recovery.Based on our results, we propose a model in which healthy neurons sustain dendritic proteostasis through the regulated localization and translation of HSPs, especially the constitutive chaperone HSPA8, providing an on-demand system to tailor the amounts of HSPs to the load of misfolded proteins.Disruptions in RBPs that impair this localization decrease neuronal proteostatic capacity and prevent synapse formation and transmission, leading to neurodegeneration (Fig. 7e).

Discussion
This study uncovered a novel mechanism to sustain neuronal proteostasis under proteotoxic stress; the partitioning of distinct HSPs in the soma and dendrites through the regulated localization of their encoding mRNAs and subsequent translation.Stress-induced changes in HSP mRNA compartmentalization indicate that the soma and dendrite's distinct proteomes are upheld by particular sets of chaperones.It also suggests that after proteotoxic damage the dendrite demands for HSPs exceed the neuron capacity to transport individual chaperones from the soma and, instead rely on the competence of an mRNA to produce tens of proteins.The mRNA encoding the constitutive HSP70, HSPA8 has the highest levels of all HSP mRNAs we detected in dendrites.HSPA8 is central to sustaining proteostasis because of its chaperonin functions in assisting co-translational folding and regulated degradation of proteins by chaperone-mediated autophagy 120 .HSPA8 also plays moonlighting functions in axonal terminals, where mediates synaptic vesicle fusion and recycling 64,121 , and in dendrites, where it regulates the shape of dendritic spines 122 .Besides these critical roles, how HSPA8 localizes to neuronal projections was understudy.Leveraging single mRNA imaging techniques in stressed mouse hippocampal and motor neurons in culture, has allowed us to discover a speci c and regulated Hspa8 mRNA targeting mechanism to dendrites.Since we did not detect HSP mRNAs in axons, they might localize HSPA8 synthesized in the soma.
This study identi ed the post-transcriptional regulation of constitutive Hspa8 mRNA in neurons, both its boosted dendritic localization and translational e ciency, as a crucial aspect of their stress response.It operates by RBPs recognizing zip codes in the mRNAs, dynein leading them to dendrites where they are transported as individual molecules, and a stress-induced translational e ciency.As such, it resembles the induced postsynaptic localization and local translation of the mRNAs encoding an activity-regulated cytoskeleton-associated protein ARC and b-ACTIN that occurs upon synaptic stimulation 12,25 .We approached uncovering the components directing Hspa8 mRNAs to dendrites by searching for RBPs that bind to the 3′ UTR sequence of Hspa8 mRNA since this region is known for its role in mRNA localization.We identi ed STAU2, well-known for its function in stabilizing and transporting speci c mRNAs to dendrites 31,107,108 .Although STAU2 knockdown in cultured motor neurons reduced the levels of Hspa8 mRNAs in the soma and dendrites, it failed to prevent the increase in dendritic Hspa8 mRNAs in response to treatment with MG132.Conversely, FUS was identi ed as an important player in the subcellular distribution of Hspa8 mRNA.FUS knockdown signi cantly impaired Hspa8 mRNA localization in dendrites but did not completely abrogate it.Thus, FUS might cooperate with additional RBPs binding Hspa8 5' UTR or CDS.It is also possible that the observed effect of FUS to be indirect and operate through the expression regulation of factors involved in the transport of mRNAs.
Besides the fundamental role of FUS in Hspa8 mRNA transport to dendrites upon stress, we explored this RBP because it is implicated in ALS 116 , in which dendritic attrition is an early sign of motor neuron damage 6,123,124 .FUS depletion correlated with a decreased Hsap8 mRNA dendritic localization and the loss of dendritic proteostasis.Since ndings in mouse neurons might not re ect the human situation because of different nucleotide sequences in the HSPA8 3′ UTR, we examined iPSC-derived motor neurons.We focused on patient-derived neurons carrying the ALS-linked HNRNPA2B1 mutation D290V because the 3'UTR of the human HSPA8 mRNA, but not the mouse, has ve putative binding sites for HNRNPA2B1.Both lines of HNRNPA2B1 D290V -derived motor neurons had signi cantly less dendritic HSPA8 mRNA than control-derived neurons, indicating that the general principles of RBP regulation of dendritic HSP mRNAs are conserved between mouse and human neurons, with common roles in the neuronal stress response.In addition, the converging actions of different disease forms on HSPA8 mRNA biogenesis could explain the common loss of proteostasis that characterizes diverse neurodegenerative diseases.
