Brain-enriched RagB isoforms regulate the dynamics of mTORC1 activity through GATOR1 inhibition

The Rag isoforms are differentially expressed in tissues

The Rag GTPases are central components of the molecular machinery regulating mTORC1 in response to nutrients. Unclear is whether RagA is functionally equivalent to RagB, and RagC to RagD, and whether different combinations of Rag proteins cause different responses of mTORC1 to nutrients. Transcriptomic data show that the relative expression of Rag GTPases varies in different human tissues, suggesting that different repertoires of Rag GTPases may be present in different cell types (Fig. 1a). Similarly, we also detected differential levels of RagA, RagB and RagC proteins in mouse tissues (Fig. 1b) (no working antibody for mouse RagD is commercially available). In particular, the main RagB isoform (hereafter named RagBshort) is expressed at low levels in most tissues, but at higher levels in the brain, where additionally a longer splice isoform with unknown function is expressed (hereafter named RagBlong). RagC levels are lower in skeletal muscle and heart, where RagD is highest according to transcriptomic data (Fig. 1a), indicating that RagD could be the predominant isoform in these tissues.

Fig. 1: RagA/B paralogues determine distinct mTORC1 responses.figure 1

a, Transcript levels of the Rag isoforms in healthy human tissues (gtexportal.org). n, biological replicates. b, Western blot for RagA, RagB and RagC in mouse tissues. Calnexin is the loading control. The experiment was repeated once. c,d, Domain organization of the Rag isoforms. Numbering indicates amino-acid positions in the human sequence. Percentages represent similarity of each domain between Rag paralogues. Ex4 is the sequence encoded by exon 4 of the Rragb gene. eh, S6K1, TFEB and 4EBP1 phosphorylation in control or RagABKO cells stably transfected with a control protein (FLAG–metap2) or with the indicated Rag isoforms. Cells were incubated in amino-acid-rich medium or starved of amino acids for 30 min: representative example (e) and quantification of three independent experiments, with unstarved control cells set to 1 (fh). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. i,j, S6K1 phosphorylation upon loss of RagA, RagB or both: representative example (i) and quantification of three independent experiments, with control cells set to 1 (j). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. One-way ANOVA and Tukey’s post-hoc test. kn, RagA (k and l) but not RagB (m and n) loss causes persistent mTORC1 activity. Cells were incubated in amino-acid-rich medium, starved of amino acids for 1 h, or starved for 1 h and re-stimulated with amino acids for 15 min (addback, ‘ab’): representative examples (k and m) and quantification of three independent experiments, with unstarved control cells set to 1 (l and n). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. op, The elevated mTORC1 activity in RagAKO cells upon amino-acid removal cannot be rescued by stable overexpression of the RagB isoforms. Cells were incubated in amino-acid-rich medium or starved of amino acids for 30 min: representative example (o) and quantification of three independent experiments, with unstarved control cells set to 1 (p). Bar height indicates average, error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. −aa, amino-acid-free DMEM + 10% dFBS. +aa, −aa medium supplemented with 1× amino acids. Exact P values are shown in the graphs. Source numerical data and unprocessed blots are available in source data.

Source data

Although RagA/B and RagC/D proteins are almost identical in the GTPase and CRD domains, they diverge substantially in the most N- and/or C-terminal regions (Fig. 1c,d). RagBshort and RagBlong have a 33-amino acid N-terminal extension that is absent in RagA. Additionally, RagBlong contains a stretch of 28 amino acids encoded by exon 4 that is inserted in the switch I region, which changes conformation upon GTP binding and is responsible for effector binding. Analogously, two poorly structured N- and C-terminal extensions present in RagC and RagD exhibit only 25% and 39% similarity, respectively, between the two paralogues. Together, these differences and the non-homogeneous tissue distribution of the four Rag GTPases raise the possibility that they might have specific functions in certain cell types and/or conditions.

