Genetic removal of p70 S6K1 corrects coding sequence length-dependent alterations in mRNA translation in fragile X syndrome mice

Results

To investigate the impact of FMRP loss on the translation of specific mRNAs, we first examined polysome profiles prepared from cortical lysates from P30 FXS mice. We observed little difference in the polysome profiles of FXS mice compared to WT littermates (SI Appendix, Fig. S1A). We proceeded to perform ribosome profiling to infer mRNA translation profiles at a genome-wide level (SI Appendix, Fig. S1B) (9). The majority (n = 153) of the 204 genes with significantly altered RF abundance in FXS cortices showed reduced footprint counts (Fig. 1A) and were enriched for genes linked to glial cell development (Fig. 1F). In stark contrast, of the 351 genes with significantly changed mRNA expression, the majority (n = 265) showed elevated expression in FXS brains (Fig. 1A, n = 4 FXS, 4 WT) and were enriched for genes implicated in synaptic transmission and development of the nervous system (Fig. 1F).

The genes with aberrant RF abundance in FXS brains overlapped significantly (P = 2.0 × 10⁻³), albeit weakly (n = 26), with previously identified binding targets of FMRP (10). In addition, all overlapping genes exhibited reductions in RF abundance in FXS. In contrast, the genes with altered mRNA expression showed a robust overlap with FMRP targets (n = 131, P = 4.5 × 10⁻⁶⁶), with almost all (n = 129) exhibiting increased mRNA expression in FXS brains.

Previous work has suggested that the translation of mRNAs with long CDSs is particularly vulnerable to FMRP loss (7, 8). To evaluate this phenomenon, we ordered mRNAs ascendingly by their CDS lengths, divided them into six color-coded bins, and evaluated their (LFCs) against their false discovery rate (FDR)-adjusted P values (Fig. 1B). Among mRNAs with significantly decreased RF abundance in FXS cortices, we observed enrichment for those with long CDSs. Conversely, mRNAs with short CDSs exhibited a significant up-regulation of RFs. Analyzing the fold changes of all mRNAs revealed that as the CDS length of an mRNA increases, it is more likely to exhibit reduced RF abundance in FXS brains, with the mRNAs harboring the shortest CDSs showing up-regulation of footprints (Fig. 1D). Meanwhile, identical analysis of mRNA expression showed opposite associations. mRNAs with long CDSs were up-regulated in FXS brains, whereas mRNAs with short CDSs were down-regulated (Fig. 1C). Analyzing transcriptome-wide fold changes revealed that the probability of elevation in mRNA expression was a positive function of CDS length: As the CDS length of mRNA is extended, it is more likely to increase in expression in FXS brains (Fig. 1E). Furthermore, the opposing effects in RF abundance and mRNA expression summate additively to elevate the LFCs in TE in FXS brains (Fig. 1G and SI Appendix, Fig. S1C). As expected, evaluating the cumulative probabilities of LFCs in TE against CDS lengths transcriptome-wide revealed a profile reminiscent to that of RF abundances but with elevated changes in each CDS length bin (Fig. 1H).

To confirm that our results were not artifacts of prespecified binning thresholds, and to directly compare the CDS length-dependent alterations in RF abundance and mRNA expression in FXS, we ordered mRNAs into 50 bins by their CDS length, with each bin harboring the same number of mRNAs. Evaluating the average LFC in each bin confirmed that mRNAs with shorter CDSs exhibit elevated RF abundance in FXS, and those with longer CDSs exhibit reduced RF abundance (Fig. 1I). It also validated the exact opposite trend for mRNA expression: mRNAs with shorter CDSs are reduced in FXS brains, while mRNAs with longer CDSs are elevated (Fig. 1I). Remarkably, both of these effects are linear: As CDS length increases, the magnitude of the alterations in RF abundance and mRNA expression in FXS brains exhibit negative and positive proportional changes, respectively. The combination of the positive-to-negative gradation in LFCs in RF abundance and the negative-to-positive variation in fold changes in mRNA expression with CDS length results in an even stronger positive-to-negative relationship between TE and CDS length in FXS brains (Fig. 1J). Importantly, we also observed strikingly similar results when we implemented our analysis on a previously published ribosome profiling and RNA-Seq dataset from P24 FXS model mouse brains (SI Appendix, Fig. S2) (7).

