Research Article

Monosomes actively translate synaptic mRNAs in neuronal processes

See allHide authors and affiliations

Science  31 Jan 2020:
Vol. 367, Issue 6477, eaay4991
DOI: 10.1126/science.aay4991

Monosomes translate proteins in neurons

Like all other cells, neurons use different proteins to process information and respond to stimuli. To meet the huge demands for new proteins in their large and complex cell volume, neurons have moved the protein templates—messenger RNAs (mRNAs)—and the protein synthesis machines—ribosomes—out to synapses to make proteins locally. During protein synthesis, multiple ribosomes can form a structure known as a polysome, which produces multiple protein copies from a single mRNA. Working in rodent preparations, Biever et al. found that solitary, mRNA-associated ribosomes, or monosomes, are a substantial source of proteins in neuronal processes. Many synaptic proteins are made on single ribosomes, which may solve the problem of limited space in tiny synaptic compartments.

Science, this issue p. eaay4991

Structured Abstract

INTRODUCTION

RNA sequencing and in situ hybridization have revealed the presence of an unexpectedly high number of RNA species in neuronal dendrites and axons, and many studies have documented the local translation of proteins in these compartments. During messenger RNA (mRNA) translation, multiple ribosomes can occupy an individual mRNA (a complex called a polysome), resulting in the generation of multiple copies of the encoded protein. Polysomes, usually recognized in electron micrographs as a cluster of three or more ribosomes, have been detected in neuronal dendrites but are surprisingly infrequent given the diversity of mRNAs present in dendrites and axons. In neuronal processes, the features and mechanisms of translation have not been explored in detail, in part because of the relative inaccessibility of dendrites and axons. In this study, we investigated how a diverse set of neuronal proteins might be synthesized from a limited population of polysomes present in small synaptic volumes.

RATIONALE

To accommodate their complex morphology, neurons localize mRNAs and ribosomes near synapses to produce proteins locally. Yet a relative scarcity of polysomes (considered the active sites of translation) detected in electron micrographs of neuronal processes has suggested a rather limited capacity for local protein synthesis. To visualize the capacity for local protein production in vivo, we profiled actively translating mRNAs in rodent hippocampal neuronal processes. To access neuronal compartments, we microdissected the neuropil and somata layer from the CA1 region of hippocampal slices, generating samples enriched in dendrites and/or axons versus cell bodies. Polysome profiling of microdissected regions was used to determine the association of axonal and/or dendritic and somatic transcripts, respectively, with monosomes or polysomes. Ribosome footprinting was then used to assess the translational activity of monosomes (single ribosomes) and polysomes. Bioinformatic analyses were used to determine the features of monosome-preferring transcripts as well as the families of protein groups that were encoded by monosome-preferring transcripts. We also compared the monosome-to-polysome (M/P) preference of transcripts between the somata and neuropil. To estimate the abundance of proteins encoded by monosome- and polysome-preferring transcripts, we measured protein levels in the neuropil by using mass spectrometry–based proteomics.

RESULTS

In the adult rodent brain, we detected substantial levels of ongoing protein synthesis in the synaptic neuropil (a region enriched in neuronal axons and dendrites) in vivo and provide direct evidence for the preferential translation of a high number of both pre and postsynaptic transcripts by monosomes. The monosomes were in the process of active polypeptide elongation in dendrites and axons. Most transcripts exhibited a similar M/P preference in both somata and neuropil, suggesting that ribosome occupancy is often an intrinsic feature of the transcript. Several transcripts exhibited a preference for monosomes or polysomes that switched depending on the compartment; these mRNAs encoded some synaptic plasticity–related proteins. Overall, neuropil transcripts exhibited a preference for monosome translation. Monosome-preferring transcripts encoded a full range of low- to high-abundance proteins in the neuropil.

CONCLUSION

In this study, we investigated the translational landscape in neuronal processes and identified local translation on 80S monosomes as an important source of synaptic proteins. Neuropil-localized transcripts exhibited a greater monosome preference than somatic transcripts, potentially allowing for the production of a more diverse set of proteins from a limited pool of available ribosomes at synapses. This finding thus bridges the gap between the relative paucity of visualized translational machinery in neuronal processes and actual measurements of local translation. Given the spatial limitations within dendritic spines and axonal boutons, synaptic activity could also regulate monosome translation to diversify the local proteome with spatial and temporal precision.

Monosomes translate synaptic mRNAs in the neuropil.

(A) Polysome profiling followed by monosome (cyan) or polysome (orange) footprinting (Ribo-seq) in microdissected somata (enriched in cell bodies) or neuropil (enriched in dendrites and/or axons) from rodent brain slices. (B) Transcripts localized to dendrites and/or axons were predominantly associated with monosomes. (C) Monosomes were in the process of active polypeptide elongation. (D) Neuropil monosome-preferring transcripts (cyan) often encoded synaptic proteins.

Abstract

To accommodate their complex morphology, neurons localize messenger RNAs (mRNAs) and ribosomes near synapses to produce proteins locally. However, a relative paucity of polysomes (considered the active sites of translation) detected in electron micrographs of neuronal processes has suggested a limited capacity for local protein synthesis. In this study, we used polysome profiling together with ribosome footprinting of microdissected rodent synaptic regions to reveal a surprisingly high number of dendritic and/or axonal transcripts preferentially associated with monosomes (single ribosomes). Furthermore, the neuronal monosomes were in the process of active protein synthesis. Most mRNAs showed a similar translational status in the cell bodies and neurites, but some transcripts exhibited differential ribosome occupancy in the compartments. Monosome-preferring transcripts often encoded high-abundance synaptic proteins. Thus, monosome translation contributes to the local neuronal proteome.

RNA sequencing (RNA-seq) and in situ hybridization have revealed the presence of an unexpectedly high number of RNA species in the CA1 neuropil (a region enriched in neuronal axons and dendrites) (1, 2), and many studies have documented the local translation of proteins in dendrites and/or axons (35). During messenger RNA (mRNA) translation, multiple ribosomes can occupy an individual mRNA (a complex known as a polysome), resulting in the generation of multiple copies of the encoded protein. Polysomes, usually recognized in electron micrographs as a cluster comprising three or more ribosomes, have been detected in neuronal dendrites (6, 7). Given the diversity of mRNAs present, polysomes are relatively infrequent in dendrites and axons [e.g., <0.5 polysomes per micrometer (7)]. In neuronal processes, the features and mechanisms of translation have not been explored in detail, partly because of the relative inaccessibility of the dendrites and axons in the neuropil. In particular, it is not clear how diverse proteins might be synthesized from a limited population of polysomes present in small synaptic volumes.

Monosomes are the predominant ribosome population in neuronal processes

To visualize the capacity for protein synthesis in the neuropil in vivo, we labeled the de novo proteome using puromycylation (8). We infused puromycin directly into the lateral ventricle of mice, waited 10 min, and then visualized newly synthesized proteins in hippocampal pyramidal neurons by coimmunofluorescence labeling of nascent protein (anti-puromycin antibody) and CA1 pyramidal neurons (anti-wolframin antibody; Wfs1). As expected, we detected an intense nascent protein signal in the somata layer (stratum pyramidale), comprising the cell bodies of pyramidal neurons (Fig. 1A and fig. S1A). Additionally, strong nascent protein was evident throughout the dendrites of pyramidal neurons in the neuropil (stratum radiatum) (Fig. 1A and fig. S1A). Co-injection of a protein synthesis inhibitor (anisomycin) abolished the nascent protein signal. Because of the very short window of metabolic labeling, these data indicate that protein synthesis also occurs in dendrites in vivo.

Fig. 1 Monosomes are the major ribosome population in neuronal processes.

(A) Immunofluorescence labeling of the nascent protein metabolic label (cyan) and the CA1 pyramidal neuron marker Wfs1 (purple) in hippocampal sections from mice that received a brief infusion of puromycin without (left) or with the protein synthesis inhibitor anisomycin (right) into the lateral ventricle. Scale bar, 20 μm. A higher-magnification image of the nascent protein signal in the boxed dendritic region is shown. Scale bar, 50 μm. so, stratum oriens; sp, stratum pyramidale; sr, stratum radiatum; slm, stratum lacunosum moleculare. (B) Scheme of a hippocampal slice showing the regions (somata and neuropil) that were microdissected for subsequent polysome profiling. Representative polysome profiles (C and D) and comparison of the monosome/polysome (M/P) ratios (E) of the microdissected somata (Soma, blue; M/P = 0.30 ± 0.03) or neuropil (Npl, purple; M/P = 0.76 ± 0.19) (n = 7 biological replicates). Areas measured to calculate the M/P ratios are shaded (see Materials and methods). ***P ≤ 0.001, Welch’s t test. AU, arbitrary units. (F) Scheme showing cortical neurons grown on a microporous membrane enabling the separation of cell bodies and neurites for polysome profiling. Representative polysome profiles (G and H) and M/P ratios (I) of the cell body (CB, blue) or neurite layer (N, purple) (n = 4 replicates). Areas measured to calculate the M/P ratios are shaded. *P ≤ 0.05, Welch’s t test. Error bars in (E) and (I) indicate SDs.