RagB isoforms are more resistant to amino-acid removal than RagA

To study if the Rag isoforms differ functionally, we used HEK293T cells, which are often employed to study mTOR signalling. We generated RagA and RagB double-knockout HEK293T cells (RagABKO) (Extended Data Fig. 1a–c), which we then reconstituted with either RagA, RagBshort or RagBlong to yield cells containing only one Rag paralogue (Fig. 1e–h). Likewise, we generated RagC/D double-knockout cells (RagCDKO) (Extended Data Fig. 1d–f) and reconstituted them with either RagC or RagD (Extended Data Fig. 1h–k). This approach yields cells expressing comparable levels of the different Rag paralogues, thereby revealing effects caused by differences in function rather than expression. Indeed, HEK293T cells endogenously express more RagA than RagB mRNA (Extended Data Fig. 1g) and protein (Extended Data Fig. 1a). Deletion of RagA/B caused a decrease in RagC/D protein levels and vice versa (Extended Data Fig. 1a,d), which was rescued upon reconstitution with single Rag paralogues (Fig. 1e and Extended Data Fig. 1h), suggesting Rag monomers are unstable.

We then assessed mTORC1 activity via phosphorylation of S6K1, TFEB and 4EBP1, direct mTORC1 substrates that respond rapidly to changes in nutrient levels30,31 (Extended Data Fig. 2a,b). Double knockout of RagA/B or RagC/D caused a decrease, but not a complete loss, of mTORC1 activity and rendered the residual mTORC1 activity largely unresponsive to amino-acid withdrawal or re-addition (Extended Data Fig. 1a–f). Consistent with Rag depletion, immunofluorescence experiments showed loss of mTOR accumulation on lysosomes in RagABKO and RagCDKO cells under nutrient-replete conditions (Extended Data Fig. 3a–d). Low levels of lysosomal mTOR and Raptor, however, could still be detected by immunoblotting-purified lysosomes (lyso-IP) from RagA/B or RagC/D double knockouts (Extended Data Fig. 3e–g), probably contributing to the residual S6K phosphorylation in double knockouts. These results are consistent with previous work showing that, in the absence of all Rag isoforms, mTOR can still be recruited to lysosomes in an Arf-1-dependent manner32 and that upon amino-acid starvation the inactive Rag GTPases not only release mTORC1 from the lysosome, but also actively recruit factors that inactivate mTORC1, such as the TSC complex, causing persistent mTORC1 activity when all Rag isoforms are missing33. We also noticed that RagC/D did not localize to lysosomes in the absence of RagA/B and vice versa (Extended Data Fig. 3e,h–k), suggesting that assembly of functional Rag dimers is necessary not only for their stabilization, but also for their delivery to lysosomes.

Interestingly, re-expression of the three RagA/B isoforms in RagABKO cells yielded distinct patterns of mTORC1 responses. In nutrient-replete conditions, reconstitution with RagA or RagBshort increased S6K1, 4EBP1 and TFEB phosphorylation to comparable levels, while these were lower in RagBlong-expressing cells (Fig. 1e–h). Furthermore, while phosphorylation of the three substrates dropped strongly upon amino-acid starvation in RagA-expressing cells, it dropped less strongly in RagBshort- and RagBlong-expressing cells, indicating persistent mTORC1 activity. This suggests that RagBshort and RagBlong keep mTORC1 more active despite amino-acid removal. All three RagA/B isoforms interact similarly with the RagC/D isoforms in co-immunoprecipitation (co-IP) experiments (Extended Data Fig. 3l–n), indicating that the phenotypic differences between RagA and RagB are not due to differences in RagC/D binding. Reconstitution of RagCDKO cells with RagC versus RagD showed differences in the phosphorylation of TFEB but not S6K1 or 4EBP1 in response to nutrients (Extended Data Fig. 1h–k), suggesting that these two isoforms might have differential effects on a subset of mTORC1 substrates. In this manuscript we focus on the functional differences between RagA and RagB, while the accompanying manuscript by Demetriades and colleagues focuses on RagC versus RagD34.

