Abstract
Fragile X syndrome (FXS) is caused by inactivation of FMR1 gene and loss of its encoded product the RNA binding protein FMRP, which generally represses translation of its target transcripts in the brain. In mouse models of FXS (i.e., Fmr1 knockout animals; Fmr1 KO), deletion of Cpeb1, which encodes a translational activator, mitigates nearly all pathophysiologies associated with the disorder. Here we reveal unexpected wide-spread dys-regulation of RNA abundance in Fmr1 KO brain cortex and its rescue to normal levels in Fmr1/Cpeb1 double KO mice. Alteration and restoration of RNA levels are the dominant molecular events that drive the observed dys-regulation and rescue of translation as measured by whole transcriptome ribosome occupany in the brain. The RNAs down-regulated and rescued in these animal models are highly enriched for FMRP binding targets and have an optimal codon bias that would predict their stability in wild type and possible instability in FMRP knock-out brain. Indeed, whole transcriptome analysis of RNA metabolic rates demonstrates a codon optimality-dependent elevation of RNA destruction in FMRP knock-out cortical neurons. This elevated RNA destruction leads to a massive reshuffling of the identities of stabilizing versus destabilizing codons in neurons upon loss of FMRP. Our results show a widespread RNA instability in FXS, which results from the uncoupling of codon optimality, ribosome occupancy, and RNA degradation mechanisms. Re-establishment of the linkage among these events is likely required by the genetic rescue of the disorder.
Introduction
FXS is the most common form of inherited intellectual disability that is caused by a single gene mutation1. In addition to mild to severe intellectual disability, individuals with FXS often have increased susceptibility to seizures, autism-like behaviors, developmental delays, among other symptoms1. FXS is caused by the expansion of a CGG trinucleotide repeat in the 5’UTR of FMR1, which results in transcriptional silencing of the gene and subsequent loss of its protein product, the fragile X mental retardation protein (FMRP)2. In the absence of FMRP, protein synthesis in the hippocampus (where most activities of FMRP have been studied) is elevated by ~20%3,4, leading to the general hypothesis that this protein represses translation, possibly in dendritic spines as well as other regions of neurons. Stimulation of post-synaptic metabotropic glutamate receptors (mGluRs) results in a form of synaptic plasticity known as long-term depression (mGluR-LTD), which while normally protein synthesis-dependent, becomes protein-synthesis independent in the absence of FMRP. This causes aberrant synaptic plasticity (i.e., exaggerated LTD) and impaired learning and memory5. In mouse brain, FMRP binds mostly to coding regions of ~850 to 1000 mRNAs6,7, and co-sediments with polyribosomes6. Because the ribosomes associated with many of these mRNAs are resistant to release by puromycin treatment and because these ribosomes translocate at faster rates in Fmr1 KO brain compared to WT6,8, FMRP is thought to repress translation by impedeing ribosme transit6.
Most genetic or pharmacologic rescue paradigms of FXS in mouse models display re-establishment of disrupted translational homeostasis3,4,8–10. We have previously shown that depletion of CPEB1, which co-localizes and co-immunoprecipitates with FMRP and activates translation in response to synaptic stimulation11, mitigates nearly all pathophysiologies associated with FXS in Fmr1/Cpeb1 double knockout (dKO) mice, including the exaggerated mGluR-LTD and elevated protein synthesis8. However, the identities of the mRNAs whose translation is disrupted in the absence of FMRP but is restored in the dKO is unknown. Here, we have used ribosome profiling and RNA-seq to investigate the mRNAs whose expression is dys-regulated in the Fmr1 KO cortex and rescued in the dKO animal. To our surprise, the apparent dys-regulation and rescue of translational activity (i.e., ribosome occupancy) in our FMRP- and CPEB1-depletion “disease” and rescue paradigm is largely driven by the dys-regulation and rescue at the RNA stability level. The RNAs that are up-regulated in Fmr1 KO cortex are enriched for those that encode ribosomal components and translational factors, and may partially explain the excessive protein production in the Fmr1 KO brain. Strikingly, the down-regulated mRNAs in the Fmr1 KO, which are enriched for those that encode factors involved in neuronal and synaptic functions, are highly enriched for FMRP binding targets6 and have a strong bias for optimal codons (i.e., codons that are favored over other synonymous codons in highly expressed mRNAs; such RNAs tend to be stable12–15), suggesting that their levels are controlled by a post-transcriptional mechanism. These observations imply that in the cortex, FMRP directly or indirectly regulates RNA stability. Indeed, RNA metabolic profiling by 5-ethynyl uridine incorporation and whole transcriptome sequencing revealed wide-spread dysregulation in RNA degradation rates in Fmr1-KO cortical neurons while synthesis and processing rates remained substantially unchanged. We identified ~700 mRNAs that degrade significantly faster in Fmr1 KO cortex compared to WT; those that favor optimal codons were particularly affected. This wide-spread codon-dependent dys-regulation in RNA degradation involves a massive reshuffle of the identities of stabilizing vs destabilizing codons, which is unlinked from codon bias. These results indicate that a primary consequence of FMRP depletion from the brain transcriptome is dys-regulated mRNA stability by uncoupling codon bias from the RNA destruction machinery. This uncoupling may be a general mechanism that underlies the FXS, and restoring the RNA stability landscape could be a key to ameliorating the disorder as implied by the restored RNA levels in the dKO brain.