Polysome profiling is a biochemical fractionation method that allows one to examine the degree of ribosome association of a transcript—that is, association with a monosome (single ribosome) or a polysome (multiple ribosomes loaded on an mRNA) (9). Using polysome profiling, we examined the ribosome occupancy of transcripts in the hippocampus by comparing somata and neuropils that were microdissected from ex vivo adult rat hippocampal slices (area CA1) (Fig. 1B). Immunoblot analysis confirmed that the microdissected neuropil was strongly de-enriched for neuronal cell bodies (fig. S1, B and C). We obtained a typical polysome profile with two ribosomal subunit peaks (40S and 60S), one monosome (single ribosome, 80S) peak, and multiple polysome peaks. No signs of altered polysome integrity (such as a shift toward lower ribosome occupancy) were observed. We assessed the relative association of transcripts with monosomes or polysomes [monosome/polysome (M/P) ratio; see Materials and methods section] in the somata and neuropil by measuring the area under the curve (AUC) of the corresponding absorbance peaks. Although a large proportion of mRNAs was associated with polysomes in the somata (Fig. 1C), the M/P ratio was more than twice as high in the neuropil (Fig. 1, D and E). The increased M/P ratio observed in the neuropil resulted from a decrease in polysome abundance when compared to the somata (fig. S1D). As expected, the M/P ratios in whole (nonmicrodissected) hippocampi (0.56 ± 0.04), comprising cell bodies and neuronal processes, occupied a position between the values obtained for the somata (0.30 ± 0.03) and neuropil (0.76 ± 0.19) (fig. S1E and Fig. 1, C and D), confirming that the microdissection procedure did not disrupt polysome stability.

To confirm the difference in the M/P ratios between somata and neuronal projections, we used a well-established in vitro system to enrich for cell bodies and neuronal processes (10). Rat neurons were cultured on microporous membranes, allowing the dendrites and axons (but not cell bodies) to extend to the area beneath the membrane (Fig. 1F and fig. S1, F to I). After 21 days in vitro, we harvested the cell bodies and dendrites and/or axons separately and again conducted polysome profiling. Consistent with the microdissected slice data, the M/P ratio was significantly higher in neurites than in cell bodies, owing to an increase and a decrease in the number of monosomes and polysomes in the neurite layer, respectively (Fig. 1, G to I, and fig. S1J).

Monosomes actively elongate transcripts in the synaptic neuropil

In mammalian cells, polysomes are thought to represent the translationally active ribosome population (1113). By contrast, monosomes, reflecting single ribosomes detected on transcripts, are presumed to represent the isolation of protein synthesis initiation and termination events but not active protein synthesis (i.e., the elongation of the polypeptide chain) (1113). We compared the translational status of somatic or neuropil-localized monosomes and polysomes by using ribosome profiling to precisely map the position of the ribosome(s) along the mRNA (14) (Fig. 2A). Monosomal or polysomal fractions from the rat neuropil or somata were collected; the purity of fractionation was independently demonstrated by the lack of polysome or monosome peaks on sucrose gradient profiles from isolated monosomal and polysomal fractions, respectively (fig. S2). After polysome profiling, ribosomal fractions were digested and monosome or polysome footprint libraries were prepared. After sequencing three replicates of monosome and polysome footprint libraries and aligning the reads to a reference genome (alignment statistics shown in fig. S3A), the classical ribosome profiling quality metrics were assessed (fig. S3). As expected, the monosome and polysome footprints peaked at a length of ~31 nucleotides (representing the area occupied by the ribosome; fig. S3, B and C) and exhibited a depletion of read densities in the untranslated regions (UTRs) and introns (fig. S3, D and E). The ribosome profiling libraries were highly reproducible between replicates, as shown by the very small within-group variance (fig. S3F) and Pearson correlation coefficients >0.95 for the majority of the samples (fig. S3G). The neuropil monosome and polysome translatomes also overlapped with the previously published neuropil de novo proteome (15) (fig. S3H).

Fig. 2 Neuronal monosomes actively elongate transcripts in the neuropil.

(A) Experimental workflow. Somata or neuropil fractions were obtained, monosomes and polysomes were isolated by polysome profiling, and ribosome profiling was performed on isolated fractions. (B and C) Metagene analyses showing the footprint density throughout the transcript ORF in the neuropil monosomes (B) or polysomes (C). The average relative normalized coverage is plotted per nucleotide position, and the standard deviation is shaded (n = 3 replicates). Genes were individually normalized. (D) To assess the translational status of neuronal monosomes or polysomes, only reads classified as “neuronal” (fig. S5) were retained for further analysis. (E) Metagene analyses showing the P-site coverage of neuronal transcripts in the neuropil monosome sample. The average normalized coverage is plotted per nucleotide position around the 5′ end (start), central portion (center), and 3′ end (stop) of the ORF. The standard deviation is shaded (n = 3 replicates).

We examined the positions of the RNA footprints obtained from neuropil monosomes (Fig. 2B) or polysomes (Fig. 2C) across the open reading frame (ORF) of transcripts. Both the monosome and polysome footprint coverages peaked at the 5′ ORF (near or at the translation initiation site); monosome footprints decreased more sharply than polysome footprints over the first 25% of the ORF before reaching a plateau. Only the monosome sample exhibited a pronounced enrichment of footprint reads around the stop codon, presumably reflecting the position of terminating ribosomes. This pattern is in good agreement with previously published metagene analyses of monosome and polysome footprint densities in yeast (16), thus confirming the purity of isolated monosomal and polysomal fractions. Surprisingly, however, a large fraction of monosome footprints occupied the center of the ORF, demonstrating that the localized monosomes were engaged in peptide elongation. A similar pattern was evident for the monosome (and polysome) footprint coverage in the somata (fig. S4, A and B) and the whole (nonmicrodissected) hippocampus (fig. S4, C and D; representative polysome profile shown in fig. S1E). Thus, the mid-ORF monosome footprints were not a result of altered polysome integrity during the microdissection procedure.

Because the somata and neuropil include neurons as well as glia and interneurons (fig. S5A), we developed a strategy to investigate the translational status of monosomes and polysomes in hippocampal excitatory neurons. We identified the translatome (ribosome-associated mRNAs) of select hippocampal excitatory neuron populations by combining RiboTag immunoprecipitation (RiboTag-IP) (17) with RNA-seq (fig. S5, B and C). Using differential expression analysis (18), we identified transcripts enriched in the RiboTag-IP from hippocampi of Camk2Cre::RiboTag mice (fig. S5, D and E) or microdissected somata (fig. S5, D and F) and neuropils (fig. S5, D and G) of Wfs1Cre::RiboTag mice (19). Combining the three datasets, we obtained a comprehensive list of 5069 mRNAs (neuronal transcripts) selectively translated in cell bodies and processes of excitatory hippocampal neurons (fig. S5H). The relative enrichment and de-enrichment of neuronal genes and glia- and/or interneuron-related genes, respectively, were validated using a previously published dataset (fig. S5I) (20). These data were used to obtain a filtered list of neuronal footprint reads in monosome or polysome libraries from the somata and neuropil (Fig. 2D). Again, a significant fraction of neuronal transcripts displayed coverage in the elongating portion of the ORF in the monosome and the polysome samples of both neuropil (fig. S6, A and B) and somata (fig. S6, C and D). The neuropil-derived monosome (Fig. 2E) and polysome (fig. S6E) footprints exhibited three-nucleotide phasing throughout the ORF, reflecting the characteristic codon-by-codon translocation of the ribosome on its mRNA (14). Thus, both monosomes and polysomes contribute to the active elongation of transcripts localized to neuronal processes.

Neuropil monosomes translate synaptic transcripts

To measure the degree to which a neuropil-localized transcript is translated by monosomes or polysomes, we focused on ribosomes that were undergoing elongation but not initiation or termination by using footprints aligned to the center of the ORF (see Materials and methods) in the monosome and polysome footprint libraries. Using differential expression analysis (18), we identified localized neuronal transcripts preferentially translated by either monosomes or polysomes. In the neuropil, we found 463 transcripts significantly enriched in the monosome fraction versus 372 transcripts enriched in the polysome fraction (Fig. 3A and table S1). By contrast, a greater number of transcripts exhibited a significant enrichment on polysomes in the somata (fig. S7A and table S1). When we examined the neuropil footprint pattern across individual transcripts, we identified transcripts that displayed increased monosome (e.g., Kif1a; Fig. 3B) or polysome (e.g., Camk2a; Fig. 3C) footprint coverage throughout the entire ORF. There was also a large proportion of transcripts (e.g., Slc17a7; Fig. 3D) that exhibited equal coverage in monosome and polysome footprint libraries.

Fig. 3 Monosomes predominate on a subset of transcripts in dendrites and axons.