The effects of RagA, RagBshort and RagBlong on mTOR localization correlated with their effect on mTORC1 activity. Stable transfection of RagABKO cells with RagA rescued mTOR localization to a large extent to wild-type behaviour, with predominantly lysosomal accumulation in nutrient-rich conditions and cytosolic localization upon amino-acid removal (Extended Data Fig. 4a,b). Cells reconstituted with RagBshort were able to recruit mTOR to lysosomes in nutrient-replete conditions, but still retained significant amounts of mTOR on lysosomes upon amino-acid removal, consistent with the persistent mTORC1 activity observed by western blot (Fig. 1e–h). In contrast, RagBlong failed to increase the lysosomal localization of mTOR in either immunofluorescence or lyso-IP experiments (Extended Data Fig. 4a–d), suggesting that RagBlong interacts poorly with mTORC1. Indeed, RagBlong co-immunoprecipitated substantially less Raptor than RagA or RagBshort, although a weak interaction could still be detected (Extended Data Fig. 4e,f).

We observed the same phenotypic differences between RagA and RagB in a complementary experimental set-up, where we knocked out either RagA or RagB, leaving the cells to express only the other endogenous paralogue. Worth noting, deletion of RagA led to a mild compensatory increase in RagBshort levels together with the appearance of RagBlong, which is undetectable in control cells (Fig. 1i). RagAKO cells had lower basal mTORC1 activity than control cells, but higher than RagA/B double-knockout cells (Fig. 1i,j), indicating that RagB can partially compensate for loss of RagA. Consistent with the possibility that RagB is more resistant to nutrient removal than RagA, amino-acid removal caused only a mild drop in S6K1 phosphorylation in RagAKO cells, but a complete loss in control and RagBKO cells (Fig. 1k–n). Likewise, mTOR accumulation on lysosomes decreased only mildly in RagAKO cells during amino-acid starvation, in contrast to the complete re-localization to the cytoplasm in control cells (Extended Data Fig. 4g,h). Importantly, the persistent mTORC1 activity in RagAKO cells could be reverted fully by re-expressing RagA, only partially by RagBshort and not at all by RagBlong (Fig. 1o,p), confirming that the persistent mTORC1 activity stems from qualitative and not quantitative differences between RagA and the RagB isoforms.

In sum, these results show that the RagA/B isoforms are not functionally redundant: (1) RagBshort and RagBlong are more resistant to nutrient withdrawal compared with RagA, and (2) RagBlong does not bind and recruit mTOR to lysosome as efficiently as the other isoforms.

RagBshort and RagBlong are resistant to GATOR1

Amino-acid removal activates GATOR1, which acts as a GAP for RagA/B to promote GTP hydrolysis and subsequent release of mTORC1 from the lysosome16. One possible explanation why mTORC1 activity remains high in RagB-expressing cells upon amino-acid removal is that GATOR1 does not stimulate the RagB isoforms to hydrolyse GTP as efficiently as it does RagA. Alternatively, the RagB isoforms efficiently hydrolyse GTP to GDP but their non-GTP-bound conformations still bind mTORC1. To test the latter option, we first performed a co-IP experiment between the three RagA/B isoforms in different nucleotide loading states and the mTORC1 subunit Raptor. GTP-locked RagA and RagBshort interacted comparably with Raptor, while GTP-locked RagBlong bound more weakly (Extended Data Fig. 5a,b), consistent with the reduced ability of wild-type RagBlong to bind mTORC1 (Extended Data Fig. 4a,b,e,f). In contrast, none of the three RagA/B isoforms in the non-GTP-bound state interacted with Raptor (Extended Data Fig. 5a,b), indicating that inactive RagBshort and RagBlong should not be retaining mTORC1 on the lysosome. Consistent with these results, strong overexpression of GDP-locked RagA, RagBshort or RagBlong rescued mTORC1 activity, enabling it to be low in RagAKO cells upon amino-acid removal (Extended Data Fig. 5c–f). Together, these data indicate that, if the RagB isoforms hydrolyse GTP to GDP, they release mTORC1 and allow it to turn off.