Results
RNA dys-regulation and recovery correlate with Fragile X Syndrome and genetic rescue in mouse models
To identify mRNAs that are translationally dys-regulated in the Fmr1-KO mouse cortex and that are rescued in the Fmr1-Cpeb1 dKO cortex8, we performed ribosome profiling16 and RNA-seq in this brain tissue from WT, Fmr1 and Cpeb1 single KO as well as dKO animals (Fig 1a). Ribosome occupancy (translational efficiency), defined as ribosome protected fragments (RPFs) normalized to mRNA levels, is a measure of translational activity16 and in this sense serves as a proxy for protein synthesis. Accumulating evidence suggests that one mechanism whereby FMRP inhibits translation is by stalling ribosome transit6,8 and indeed there is a moderately (10-15%) higher rate of protein synthesis in FMRP-deficient brain3,4,9. Using Xtail17, an algorithm that tests for differential ribosome occupancies (DROs) between samples, we identified 651 genes with DROs among the four genotypes (FDR < 0.05; Fig 1b, left). Consistent with FMRP acting as a translation repressor, 345 out of 431 genes (80%) with DRO between Fmr1 KO (FK) and WT were up-regulated. Importantly, 425 of these DROs were rescued in the dKO cortex. Unexpectedly, more than 50% of genes with DRO in Cpeb1 KO (CK) (204 out of 359) also had DROs in FK, and were changed in the same direction (i.e., up or down). These molecular data are consistent with previous observations such as dendritic spine number and morphology, which are similarly aberrant in the two single KOs but rescued to normal in dKO animals. In this same vein, mGluR-LTD is enhanced in both of the single KOs but restored to WT in levels in dKO animals8. Because of the molecular similarities between WT and dKO, and between FK and CK, we henceforth refer to these two groups as “normal” and “FXS-like.”
To determine the underlying cause of DRO among the genotypes, we analyzed our RPF and RNA-seq data separately. Surprisingly, most RPFs were indistinguishable among FK, CK and WT. Only 23 and 21 RPFs were significantly different (padj < 0.05) between FK and WT and between CK and WT. Conversely, the dKO was the most different from WT with 410 and 333 RPFs that were significantly higher or lower (Fig 1b, middle).
In contrast, the RNA-seq heatmap displayed a reverse mirror image of the DRO heatmap (Fig S1a, Fig 1b right). Compared to WT, the expression of 50 genes was dys-regulated in FK (padj < 0.05; 10 up-regulated, 40 down-regulated), 145 in CK (padj < 0.05; 13 up-regulated, 132 down-regulated), but only 2 in dKO (padj < 0.05; Cpeb1 and Fmr1) (Fig S1a). The differentially expressed (DE) genes in FK and CK were largely identical. Among the 10 and 13 genes up-regulated in FK and CK, 7 overlap (p = 8.72 × 10−25, hypergeometric test, upper tail); among the 40 and 132 genes down-regulated, 35 overlap (p = 3.91 × 10−72, hypergeometric test, upper tail). Because the transcriptome profiles in FK and CK are so similar as are the WT and dKO (Fig S1a), we performed an unsupervised hierarchical clustering to test for sample to sample similarities (Fig S1b). FK and CK formed one cluster while WT and dKO formed another, validating the “FXS-like” vs “normal” grouping at the RNA level.
Having validated the grouping, we tested for DE genes in the FXS-like group (FK and CK) relative to the normal group (WT and dKO). The DE genes identified between the groups are changed the same way (i.e., up or down) in the single KOs and are rescued in the dKO to WT levels. We identified 733 genes dys-regulated in the FXS-like group (padj < 0.01), 162 (22.1%) up-regulated and 571 (77.9%) down-regulated (Fig 1b right). Strikingly, over 77% of the genes with up-regulated ROs in FK vs WT (267 out of 345) were significantly reduced at the RNA level in the FXS-like group (p-value = 0, hypergeometric test, upper tail). Similarly, 42% of the genes with down-regulated ROs in FK (36 out of 86) were significantly increased at the RNA level (p-value = 3.28 × 10−51, hypergeometric test, upper tail) (Fig 1c). We conclude that the observed dys-regulation and rescue ostensibly occurring at the translational level is largely driven by dys-regulation and rescue at the RNA level.