(A) MA plot showing transcripts with significantly enriched monosome (cyan) or polysome (orange) footprint coverage in the central portion of the ORF (region spanning 15 codons from the start site to 5 codons before the stop site). DESeq2 was used for analysis, with a threshold of 0.05 on the adjusted P value (see Materials and methods). Gray dots denote nonenriched transcripts. (B to D) Genome browser views representing the average monosome (top) or polysome (bottom) footprint coverage for three transcripts: Kif1a (B), Camk2a (C), and Slc17a7 (D). The y axis indicates the number of normalized reads. (E) Metagene analysis showing the monosome P-site coverage of transcripts that exhibit significant monosome enrichment in the neuropil. The average normalized coverage is plotted per nucleotide position around the 5′ end (start), central portion (center), and 3′ end (stop) of the ORF. The standard deviation is shaded (n = 3 replicates). The inset shows the observed (obs)–to–expected (exp) ratio of the footprint distribution in different reading frames. P = 2.26 × 10−4, ANOVA. (F) A pause score was computed for each codon located in the elongating ORF portion of the 463 monosome-enriched transcripts: pause score (z-score–like) = (normalized footprint coverage in monosome library – normalized footprint coverage in polysome library)/(normalized footprint coverage in polysome library)1/2 (n = 3 replicates). The graph shows the fraction of codons per pause score. Dashed lines highlight pause score values of ±1.96 (P = 0.05), values between these lines represent codons exhibiting similar coverage in monosome and polysome libraries.

Footprints from monosome-enriched mRNAs exhibited strong three-nucleotide periodicity, reflecting the stepwise movement of active individual ribosomes during the elongation of this transcript subset (Fig. 3E). During translation elongation, however, ribosomes can pause as a result of local RNA structures, the presence of rare codons, interactions between nascent chains, or association with trans-regulatory factors (2124). The predominant association of an mRNA with monosomes could thus result from increased pausing at individual codons when compared with the same mRNA’s association with polysomes. To test this for the 463 monosome-enriched transcripts, we computed a pause score by comparing the normalized footprint coverage at individual codons in the monosome and polysome samples (see Materials and methods). We found that most codons did not exhibit significant differences in pausing between the monosome and polysome libraries (Fig. 3F). To further investigate the translational activity status of monosome-preferring transcripts, we used harringtonine (an initiation inhibitor) to analyze a time series of ribosome run-off during elongation (25) in hippocampal cultures. Metagene analysis revealed a progressive loss of ribosomes from the 5′ end of monosome-preferring transcripts after the harringtonine treatments (fig. S8). Thus, monosome-preferring transcripts are actively elongated and do not exhibit differential pausing when associated with single or multiple ribosomes.

What transcript properties influence the neuropil M/P preference? We detected a positive correlation between the neuropil M/P preference and ORF length, 3′UTR folding energy, and 5′UTR length (fig. S9A and table S1). On the other hand, a negative correlation was observed between the M/P ratio and the mean of the typical decoding rate (MTDR) index [an estimate of the elongation efficiency (26)], GC content, codon adaption index (CAI), and initiation rate (fig. S9A and table S1). We also observed an overrepresentation of upstream ORF-containing transcripts (73 mRNAs) among monosome-enriched genes (fig. S9B). Although a previous study in yeast reported that monosomes occupy nonsense-mediated mRNA decay (NMD) targets (27), no relationship was found between the neuropil M/P preference of transcripts and their likelihood of classification as NMD targets (fig. S9C). The fine-tuning of translation rates may allow for the optimization of the nascent polypeptide folding during protein synthesis (28, 29). We thus explored how the M/P preference related to the structural complexity of the encoded polypeptide. Indeed, an increased number of secondary structures (α helix and β strand) were predicted for monosome-preferring transcripts (fig. S9D). Furthermore, monosome-preferring transcripts encoded proteins displaying longer structural domains (fig. S9E).

To examine whether particular protein function groups are encoded by monosome- versus polysome-preferring transcripts in the neuropil, we used gene ontology (GO) (Fig. 4A; see fig. S7B for the somata). Monosome-preferring transcripts exhibited a more significant association with GO terms such as “synapse,” “vesicle,” or “dendritic tree” than did polysome-preferring transcripts in the neuropil. In accordance with this finding, synaptic genes [SynGO annotation (30)] displayed higher mean M/P ratios compared with nonsynaptic genes (fig. S10, A and B). Polysome-preferring transcripts often encoded proteins involved in actin cytoskeleton remodeling (Fig. 4, A and B). Because functional and morphological changes in synapses rely on the dynamic actin cytoskeleton remodeling (31), polysome translation may be required to supply synapses with high copy numbers of cytoskeletal proteins. Thus, in dendrites and axons, a significant proportion of transcripts important for synaptic function are principally translated by monosomes.

Fig. 4 Monosomes translate many key synaptic transcripts in dendrites and axons.

(A) GO terms representing the top 10 significantly enriched protein function groups for monosome-enriched (cyan) or polysome-enriched (orange) transcripts. (B) Scheme of pre- and postsynaptic compartments highlighting some of the transcripts preferentially translated by monosomes (cyan) or polysomes (orange). Asterisks denote key synaptic components that were manually added, owing to their exclusion by the excitatory neuron-specific filter. (See table S1 for information about the fold changes.)

Dissecting neuronal monosome translation

To address whether the M/P preference is intrinsic to the transcript or influenced by its environment (i.e., its subcellular localization), we compared the relative M/P enrichment of each transcript in the neuropil and somata. We observed a high correlation (coefficient of determination R2 = 0.6) between the somata and neuropil M/P ratios, indicating that a large proportion of transcripts prefer the same type of ribosome occupancy in both compartments (Fig. 5A, monosome-enriched in quadrant I or polysome-enriched in quadrant III). An overlap of the genes classified as monosome- or polysome-preferring in the somata (fig. S7A) and neuropil (Fig. 3A) revealed that many but not all genes exhibited a similar preference between compartments (fig. S11, A and B). Using DESeq2 (18), we identified transcripts that exhibited significant differences in the M/P ratio between somata and neuropil (Fig. 5A and table S1). Only a handful of transcripts (e.g., Arc; Fig. 5B) exhibited a significantly lower M/P ratio in the neuropil than in the somata (Fig. 5A, purple dots, and Fig. 5C). The majority of transcripts (e.g., Serpini1; Fig. 5D) with differential ribosome occupancy between compartments displayed significantly elevated M/P fold changes in the neuropil (Fig. 5A, cyan dots, and Fig. 5E). Overall, we observed a shift toward a higher monosome preference in the neuropil (Fig. 5F and fig. S11C), which could result, at least in part, from a lower ribosome abundance in the neuropil than in the somata (fig. S12, A and B).

Fig. 5 Localization influences the translational status of selective transcripts.

(A) M/P log2 fold changes (FC) in the neuropil (y axis) versus the somata (x axis). The majority of transcripts exhibited correlated (R2 = 0.6, P < 2.2 × 10−16) M/P enrichments between both compartments. Colored dots highlight transcripts that exhibit significantly increased (cyan, n = 136 replicates) or decreased (purple, n = 36 replicates) M/P log2 fold changes in the neuropil compared with the somata. DESeq2 was used for analysis, with a threshold of 0.05 on the adjusted P value (see Materials and methods). Numerals represent the different quadrants. Gray dots denote transcripts that do not exhibit significant changes in M/P log2 fold changes between compartments. (B and C) Example (Arc) (B) and cumulative distribution frequency of the M/P log2 fold changes (C) of transcripts exhibiting significantly higher M/P ratios in the somata (purple) compared with the neuropil (dark purple). P = 6.128 × 10−5, Kolmogorov-Smirnov test. ecdf, empirical cumulative distribution function. (D and E) Example (Serpini1) (D) and cumulative distribution frequency of the M/P log2 fold changes (E) of transcripts exhibiting significantly higher M/P ratios in the neuropil (dark cyan) compared to the somata (cyan). P = 9.215 × 10−15, Kolmogorov-Smirnov test. (F) Cumulative distribution frequency depicting the M/P log2 fold changes of all neuronal genes [cyan, purple, and gray dots in (A)] in the somata (red) and neuropil (blue), indicating an overall tendency toward higher M/P ratios in the neuropil. P = 1.692 × 10−8, Kolmogorov-Smirnov test.

Notably, we also identified some transcripts with opposing M/P ratios between somata and neuropil (i.e., monosome-preferring in one compartment and polysome-preferring in the other), some of which were key regulators of synaptic plasticity (Fig. 5A, quadrants II and IV; fig. S12, C and D; and table S1) (3241). Thus, neuropil-localized transcripts are, in general, more likely to be translated on monosomes than somatic transcripts.

