Neuroprogenitor cells differentiation into cortical neurons as a model to study translation regulation of genes involved in neuronal processes
Neuroprogenitor cells (NPC) were differentiated from H9 human embryonic stem cells using a SMAD dual inhibition strategy [8] (Additional file 2: Fig. S1A). The generated NPCs were positive for the progenitor marker Nestin and upon differentiation induction by FGF removal, developed neural projections, and expression of synaptic neuronal markers (Figs. 1A, 3A; Additional file 2: Fig. S1B).
For translation efficiency analysis we selected 3 conditions: NPC, 3 days (Early differentiation—ED), and 30 days (neuron) after differentiation induction (Fig. 1B). We simultaneously collect material to perform transcriptome (RNAseq) and translatome (Ribosome Profiling—Riboseq) analysis to be able to measure translation efficiency (TE = Riboseq reads/RNAseq reads) of each expressed gene. The quality parameters for the Riboseq and RNAseq libraries confirmed a high correlation between biologic replicates (Additional file 4: Fig. S3).
We first performed a transcriptome analysis to validate our model (Fig. 1C). During the first 3 days of differentiation, only a few genes (222) had their expression significative modified, in contrast to 30 days of differentiation (around 6000 genes). As expected, differentiation markers analysis (Fig. 1E; Additional file 3: Fig. S2) indicated a gradual reduction of neuroepithelial markers, greater induction of early differentiation markers in 3 days, and high induction of neuronal, synaptic, and glutamatergic makers after 30 days of differentiation. The high expression of doublecortin suggests that even after 30 days of differentiation, the generated neurons are still going through the maturation process. Comparison between transcriptome data from previously published works with ours, and among themselves [10, 11] (Additional file 5: Fig. S4), pointed out little correlation between regulated genes in the different time points, indicating that the transcriptome is dynamic regulated during differentiation. Therefore, the chosen 30 days differentiation time point could reveal new translation-regulated genes not encountered in the previously published work.
Gene Ontology analysis of differentially expressed genes (Fig. 1D) shows a significant induction of genes functionally related to nervous system development, dendrite morphogenesis, axonogenesis, axon guidance, and synaptogenesis. Together with the cellular morphology changes observed in cultured cells (Additional file 2: Fig. S1B), these results indicate that translation efficiency analysis in the selected time point could help elucidate how translation control contributes to the regulation of the above neuronal processes.
Translation control significantly contributes to differential gene expression between NPC and neurons
Next, we compared mRNA and translation levels of 14,159 genes for which we obtained a significant number of counts, after sequencing a minimum of 20 million unique mapped reads per library (Additional file 8: Table S1). In NPC, RNAseq and Riboseq reads showed a high Pearson correlation of 0.77 (R = 0.77), which is similar to the previously reported value for NPC [10] and mammalian cells in general [23]. However, this correlation progressively and drastically modified from NPC, Early Differentiation (ED) to neurons (R = 0.42), indicating that neurons greatly rely on translation control to regulate their protein expression (Fig. 2A).
Since none of the regulated genes in ED samples showed modifications in translation efficiency (TE), we will discuss the neuron to NPC comparison, where thousands of genes changed their TE (Fig. 2B). A summary of the different translation regulation classification and criteria used in our study is provided in Fig. 2B. For 2145 differentially expressed genes between these two cells, there is a perfect direct correlation between their RNAseq and Riboseq regulation data (genes up or down-regulated, TE constant category). For 3029 genes, however, there is a statistically significant TE modification between the 2 conditions. Surprisingly, for more than half of these genes (1734 genes), Riboseq counts did not change significantly meanwhile mRNA abundance either increased (RNA up, TE down category) or decreased (RNA down, TE up category). The elevated number of genes found in this category, known as “translation buffering” [24] could be due to the presence of mRNA stored translationally inhibited in neuronal granules in neurons. For 1295 genes, their TE significantly changed in neurons due to translation up or down-regulation in relationship to corresponding mRNA levels (genes translationally induced and translationally repressed categories, respectively).