Gene Ontology (GO) analysis shows that many up-regulated RNAs have protein synthetic functions including ribosome biogenesis, translation, and protein folding, while the down-regulated RNAs have cell projection, synaptic transmission, as well as transcription and chromatin functions (Fig 1d; Tables S1, S2). Several important points come from this analysis. First, the down-regulation of many mRNAs may be “buffered” or compensated at the translational level by the up-regulation of other mRNAs that promote protein synthesis. Hence this could explain the net increase in protein output in FXS brain3,4,9. Indeed, we do observe increased ribosome occupancy of these RNAs (Fig 1c). Buffering in FMRP-deficient cells has been observed previously18. Second, we find that FMRP regulates the levels of mRNAs that encode chromatin modifying factors, which is reminiscent to other observations showing that FMRP controls the synthesis of epigenetic regulators in young neurons, albeit at the translational level19. Third, the brain and neuron components enriched in the GO terms of the down-regulated genes reflect the neural dysfunction that occurs in FXS. Indeed, the down-regulated genes are also significantly enriched for autism genes as compiled by SFARI20 (Fig 1e left; p = 3.56 × 10−7, hypergeometric test, upper tail).
We examined whether FMRP might have a direct effect on the steady state levels of the brain transcriptome. Significantly, 199 genes out of 733 DE mRNAs are bound (i.e., by CLIP, UV Crosslink and Immunoprecipitation) by FMRP6, and 194 of these were down-regulated, which is 34% of all the down-regulated genes in the FXS-like group (p = 1.01 × 10−104, hypergeometric test, upper tail; Fig 1e right). This result indicates that loss of FMRP may have a direct impact on the levels of a subset of the transcriptome important for the proper brain functions.
We found that the genes down-regulated in this study, as well as FMRP target mRNAs6, were also reduced in other studies that examined various FMRP-deficient cell and tissue types from mouse to human (Fig S1c). Given that this dys-regulation is widespread in other FXS paradigms and that the RNA rescue parallels phenotypic rescue8, it is axiomatic that investigating RNA dys-regulation is fundamental to understanding and perhaps mitigating the disorder.
Down-regulated mRNAs have a strong bias for optimal codons
Because of its strong cytoplasmic localization21 and likely direct control of steady state levels of its binding target, we surmised that FMRP would regulate mRNA stability22. How FMRP could stabilize target RNAs is suggested by its role in stalling ribosomes during translation elongation6,8,23 (Fig 2a). In yeast, Dhh1p (DDX6) destabilizes mRNAs with low codon optimality by sensing their slow ribosome decoding rate24. Codon optimality, a measure of the balance between the demand and supply of charged tRNAs25, is a major determinant of mRNA stability from yeast to vertebrates12–15,24,26. Generally, mRNAs with more optimal codons (presumably with faster decoding rates) are more stable than mRNAs using less optimal codons, connecting translation regulation to mRNA stability. Consequently, we calculated the codon Adaptation Index (cAI)27 from our WT mouse cortex transcriptome data, which describes the codon usage bias among synonymous codons for the highly expressed genes. We then derived the geometric mean of the cAI of each codon in each gene, which is referred to as the gene cAI score (see Materials and Methods). We considered the codon cAI score as a proxy of codon optimality and the gene cAI score as a predictor of mRNA stability in WT mouse cortex; high gene cAI scores predict stable mRNAs. In WT cortex, the transcripts have gene cAI score ranging from 0.62(Gm14431) to 0.95 (Rpl41) (0.77 ± 0.04, mean ± S.D.).
Surprisingly, the RNA down-regulated genes (Fig 2b, blue) were significantly more optimal than the overall transcriptome (p = 1.51 × 10−92, Wilcoxon test, two tail), and the RNA up-regulated genes (red, Fig 2b) were significantly less optimal (p = 5.80 × 10−17, Wilcoxon test, two tail). Indeed, the down-regulated genes were among the most optimal (gene cAI of 0.8 ± 0.03, mean ± S.D.) in the transcriptome, while the up-regulated genes among the least optimal (cAI 0.74 ± 0.04, mean ± S.D.) (Fig S2a, left). These cAI scores of the DE genes are not mere reflections of their transcript levels (Fig S2a, right); both gene groups were highly expressed. These values predict that the mRNAs down-regulated in FK cortex would be among the most stable in the WT cortex, while the up-regulated mRNAs the least stable.