Monosome translation contributes to the neuropil proteome

Individual synapses are small independent information processing units, each endowed with their own complement of proteins, ranging in copy numbers from tens to a thousand or so (42, 43). We observed that previously published protein copy numbers in the rat presynapse (Fig. 6A) (43) and postsynapse (Fig. 6B) (42) were poorly correlated with the neuropil M/P preference. To understand the contribution of monosome and polysome translation to the overall proteome composition, we conducted mass spectrometry of neuropil proteins (see Materials and methods) and estimated their absolute protein abundances using iBAQ (intensity-based absolute quantification) (44) (fig. S13A). As might be expected, we observed higher median iBAQ values for proteins encoded by polysome-preferring transcripts when compared with proteins encoded by monosome-preferring transcripts (Fig. 6C; see fig. S14A for the somata). When we examined the relationship between the abundance of neuropil proteins and their respective M/P ratios, however, we observed a surprisingly weak correlation (R2 = 0.021; P value = 2.944 × 10−11; Fig. 6D; see fig. S14B for the somata). Around half of the 326 proteins encoded by monosome-preferring transcripts exhibited protein abundances that were greater than the average (Fig. 6D and table S1), indicating that monosome-preferring transcripts can also encode highly abundant proteins. We next examined the properties of the high-abundance proteins encoded by monosome-preferring transcripts (“mono-high”; n = 177). To investigate whether the mono-high protein abundance is related to mRNA abundance in the neuropil, we used RNA-seq to estimate local transcript levels (fig. S13B). Consistent with the correlation between the local transcriptome and proteome (R2 = 0.26; P value < 2.2 × 10−16), the mono-high genes had higher mRNA levels (Fig. 6E; see fig. S14C for the somata). We then looked at the relationship between mono-high protein abundance and local translation rates, a measurement obtained from neuropil total footprint libraries (without biochemical fractionation) (fig. S13C). Perhaps predictably, we observed that mono-high transcripts were among the most highly translated mRNAs within the neuropil, which agrees with the overall positive correlation between the neuropil proteome and local translatome (R2 = 0.33; P value < 2.2 × 10−16) (Fig. 6F; see fig. S14D for the somata). Thus, monosome-translated transcripts can contribute to the neuropil proteome composition by encoding a full range of low- and high-abundance proteins, depending on their expression level and translation rate.

Fig. 6 Monosome translation can contribute to the maintenance of the local proteome.

(A and B) M/P fold changes in the neuropil were not correlated with the copy numbers of some key presynaptic (43) (A) and postsynaptic proteins (42) (B). Regression lines and corresponding adjusted R2 values are represented (presynapse P = 0.1488, postsynapse P = 0.07145). (C) Box plots of protein (log2 iBAQ) measurements in the neuropil for monosome-enriched (mono, cyan) or polysome-enriched (poly, orange) genes. P = 2.735 × 10−6, Wilcoxon rank-sum test. Of 463 and 372 monosome- and polysome-preferring transcripts in the neuropil, 326 and 242, respectively, passed the stringent proteomics filtering criteria (see Materials and methods). (D) Scatter plot of the protein abundance (log2 iBAQ) versus M/P fold changes for monosome-enriched (cyan), polysome-enriched (orange), and nonenriched (gray) genes (R2 = 0.021, P = 2.944 × 10−11). The dashed line indicates the mean log2 iBAQ value. Monosome-preferring transcripts encoding proteins with abundances greater than average are highlighted by dark cyan dots (mono-high). (E and F) The local proteome correlates with the local transcriptome and translatome. Scatter plots of the protein abundance (log2 iBAQ) versus RNA (log2 TPM) (R2 = 0.26, P < 2.2 × 10−16) (E) and translation rate (obtained from total footprints, without biochemical fractionation) (R2 = 0.33, P < 2.2 × 10−16) (F) measurements for all neuronal genes are shown. Monosome-preferring genes encoding high-abundance proteins are highlighted by dark cyan dots.

Discussion

In this work, we investigated the translational landscape in neuronal processes and identified local translation on 80S monosomes as an essential source of synaptic proteins. To date, knowledge about the conformation of the translational machinery near synapses has originated primarily from electron micrographs. In these studies ribosomes are unambiguously identified when organized as a polysome cluster formed by more than three ribosomes (45). The sparse distribution of polysomes in dendrites and spines apparent in electron micrographs has led some to suggest that local protein synthesis represents a minor source of synaptic protein under basal conditions (46). Indeed, until the recent detection of mRNAs and the machinery needed for their translation (5, 47, 48), the inability to identify polysomes in electron microscopy (EM) images from mature axons led to assertions that mature axons obtain protein exclusively via intracellular transport from the soma. Although a previous EM study suggested the putative visualization of monosomes in dendritic spines (45, 49), monosomes have not been identified with certainty, because their small size (10 to 25 nm) makes it difficult to distinguish them from other dark-staining cytoplasmic particles (45). A previous study using a fluorescent reporter suggested that monosome translation might be associated with sporadic (isolated) translation events in cultured neuron processes (50). In this study, we detected substantial levels of ongoing protein synthesis in the synaptic neuropil in vivo, and here we provide direct evidence for the preferential translation of many pre- and postsynaptic transcripts by monosomes. This finding thus bridges the gap between the relative paucity of visualized translational machinery in neuronal processes and actual measurements of local translation.

Dendritic spines and their associated presynaptic boutons that comprise the excitatory synapse are small subcellular compartments, often <100 nm3 for spines (51). The relatively large dimensions of a polysome [~100 to 200 nm (7)] limit the possibilities for high ribosome occupancy in spines and axon terminals. Indeed, each dendritic spine has been estimated to contain, on average, one polyribosome (52). The observed low density of polysomes at synapses could be due to a limited pool of available ribosomes in neuronal processes compared with cell bodies. In agreement with this concept, we observed a decrease in the percentage of ribosomal RNA (rRNA) relative to total RNA as well as a de-enrichment of ribosomal proteins in the neuropil compared with the somata. Translation via smaller machines (i.e., monosomes) allows for more protein synthesis sites within synaptic compartments. Polysomes have been reported to move within cells at an average speed of 2 μm/s (53), and potentially greater mobility of translating monosomes may allow them to patrol and serve a larger number of synapses. Given that one polysome translates a single mRNA resulting in multiple copies of a single protein, the relative scarcity of ribosomes imposes constraints on both the timing and diversity of locally synthesized proteins. Neuropil-localized transcripts exhibited a greater monosome preference than somatic transcripts, potentially allowing for the production of a more diverse set of proteins from a limited pool of available ribosomes at synapses.

Monosome-preferring transcripts encoded proteins that span a broad range of abundances in the neuropil. Because many synaptic proteins are present at very low copy numbers within the pre- and postsynaptic compartments [e.g., AMPA receptors; estimated ~15 to 20 per postsynaptic density (PSD)] (54), their local translation by single ribosomes may suffice to maintain or even alter synaptic activity. We also uncovered a subset of monosome-preferring transcripts that encode surprisingly high-abundance proteins, including the scaffolding proteins Bsn and Dlg3. This subset also exhibited increased RNA levels and translation rates within the neuropil. These features might underlie the ability of these monosome-preferring transcripts to encode abundant proteins. On the other hand, predominant polysome translation was observed for key signaling, scaffolding, or cytoskeletal proteins (e.g., Camk2a, PSD95, and actin), which are present at very high copy numbers within synapses (54). Many studies investigating translational control in synaptic plasticity or neurological disorders have focused their analysis on transcripts that cosediment with polysomes (9, 5557). Given that monosomes are key contributors to the neuronal translatome, focusing on polysome-associated transcripts may provide an incomplete picture of translational regulation.

Most transcripts exhibited a similar M/P preference in both the somata and neuropil, suggesting that ribosome occupancy is often an intrinsic feature of the transcript. Consistent with this finding, we detected a positive correlation between M/P ratio and ORF length, in agreement with previous studies reporting decreased ribosome density and protein production for long ORFs (5860). In part, this observation can be explained by reduced initiation rates of longer transcripts [correlation coefficient r = −0.29; P value < 2.2 × 10−16; see also (61)]. Contrasting observations, however, have been made in yeast, where monosomes preferentially occupy short ORFs (27). This discrepancy might be explained by differences in the translational regulatory mode between organisms, such as an expansion in the UTR length and/or complexity during evolution from lower to higher eukaryotes (2, 62, 63).

We also observed that monosome-preferring transcripts were often subject to a negative translational regulation, with moderate initiation and elongation kinetics. Notably, proteins predominantly encoded by monosome-preferring transcripts were not only longer but also structurally more complex. A “quality mode” slow translation of the monosome-preferring transcripts might allow the fine-tuning of cotranslational folding events, ensuring the functionality and preventing the aggregation of the encoded proteins. On the other hand, we found that polysome-preferring transcripts displayed increased initiation and elongation rates, allowing a more efficient translation. Polysome-preferring transcripts may thus encode proteins of lower structural complexity, which require less de novo protein folding fidelity, potentially allowing their translation in a fast “productivity mode” (28, 29).

Some transcripts exhibited a differential M/P preference between the somata and neuropil. Neurons differentially localize 5′ and/or 3′ UTR isoforms between subcellular compartments (64). Because these cis-regulatory mRNA elements regulate initiation efficiency (62, 63), neurons may fine-tune their M/P preference through selective targeting of competitive UTR isoforms between compartments. Notably, Arc, a previously reported natural NMD target that contains 3′UTR introns (65), was monosome-preferring in the somata but polysome-preferring in the neuropil. According to the model proposed by Giorgi et al. (65), Arc may be silenced by NMD in the somata, whereas in the neuropil, synaptic activity could trigger its release from NMD, resulting in a translational up-regulation (i.e., polysome translation).