A straightforward interpretation of the reduction in RF counts in the majority of altered mRNAs would be that FMRP loss reduces mRNA translation (8). However, examination of cortical protein levels of several altered mRNAs with particularly long or short CDSs revealed either significant up-regulation or a trend toward up-regulation in FXS mice (Fig. 2A). Given evidence suggesting that FMRP directly binds to the ribosome (11) and stalls its translocation on specific mRNAs (10), we decided to further investigate the molecular basis of these observations. First, we examined de novo protein synthesis in embryonic cortical neurons from FXS mice. To label newly synthesized proteins, we used puromycin, an aminoacyl-transfer RNA (tRNA) analog that covalently binds nascent peptide chains and causes their premature release. Quantifying these truncated peptides measures de novo protein synthesis (12), which we observed to be elevated in FXS primary neurons (Fig. 2 B and C).

To examine protein synthesis specifically during the elongation phase of translation, we pulsed FXS neurons with Harringtonine (HHT) for 0, 8, or 16 min, then chased them with puromycin (Fig. 2F). HHT inhibits translation initiation rapidly (13). Therefore, after its addition, puromycin labeling specifically measures the elongation phase of protein synthesis. Using this approach, hereafter called “elongation SUnSET” (ES), we observed elevated elongation-specific protein synthesis at baseline (where HHT and puromycin are added simultaneously) in FXS neurons (Fig. 2 D and E). Conversely, we observed a mild reduction in elongation-specific protein synthesis at baseline in FXS mouse embryonic cortical neurons transduced with human homolog of full-length FMRP (SI Appendix, Fig. S3 A and B). These results establish that de novo protein synthesis is elevated in FXS neurons even after inhibition of translation initiation, consistent with previous work showing that FMRP loss leads to a global elevation in elongation-specific protein synthesis in cortical lysates (14). However, one might also expect a similar outcome if FMRP loss led to increased initiation, in which mRNAs are already engaged with an increased number of ribosomes before HHT addition (SI Appendix, Fig. S4B). Intriguingly, comparing the elongation-specific protein synthesis in each condition against its own baseline (to examine the rate of decrease) revealed a significant reduction in puromycin signal in FXS neurons after 16 min of runoff elongation (SI Appendix, Fig. S4A). Conversely, we observed a mild increase in elongation-specific protein synthesis after 16 min of runoff elongation in FXS neurons transduced with human FMRP (SI Appendix, Fig. S3C). These results raised the possibility that there are far fewer ribosomes remaining to carry out protein synthesis after 16 min of runoff elongation in FXS neurons, which may explain why, despite elevation at baseline, FXS neurons show reduction in elongation-specific protein synthesis after runoff.