We grouped the detectable transcriptome into 10 equal sized bins of increasing gene cAI score and examined whether there was a global correlation with the change of the mRNAs in FK relative to WT for each bin (Fig 2c, left). We observed a global depletion of high relative to low cAI score RNAs upon loss of FMRP (Fig 2c, left). Globally, the log2FC (log2 Fold Change) of RNA in the FXS-like group relative to the normal group has a strong negatively correlation with gene cAI scores (Fig S2b, left; Pearson’s correlation coefficient = −0.34). Other transcript features that are often associated with RNA stability regulation28, including coding sequencing (CDS) length, 5’ and 3’ UTR length, also correlated with mRNA level changes albeit not as strongly, except for coding sequence GC content (Fig S2b), which is a known confounding factor with codon optimality29. Given that FMRP target mRNAs strongly overlap with the down-regulated genes (Fig 1d), they are, not surprisingly, reduced in all cAI bins. However, the FMRP targets with higher cAI scores are even more reduced than those with lower cAI scores (Fig 2c, right; Fig S2c). These results show that loss of FMRP in mouse cortex leads to depletion of its target mRNAs as well as mRNAs showing higher codon optimality. These results also predict a global trend of destabilization of FMRP target mRNAs as well as stable mRNAs in FK brain cortex.
RNA metabolism profiling reveals major disruption in RNA stability in FMRP-deficient neurons
To determine whether loss of FMRP destabilizes mRNA, we incubated WT and FK mouse cortical neurons (14 DIV) with 5-ethynyl uridine (5EU) for 0 (i.e., unlabeled control, or “unlab”), 20 (library A), or 60 min (library B), after which the RNA was “clicked” to biotin and purified by streptavidin chromatography. The RNA was mixed with 5EU-labeled fly RNA and unlabeled yeast RNA as a control and sequenced together with total unenriched RNAs as input samples (Fig 3a). The spike-in RNAs for the libraries were used as quality control measures, showing that the WT and FK libraries were of equal quality (Fig S3a-c). After filtering (Fig S3d), we calculated RNA metabolism rates (synthesis, processing and degradation rates) by comparing nascent and mature RNA concentrations in the 5EU-labeled and input total RNA libraries using INSPEcT30,31. We obtained metabolism rate information for 8590 RNAs, which include 412 FMRP target mRNAs. The rates follow lognormal distributions with medians of 1.12 and 1.04 RPKM/hr, 6.84 and 6.60 hr−1, and 0.13 and 0.14 hr−1 for synthesis, processing, and degradation for libraries A and B, respectively (Fig 3b), demonstrating the reproducibility of the assay.
We calculated Spearman’s correlation coefficients for all three metabolism rates per genotypes for both libraries (Fig 3c). For synthesis, processing, and degradation rates, we observed decreasing correlation coefficients between WT and FK. For synthesis rates, WT and FK cluster together for the same labeling parameter (library A or B), indicating that there is little genotype difference. For libraries A and B, the correlation coefficients were 0.97 and 0.88 between WT and FK, again demonstrating that the synthesis rates between the 2 genotypes are similar. For processing rates, the two genotypes were also similar despite slightly lower Spearman’s correlation coefficients between WT and FK (0.79 and 0.61 for libraries A and B). Strikingly, the correlation coefficients of degradation rates between WT and FK were substantially lower (0.22 and 0.36 for libraries A and B), indicating that there is a major difference in degradation between genotypes. The Spearman’s correlation coefficients between libraries A and B for each genotype (0.87 and 0.72, respectively) indicates high reproducibility. Therefore, the degradation rates for the four libraries are separated by genotype (Fig 3c), demonstrating that RNA stability in FK neurons is disrupted.
We determined whether the RNA DE genes identified in the FXS-like group (Fig 1b, right) have altered RNA degradation rates. As predicted by gene cAI scores in WT cortical tissue (Fig 2b) and cultured cortical neurons (Fig S3e), the down-regulated RNAs (blue) are among those with the lowest degradation rates, i.e., the most stable (Fig 3d, left). On the contrary, in FK neurons, the down-regulated RNAs degrade significantly faster than the transcriptome in general (Fig 3d, right; p = 0.00029 and 0.026 for libraries A and B respectively; Wilcoxon test, one tail). The up-regulated RNAs (red) do not show a significant change in degradation rate (Fig 3d). Therefore, many mRNAs with optimal codons that are stable in WT cortical neurons become unstable in FK neurons.