Alternatively, differences in the monosome preference between somata and neuropil could also arise from differential localization or activity of specific translational regulators, including RNA-binding proteins (RBPs) (66, 67), microRNAs (68, 69), initiation and elongation factors (57), or the ribosome itself (70). For instance, the RBP FMRP is thought to inhibit the translation of selective transcripts in neuronal processes by pausing the translocation of polysomes or by directly interacting with the RNA-induced silencing complex (71, 72). Synaptic activity has also been reported to regulate the local translational machinery through changes in the phosphorylation status of initiation and elongation factors (57). Thus, local activity-induced signaling events may also control the flow of ribosomes on an mRNA and dictate its M/P preference.

A rapid up-regulation in the number of polysomes has been observed in electron micrographs of dendritic shafts and spines after synaptic plasticity induction (7). Our data show that, for many transcripts, monosome translation is the preferred mode of protein synthesis in neuronal processes and presumably satisfies the local demands under basal conditions. The formation of polysomes, however, could be required to supply synapses with de novo plasticity-related proteins in response to stimulation. We identified transcripts that prefer the predominant ribosome population present in either somata (polysomes) or neuropil (monosomes) and thus represent candidates that may shift toward higher polysome occupancy in response to synaptic stimulation. Furthermore, given the spatial limitations within dendritic spines and axonal boutons, synaptic activity could also regulate monosome translation to diversify the local proteome with spatial and temporal precision.

Materials and methods

Experimental procedures

Animals

Homozygous RiboTag Rpl22HA/HA mice (The Jackson Laboratory, 011029) were crossed with Camk2Cre (The Jackson Laboratory, 005359) or Wfs1CreERT mice (The Jackson Laboratory, 009103). Male 8-week-old C57Bl/6, Wfs1CreERT::RiboTag, and Camk2Cre::RiboTag mice were housed in standard cages and fed standard lab chow and water ad libitum. Male Wfs1CreERT::RiboTag mice were treated for 3 days with tamoxifen [100 mg/kg, intraperitoneally (i.p.), Sigma], dissolved in sunflower oil/ethanol (10/1) to a final concentration of 10 mg/ml and used 1 week later for immunostaining or IP studies (19).

Adult male 4-week-old Sprague Dawley SPF (specific pathogen–free; Charles River Laboratories) rats were housed on a 12/12-hour light/dark cycle with food and water ad libitum until euthanasia. Timed pregnant SPF (Charles River Laboratories) females were housed in the institute’s animal facility for 1 week on a 12/12-hour light/dark cycle with food and water ad libitum until the litter was born. Cultured neurons were derived from P0 (postnatal day 0) Sprague-Dawley rat pups (CD Crl:CD, both male and female, RRID: RGD 734476). Pups were euthanized by decapitation.

The housing and euthanasia procedures involving animal treatment and care were conducted in conformity with the institutional guidelines that are in compliance with national and international laws and policies (DIRECTIVE 2010/63/EU; German animal welfare law; FELASA guidelines). The animals were euthanized according to annex 2 of § 2 Abs. 2 Tierschutz-Versuchstier-Verordnung. Animal numbers were reported to the local authority (Regierungspräsidium Darmstadt, approval numbers: V54-19c20/15-F126/1020 and V54-19c20/15-F126/1023).

Hippocampal tissue collection and microdissection

After sacrifice, the heads of 4-week-old male rats (for polysome/ribosome profiling experiments) or 8-week-old male mice [for translating ribosome IP (RiboTag) experiments] were immediately immersed in liquid nitrogen for 6 s to cool the brains. The brains were removed and the hippocampi were rapidly dissected on an ice-cooled disk. Hippocampal slices (500 μm) were prepared in a drop of ice-cold RNase-free phosphate-buffered saline (PBS) containing 100 μg/ml cycloheximide using a manual tissue slicer (Stoelting). Each slice was immediately passed to a second experimenter who microdissected the CA1 somatic and the neuropil layer at 0° to 4°C on a cold plate (TCP50, Thermoelectrics) in a drop of ice-cold RNase-free PBS containing 100 μg/ml cycloheximide. To ensure the purity of the microdissected neuropil, only slices located in the middle portion of the dorso-ventral axis of the hippocampus were used (approximately six slices per hippocampus). Somata and neuropil sections were immediately snap-frozen after their dissection and kept at −80°C until lysis. The microdissection procedure described here maintained the polysome integrity in the somata and neuropil regions. By contrast, signs of ribosome run-off were observed when the microdissection was carried out after 1 hour of slice recovery in artificial cerebrospinal fluid (1, 2).

Primary hippocampal and cortical cultures

Dissociated rat hippocampal or cortical neurons were prepared from P0 day-old rat pups, as previously described (73). For hippocampal cultures, neurons were plated at a density of 31.250 cells/cm2 onto 100-mm culture dishes and cultured for 21 days in vitro (DIV) in preconditioned growth medium (Neurobasal-A supplemented with B27 and GlutaMAX, 30% glia-culture supernatant, 15% cortex-culture supernatant). Cortical neurons were plated at a density of 100.000 cells/cm2 onto poly-d-lysine-coated 75-mm, 3-μm-pore polycarbonate membrane culture inserts (Corning 3420). At 1 DIV, AraC was added to a final concentration of 5 μM. After 2 days, medium was exchanged to preconditioned growth medium and neurons were cultured until 21 DIV. All cultures were maintained in a humidified incubator at 37°C and 5% CO2. The sex of animals from which the cells were obtained was not determined.

Run-off experiment in primary hippocampal culture

At 24 hours before drug treatment, cell medium was adjusted to 8 ml per dish. Harringtonine (LKT Laboratories) was added to a final concentration of 2 μg/μl from a 2-mg/ml stock in 100% ethanol. Cells were returned to the incubator at 37°C for 30 or 90 s. Cycloheximide was added to a final concentration of 100 μg/ml from a stock of 50 mg/ml in 100% ethanol. After drug addition, cells were returned in the incubator at 37°C for 1 min. After the incubation with cycloheximide, the cells were immediately placed on ice and washed twice with ice-cold PBS plus 100 μg/ml cycloheximide and lysed in polysome lysis buffer as described below. Total footprint libraries were prepared as described below.

Immunolabeling of cortical neurons cultured on membrane inserts

At 21 DIV, a part of the membrane was excised, briefly submerged in PBS, and fixed for 20 min in PFA (4% paraformaldehyde in PBS pH 7.5). Cells were permeabilized with 0.5% Triton X-100 in PBS supplemented with 4% goat serum for 15 min and blocked with blocking buffer (4% goat serum in PBS) for 1 hour. Dendrites were stained using an anti-MAP2 antibody (SySy 188004, 1/1000) in blocking buffer overnight at 4°C. After washing the cells three times for 5 min in PBS, the secondary antibody (ThermoFisher A488 A-11073, 1/1000) was incubated in blocking buffer for 45 min at room temperature. Cells were washed three times for 5 min in PBS with DAPI (4′,6-diamidino-2-phenylindole) added to the second wash. Membranes were mounted on glass slides using Aqua-Poly/Mount and imaged from the top (cell body layer) or bottom (neurite layer).

Tagged ribosome immunoprecipitation

Hemagglutinin (HA)–tagged ribosome IP of hippocampi from male Camk2Cre::RiboTag or somata/neuropil sections from male Wfs1Cre::RiboTag mice was performed as described previously (17, 74), with slight modifications. Tissue sections were homogenized in a glass homogenizer containing ice-cold RiboTag lysis buffer [50 mM Tris pH 7.4, 100 mM KCl, 12 mM MgCl2, 1% NP40, 1 mM DTT, 20 U/ml SUPERaseIN*RNase inhibitor (Ambion), 200 U/ml RNasin (Promega), 100 μg/ml cycloheximide, 10 U/ml TurboDNase, and protease inhibitor cocktail (Roche)]. After triturating the lysate 10 times using a 23-gauge syringe, samples were chilled on ice for 10 min and cleared by centrifugation at 16,100g for 10 min. Ten percent of the supernatant was kept as an input. HA IP was performed by incubation of the remaining supernatant with 5 μl of anti-HA antibody (abcam ab9110) overnight at 4°C with gentle rotation. Incubation of the samples with magnetic beads (Dynabeads protein G, Invitrogen), washes, and elution were performed according to (74). Total RNA was extracted from both the input and immunoprecipitated ribosome-mRNA complexes using the RNeasy MinElute kit (Qiagen). RNA integrity was assessed using the Agilent RNA 6000 Pico kit.

Lysate preparation for polysome and ribosome profiling

Tissue: Rat tissue samples were homogenized in polysome lysis buffer [20 mM Tris pH 7.5, 150 mM NaCl, 5 mM MgCl2, 24 U/ml TurboDNase, 100 μg/ml cycloheximide, 1 mM DTT, 1% Triton X-100, and protease inhibitor cocktail (Roche)] (25) by douncing in a glass homogenizer. For the experiments including RNase inhibitors, the polysome lysis buffer was supplemented with 200 U/ml RNase inhibitors (Promega). After triturating the lysate 10 times using a 23-gauge syringe, samples were chilled on ice for 10 min and cleared by two centrifugations at 16,100g for 6 min.