To gain more insight into the subcellular distribution of translation-regulated transcripts, we compared our data with the following publicly available datasets: axonal transcriptome [25], synaptic bouton proteome [26], neuronal soma, and projections Riboseq (neuronal neuropil) [27], (Fig. 2C; Additional file 10: Table S3). After bioinformatics comparative analysis of these datasets, we found a statistically significant enrichment of translationally regulated genes whose transcripts are found in axons in almost all categories. Around 11% of the 2192 genes present in axons are translationally induced, corresponding to 36% of all translationally induced genes (Fig. 2B). More than half of the proteins detected in the synaptic bouton proteome are either upregulated (TE constant) between NPC and neuron or are translationally induced. Interestingly, there is an enrichment of translationally inhibited transcripts in monosomes in general (translationally repressed and RNA up, TE down); meanwhile, polysomes are enriched in transcripts upregulated or translationally upregulated (Translationally induced and RNA down, TE up).
Next, we performed a GO enrichment analysis (using Ingenuity Pathways Analysis system—IPA) on the genes that changed translation efficiency between cells to investigate if any neuronal pathway or process is particularly regulated translationally (Fig. 2D; Additional file 9: Table S2). Interestingly, the processes of oxidative phosphorylation, synaptogenesis, and pathways that modulate actin polymerization for neuritogenesis and axon guidance are highly enriched in translationally regulated genes. In addition, pathways that participate in translation control such as mTOR and EIF2 signaling are also translationally regulated and may be in part responsible for the intense translation program modification observed between compared cells. Translationally controlled genes on these pathways and processes are discussed in the next sections.
Therefore, translation regulation significantly contributes to differential gene expression between NPC and 30 days Neurons and seems to participate in regulating important neuronal pathways and processes involved in neuronal development.
Translation control regulates essential synaptic genes necessary for synaptic transmission
In the selected time point, synaptic markers are detected in a pattern that indicates that synaptic connections are forming (Fig. 3A). To better understand which synaptic genes and processes are translationally modulated, we reanalyzed our data using SynGO, a GO database focused on synaptic processes [22]. Most of the genes involved in synapse transmission are strongly induced during differentiation but not translation regulated (Fig. 3B). However, almost half of the identified differentially expressed synaptic genes are translationally regulated. For 90 of these genes, translation efficiency is increased with a significant increase in Riboseq reads (translationally induced category), meanwhile, for 34 genes, translation is down-regulated (translationally repressed category). For 117 genes, there is either mRNA up or down-regulation, without a significant change in riboseq reads (RNA up, TE down; RNA down, TE up categories).
Synaptic processes are significantly enriched in induced genes without translation regulation (upregulated TE constant) and genes translationally induced (Fig. 3C). For the GOs “synaptic vesicle” and “synaptic vesicle proton loading”, translationally induced genes are even more enriched than the set of upregulated genes with constant TE. Some translationally induced genes on these GOs code for proteins related to the SNARE complex (N-ethylmaleimide-sensitive factor attachment protein receptors) (Fig. 3D), which mediate vesicle trafficking and membrane fusion. In neurons, the SNARE complex is essential for calcium-triggered synaptic vesicles exocytosis and works together with the small GTPase RAB family members (Rab3a/b/c and Rab27b) [28, 29]. This complex is formed by 3 families of proteins: synaptobrevin, syntaxin, and synaptotagmin. In our data, synaptobrevin 2 (Vamp2), syntaxin 12 (stx12), synaptotagmin 11 (Syt11), and all members of SNARE disassembly/recycling complex formed by NSF, NAPA (alpha-snap), NAPB (beta-snap), and NAPG (gamma-snap) are translationally induced (Fig. 3D). Among these genes, Vamp2 is essential for synaptic transmission [30] and it is very strongly translation induced (TE = 1.5 in NPCs and TE = 14 in neurons). Other synaptic SNARE genes that work together with Vamp2, such as Syt1A and SNAP25, are strongly upregulated without translation regulation.