To perform gene level comparisons of RNA metabolism rates, we normalized the values between libraries A and B for WT and FK (Fig S3f). At an adjusted p-value cut-off of 0.01, we identified no RNA with different synthesis or processing rate, but 748 RNAs with different degradation rates, of which 688 (92%) degraded faster in FK compared to WT (Fig 3d, Fig S3g). Significantly, the RNAs that degraded faster in FK were highly enriched for FMRP targets and down-regulated RNAs in the cortex (Fig 3f), showing that faster degradation is mostly responsible for the reduced RNA levels in FK brain. We also determined the influence of codon optimality on global mRNA degradation upon the loss of FMRP. Consistent with the high gene cAI score-dependent reduction of steady state mRNA in the cortex as we observed previously (Fig 2c, left), there was indeed a preferential destabilization (higher degradation rate) of genes with high cAI scores (Fig 3g, Fig S3h).
Disruption in RNA stability leads to massive reshuffling of stabilizing vs destabilizing codon identities in FMRP-deficient neurons
The codon-stability coefficient (CSC), which describes the link between mRNA stability and codon occurrence, has been calculated for each codon from yeast to human 12–15. We determined whether this relationship is maintained in FK neurons by first calculating CSC in WT neurons. Here CSC values ranged from < −0.2 to > 0.2, which is comparable to previously reported CSC values for human cell lines and mouse embryonic stem cells (Fig 4a). This is unlike what has been described in fly where the neuronal CSC is attenuated relative to somatic cells13. Of the 60 non-start or -stop codons, 29 had CSCs greater than 0 (stabilizing codons) and 31 less than 0 (destabilizing codons) (Fig 4b, upper). Strikingly, 17 codons that are stabilizing in WT are destabilizing in FK neurons, and 21 codons changed the opposite way (Fig 4b, lower). Because optimal codons are generally frequently used in highly expression genes and are associated with positive CSCs (i.e., are stabilizing codons) 12–15, this reshuffling of the identities of stabilizing vs destabilizing codons in FMRP-deficient neurons could reflect changes in codon usage bias in highly expressed genes accordingly, i.e., the correlation between CSCs and codon usage bias is maintained, or alternatively, this reshuffling could result from a uncoupling of the link between codons’ stabilizing or destabilizing properties and their usage bias. To examine which of the possibilities is the case, we tested Spearman’s rank correlation between CSCs and codon usage frequencies in the top 10% expressed mRNAs in WT and FK neurons. As expected, in WT neurons there is a positive correlation (r = 0.55, p-value = 1.92 × 10−5). However this correlation is largely lost in FK neurons, i.e., the correlation is almost random (r = 0.067, p-value = 0.612) (Fig 4c). The codon usage frequency in the top expressed genes in WT and FK neurons was largely unchanged (data not shown). These results show that the link between codons’ stabilizing vs destabilizing properties and their usage bias is uncoupled in FMRP-deficient neurons.
Discussion
Although it is widely assumed that FXS is caused by excessive protein synthesis32, our study shows this postulate is over-simplistic. We find that steady state RNA levels are globally disrupted in the disorder, and that genetic rescue by Cpeb1 deletion, and possibly in other rescue paradigms as well, mitigates this molecular dys-regulation. The loss of FMRP results in enhanced instability not only of its direct target substrates, but also of other mRNAs with an optimal codon bias transcriptome-wide. Our data show that RNA stability conferred by optimal codons requires trans-acting factors such as FMRP. This requirement leads to the massive reshuffling of the identities of stabilizing versus destabilizing codons in FMRP-deficient neurons.
FMRP could regulate codon dependent mRNA stability either directly or indirectly. Because FMRP seems to target transcripts with a bias for optimal codons (Fig S2c), and FMRP CLIP targets are generally reduced in FMRP deficient cortex (Fig 2c, right), FMRP could be recruited to optimal codons and its binding directly stabilize the target transcripts. Alternatively, FMRP could regulate mRNA stability indirectly through translation regulation. Translation and RNA decay are closely linked; aberrant translation activity could lead to accelerated mRNA decay33. Indeed, the genes with down-regulated transcript levels in FK cortex had generally up-regulated ribosome occupancy (Fig 1c), a measure for translation activity. In particular, repressing translation elongation by applying translation elongation inhibitors such as cyclohexamide34,35 and sordarin35, by mutating the gene encoding eIF5A36,37, or by simulating histidine starvation by treating cells with 3-amino-1,2,4-triazole35, has been shown to stabilize mRNAs. This is reminiscent of the model where FMRP stalls translation elongation. Loss of FMRP could lead to derepression of translation elongation of its target transcripts and with it enhanced mRNA decay. However, these scenarios cannot explain the dys-regulated RNA degradation rates for mRNAs that are not FMRP CLIP targets. One possibility is that FMRP binds far more mRNAs than can be covalently crosslinked by UV irradiation. The FMRP CLIP experiments in the mouse cortex6 did not use a nucleoside analog such as 4-thio uridine to enhance UV crosslinking38, nor did they use formaldehyde39, which does not rely on short-range proximity of FMRP to RNA to detect an association. Thus, the FMRP CLIP RNAs may be an underestimate of the number of transcripts bound by this protein.