Neurons last the lifespan of an organism, and loss of neuron proteostasis is a feature of many agingrelated neurodegenerative diseases.The localized synthesis of HSPs provides dendrites with the folding resources to aid their protein synthesis and degradation mechanisms in sustaining proteostasis.
Perturbations in any of the proteostasis components preclude dendritic homeostasis and lead to neurodegeneration.As such, boosting HSP transcription has been considered as a therapeutic strategy for neurodegenerative diseases, but has had limited clinical success [125][126][127][128] , and has focused particularly on HSPA1A, which is not upregulated in neurons under most conditions 72,125,126 .Our work stresses the importance of regulating not only the levels of constitutive HSPs, but also the dynamics of their localization to vulnerable regions, such as dendritic spines.These dynamics are crucial to consider when testing potential therapeutics.Sustaining functional synapses is essential for neuronal network functions, and stress on their proteomes would contribute to the impaired connectivity that underlies loss of function early in neurodegenerative disorders, prior to neuronal death [129][130][131] .

Methods
Neuronal cultures Mouse primary hippocampal neurons were obtained from postnatal day 0 C57BL/6 and FVB mice and prepared as previously described 78,85 .Housing and euthanasia were performed in compliance with the Canadian Council on Animal Care.Neurons used in imaging experiments were cultured at low density (50,000 neurons per 35 mm (14 mm glass) dish (MatTek, # P35G-1.5-14-C)).
Neurons used in RNA-seq and Simple Western experiments were cultured at 300,000 neurons per well in 100 µm Transwell membranes (Thermo Fisher Scienti c, # 35102) in 6-well dishes 80 .Dissociated spinal cord cultures from E13 CD1 mice were prepared as previously described 96 .Cells were plated on poly-D-lysine (Sigma, #P6407) -and Matrigel-coated glass coverslips in 6-well dishes.The culture medium was as described 96 with the addition of 1% B27 (Gibco Life Technologies, Burlington, ON, Canada, #17504044), 0.032 g/mL sodium bicarbonate and 0.1 g/mL dextrose.Cultures were maintained for at least 3 weeks to ensure motor neuron maturation.
Human motor neurons were differentiated from iPSCs as previously described 117,118 .CV-B (wild type) iPSCs were a gift from the Zhang Lab 132 and HNRNPA2B1 D290V-1.1 and − 1.2 human iPSCs were generated in the Yeo lab 118 .Human iPSCs were grown on Matrigel-coated 10 cm tissue culture plates.
Neuronal manipulation Neurons were stressed via 10 µM MG132 for the indicated times, hypoxiareoxygenation (1% O 2 for 3 h and 4 h recovery at 5% O 2 ) using a hypoxia glove box (BioSpherix Xvivo System Model X3), or incubation with oligomers made from 500 nM amyloid-β (1-42) monomers (rPeptide, #1163-1) 133 .As mature neurons cannot be transfected, plasmids were introduced into primary cultured mouse motor neurons by intranuclear microinjection.The injectate (the plasmid in 50% Trisethylenediaminetetraacetic acid (EDTA), pH 7.2) was clari ed by centrifugation prior to insertion into 1 mm diameter quick-ll glass capillaries (World Precision Instruments) pulled to ne tips using a Narishige PC-10 puller (Narishige International USA, Inc., NY, USA).Cultures on coverslips were bathed in Eagle's minimum essential medium without bicarbonate, supplemented with 5 g/L glucose, and adjusted to pH 7.4 in 35 mm culture dishes on the stage of a Zeiss Axiovert 35 microscope (Carl Zeiss Microscopy, LLC, USA) and microinjected using a Transjector 5246 or a FemtoJet Transjector and a Micromanipulator 5171 (all from Eppendorf, Hamburg, Germany).Following microinjection, coverslips were placed in regular culture medium containing 0.75% Gentamicin (Gibco) and maintained at 37°C in a 5% CO 2 environment until analysis.