From the RABs involved in synaptic transmission, all 3 Rab3s (A, B, and C) are strongly translationally induced meanwhile Rab27a/b are very low expressed and do not seem to be translation-regulated. In addition, several other Rab family members are highly translationally regulated, constituting one of the most translation-regulated gene families in our dataset: RABs 1A/B, RAB5c, Rab7A, RAB11B, RAB33B, and RAB35 (Fig. 3D). The regulated RABs are involved in several aspects of neuronal biology: synaptic vesicle recycling (RAB5C, RAB33B, and RAB35), retrograde neurite transport (RAB5c and RAB7), and neurite outgrowth (RAB7, RAB11, and RAB35) [29].
Almost all regulated genes in “Synapse vesicle proton loading” GO are translationally induced. This GO comprises vacuolar ATPases that promote vesicle acidification necessary for neurotransmitter loading into synaptic vesicles [31] (Fig. 3D).
Next, to confirm some of these results, we randomly selected translation-induced genes for further validation experiments. In these experiments, we used two different methods on a non-redundant list of genes, to maximize the number of interesting candidates tested. Confirmation in any of these methods provides further evidence of the translation regulation.
The mRNA distribution of ATP6VOD1, NAPG, and RAB3A was analyzed in sucrose gradient fractions obtained from 3 new NPCs and neuronal samples (Fig. 3E). As expected for differentially translated regulated genes, they were enriched in the high polysomal fraction in neurons but not in NPCs, suggesting that they have a higher translation efficiency in neurons (Fig. 3F). For VAMP2, we have compared protein and mRNA levels in NPCs and neurons (Fig. 3G; Additional file 7: Figure S6). Although mRNA was induced only around 3 folds during differentiation, protein level increased more than 70X, corroborating with the results of the translation regulation analysis.
Altogether, our data indicate that translation control is specially used to modulate synaptic gene expression, regulating essential proteins and processes necessary for synaptic transmission.
Translation control contributes to the regulation of neuronal cell metabolism
The obtained RNAseq data confirm downregulation of glycolysis genes, with the absence of up-regulation of tricarboxylic acid (TCA) genes, previously reported to occur during the metabolic shift (glycolysis—NPC to OxPhos—Neurons) that follows neural differentiation [32]. However, our Riboseq data indicates that several of these genes have their translation efficiency and total translation increased upon differentiation (Fig. 4A, B). The glycolysis and TCA genes under translational regulation in our dataset are different from those reported by Rodrigues et al. [11] (Additional file 11: Table S4), which probably is due to the different stages of neuron differentiation analyzed. Both results suggest that translation control is an important layer of regulation that contributes to adjusting the OxPhos pathway in neurons during development. Besides glucose catabolism, other metabolic enzymes are either translationally induced or repressed in neurons in our dataset, particularly genes involved in the ornithine cycle (Urea cycle) and aspartate metabolism.
In the Urea cycle, there is a strong translation induction of ARG2 and OAT (translationally induced category) enzymes with a simultaneous translation down-regulation of SRM (translationally repressed category) and transcriptional downregulation of ODC1 and SMS enzymes (downregulated TE constant category). These regulations suggest that in neurons there is an increase in ornithine production from arginine, with inhibition of its usage in polyamines synthesis (putrescine, spermidine, and spermine) and prioritization of its use to produce glutamate-semialdehyde in a reaction that may convert alpha-ketoglutarate into glutamate. Glutamate-semialdehyde can also be converted into glutamate by the ALDH4A1 enzyme.
Translational up-regulation of GOT1/2 enzymes (translationally induced category) with transcriptional downregulation of CAD (downregulated TE constant category) suggests that aspartate usage to generate oxaloacetate is increased in neurons, in a reaction that simultaneous convert alpha-ketoglutarate into glutamate. Although alpha-ketoglutarate from TCA can be used as a glutamate source to glutamatergic neurons, it is believed that most of the glutamate used as a neurotransmitter by these neurons come from glutamine produced by astrocytes (Glutamine-Glutamate cycle) [33]. This happens due to the absence of pyruvate carboxylase enzyme expression in neurons and consequently the inability to synthesize oxaloacetate to maintain the TCA cycle depleted of alpha-ketoglutarate [34]. In neuronal cultures derived from NPC, astrocytes differentiate and accumulate much later. Therefore, we speculate that in the absence of sufficient glutamine provided by astrocytes, neuronal cells may increase metabolic usage of amino acids as a source of glutamate.