Another possibility could be that FMRP regulates the codon-dependent stability of the transcriptome via its interaction with other protein binding partners. In yeast, nonoptimal codons induce ribosome pileup, which is recognized by Dhh1p, an RNA helicase that leads to mRNA destruction24. However, we have no evidence for ribosome pileup in mouse brain cortex. Moreover, in FMRP-deficient brain, the destabilized RNAs have increased ribosome occupancies (Fig 1c), not decreased. Consequently, a cause-and-effect relationship among ribosome occupancy, codon optimality, and RNA destruction as illustrated in yeast may not precisely apply to the mammalian brain. However, it is curious to note that FMRP interacts with the mammalian ortholog of Dhh1p, DDX640,41, and that DDX6/Dhh1p CLIPs predominantly to mRNA coding regions and 3’UTRs42,43. It is worth noting the strong correlation between CDS GC content and RNA changes in the FXS-like group (Fig S2b), as well as the high GC content of the most destabilizing codons in FK neurons (Fig 4b, lower). Several factors have been found to regulate mRNA stability depending on GC content associated with codon optimality, including DDX644 and ILF245 in human cells. It is tempting to speculate that in FMRP-deficient brain, DDX6 might mediate the destabilization of the down-regulated RNAs.
Lastly, we cannot exclude the possibility that loss of FMRP could impact the availability of charged tRNAs of certain anticodons. If such were the case, the balance between supply (charged tRNA) and demand (codon usage) would be lost, leading to the dys-regulation of translation elongation and mRNA decay, and therefore the uncoupling of the link between stabilizing/destabilizing codon and codon usage bias.
Our study establishes FMRP as a link between stabilizing/destabilizing codons and codon usage bias in the neuronal transcriptome (Fig 5). Given the broad similarity between the WT and dKO transcriptional profiles, we speculate that genetic rescue by CPEB1 ablation likely causes a realignment of CSC to resemble that of WT. This realignment could be a key to ameliorating the Fragile X disorder.
Methods
No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.
Animals
WT, FK, CK and dKO mice were as used previously8. Specifically, FK (JAX stock# 004624) and its WT controls (JAX stock# 004828) were purchased from the Jackson Lab. CK were created in-lab46. Mice were bred as previously described8. All mice were maintained in a temperature-(25°C), humidity-(50–60%) and light-controlled (12 hr light-dark cycle) and pathogen-free environment. Animal protocols were approved for use by the University of Massachusetts Medical School Institutional Animal Care and Use Committee (IACUC).
Ribosome profiling and RNA-seq in cortex
Two mice per genotype were used for ribosome profiling and RNA-seq. The brain was rapidly removed from P28–P35 mice, rinsed in ice-cold dissection buffer (1× HBSS + 10 mM HEPES-KOH), rapidly dissected in dissection buffer ice-liquid mixture to collect cerebral cortex as described previously47. Both cortex hemispheres were homogenized in 900µl of homogenization buffer47 (10 mm HEPES-KOH, pH 7.4, 150 mm KCl, 5 mm MgCl2, 0.5 mm DTT, 100 µg/ml cycloheximide and 2 µg/ml harringtonine), containing protease and phosphatase inhibitors (cOmplete, EDTA-free Protease Inhibitor Cocktail and PhosSTOP from Roche/Sigma, cat. no. 11836170001 and 4906837001), in a pre-chilled 2-mL glass Dounce homogenizer, 20 strokes loose, 20 strokes tight, and centrifuged at low speed (2000 rcf 10 min 4°C) to pellet insoluble material8. Five hundred microliters of the resulted ~700µl supernatant (cytoplasmic lysate) were used for ribosome profiling, the rest for RNA-seq. For ribosome profiling, the lysate was digested by 60 U RNase T1 (Thermo Scientific, cat. no. EN0541) and 100ng RNase A (Ambion, cat. no. AM2270) per A260 unit48 for 30 min at 25°C with gentle mixing. Digestion was stopped by adding 30 µl SUPERase⋅In (Invitrogen, cat. no. AM2694). Digested lysate was separated by sedimentation through sucrose gradients. Monosome fractions were identified, pooled, and extracted with TRIzol LS (Invitrogen, cat. no. 10296028).