To examine ribosome engagement specifically, we employed the ribopuromycylation method (RPM) (15). This assay uses puromycin to tag nascent proteins in the presence of emetine, a translation elongation inhibitor that freezes ribosomes in place on mRNAs without affecting its catalytic activity. Because emetine rescues the ribosome from puromycin-induced dissociation, each translating ribosome is tagged with a single molecule of puromycin. We reasoned that pulsing cultured neurons with HHT and chasing with RPM would permit measurement of the number of translating ribosomes remaining after defined periods of runoff elongation (Fig. 2F). To examine whether this approach was feasible, we first carried out RPM on human embryonic kidney (HEK) 293 cells. Probing for puromycin revealed a clear cytoplasmic signal that was dramatically reduced by a 15 min preincubation with anisomycin or HHT (SI Appendix, Fig. S4C). Pulsing HEK293 cells with HHT for 0, 5, 10, or 15 min, chasing with RPM, and probing for puromycin on immunoblots revealed an anisomycin and runoff elongation-dependent decrease in puromycin signal (SI Appendix, Fig. S4D). These results suggested that the combination of HHT and RPM could be used to probe actively translating ribosomes during runoff elongation in cultured neurons, which also showed an anisomycin and runoff elongation-dependent reduction in RPM signal (SI Appendix, Fig. S4E). Using this approach, hereafter termed “runoff-RPM,” we observed reduced levels of translating ribosomes in FXS neurons at baseline (Fig. 2 G and H and SI Appendix, Fig. S4F). With increasing runoff durations of 3, 6, 9 and 12 min, we observed further reductions in the number of translating ribosomes in FXS neurons (Fig. 2 G and H and SI Appendix, Fig. S4F). Determining the rate of loss in the number of translating ribosomes, which we computed by comparing the number of translating ribosomes remaining after runoff elongation in each condition against its own baseline, revealed a faster rate of decrease in the RPM signal with runoff elongation in FXS neurons (Fig. 2I). Conversely, with increased runoff elongation, we observed a slower rate of decrease in the number of translating ribosomes in FXS neurons transduced with human FMRP (SI Appendix, Fig. S5 A–C). These results strongly suggest that FMRP impedes ribosome translocation and that its loss causes a global increase in the rate of translation elongation in neurons.

To better understand how S6K1 regulates translation in FXS, we carried out runoff-RPM followed by Western blotting in WT and FXS cortical neurons treated with the S6K1 inhibitor PF-4708671. Consistent with the results from immunofluorescence studies, we observed a significant reduction in the number of ribosomes in FXS neurons treated with vehicle at baseline compared to WT neurons (Fig. 2 J and K). This effect was amplified with runoff elongation: Fewer translating ribosomes remained after 8 and 16 min of runoff elongation in FXS neurons. In contrast, the expression of human FMRP in FXS neurons elevated the number of ribosomes remaining after runoff elongation (SI Appendix, Fig. S5 D–F). Inhibition of S6K1 with PF-4708671 dramatically reduced the number of translating ribosomes at baseline in both WT and FXS neurons. However, with runoff elongation, the reductions observed in FXS neurons treated with PF-4708671 were not as substantial as those observed in vehicle-treated FXS neurons. Evaluating the rate of loss of the runoff-RPM signal, which quantifies the inverse of the rate of translation elongation, revealed that pharmacological inhibition of S6K1 rescues the excessive rate of translation elongation in FXS neurons (Fig. 2L). Combined, these results establish that an increase in the rate of ribosome translocation largely mediates elevated de novo protein synthesis in FXS neurons. This increase is captured by the reduced number of translating ribosomes with runoff elongation in FXS neurons, which in turn explains why the majority of altered genes in FXS brains have decreased RF counts.

Given our observation that inhibition of S6K1 rescues the rate of translation elongation in FXS neurons, we asked whether the previously observed genetic rescue with S6K1 deletion also was mediated by reducing the rate of elongation on specific mRNAs in vivo. We examined the RF abundance (n = 3) and mRNA expression (n = 4) of DKO mice that harbor genetic deletions of both Fmr1 (FMRP) and Rps6kb1 (S6K1, Fig. 3A and SI Appendix, Fig. S6E). To our surprise, the DKO mice exhibited significant departures from WT mice in both RF abundance (236↑ and 355↓, Fig. 3B and SI Appendix, Fig. S6A) and mRNA expression (945↑ and 666↓, Fig. 3C and SI Appendix, Fig. S6A) for a large number of genes. However, we did not observe any clear CDS length dependency among mRNAs with significantly altered RF abundance or mRNA expression in DKO brains. We confirmed this transcriptome-wide by analyzing cumulative probabilities against LFCs of RF abundance (Fig. 3D) and mRNA expression (Fig. 3E) in DKO cortices. To compare DKO mice directly to their FXS littermates, we ordered mRNAs into 50 bins by their CDS lengths and examined their fold changes compared to WT mice. We observed that genetic reduction of S6K1 in FXS mice specifically normalized the aberrant bin-wise deviations in mRNA expression (Fig. 3F) and TE (Fig. 3G and SI Appendix, Fig. S6B), particularly for genes with especially long or short CDSs. Furthermore, examining the RF abundance and mRNA expression in FXS and DKO brains of the genes in the final four bins revealed that genetic reduction of S6K1 corrects overall alterations in gene expression by moving the RF abundance (Fig. 3H), and especially mRNA expression (Fig. 3I), of the entire bulk of genes toward the WT, rather than by rescuing specific genes. Identical evaluation of genes in the first four bins also showed a strikingly concordant effect (SI Appendix, Fig. S6 C and D).