To validate the translation regulation of ARG2, GOT1, OAT, and SRM, we analyzed their mRNA distribution in a sucrose gradient. The results confirmed that ARG2, GOT1, and OAT mRNA are more associated with high polysomal fractions in neurons than in NPCs. Furthermore, as expected for a translation inhibited transcript, SRM mRNA was detected mostly in the free and 80S fractions in neurons (Fig. 4C).
Overall, our results show that neurons rely on the translation regulation of metabolic enzymes to fine adjust their metabolism according to cell demand.
Translation control participates in the regulation of actin and microtubule cytoskeleton pathways critical for neuronal projections
The neuronal cytoskeleton is actively modeled to allow dendritogenesis, axon growth, and guidance during development. Actin in the growth cone, present on the tip of axons or dendrites, can be polymerized or depolymerized in response to extracellular cues to guide neurites' extension to their right destination. In mature neurons, cytoskeleton structure changes in response to synapse activity and is implicated in new synapses formation.
Interestingly, IPA GO enrichment analysis on translation-regulated genes shows strong enrichment of GOs associated with actin cytoskeleton regulatory pathways and upstream axon/dendrite development and guidance pathways (Neuregulin, Reelin, Semaphorin, CXCR4, and Ephrin signaling) (Figs. 2D, 5A, B).
To expand these findings, we performed a careful search of all genes belonging to the AMIGO [35] GO terms axogenesis, dendritogenesis, and axon guidance and investigated their translation regulation. This analysis revealed that an important fraction of these genes is translationally regulated representing more than half (592 genes) out of 1136 differentially expressed genes in these processes. Furthermore, around 25% (265 genes) are in the translationally induced or repressed category (Fig. 5B).
Many of the translated regulated genes are key hubs on the investigated processes, suggesting that translation control is specially used by neurons to modulate cytoskeleton structural changes necessary to neuritogenesis and neurite guidance (Fig. 5C).
Two of the 3 small GTPases from the Rho family that play crucial roles in mediating actin cytoskeleton remodeling are translation modulated: CDC42 (translationally induced) and RhoA (RNA down, TE up). In neurons, RhoA activation is associated with growth cone collapse, meanwhile, Cdc42 is downstream of attractive clues [36]. RhoA activates ROCK, which phosphorylates LIMK, promoting CFL phosphorylation inhibiting its actin depolymerization activity. LIMK can also be phosphorylated by Pak1, which is downstream of Rac1 GTPase (revied in [37]). Our data show that both ROCK1 and Pak1 are translationally repressed while CFL is translationally induced. CFL has an essential role in regulating the actin cytoskeleton during growth cone elongation [37]. Besides this, 2 GTPases close related to RhoA, and also involved in actin cytoskeleton remodeling, RhoB and RhoC, are equally translationally induced. Several Rho family modulating GEFs and GAPs are translation regulated, such as TRIO, ARHGAP1, ARHGAP5, and FARP1. FARP1, a RAC1 GEF, is implicated in the assembly and disassembly of dendritic filopodia, formation of dendritic spines, and regulation of dendrite length [38, 39].
Some transcripts from proteins directly involved in actin polymerization are also regulated. Profilin binds monomeric actin and catalyzes the exchange of ADP for ATP, promoting actin polymerization into actin barbed ends. This reaction is stimulated by ARP 2/3, WASP, and WAVE complexes. Arp2/3 complex is one of the most important regulators of dendrite spine growth [40]. Profilin, BRK1 (part of WAVE complex), ARPC4, ACTR2, ACTR3 (part of ARP complex) have increased TE; meanwhile, the Wasp complex members WASF2 and WASF3 have a slight decrease in TE. Proteins from the thymosin family sequester monomeric actin, inhibiting actin polymerization [41]. In our data, the thymosin family members TMSB4X and TMSB10 have increased TE in neurons (translationally induced category).