We therefore tested the alternate explanation, that the RagB isoforms are more resistant to GATOR1 than RagA. We first tested this in nutrient-replete conditions by transfecting cells expressing the single RagA or B isoforms with increasing amounts of GATOR1 (Fig. 2a,b). Although high levels of GATOR1 inhibited mTORC1 in all cells, lower levels of GATOR1 overexpression induced a stronger reduction of mTORC1 activity in RagA- than in RagBshort-expressing cells, while high levels of GATOR1 were required to cause an appreciable drop in S6K1 phosphorylation in RagBlong-expressing cells (Fig. 2a,b, Extended Data Fig. 6a,b). Together, these data indicate that the RagB isoforms are comparatively resistant to GATOR1.

Fig. 2: RagBshort and RagBlong are resistant to GATOR1.figure 2

a,b, RagABKO cells expressing each RagA/B isoform were transiently transfected with increasing amounts of GATOR1 plasmids (5 ng, 25 ng or 100 ng of each GATOR1 subunit) or metap2 (100 ng) as negative control: representative example (a) and quantification of four independent experiments, with RagA-expressing cells transfected with metap2 set to 1 (b). Bar height indicates average, and error bars represent standard deviation; n = 4 biological replicates. Two-way ANOVA and Tukey’s post-hoc test. c,d, Control and RagAKO cells were transiently transfected with high (200 ng) levels of each GATOR1 subunit or metap2 as negative control and treated with amino-acid-rich medium or starved of amino acids for 30 min before lysis: representative example (c) and quantification of three independent experiments, with unstarved metap2-transfected control cells set to 1 (d). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. e, Schematic representation of the two binding interfaces of GATOR1 to RagA/B. fh, Rag interaction with the inhibitory interface (depicted in f) is assessed by co-immunoprecipitating the whole GATOR1 complex: representative example (g) and quantification of three independent experiments, with the inactive mutant of RagA set to 1 (h). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. One-way ANOVA and Tukey’s post-hoc test. ik, Rag interaction with the GAP interface (depicted in i) is assessed through co-IP with the Nprl2/3 dimer in DEPDC5KO cells: representative example (j) and quantification of three independent experiments, with the inactive mutant of RagA set to 1 (k). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. One-way ANOVA and Tukey’s post-hoc test. −aa, amino-acid-free DMEM + 10% dFBS. +aa, −aa medium supplemented with 1× amino acids. Exact P values are shown in the graphs. NS, not significant. Source numerical data and unprocessed blots are available in source data.

Source data

To confirm that relative resistance to GATOR1 is the cause of high mTORC1 activity in RagB-expressing cells upon amino-acid removal, we performed two epistasis experiments. First, we verified that overexpression of GATOR1 rescues this phenotype. Indeed, GATOR1 overexpression rescued the persistent mTORC1 activity observed in RagAKO cells upon amino-acid deprivation, causing it to decrease to the same level as control cells (Fig. 2c,d). Second, the phenotypic difference between RagA and RagB should be gone in cells lacking GATOR1. To this end, we knocked out the GATOR1 subunit DEPDC5 in RagABKO cells and stably transfected them with RagA, RagBshort, RagBlong or metap2 as a control. Indeed, cells expressing RagA or RagBshort in a DEPDC5KO background have high mTORC1 activity upon amino-acid starvation, as expected, but without noticeable differences between these two Rag isoforms (Extended Data Fig. 6c–e), indicating that differential resistance to GATOR1 is the main functional difference between RagA and RagBshort. Consistent with the low binding of Raptor to RagBlong, mTORC1 activity in RagBlong-expressing cells remained substantially lower in all nutrient conditions (Extended Data Fig. 6c–e).

In sum, these results suggest that the three RagA/B isoforms have different resistance to GATOR1 in the order RagBlong > RagBshort > RagA and that this different resistance to GATOR1 is probably the main functional distinction between RagBshort and RagA.