Discussion

The TE of an mRNA historically has been deduced by the number of polysomes bound to it (16). Remarkably, an inverse correlation between CDS length and TE—the longer the CDS length of a transcript, the lower its ribosome density—has been observed in multiple model systems ranging from the protozoan Plasmodium falciparum to human cells (17). This relationship therefore suggests a CDS length-dependent effect on translational control that might be shared across phyla.

Because each RF reports on a single ribosome bound to an mRNA, the total number of RF reads aligning to a transcript directly reflects the overall number of polysomes bound to that mRNA species. Using ribosome profiling and RNA-Seq, we uncovered a positive-to-negative gradation of alterations in TE in the cortices of FXS model mice. This gradation suggests that the mechanisms regulating CDS length-dependent control of translation go awry with the loss of FMRP.

How does CDS length control translation? Using an elegant mathematical model, Fernandes et al. (17) proposed that mRNAs with short CDSs exhibit an increased rate of translation initiation. They assumed that the local concentration of ribosomal subunits around the 5′ untranslated region (UTR) of an mRNA determines its initiation rate. The shorter the CDS, the closer the 5′ end of an mRNA is to its 3′ end. After translation terminates, the concentration of ribosomal subunits at the 5′ UTR is therefore higher for short CDS mRNAs, which increases the initiation rate on those mRNAs. Ingolia et al. (9), meanwhile, proposed that mRNAs with long CDSs exhibit an increase in the rate of translation elongation. Consistent with previous polysome profiling studies, these researchers observed that shorter genes tended to have higher ribosome densities. Examination of the average ribosome densities of thousands of mRNAs revealed a high ribosome density toward the 5′ end of their CDSs, which tapered off and persisted at a uniform level toward the 3′ end. This trend suggested that the reduction in ribosome density with CDS length may be mediated by a relative increase in the rate of elongation on distal parts of the transcript on long CDS mRNAs (9). Therefore, in the absence of any exogenous changes, mRNAs with short CDSs are likely to exhibit increased rates of initiation, while those with long CDSs are likely to exhibit increased rates of elongation.

If these two proposals were to be true, what would be the consequences if FMRP were genetically removed from this system? Using molecular assays, we have provided substantial confirmatory evidence that the rate of translation elongation is globally elevated in FXS. FMRP loss therefore likely causes elevated rates of elongation on mRNAs with short or long CDSs. On mRNAs with especially short CDSs, the increase in the rate of elongation would in turn increase the rate of termination and subsequently elevate the local concentration of ribosome subunits around their 5′ UTRs. This increase would cause further enhancement in the rate of translation initiation on those mRNAs. On mRNAs with long CDSs, the elevated rate of elongation would cause additional reductions in ribosome density. Importantly, both of these effects would be determined directly by CDS length. This model would therefore predict that loss of FMRP causes an elevation in the TEs of mRNAs with short CDSs, a reduction in the TEs of mRNAs with long CDSs, and a CDS length-dependent gradation between the two extremes. Remarkably, the predictions of this model exactly match the alterations we observed in the cortices of FXS mice. The positive-to-negative linear gradation of alterations in RF abundance and TE with CDS length may, therefore, be an emergent property of an increased rate of ribosome translocation in FXS.