Plasmid transfection and analysis Plasmids expressing shRNAs in a lentiviral backbone were obtained through the McGill University library (https://www.sidonghuanglab.com/pooled-screeninglibraries/service-request/)(Table S2).They were transfected by calcium phosphate into 293T cells and knockdown e ciency was tested 72 h later by western blotting (Table S2).
IF and smFISH Detailed protocols for these methods have been previously described 78 .RNA FISH probes were designed using the Stellaris Probe Designer (LGC Biosearch Technologies; masking level: 5, oligo length: 20, minimum spacing: 2) (Table S2).
Image acquisition and analysis Images were acquired using a Nikon eclipse Ti-2 inverted wide eld microscope equipped with a SPECTRA X Light Engine (Lumencor) and an Orca-Fusion Digital CMOS Camera (Hamamatsu) controlled by NIS-Elements Imaging Software.A 60× 1.40 NA oil immersion Plan Apochromat objective lens (Nikon) was used with a xy pixel size of 107.5 nm and a z-step of 200 nm.Chromatic aberrations were measured before imaging using 100 nm TetraSpeck™ Fluorescent Microspheres (Invitrogen, #T14792) and considered in the downstream pipeline.
Single mRNAs, peptides, and postsynaptic densities were identi ed with the MATLAB version of the FISHquant (v3) 86 .Post-detection analyses of subcellular mRNA distributions in neurons and simulations were performed with the second version of ARLIN 88 .The code for ARLIN v1.0 and ARLIN v2.0 can be found here: https://github.com/LR-MVU/neuron.See the corresponding documentation for a detailed explanation on ARLIN's functionalities.Brie y, in ARLIN v2.0, simulations were improved by mimicking the distributions of real mRNAs when selecting simulated mRNAs.To do this, the dendrite was divided into "bins" of 25 µm.The program rst counts x real mRNAs found in the rst bin of the dendrite (i.e., 0-25 µm from the soma).Then, the program selects x "simulated mRNAs" (i.e., randomly selected pixels) from only the rst bin of the dendrite.This ensures that the concentration of simulated mRNAs near the soma matches the concentration of real mRNAs, but with random distributions within the bin.The program counts the number of real mRNAs found in each bin, then randomly selects that number of pixels within it as "simulated mRNAs".With this improved simulation, the statistical likelihood of an mRNA localizing to the synapses or to another mRNA can be calculated.statistics are calculated for the localization of mRNAs to synapses or to another mRNA.The simulation is repeated 100 times and the localization statistics are averaged.This provides a more accurate comparison between random and biologically driven colocalization patterns than in the rst version of ARLIN.
To quantify translation e ciency, cell and dendrite segmentations were performed manually using the "De ne outlines" tool in FISH-quant.The smFISH and peptide spots in the cells were t to a 3D Gaussian model based on the point spread function and the analysis was run in batch mode.The x, y, and z coordinates of the mRNAs and peptides in cells were exported as tabulated text les (.txt) recording the identity of each cell in each image le analyzed.We designed a Python pipeline to rst calculate the number of nascent peptides in spots with a signal brighter to that corresponding to the the average intensity of one peptide.Secondly, the pipeline assigns each peptide to the closest mRNA in the cell.If the distance between an mRNA and its closest peptide exceeded the threshold (200 nm plus the chromatic aberration), then we considered it a non-translating mRNA.To remove repeated mRNAs and peptides within the threshold, we selected the brightest (i.e., brightest) peptide signal, and then the closest to a single mRNA.The percentage of translating mRNA was calculated by dividing the number of translating mRNAs by the total number of mRNAs per cell (https://github.com/LR-MVU/neuron).
We used Fiji (ImageJ 2.14.0,Java 1.9.0_322) to calculate the granularity of the FLUC-GFP signal.First, we generated a maximum projection for the GFP channel.Then, we outlined a region of interest to de ne each dendrite in a neuron and measured its area, mean uorescent signal, and standard deviation (SD).
The coe cient of signal variation-the SD divided by the mean-was used as a readout for GFP.