RagBshort inhibits GATOR1 through binding via DEPDC5

The relative resistance of the RagB isoforms to GATOR1 suggests that either they are poor substrates of the GATOR1 complex or they actively inhibit GATOR1, or both. The interaction between GATOR1 and the Rag GTPases consists of two binding interfaces (Fig. 2e)17. At the GAP interface, the Nprl2/3 subunits of GATOR1 bind with low affinity to RagA/B to provide the arginine finger (R78 of Nprl2) necessary for GTP hydrolysis. Unlike other known GAPs and their target GTPases, GATOR1 and the Rag GTPases also have an additional, high-affinity interaction between the DEPDC5 subunit of GATOR1 and switch I of RagA/B, which does not execute any GAP activity. As expression of a DEPDC5 mutant that does not bind to the Rag GTPases results in stronger mTORC1 suppression than its wild-type counterpart17, this binding mode is thought to inhibit the GAP activity of GATOR1 and has therefore been named the inhibitory interface. Recent structural studies suggest that binding of GATOR1 to Rags via the inhibitory interface holds GATOR1 in an orientation relative to the lysosomal surface that is unfavourable for acting as a GAP on adjacent Rag molecules35.

We first tested how the three isoforms interact with GATOR1 at the two binding interfaces. As Rag binding to the GAP interface is approximately 40-fold weaker than binding to the inhibitory interface17,18, co-IP of the Rag GTPases with the entire GATOR1 complex reflects mainly binding to the inhibitory interface (Fig. 2f–h). In parallel, we co-immunoprecipitated the Rag GTPases with Nprl2/3 from DEPDC5KO cells to exclude binding via the inhibitory interface, thereby specifically assessing the GAP interface (Fig. 2i–k). Interestingly, each RagA/B isoform exhibited a distinct profile of GATOR1 interaction. RagBshort interacted less than RagA with the inhibitory interface, but interacted similar to RagA with the GAP interface. RagBlong interacted even less with the GATOR1 inhibitory interface as compared with RagA and RagBshort, but more strongly than the other two isoforms with the GAP interface, when non-GTP bound.

We focused first on the differences between RagA and RagBshort. To compare how strongly RagA versus RagBshort inhibit GATOR1 via the inhibitory interface, we compared cells expressing wild-type DEPDC5 with cells expressing a DEPDC5 mutant (Y775A) that does not bind the Rag GTPases on the inhibitory interface17. We did this by reconstituting RagA/B–DEPDC5 triple-knockout cells with single Rag isoforms and either wild-type or mutant DEPDC5. As expected, expression of wild-type DEPDC5 in RagA- or RagBshort-expressing cells caused a reduction in mTORC1 activity, because this reconstitutes the GATOR1 complex (lanes 1–4 in Fig. 3a–d). In RagBshort-expressing cells, DEPDC5Y775A expression led to an even stronger inhibition of mTORC1 compared with wild-type DEPDC5, because the DEPDC5Y775A mutant cannot bind RagBshort protein and thus cannot be inhibited by it17 (lane 2 versus lane 5, Fig. 3c,d). Thus, the difference between lane 2 and lane 5 of Fig. 3c,d reflects the inhibitory activity of RagBshort on GATOR1 through DEPDC5 binding. In RagA-expressing cells, however, DEPDC5Y775A expression caused the same degree of mTORC1 inhibition as wild-type DEPDC5 (Fig. 3a,b). Hence, although RagA binds DEPDC5, it does not cause GATOR1 inhibition as much as when RagBshort binds DEPDC5.