Given that CDS length-dependent control of translation may be a general feature across phyla, it is worthwhile to consider what evolutionary advantages it may serve. It has long been known that mRNAs of housekeeping genes have particularly short CDSs (18). Housekeeping genes, by definition, are constitutively, and often robustly, expressed in all human cells. Given the large energetic costs of transcription, translation, and protein folding (19–21), it is possible that natural selection has favored shorter CDSs for constitutively expressed genes. The short length of the mRNA also might allow cells to respond to stimuli by rapidly increasing the transcription and/or translation of these genes, thereby providing a benefit to respond rapidly to the environment.

Interestingly, housekeeping genes are enriched in mRNAs coding for components of the ribosome (18). Consistent with the model that FMRP exerts CDS length-dependent translational control, the short CDS mRNAs coding for ribosomal proteins are elevated in RF abundance and TE in FXS cortices (SI Appendix, Fig. S7). Conversely, there is evidence that FMRP binding partners are enriched in long mRNAs, which tend to harbor proportionately long CDSs (22). Again, consistent with the proposed model, FMRP binding partners exhibit reductions in RF abundance and TE in the cortices of mice lacking FMRP (SI Appendix, Fig. S7).

One limitation of the model proposed above is that the individual fold changes in RF abundance and mRNA expression are often small and are largely between ±1.3-fold. It is notable that despite the small individual effect sizes, we observe a linear gradation of alterations in RF abundance and mRNA expression. Because this analysis was carried out by examining the LFCs in over 12,000 mRNAs, it is very unlikely that such a gradation of effects would occur if the translation of only a limited number of mRNAs were affected in FXS. Our model therefore predicts that the vast majority of mRNAs are excessively translated in FXS cortices. This idea is consistent with the failure to find a specific motif or secondary structure for FMRP binding. Although studies have suggested that the ACUK motif is overrepresented in FMRP binding partners, it is improbable that it confers specificity to FMRP binding because it is present in every mRNA (11). Similarly, although it has been shown that the RGG box of FMRP binds to G-quadruplexes in vitro, the in vivo binding partners of FMRP do not show any enrichment for these secondary structures (10). It is therefore likely either that FMRP binds to a far larger number of mRNAs than is currently accepted or that its loss causes an indiscriminate increase in the rate of ribosome translocation on all mRNAs. Our data shows that this elevation is captured by the increase in de novo protein synthesis and is mediated by an overall increase in the rate of translation elongation.

It should also be noted that although we observed trends toward elevated protein expression of mRNAs with particularly long or short CDSs, we observed statistically significant elevations in the protein levels of only a small subset of the mRNAs we examined. We speculate that because de novo translation is elevated as a whole across the translatome, the effects on individual proteins are likely to be small; only when summated across the entire translatome are the effects robust. This reduction in the magnitude of effect sizes of individual proteins is evident in the small fold changes in RF abundance and mRNA expression, which are largely between 1.3- and −1.3-fold. Furthermore, using ribosome profiling and RNA-Seq, we only measured alterations in de novo translation in FXS cortices. The steady-state levels of individual proteins are determined by other factors as well, including their degradation kinetics, which make it challenging to validate the alterations by examining total protein levels in tissue lysates.

Another limitation of our model is that the alterations observed in both runoff-RPM and elongation SUnSET are more muted in the data comparing FXS neurons transduced with FMRP and those with GFP, as opposed to data comparing WT and FXS neurons. For instance, we observe a significant reduction in runoff-RPM signal at baseline in FXS neurons compared to WT (SI Appendix, Fig. S5 D and E). However, in FXS neurons transduced with human FMRP, the runoff-RPM signal at baseline is elevated, albeit without reaching statistical significance, compared to murine FXS neurons transduced with GFP. There are three reasons why this may happen. First, it is possible that human FMRP does not fully recapitulate the functions of mouse FMRP. Second, the amount of FMRP expressed virally may not be sufficient to fully rescue the deficits in translation observed in FXS neurons. Although we used strong, ubiquitously expressed promoters and observed robust Fl-FMRP expression in both immunofluorescence images and Western blots, we can only comment on the relative amount of FMRP expression and not on absolute amounts. Third, it is possible that FMRP is not expressed for long enough to fully recapitulate the effects of endogenous FMRP expression. In the experiment comparing WT and FXS neurons, endogenous FMRP is present throughout the duration of culture in FXS neurons, whereas in the viral transduction experiment, FMRP is present only for approximately 4 d. If FMRP loss causes systemic secondary alterations, it is possible that this duration of FMRP expression is too short to fully correct those changes.