RNA extraction and RNA-seq After 17 days in culture 134 , primary hippocampal neurons were washed with PBS, and the somas were scraped from the membrane and placed into a tube.Somas were centrifuged for 2 min at 2,000 × g and resuspended in 400 µL ice-cold PBS.The somas were divided into two tubes, and 750 µL of Zymo RNA Lysis Buffer (ZymoResearch, # R1013) was added to each.To harvest the neurites, membranes were cut from the Transwell, placed face down in a 6 cm plate containing 750 µL of Zymo RNA Lysis Buffer, and incubated for 15 min on ice while tilting the plate every few minutes.The solution was transferred into an Eppendorf tube.RNA was isolated using a Zyma Quick RNA Miniprep Kit (ZymoResearch, #R1054).RNA library generation and Illumina sequencing was performed by the University of Montreal Genomic Platform.PolyA capture, Nextseq High Output paired-end run (2 × 75 bp, coverage ~ 50M paired-ends per sample).The raw data have been deposited in the Gene Expression Omnibus under accession number GSE202202.RNA-seq analysis was performed on usegalaxy.org.Adaptors and reads with a quality below 20 within 4base sliding windows were removed using Trimmomatic (galaxy version 0.38.0;https://doi.org/10.1093/bioinformatics/btu170).Trimmed single-end reads were aligned to the mouse mm10 genome using STAR (galaxy version 2.7.8a + galaxy0; https://doi.org/10.1093/bioinformatics/bts635)with default parameters, and the number of reads per transcript was determined using featureCounts (galaxy version 2.0.1 + galaxy2; https://doi.org/10.1093/bioinformatics/btt656)using default parameters.Differential gene expression was determined using DESeq2 (galaxy version 2.11.40.7 + galaxy1; https://doi.org/10.1186/s13059-014-0550-8)using default parameters.Gene ontology analysis to identify biological processes enriched in differentially expressed genes was performed using geneontology.org.
To validate the data by RT-qPCR, 25 ng of RNA isolated from the soma or neurites was reverse transcribed into cDNA using iScript™ Reverse Transcription Supermix (Bio-Rad) following the manufacturer's instructions.For qPCR, cDNAs were diluted two-fold in water.PCR was performed in 5 µL reactions consisting of 1 µL DNA, 2.5 µL PowerUp SYBR Green Master Mix (Thermo Fisher Scienti c) and 0.25 µL of each primer (at 1 µM) on a Viaa 7 Real-Time PCR System (Thermo Fisher Scienti c; 45 cycles).Standard curves were generated using a log titration of N2A genomic DNA (50-0.05ng) and used to quantify the cDNA.The primers used are listed below.
Simple Westerns Primary hippocampal neurons grown on Transwell membranes were treated with 10 µM MG132 for 7 h at 37°C.After washing with ice-cold PBS, somas were rst scrapped from the upper part of the membrane and then neurites were scraped from the bottom membrane and resuspended in ice-cold PBS.After centrifugation, somas and neurites were resuspended in radioimmunoprecipitation buffer (150 mM NaCl, 10 mM Tris-HCl pH 7.5, 0.1% sodium dodecyl sulfate (SDS), 1% Triton X-100, 1% sodium deoxycholate, and 5 mM EDTA pH 8.0).Protein extracts from somas and neurites were stored at -80°C.Simple Westerns (Bio-Techne) were performed in multiplex on a Jess automated western blot system (Bio-Techne) following the manufacturer's instructions (Table S2).
Identi cation of RBPs binding to the mouse HSPA8 3′ UTR PP7-HSPA8 and PP7-LacZ RNA were rst PCR ampli ed from N2A genomic DNA or a plasmid (donated by Dr. Jerry Pelletier) using the primers listed below, and then in vitro transcribed with a MEGAshortscript™ T7 Transcription Kit (Invitrogen, # AM1354) following the manufacturer's instructions.After transcription, RNA was treated with 2 units of Turbo DNase and then puri ed by phenol-chloroform extraction and ethanol precipitation.RNA was resuspended in 10 mM Tris containing 0.2 U/mL RNaseOUT.Small samples were resolved on a 0.5X TBE agarose gel and their A260/A280 ratios were measured using a nanodrop to verify the purity of the RNAs.PP7-HSPA8 and PP7-LacZ RNA were heated for 2 min at 95°C, allowed to cool to room temperature to allow PP7 loops to form, and stored at -80°C.