Neuronal culture: At 21 DIV, rat cortical primary neurons were washed twice in ice-cold PBS supplemented with 100 μg/ml cycloheximide. Neurons were collected with a scraper in polysome lysis buffer [20 mM Tris pH 7.5, 150 mM NaCl, 5 mM MgCl2, 24 U/ml TurboDNase, 100 μg/ml cycloheximide, 1 mM DTT, protease inhibitor cocktail (Roche), and 8% glycerol]. After scraping, the lysates were supplemented with Triton X100 to a final concentration of 1% and chilled on ice for 10 min. After triturating the lysates 10 times using a 23-gauge syringe, samples were chilled on ice for 10 min and then cleared by centrifugation at 16,100g for 10 min.

Polysome profiling

Samples were loaded onto 6-ml 10 to 50% sucrose density gradients that were prepared w/v in the following gradient buffer: 20 mM Tris pH 7.5, 150 mM NaCl, 5 mM MgCl2, 100 μg/ml cycloheximide, and 1 mM DTT. For polysome profiling from neuronal cultures, the gradient buffer was supplemented with 8% glycerol. To ensure proper RNase digestion during ribosome profiling on the sucrose gradient fractions, RNase inhibitors were omitted from the polysome lysis buffer. RNase-free reagents were used and samples were handled on ice during the entire procedure. The similarity of the neuropil polysome profiles in the presence or absence of RNase inhibitors indicated that this procedure did not affect RNA integrity (fig. S15). Gradients were centrifuged for 2 hours 45 min at 36,000 revolutions per minute (rpm) at 4°C in a SW41 Ti swing-out rotor. Polysome profiling was performed using a density gradient fractionation system (Brandel) with upward displacement and continuous monitoring at 254 nm using a UA-6 detector. The AUC of individual absorbance peaks was quantified. A M/P ratio was calculated by relating the monosome AUC to the sum of the AUCs of all polysome peaks. Somata and neuropil polysome profiles loaded with an equal amount of RNA were used for representation and the comparisons of the monosome or polysome AUC separately between compartments. Fractions of 125 μl corresponding to the monosome or the polysome peaks were collected and combined in a monosome and polysome pool, respectively.

Monosome and polysome footprint isolation

For the whole-hippocampus monosome and polysome footprinting, three replicates were used, each comprising the hippocampi from three rats, yielding ~150 μg of RNA. For the somata and neuropil monosome and polysome footprinting, three replicates were used, each comprising a pool of microdissected tissue from 55 rats, yielding ~110 μg of RNA. For each replicate, microdissected tissue was lysed as described above, and aliquots containing 20 or 10 μg of RNA were retained for total ribosome footprinting and total RNA-seq, respectively. The remaining lysate was loaded onto 10 to 50% sucrose gradients and centrifuged as described above. To prevent masking of the ribosome peaks by myelin (75), each replicate was loaded onto two to three gradients, and monosome or polysome fractions from different gradients were pooled after polysome profiling. A volume of M/P fraction containing 10 μg (hippocampi) or 2 to 5 μg (somata/neuropil) of RNA was diluted with gradient buffer and digested with 7.5 U/μg RNA of RNase I (Epicentre), rotating for 45 min at 24°C (a range of RNase I concentrations was tested beforehand to optimize the digestion conditions; table S1). Nuclease digestion reactions were promptly cooled and spun, and 10 μl of SUPERaseIN*RNase inhibitor was added. Samples were then layered onto a 34% sucrose cushion, prepared w/v in gradient buffer supplemented with 20 U/μl of SUPERaseIN*RNase inhibitor. 80S particles were pelleted by centrifugation in a SW55Ti rotor for 3 hours 30 min at 55,000 rpm at 4°C.

Total ribosome footprint isolation

Neuropil lysates from three biological replicates (see section “Monosome and polysome footprint isolation”) containing 20 μg of RNA were digested with 0.5 U/μg RNase I (Epicentre), shaking for 45 min at 400 rpm at 24°C (76). Nuclease digestion reactions were promptly cooled and spun, and 10 μl of SUPERaseIN*RNase inhibitor was added. Samples were then layered onto a 34% sucrose cushion, prepared w/v in gradient buffer supplemented with 20 U/μl of SUPERaseIN*RNase inhibitor. 80S particles were pelleted by centrifugation in a SW55Ti rotor for 3 hours 30 min at 55,000 rpm at 4°C.

Ribosome footprint library preparation

Footprint libraries were prepared according to (76) with the following modifications: After RNA extraction from the ribosomal pellet, rRNAs were depleted using the Ribo-Zero Magnetic Gold Mammalian kit (Illumina), followed by footprint purification using the RNA Clean & Concentrator-5 kit (Zymo). Footprint fragments were purified by polyacrylamide gel electrophoresis (PAGE) on a 15% tris-borate EDTA (TBE)–urea gel, and fragments from 26 to 34 nucleotides were isolated. After footprint dephosphorylation and linker ligation, the ligation reaction was depleted of unligated linker by incubation with 0.5 μl of 5′ yeast deadenylase 10 U/μl (NEB) and 0.5 μl of RecJ exonuclease 10 U/ul (Epicentre) for 45 min at 30°C. In addition, ligation products were purified by PAGE purification on a 15% TBE-urea gel. Reverse transcription was performed as described previously, with the following modification: The reverse transcription reaction was directly incubated with 2 μl of exonuclease I (NEB) at 37°C for 1 hour followed by 15 min at 80°C. cDNA was gel purified by PAGE on a 15% TBE-urea gel. After circularization, circDNA was submitted to an additional rRNA depletion using 14 custom biotinylated rat rRNA oligos (table S1) according to (77). After amplification, the libraries were run on an 8% nondenaturing TBE gel, and 160–base pair (bp) products were isolated and characterized using the Agilent High Sensitivity DNA assay. Libraries were sequenced on an Illumina NextSeq500, using a single-end, 52-bp run.

RNA isolation and library preparation

RNA was isolated from tissue lysates using the Direct-zol RNA micro prep kit (Zymo). RNA integrity was assessed using the Agilent RNA 6000 Nano kit. Rat neuropil total RNA-seq libraries were prepared starting from ~200 ng of total RNA using the TruSeq stranded total RNA library prep gold kit (Illumina). For the input/IP samples from Camk2Cre::RiboTag hippocampi or Wfs1Cre::RiboTag somata and neuropil, mRNA-seq libraries were prepared, starting from ~100 ng of total RNA, using the TruSeq stranded mRNA library prep kit (Illumina). Libraries were sequenced on an Illumina NextSeq500, using a single-end, 75-bp run.

rRNA–to–total RNA percentage

Total RNA was isolated from rat somata and neuropil (n = 4) as described above and measured using the Agilent RNA 6000 Nano kit. The ratio of rRNA to total RNA was obtained by summing the 18S rRNA and 28S rRNA percentages of total RNA calculated by the Agilent Bioanalyzer.

Immunoblotting

Neurite and cell body layers were collected in ice-cold PBS and centrifuged, and pellets were lysed in lysis buffer (1% Triton X-100, 0.5% SDS in PBS) supplemented with TurboDNase 24 U/ml at 70°C for 15 min. Lysates were cleared by centrifugation and stored at −80°C until use. Somata and neuropil lysates were prepared in polysome buffer. Lysates were resolved by SDS-PAGE in 4 to 12% Bis-Tris gels (Invitrogen) and analyzed by immunoblotting using far-red fluorescent dyes and a Licor Odyssey scanner [mouse anti-NeuN (1/1000, MAB377); rabbit anti-bActin (1/2000, ab8227); anti-mouse IR800 (1/5000, Licor); anti-rabbit IR680 (1/5000, Licor)]. Protein levels in bands of interest were quantified using ImageJ (NIH). Western blot normalization was conducted according to the Revert Total Protein Stain (Licor) manufacturer’s instructions.

Mass spectrometry data acquisition

Three replicates of rat neuropil were microdissected as described above. Tissue pieces were snap-frozen and kept at −80°C until lysis. Tissue pieces were lysed in 4% Chaps, 8 M urea, 0.2 M Tris HCl, and 1 M NaCl. All samples were digested, reduced, and alkylated according to a previously published filter-aided ample preparation protocol (78). Dried peptide pellets were stored at −20°C until liquid chromatography–tandem mass spectrometry (LC-MS/MS) analysis. Proteolytic digests were analyzed via Nano-LC-MS/MS on an Ultimate 3000 nanoUPLC (Thermo Fisher Scientific, Bremen) coupled to a Orbitrap Fusion Lumos (Thermo Fisher Scientific, Bremen).

After dissolving the dried peptides in 20 μl of 0.1% FA in 5% acetonitrile, samples were separated using an Acclaim pepmap C18 column (50 cm by 75 μm, particle size: 2 μm) after trapping on an Acclaim pepmap C18 precolumn (2 cm by 75 μm, particle size: 3 μm). Trapping was performed for 6 min with a flow rate of 6 μl/min using a loading buffer (98/2 water/acetonitrile with 0.05% triflouroacetic acid). Peptides were then eluted and separated on the analytical column at a flow rate of 300 nl/min with the following gradient: from 4 to 33% B in 150 min, 33 to 48% B in 20 min, 48 to 90% B in 1 min, and constant 90% for 13 min (buffer A: 0.1% FA in water, buffer B: 0.1% FA in 80/20 acetonitrile/water). All LC-MS–grade solvents were purchased from Honeywell/Riedel del Häen.