It is also important to acknowledge that the bulk of evidence demonstrating an elevated rate of ribosome translocation in FXS was acquired from experiments on primary neurons lacking FMRP, while that determining altered CDS length control was obtained from cortical lysates. We chose to carry out cell biological experiments on cultured primary neurons for two reasons. First, the magnitudes of the effects on individual mRNAs were quite small. In the cerebral cortices of mice, neurons strongly express FMRP at P30; glial cells do not (23). Because glia are thought to be at least as abundant as neurons in cortical tissue (24), we reasoned that our signals were being diluted by glial cells, which—given their comparative absence of FMRP—likely do not show alterations between genotypes. We therefore moved to an ex vivo cortical culture model system. We carried out our cultures at E16 and E17, where the proportion of glial cells is minuscule. As a result, our cultures are almost entirely neuronal (for instance, in Fig. 2G, almost every cell stained for puromycin is also positive for Map2, which is a marker for neurons). Second, puromycin does not cross the blood brain barrier, which makes in vivo experiments far more challenging. Given the number of drugs, sequential steps involved and the constraint that intracerebroventricular infusions must be carried out very slowly, we were not able to implement the runoff-RPM assay in vivo.

Given our proposition that the loss of CDS length control of translation is likely the emergent mechanism for altered protein synthesis in FXS brains, it is remarkable that the genetic removal of S6K1 in FXS brains corrects this aberration. A reduction in the rate of elongation in DKO cortices is a molecular mechanism that would predict the rescue of alterations in CDS length control of translation in FXS. Consistent with this model, pharmacological inhibition of S6K1 corrected the increase in the rate of elongation in FXS neurons. However, the excitement of this finding was tempered by the observation that the LFCs in RF abundance in DKO cortices closely tracked those observed in FXS cortices, especially for mRNAs with the longest CDSs. Because reduced RF abundance likely reflects an increase in the rate of elongation, this correspondence suggests that the rate of ribosome translocation is not dampened by genetic removal of S6K1 in FXS mice. Instead, the rescue in CDS length control of translation in DKO cortices is primarily driven by the correction of the alterations in CDS length dependency of mRNA expression in FXS. This normalization is especially evident in mRNAs with the longest CDSs, the majority of which exhibit increases in FXS cortices but reductions in DKO cortices. However, it remains unclear how the rescue of CDS length dependency in mRNA expression arises.

The reduction of CDS length dependency of alterations in mRNA expression in DKO cortices suggests that S6K1 signaling modulates mRNA expression in FXS model mice. Increased phosphorylation of S6K1 and its canonical target S6 have been observed in the hippocampus of FXS brains, suggesting that S6K1 is hyperactivated (25). Previous studies suggest that S6K1 is required for ribosome biogenesis (26), and tissues that lack both S6K1 and S6K2 display widespread reductions in the mRNA expression of genes regulating ribosome biogenesis (27). Given the decrease in mRNA expression of genes encoding ribosomal proteins in FXS mice and its correction in DKO mice, it is possible that a rescue of this transcriptional program plays a role in correcting de novo protein synthesis in DKO mice. It should be noted that although this mechanism may account for the correction of LFCs in mRNA expression in genes with the shortest CDSs, it does not explain how the overall LFCs in the longest genes are also corrected in DKO mice.