Besides actin cytoskeleton remodeling, proper axon and dendritic arbor morphology rely on careful microtubule polymerization, stabilization, severing, and depolymerization. Likewise, we found the translation regulation of several key proteins that participate in these processes (Fig. 5C).
CRMP2 (DPYSL2) and CRMP3 (DPYSL4), both translationally induced in neurons, work on microtubule cytoskeleton polymerization [42] and are essential for semaphorin class 3 signaling and downstream remodeling of actin and microtubules [43]. In addition, they are major modulators of dendrite development, neuronal growth cone collapse, and axon guidance [44].
Doublecortin is a microtubule-binding protein localized at the ends of neuritic and leading processes, where it regulates microtubule organization and stability, regulating neuron migration and dendrite growth during development[45, 46]. In our data, it is translationally induced in developing neurons.
IQGAP1 is a key regulator of dendritic spine number, morphology, and extension. In the microtubule cytoskeleton, it works cooperatively with its interaction partner Clip170 (Clip1) and APC to regulate microtubule dynamics and stabilization [47]. All three genes have a decreased TE in neurons (IQGAP1 and Clip170 translationally repressed; APC RNA up, TE down). In the actin cytoskeleton, IQGAP1 works in complex with Cdc42 and Rac1 to stimulate actin assembly by N-WASP and the Arp2/3 complexes [48].
As shown in Fig. 5C, transcripts from the Stathmin family (STMN3, STMN2, STMN1) increase their TE (translationally induced category) in developing neurons. Proteins from this family constitute a hub that relays and integrates various signaling pathways and is involved in microtubule dynamics by promoting depolymerization of microtubules or by preventing polymerization of tubulin heterodimers [49]. Interestingly the neuronal-specific STMN3 (also known as SCLIP) [50] is regulated exclusively translationally between NPC and Neuron in our dataset. This gene has been implicated in controlling growth cone expansion, axon morphogenesis, specification, branching [51], and dendritic maturation [52].
Members of the Par polarity complex—PARD3, PARD3B, and PARD6A—are translationally repressed in developing neurons. This complex is essential to establish neuronal polarity, axon specification, and dendritic spine formation [53, 54]. Pard3 is localized in the axon, especially at the growth cone, but is excluded from neurites that will become dendrites [55]. PARD3 local translation is also required for NGF-induced axon outgrowth [56].
Some genes with a less clear molecular mechanism of action but fundamental in controlling neuritogenesis were also found to be translation regulated. Some examples of translation-induced genes in developing neurons in relationship to NPCs are GPM6A, SS18L1 (also known as CREST), and RNF10. GPM6A is required for normal axonal extension and guidance [57], filopodia outgrowth, motility [58], spines, and synapse formation [59]. The transcriptional activator SS18L1 (CREST) is part of the neural progenitor Brg/Brm-associated factor (npBAF) complex and is required for calcium-dependent dendritic outgrowth and branching of cortical neurons [60]. RNF10 is a ring finger protein involved in synapse-to-nucleus signaling downstream to NMDA receptors. It regulates spine density, neuronal branching, and dendritic architecture in hippocampal neurons [61].
To validate some of our findings, we randomly selected 5 translated regulated genes (CFL1, DPYSL2, SS18L1, STMN3, and GPM6A). The first 3 were tested by qPCR analysis of sucrose gradient fractionations of ribosomal complexes (Fig. 5D). They all showed a greater mRNA association with heavy polysomal fraction in neurons when compared to NPCs. The last 2 genes were tested by comparing protein induction with mRNA induction in neurons and NPCs (Fig. 5E). Both seem to be greater induced in protein levels than in mRNA levels in neurons, suggesting that they are translated regulated.
The IPA GO analysis also identified translation modulation enrichment in pathways known to control neuronal circuit formation through cytoskeleton reorganization, such as mTOR, Wnt, NGF, and CREB (Additional file 6: Fig. S5), indicating that translation control mechanisms may participate in tuning these pathways for proper neuron development.
Taken together, our data suggest that translation control mechanisms play an important role in regulating actin and microtubule cytoskeleton pathways that are critical to neurite generation, spine formation, polarization, axon guidance, and circuit formation (Additional files 7, 8, 9, 10).