Peptides eluting from the column were ionized online using a Nano Flex ESI source and analyzed with an Orbitrap Fusion Lumos mass spectrometer in data-dependent mode. Survey scans were acquired over the mass/charge ratio range of 350 to 1400 in the Orbitrap (maximum injection time: 50 s, automatic gain control (AGC), fixed at 2 × 105 and R = 120,000) and sequence information was acquired by a “Top-Speed” method with a fixed cycle time of 2 s for the survey and after MS/MS scans. MS/MS scans were performed on the most abundant precursors exhibiting a charge state from two to five with an intensity minimum of 5 × 103. Selected precursors were isolated in the quadrupole at 1.4 Da and fragmented using higher-energy C-trap dissociation at normalized collision energy = 30%. For MS/MS, an AGC of 104 and a maximum injection time of 300 s were used. Resulting fragments were detected in the ion trap using the rapid scan mode. The dynamic exclusion was set to 30 s with a mass tolerance of 10 parts per million (ppm). All samples were measured in technical triplicates.

Intracerebroventricular puromycin administration

Mice (n = 3 per group) were anesthetized with isoflurane (induction: 4%, maintenance: 2%) in oxygen-enriched air (Oxymat 3, Weinmann, Hamburg, Germany) and fixed in a stereotaxic frame (Kopf Instruments, Tujunga, USA). Core body temperature was maintained at 37.5°C by a feedback-controlled heating pad (FHC, Bowdoinham, ME, USA). Analgesia was provided by local injection of ropivacain under the scalp (Naropin, AstraZeneca, Switzerland) and systemic injection of metamizol (100 mg/kg, i.p., Novalgin, Sanofi) and meloxicam (2 mg/kg, i.p., Metacam, Boehringer-Ingelheim, Ingelheim, Germany) (79). A stainless steel 26-gauge guide cannula (PlasticsOne, Roanoke, VA) was implanted vertically toward the right lateral ventricle (A/P: −0.22 mm, M/L: 1 mm, D/V: −2 mm). Guide cannulas were fixed onto the skull with instant adhesive (Ultra Gel, Henkel, Düsseldorf, Germany) and dental cement (Paladur, Heraeus, Hanau, Germany). An obturator was inserted into each guide cannula and remained in place until the drug infusion when it was removed and replaced with an injector that extended 0.5 mm beyond the tip of the guide cannula. After surgery recovery, 3 μl of puromycin solution (9 mg/ml, 10% DMSO/90% saline) or vehicle were infused for 1 min into the cannula through polyethylene tubing using an infusion pump (Stoelting) (80). The protein synthesis inhibitor control received an infusion of 3 μl of anisomycin (25 μg/μl, initially dissolved in 3 N HCl and brought to pH 7.3 by addition of 3 N NaOH) (81, 82). At 30 min after the anisomycin infusion, mice were infused with 3 μl of puromycin (9 mg/ml) supplemented with 75 μg of anisomycin. After drug infusions, the tubing remained in place for one extra minute to ensure proper delivery of the solution. All mice were previously handled to ensure proper immobility during intracerebroventricular administration. At 10 min after puromycin infusion, mice were transcardially perfused as described below.

Immunolabeling of hippocampal slices

After anesthesia with isoflurane, mice were rapidly euthanized and transcardially perfused for 1 min with PBS followed by 2 min with 4% (w/v) paraformaldehyde in PBS. Brains were post‐fixed overnight in the same solution and stored at 4°C. Sections of 30-μm thickness were cut with a vibratome (Leica) and stored at 4°C in PBS until they were processed for immunofluorescence. Hippocampal sections were identified using a mouse brain atlas, and sections including −1.34 to −2.06 mm from bregma were included in the analysis.

Hippocampal sections from Wfs1Cre::RiboTag and Camk2Cre::RiboTag mice were processed as follows: Free‐floating sections were rinsed three times for 10 min with PBS. After 15 min incubation in 0.2% Triton X‐100 in PBS, sections were rinsed in PBS again and blocked for 1 hour in a solution of 3% BSA in PBS. Finally, they were incubated for 72 hours at 4°C in 1% BSA, 0.15% Triton X‐100 with the anti-HA antibody (abcam Ab9110, 1/500).

In vivo puromycylated brain slices were immunostained as described previously (80). Briefly, sections were incubated for 20 min with coextraction buffer [50 mM Tris-HCl, pH 7.5, 5 mM MgCl2, 25 mM KCl, protease inhibitor cocktail (Roche), and 0.015% digitonin (Wako Chemicals)]. After three rinses with PBS, sections were incubated for 72 hours at 4°C with puromycin (Milipore MAB E343, 1/1000) and Wfs1 (Proteintech 11558-1-AP, 1/1000) antibodies in a solution containing 0.05% saponin, 10 mM glycine, and 5% fetal bovine serum in PBS. After primary antibody incubation, sections were rinsed three times for 10 min in PBS and incubated overnight at 4°C with the secondary antibody (ThermoFisher A546 A-11030, A647 A-21245, 1/500). Sections were rinsed three times for 10 min in PBS and mounted in Aqua-Poly/Mount.

Fluorescence imaging was performed with an LSM880 confocal microscope (Zeiss) using a 20x air objective (Plan Apochromat 20x/0.8 M27) with appropriate excitation laser lines and spectral detection windows. Laser power and detector gain were adjusted to avoid saturated pixels. Imaging conditions were held constant within experiments. Single images were acquired at the same depth. For better visualization, brightness and contrast were adjusted. Processing was kept constant between conditions. The brightness and contrast of the zoom-in was additionally enhanced for better visualization.

Data analyses

Proteomics data analysis

Raw data were processed using the Max Quant software version 1.6.2.2 (83). MS/MS spectra were searched against the UniprotKB database from Rattus norvegicus (36080 entries, downloaded on 21 December 2017) and additionally against a database containing common mass spectrometry contaminations using the probabilistic based algorithm from the Andromeda search engine. The set of stringent constraints allowed only peptides with full tryptic specificity allowing N-terminal cleavage to proline and up to two missed cleavages. Carbamidomethylation of cysteine was set as a fixed modification. Oxidation of methionine and acetylation of the protein N terminus were set as variable modifications. Minimum peptide length was set to seven amino acids. The first search was performed with 20-ppm precursor tolerance for mass recalibration, and the main search mass tolerance was set to 4.5 ppm. The fragment mass tolerance was 0.5 Da, and the “match between runs” option was enabled. Peptides and proteins were identified on the basis of a 1% false discovery rate (FDR) with the use of a decoy strategy, and only those protein groups identified with at least one unique peptide were used for further analysis.

Proteomics postprocessing

The Perseus package v1.6.2.2 (84) was used for further bioinformatic analysis of the resulting expression data from MaxQuant. Before further processing, decoy and contaminant database hits as well as proteins only identified using modified peptides (“identified by site”) were excluded. Additionally, only those protein groups identified in at least two of three technical replicates and in two of three biological replicates were considered for further analysis.

Footprint genome and transcriptome alignment

Adapters were removed with Cutadapt v1.15 (85) (--cut 1 --minimum-length 22 --discard-untrimmed --overlap 3 -e 0.2). An extended unique molecular identifier was constructed from the two random nucleotides from the RT primer and the five random nucleotides from the linker and added to the description line using a custom Perl script. Trimmed reads that aligned to rat noncoding RNA were removed using Bowtie2 v2.3.4.3 (86) (--very-sensitive). Remaining reads were aligned to the rat genome (rn6) with the split-aware aligner STAR v2.6.1a (87) (--twopassMode Basic --twopass1readsN -1 --seedSearchStartLmax 15 --outSJfilterOverhangMin 15 8 8 8 --outFilterMismatchNoverReadLmax 0.1). When required, STAR --quantMode was used to retrieve transcript coordinates. Transcriptome alignments were used for all analyses, except for differential expression and genomic feature analysis. The STAR genome index was built using annotation downloaded from the UCSC table browser (88). Polymerase chain reaction duplicates were suppressed using a custom Perl script, and alignments flagged as secondary alignment were filtered out.

RNA genome alignment

Adapters and low-quality nucleotides were removed with Cutadapt v1.15 (85) (--minimum-length 25 --netseq-trim=20). Reads were aligned to the rat (rn6) or the mouse (mm10) genome with STAR v2.6.1a (87).

Assigning footprint reads to genomic features

Genomic feature coordinates [coding sequence (CDS), 3′UTR, 5′UTR, intron] were downloaded from the UCSC table browser as BED files (88). Bedtools v2.26.0 (89) was used to first convert BAM files into the BED format and then identify reads overlapping with the individual features.

Counting and differential expression analysis

M/P ratios: Counts per gene were calculated from reads mapped to the genome using featureCounts v1.6.3 (90). Only a single transcript isoform, with the highest possible APPRIS score (91), was considered per gene. Only footprint reads aligned to the central portion of the ORF—by convention, 15 codons from the start until 5 codons before the stop codon—were counted (76). Raw counts were fed into DESeq2 (18) for differential expression analysis. Log fold change (LFC) shrinkage was used to generate more accurate log2 fold-change estimates (92). To test if the M/P fold-change differs across compartments, an interaction was added to the design formula. In this analysis, unshrunken log2 fold changes were used.