There is also evidence that S6K1 controls transcription by phosphorylating the cAMP response element binding (CREB) protein isoform CREM (28). In addition, hyperactive S6K1 alters mRNA expression by modulating the activity of S6K1 Aly/REF-like target (SKAR), a nuclear protein that has been proposed to couple transcription with pre-mRNA splicing and mRNA export (29). Thus, a reduction in S6K1-SKAR signaling in the DKO mice also could participate in the rescue of altered mRNA expression in the FXS mice.

Because FMRP is thought to bind only to ∼1,000 mRNAs, it is also possible that the global increase in elongation rate is caused by a generalized increase in ribosome translocation due to hyperactivation of mammalian target of rapamycin complex 1 (mTORC1) to S6K1 signaling, which has been previously noted in FXS brains (25). This framework is consistent with the limited, albeit significant, overlap between genes with altered RF abundance and FMRP-binding targets and the reduction in the rate of translation elongation with inhibition of S6K1 in FXS neurons. Meanwhile, increased translation initiation due to elevated mTORC1 signaling to its other major downstream targets (4E-BP and/or CYFIP1) and resulting increases in eIF4F formation (30) explain the increase in footprint expression in the shortest genes. These genes are highly enriched in 5′ terminal oligopyrimidine mRNAs, which code for ribosomal proteins and are established translational targets of mTORC1 activation (SI Appendix, Fig. S7) (31). The positive-to-negative gradation in RF abundance and TE with CDS length argues for a model in which elevated rates of both initiation and elongation act in concert to elevate mRNA translation indiscriminately in FXS brains. It is possible that precisely because initiation is rate limiting, its increase is best captured in mRNAs with short CDSs. Meanwhile, the higher rate of elongation dominates in longer transcripts because ribosomes must translocate through a greater number of codons, with the net effect of each increased translocation event aggregating over the length of the CDS. Our proposed models are consistent with and provide mechanistic insights into previously proposed models of FMRP function (7, 11, 25, 30). Alternatively, it is possible that FMRP binding is more promiscuous than appreciated and biased toward longer mRNAs, which are enriched in neurons (32). Therefore, the increased rate of elongation is primarily observed in longer transcripts, and the secondary effect of increased rate of initiation is observed in shorter transcripts, which are translational targets of mTORC1 (SI Appendix, Fig. S7B).

Elevated rates of ribosome translocation also have been associated with increased mRNA stability (33). Specifically, an increase in the rate of decoding of cognate tRNAs at the ribosomal A site is thought to drive increases in mRNA stability. Because this process is likely elevated in every codon in FXS (7), the effect on mRNA stability is naturally summated over the length of a transcript. Notwithstanding the limitation that our experiments examining translation elongation were carried out in cultured neurons, this model accounts for the increased expression of mRNAs with long CDSs in FXS brains, as well as the negative-to-positive gradation in mRNA expression by CDS length.

Finally, based on our ex vivo findings with inhibition of S6K1, we speculate that genetic reduction of S6K1 causes destabilization of mRNAs in FXS mice. Recent work has demonstrated that reduction of cap-binding complex formation, which reduces translation initiation, causes a global reduction in mRNA stability (34). These findings are consistent with studies that showed a reduction in the stability and elevation of nonsense-mediated decay of specific transcripts upon inhibition of mTORC1 (35–38). Because it stimulates both initiation and elongation, it is possible that the genetic reduction of S6K1 causes a similar destabilization of mRNAs in DKO mice.

In addition to establishing bona fide targets for pharmacological interventions in FXS, our results ascertain the underlying biological principles of FMRP function in health and disease. Our work also suggests that drugs that reduce translation elongation on specific target mRNAs (39) may offer a therapeutic avenue in FXS.

Materials and Methods

Supplementary Material

Data Availability

All data used in the analysis have been deposited under record number GSE143659 in the National Center for Biotechnology Information Gene Expression Omnibus database. All other study data are included in the article and/or SI Appendix.