For RNA-seq, cytoplasmic RNA was extracted from the lysate using TRIzol LS. Ten micrograms of RNA were depleted of rRNA using Ribo-Zero Gold rRNA Removal Kit, Human/Mouse/Rat (illumina, discontinued), and fragmented by incubating with PNK buffer (NEB, cat. no. M0201S) for 10 min at 94°C. Fragmented RNA as separated on 15% urea-polyacrylamide gel, and 50-60nt fraction was collected.
Ribosome profiling and RNA-seq libraries were prepared following published protocols49 and sequenced with Illumina NextSeq.
Spike-in RNA for RNA metabolism profiling
D. melanogaster (fly) Schneider 2 (S2) cells were grown in 12 ml Schneider’s insect medium (Sigma-Aldrich, cat. no. S0146) containing 10% (v/v) of Fetal Bovine Serum (FBS, Sigma-Aldrich, cat. no. F2442) at 28°C until confluent. Cells were incubated with 200 µM 5-EU for 24 hr, and were washed, pelleted and snap frozen in liquid nitrogen. RNA was extracted using TRIzol.
S. cerevisiae (yeast) cells were grown in 10 ml YEP medium containing 3% glucose at 30°C until OD600 nm reaches 0.5. Cells were then pelleted and RNA was extracted using hot acidic phenol50.
RNA metabolism profiling with cortical neuron cultures
Cortical cell suspension were obtained by dissociating cerebral cortices from E18 embryos using the Papain Dissociation System (Worthington, cat. no. LK003150). One million live cells were plated in 5 ml complete Neurobasal culture medium (Neurobasal™ Medium (Gibco, cat. no. 21103049), 1x B-27 supplement (Gibco, cat. no. 17504044), 1x Antibiotic-Antimycotic (Gibco, cat. no. 15240096), 1x GlutaMAX (Gibco, cat. no. 35050061)) per 60 mm poly-L-lysine treated cell culture dish. Neurons were fed by half-replacing the complete Neurobasal culture medium twice per week. DIV14 neurons were incubated with 200 µM 5-EU (Click-iT™ Nascent RNA Capture Kit, Invitrogen, cat. no. C10365) for 0 (input and “unlab”), 20 (“A”), and 60 (“B”) min. Neurons were then washed with ice cold 1x PBS buffer, and RNA was extracted using TRIzol (Invitrogen, cat. no. 15596018). Five-EU labeled RNA was enriched and RNA-seq library was prepared by adapting the Coller lab (Case Western Reserve University) protocol (personal communication). Specifically, mouse neuron RNA was spiked-in with 10% (w/w) 5-EU labeled fly RNA and 10% (w/w) yeast RNA. The mixed RNA was depleted of rRNA using the Ribo-Zero Gold rRNA Removal Kit (Human/Mouse/Rat) and fragmented using NEBNext® Magnesium RNA Fragmentation Module (NEB, cat. no. E6150S) for 5 min. RNA samples for libraries unlab, A and B were biotinylated by Click-iT chemistry and pulled-down using the Click-iT™ Nascent RNA Capture Kit. The pulled-down samples, together with the input sample, were subjected to library construction the same as for ribosome profiling libraries49. Here, for the pulled-down samples, all reactions were performed directly on beads until after the reverse transcription step. Sequencing library fraction with insert size 50-200nt was collected and sequenced with Illumina NextSeq. For each WT and FK, two independent batches of neurons were prepared, and each batch resulted in one of each input, unlab, A and B libraries (Tab S3).
Differential translation and RNA expression analysis
Brain cortex ribosome profiling and RNA-seq reads were processed as previously described11, which includes the following steps: 1) reads were separated based on sample barcode sequences; 2) known 3’ adapter sequences and low quality bases were removed with Cutadapt51 using parameters -O 2 -q 15 -a “TGGAATTCTCGGGTGCCAAGGAGATCGGAAGAGCGGTTCAGCAGGAATGCCGAGACCG”; 3) reads mapped to rRNA and tRNA genes were removed using Bowtie252 with parameter -N 1; 3) remaining reads were mapped to the mm10 genome using TopHat253; and 4) PCR duplicates were removed based on Unique Molecular Identifier sequences.
Uniquely mapped reads were then used as input to RSEM54 for quantification of gene expression and either mapped to RefSeq (v69) mouse coding sequences (RPF) or to whole-transcriptome (RNA-seq). Genes were filtered to have a minimum of 10 TPM (transcripts per million) in at least one sample. We used log transformed TPM expression values to correct for batch effects using ComBat55 (v3.18.0). Corrected values were transformed back to read counts using the expected size of each transcript informed by RSEM. Batch-corrected counts were used to identify differentially translated/expressed genes with DESeq256 (RPF and RNA) or Xtail10 (ribosome occupancy, RO).