To prepare crude N2A extracts, cells were differentiated into the neuronal phenotype for 3 days: one day in DMEM supplemented with 5% fetal bovine serum (FBS) and 20 µM retinoic acid, one day in 2.5% FBS and 20 µM retinoic acid, and one day in 1.25% FBS and 20 µM retinoic acid.Half of the cells were treated with 10 µM MG132 for 7 h at 37°C.Cells were washed once with ice-cold 1× PBS and pelleted by centrifugation.The cell pellets were washed once with 1× PBS and 1 mM phenylmethylsulfonyl uoride.The supernatant was removed and cells were stored at -80°C.The pellets were thawed on ice, resuspended in three volumes of N2A lysis buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1 mM MgCl 2 , 0.1 mM CaCl 2 , 1% IGEPAL CA-360, 0.5% deoxycholic acid, 0.1% SDS, 1 mM phenylmethylsulfonyl uoride, 1 mM dithiothreitol, 1× Complete Protease Inhibitor (Roche), and 100 U/mL RNaseOUT) and incubated on ice for 10 min.Cells were snap frozen in liquid nitrogen and thawed on ice twice before 10 min centrifugation at max speed.The crude extract (supernatant) was transferred to new tubes and stored at -80°C.Protein concentration was determined by Bradford assay and a small sample of crude extract was run on SDS-PAGE stained with Coomassie Blue to ensure no protein degradation.
In 100 µL reaction, 1.5 µM of PP7-HSPA8 3′ UTR or PP7-LacZ RNA were incubated with 2 µM MBP-PP7 in RNA-IP buffer (20 mM Tris pH 7.2, 200 mM NaCl, 1 mM EDTA pH 8.0, 5 mM dithiothreitol, and 0.01% IGEPAL CA-360) for 1 h on ice.Magnetic amylose beads (100 µL) were washed twice with RNA-IP buffer, then rotated with PP7-HSPA8 3′ UTR or PP7-LacZ bound to MBP-PP7 for 1 h at 4°C.The beads were washed twice with RNA-IP buffer then resuspended in 5 mL RNA-IP buffer supplemented with 0.01 mg/mL tRNA (Sigma, #10109541001) and 5-10 mg N2A crude extract for MS or 2 mg of crude extract for western blots.After rotating the beads and N2A crude extract for 2 h at 4°C, the beads were washed ve times with RNA-IP buffer, resuspended in 50 µL RNA-IP buffer and 6 µg of TEV protease, and rotated for 3 h at 4°C.The cleaved PP7 proteins bound to the HSPA8 3′ UTR or LacZ RNA and their interactors was collected and the beads were incubated in fresh RNA-IP buffer containing TEV protease overnight.
The elutions were pooled, and the proteins were analyzed by MS as previously described (Proteomics RIMUHC-McGill University) 135 .Proteins with fold change values > 1.5 and P-values < 0.01 compared to in the control sample were considered HSPA8 3′ UTR interactors.Statistics performed using total spectral count and a T-test analysis.

Declarations
been reviewed and approved by the University of California, San Diego in accordance with its con ict-ofinterest policies.The authors declare no other competing nancial interests.Three days after microinjection, stress was induced with MG132, and HSPA8 mRNA expression was detected by smFISH.Scale bars = 5 mm.(e-h) Quanti cation of the densities of Hspa8 (d, g) and eEf1a1(f, h) mRNAs per pixel of soma or dendrite area in Ctrl and MG132-stressed motor neurons expressing GFP with and without the indicated shRNA expression plasmids.(i) Representative dendrites from Ctrl and MG132-stressed motor neurons expressing the proteostasis reporter plasmid FLUC-GFP with and without FUS knockdown.GFP aggregation is proportional to proteostasis loss.(j)Quanti cation of the GFP signal granularity (the coe cient of variation) in each dendrite in I. Two independent experiments were performed (n = 44-96 dendrites).

Figures Figure 1
Figures

Figure 3 Stress
Figure 3

Figure 4 Individual
Figure 4

Figure 7 An
Figure 7