RiboTag-IP–to–input ratios: Counts per gene were calculated from reads mapped to the genome using featureCounts v1.6.3 (90). All transcript isoforms were considered. Raw counts were fed into DESeq2, and LFC shrinkage was used.

Classification of neuronal genes

A classifier to identify excitatory neuron-enriched genes was developed. The union of genes with significantly enriched RiboTag-IP to input fold changes (threshold of 0.05 on the adjusted P value and a 30% enrichment) was formed from the three RiboTag experiments (Hippocampus Camk2Cre::RiboTag, somata/neuropil Wfsr1Cre::RiboTag).

Classification of NMD targets

Genes with the Ensembl biotype annotation “nonsense_mediated_decay” or “retained_intron” were classified as possible NMD targets.

Translation rate calculations

The translation rate was computed from three biological replicates of neuropil/somata total ribosome footprinting, as previously described in (25). In brief, the number of footprint reads in the gene’s CDS was divided by its CDS length in kilobases. This value was then normalized to the total number of footprint reads mapping to any region of the gene. Only reads with a minimum of 10 raw reads in all footprint libraries were used for analysis.

Translational efficiency calculations

Translational efficiency was computed from three biological replicates of neuropil total ribosome footprinting. The translational efficiency of a gene was calculated as the ratio of normalized footprint reads [transcripts per million (TPM)] to normalized RNA-seq reads (TPM).

Integration of proteomic and transcriptomic data

Protein and RNA data were matched as described in (93). A protein centric view was taken. For each protein in the protein group, the corresponding RNA measures in TPM or the corresponding translation rates were summed and the mean of the corresponding monosome to polysome log2 fold change was determined. In a functional group, at least half of the genes had to be classified as “neuronal” to pass the RiboTag filter. A functional group was determined as “monosome enriched” or “polysome enriched” if more than half of its transcripts were classified as “monosome enriched” or “polysome enriched,” respectively. In all other cases, the functional group was classified as “nonenriched.”

Metagene analysis

Metagene plots represent the accumulated footprint coverage over the length-normalized ORF. The normalized footprint coverage was generated for each gene (footprint coverage divided by the average codon coverage). Edge positions were defined relative to the ORF start and stop codons and divided into 100 bins. Each gene contributed with its average normalized footprint coverage per bin.

Harringtonine depletion profile analysis

ORF footprint coverages per gene were generated for each time point. Analysis was performed on well-translated genes with at least 0.1 reads per codon. Profiles were scaled by the average coverage between codons 400 and 450. Transcripts shorter than 460 codons were excluded from the analysis. For each time point, the metagene profiles were smoothed in 30-codon windows and normalized to the 0-s time point. Only transcripts with a monosome preference in both hippocampal culture and tissue were considered.

Three-nucleotide periodicity analysis

First, the P-site offset was defined for individual footprint lengths. For this, all reads spanning the ORF start were used, and the most probable offset from the start and end of the read was defined for each length. Second, the P-site position per read was determined on the basis of its length and the previously defined offset. All P-site positions were projected for 100 nucleotides around the ORF start, stop, and center. The P-site coverage of each gene was normalized to its average footprint coverage. The nucleotide coverage at frame positions 0, 1, and 2 was assessed. To determine if the observed frame fraction differed from the expected frame fraction, a one-way analysis of variance (ANOVA) was performed. A significant P value rejects the null hypothesis that all frames exhibit the expected P-site coverage.

Genome browser track visualization

Footprint coverage was visualized as custom tracks on the UCSC Genome Browser (94). Footprint alignments were converted into BedGraph files (https://genome.ucsc.edu/goldenPath/help/bedgraph.html) using Bedtools v2.26.0.

Gene ontology analysis

GO enrichment of monosome- or polysome-preferring genes was performed using the R package clusterProfiler (95) with a Benjamini-Hochberg multiple testing adjustment and a FDR cutoff of 0.05, using all expressed neuronal genes in the neuropil or somata as background, respectively. The simplify function with a cutoff of 0.7 was used to remove redundancy from enriched GO terms.

Correlation between the M/P fold change and transcript attributes

DNA sequences were extracted from the rat (rn6) genome. Only genes with valid values for all transcript attributes were used for analysis. The length of 3′UTRs and 5′UTRs was set to a minimum of 10 nts.

GC content: The GC content was assessed by counting the number of G or C bases in the sequence and then dividing by the number of bases in the predicted 5′UTR, CDS, or 3′UTR.

Minimum free energy (MFE): The ViennaRNA package version 2.0 with RNAfold was used to calculate the MFE per 5′UTR or 3′UTR sequence (96). A method described by Trotta (97) was adapted to normalize MFE units to the sequence length. The sequence length was restricted to a maximum of 500 nucleotides in proximity to the start and stop codons.

Codon adaptation index: CAI values in the neuropil were obtained for neuronal genes only, following the procedure described in (98).

Initiation rate: The initiation rate per gene was calculated on the basis of the neuropil total ribosome footprint and RNA coverage, as previously described (99). In short, the initiation rate depends on the translational efficiency (defined as described above), CDS length, average time for a ribosome to traverse the CDS, and normalized ribosome occupancy in the initial 10 codons of the CDS. The average elongation rate was assumed to be 4 codons/s (53). A ξ value of 0.0084 was determined from the best-fit line to the average ribosome density of a transcript (from polysome profiling) versus its translational efficiency (from ribosome profiling and RNA-seq).

Mean typical decoding rate: A per-gene MTDR was calculated on the basis of the neuropil total ribosome footprint coverage, as previously described in (26). In short, each amino acid decoding time was defined as a convolution of an average decoding time (a Gaussian component with the parameters μ and σ) and a pausing decoding time (an exponential component with the parameter λ). A model-fitting procedure was used to deconvolve the two distributions and identify the three parameters per amino acid. The geometric mean of all average decoding times (μ) was calculated to determine the per-gene MTDR.

Upstream open reading frame (uORF)

To identify transcripts containing uORFs, neuropil total ribosome footprint libraries from three replicates were used. Only genes with annotated 5′UTRs were considered. A string match algorithm was used to identify sequences within annotated 5′UTRs that are flanked by canonical in-frame start and stop codons. Only sequences with a minimum length of three codons and at least 10 raw footprints in all three replicates were considered as uORFs.

Prediction of protein secondary structure and protein domains

Appris transcript isoforms were translated into amino acid sequences and used to predict secondary structures and protein domains. Porter 5 was used to predict protein secondary structures in three classes (α helix, β strand, and coil) (100). Spans of coils were defined as unstructured, whereas helices and strands were defined as structured sequences. Transitions from structured to unstructured, and vice versa, were counted and normalized to the sequence length. Protein domains were predicted using InterProScan5 based on the Pfam database (101). Functional domains per protein were merged into unique regions, and their average length was compared between monosome- and polysome-enriched genes.

Codon pause score analysis

For each codon located in the elongating ORF portion (15 codons from the start until 5 codons before the stop codon) of neuropil monosome-enriched genes, a pause score was calculated based on a z-score–like quantity: pause score = (normalized footprint coverage in monosome library – normalized footprint coverage in polysome library)/(normalized footprint coverage in polysome library)1/2.

Supplementary Materials

References and Notes

Acknowledgments: We thank D. Vogel for assistance with the preparation of cultured neurons; M. Heumüller and J. J. Letzkus for assistance with the intracerebroventricular injections; I. Wüllenweber and F. Rupprecht for assistance with the proteomics analysis; N. T. Ingolia and M. J. McGlincy (Department of Molecular and Cellular Biology, University of California, Berkeley) for advice on bioinformatic analysis of footprint libraries; and E. Valjent (IGF, CNRS, INSERM, University of Montpellier) for providing the Wfs1Cre transgenic mice. Funding: A.B. is supported by an EMBO long-term postdoctoral fellowship (EMBO ALTF 331-2017). E.M.S. is funded by the Max Planck Society, an Advanced Investigator award from the European Research Council (grant 743216), DFG CRC 1080: Molecular and Cellular Mechanisms of Neural Homeostasis, and DFG CRC 902: Molecular Principles of RNA-based Regulation. Author contributions: A.B. and C.G. designed and conducted experiments and analyzed results. G.T. analyzed results. E.C. and T.D. conducted experiments. J.D.L. acquired the proteomics data. E.M.S. designed experiments and supervised the project. A.B. and E.M.S. wrote the manuscript, and all authors edited the manuscript. Competing interests: The authors declare no competing interests. Data and material availability: All data are available in the main text or the supplementary materials. The accession number for the raw sequencing data reported in this paper is NCBI BioProject: PRJNA550323. The mass spectrometry proteomics data are deposited at the ProteomeXchange Consortium via PRIDE (102) partner repository with the dataset identifier PXD016552. All bioinformatic tools used in this study are contained in one modular C++ program called RiboTools. The source code and further notes on the algorithms can be found on our GitHub repository (103). Other analysis scripts and codes are available upon request.

Stay Connected to Science

Navigate This Article