GO analysis
GO enrichment analysis was performed using Cytoscape with the ClueGo57 plug-in (v2.3.3), with genes that are expressed in the mouse cortex as the reference gene set. Specifically, biological function GO terms of levels 6-13 were tested for enrichment at adjusted p-value < 0.05 (Right-sided hypergeometric test). Enriched GO terms that are similar were then fused to a group based on their Kappa score which quantifies percentage of common genes between terms. The leading group terms, which are the terms with highest significance in each group, are presented in Fig 2c. All enriched terms are in Tab. S1 and S2.
Codon adaptation index (cAI)
The codon adaptation index was calculated for a given sample as described by Sharp & Li (ref 1987). Briefly, for each sample, a set of the top 10% expressed genes was defined using batch-corrected TPM; the relative synonymous codon usage was then calculated, dividing the observed frequency of each codon by the frequency expected assuming all synonymous codons for a given amino acid are used equally; the codon adaptation index (cAI) is then calculated by comparing the frequency of each codon to the frequency of the most abundant (or optimal) codon for a given amino acid. All codes used to perform this analysis are available on GitHub (https://github.com/elisadonnard/CodonOPT).
RNA metabolism profiling analysis
Reads generated from the RNA metabolism profiling libraries were processed as for cortical RNA-seq libraries described above, except that a mouse-fly-yeast merged genome (mm10 + dm6 + sacCer3) was used as the reference genome for reads mapping by hisat2. The mapping statistics here were used for quality control and filtering purposes (Fig S3a-d). Uniquely mapped reads that are depleted of rRNA, tRNA sequences and PCR duplicates were again mapped to mm10 genome with hisat2. Intron and exon read quantification, and RNA metabolism rates (synthesis, processing and degradation) estimation was performed using INSPEcT30 (v1.10.0), with the degDuringPulse parameter set to TRUE. One set of libraries, which was of low complexity (Tab S3), was still used to confirm the global shift of degradation rates in FK neurons. This reproducible global shift allowed us to normalize WT and FK libraries separately for our gene level analysis (Fig S3g). Specifically, raw RNA metabolism rates estimated by INSPEcT were normalized between libraries A and B for WT and FK neurons separately using the limma package58 with the “cyclicloess” method. After normalization, genes with different metabolism rates were tested using the limma package.
Codon-stability coefficient (CSC) analysis
CSCs were calculated as previously described8,9,12,14. Specifically, a Pearson’s correlation coefficient was calculated for each of the 60 non-start and -stop codons between the frequencies of this codon in all the genes that use this codon, and the stability of these genes. The stability of a gene (y), which is the inverse of its degradation rate (x), is expressed as follows:
Here xA and xB are the normalized degradation rates from library A and B respectively. The highest expressed isoform of each gene was used to calculate the usage frequencies of each codon.
Quantification and statistical analysis
Blinding or randomization was not used in any of the experiments. The number of independent biological replicates used for an experiment is indicated in the respective figure legends or main text. The statistical tests and P values used for the interpretation of data are mentioned in the figure legends or main text.
Code availability
All codes used to perform cAI analysis are available on GitHub (https://github.com/elisadonnard/CodonOPT). Other customized R scripts for data analysis are available from the corresponding authors upon request.
Data availability
The data supporting the findings of this study have been deposited in the Gene Expression Omnibus (GEO) repository with the accession code GSE0000000. All other data are available from the corresponding authors upon reasonable request.
Author information
Contributions
H.S. and J.D.R. conceived the project and designed the experiments. H.S. performed most of the experiments, B.L. generated RNA-seq libraries. H.S. and E.D. performed the bioinformatic analysis with input from B.L. H.S. and J.D.R. wrote the manuscript with input from all authors.
Ethics declarations
Competing interests
The authors declare no competing interests.
Acknowledgements
We thank Emmiliano Ricci for sharing his experience and protocol for ribosome profiling, Mariya Ivshina for advice and help with the mouse breeding and Ruijia Wang for help with the bioinformatics. We thank Lindsay Romo for sharing her protocol for 5-EU labeling of neuron culture, and Jeff Coller (Case Western Reserve University) for Click-iT EU tagging – RNA-seq protocol. We also thank Ariel Bazzini for his very helpful discussion and input. Nathan Gioacchini and Yongjin Lee helped preparing yeast and fly spike-in RNA samples. This works was supported by the National Institutes of Health (U54HD82013); the Simons Foundation, and the Charles H. Hood Foundation (to JDR).