Fig. 3: RagBshort inhibits GATOR1 through interaction with DEPDC5.figure 3

ad, Transient expression of increasing amounts (5 ng, 25 ng or 100 ng DNA) of a non-Rag binding mutant of DEPDC5 (Y775A) suppresses mTORC1 more strongly than wild-type DEPDC5 in RagBshort-expressing cells (c and d) but not in RagA-expressing cells (a and b). Cells were subjected to amino-acid starvation (amino-acid-free DMEM + 10% dFBS) for 30 min to activate GATOR1: representative examples (a and c) and quantifications of three independent experiments, with metap2-transfected cells set to 1 (b and d). Circle indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. eh, Transient expression of increasing amounts (5 ng, 25 ng or 100 ng DNA) of wild-type DEPDC5 (e and f) inhibits mTORC1 activity more strongly in RagA-expressing cells than in RagBshort-expressing cells, whereas expression of a non-Rag binding mutant of DEPDC5 (Y775A) (g and h) inhibits mTORC1 activity equally well in the presence of RagA or RagBshort. Cells were subjected to amino-acid starvation (amino-acid-free DMEM + 10% dFBS) for 30 min to activate GATOR1: representative examples (e and g) and quantification of three independent experiments, with RagA-expressing cells transfected with metap2 set to 1 (f and h). Circle indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. i, Schematic representation of GATOR1 binding to RagA or RagBshort via DEPDC5. DEPDC5 binding to RagBshort but not to RagA inhibits GATOR1 activity. Exact P values are shown in the graphs. NS, not significant. Source numerical data and unprocessed blots are available in source data.

Source data

If this is the case, then the difference between RagA and RagBshort should be gone if DEPDC5 cannot bind the Rag proteins. Indeed, re-expression of wild-type DEPDC5 inhibited mTORC1 more strongly in RagA-expressing cells than in RagBshort-expressing cells (lane 3 versus lane 7, Fig. 3e,f), consistent with RagBshort being comparatively resistant to GATOR1, while this difference was gone in cells expressing DEPDC5Y775A (Fig. 3g,h). These results indicate that RagBshort is resistant to GATOR1 activity not because it is less sensitive to the GAP activity (indeed, RagBshort binds equally well to the GAP side of GATOR1 as RagA, Fig. 2i–k), but rather because RagBshort, but not RagA, is able to inhibit GATOR1 through DEPDC5 binding (Fig. 3i).

RagBshort differs from RagA at the N-terminal extension and at five other amino acids, four of which are located in the CRD domain and one in the C-terminal part of the GTPase domain (Extended Data Fig. 6f). To determine which of these features is responsible for the functional difference between RagA and RagBshort, we generated a mutant of RagBshort lacking the N-terminal extension (ΔN mutant) or a mutant of RagBshort where the five amino acids are swapped to the RagA version (AQVHS mutant) (Extended Data Fig. 6f). Only removal of the N-terminal extension restored the interaction with GATOR1 to the same level as RagA (Extended Data Fig. 6g,h), as well as the ability to inactivate mTORC1 upon amino-acid removal (Extended Data Fig. 6i,j). This is consistent with the spatial proximity between the N-terminal extension of RagB and the switch I region that mediates GATOR1 binding via DEPDC5.

We aimed to recapitulate these effects in an in vitro GAP assay. We purified GATOR1 from HEK293T cells and the two Rags from bacteria as dimers with a RagC mutant (S75N) that abolishes GTP binding so that only the GTPase activity of the RagA or RagB isoforms would be measured (Extended Data Fig. 7a,b). We used a multiple-turnover GAP assay with malachite green to detect the phosphate released from GTP (Extended Data Fig. 7c), and we immobilized the purified RagA•RagCS75N and RagBshort•RagCS75N dimers on the surface of beads, since the spatial orientation of GATOR1 and neighbouring Rag dimers could be important for the inhibitory mechanism through DEPDC5 binding, as discussed above35. Using this set-up, we found that also in vitro GATOR1 had reduced activity towards RagBshort compared with RagA (Extended Data Fig. 7d) and this difference was reduced when using GATOR1 containing DEPDC5Y775A (Extended Data Fig. 7e), consistent with the in vivo results (Fig. 3a–h).

In sum, these results suggest that the N-terminal extension of RagBshort enables it to inhibit GATOR1 via DEPDC5.

RagBlong has low affinity for GTP

We next turned our attention to RagBlong. RagBlong binds weakly to Raptor (Extended Data Figs. 4e,f and 5a,b) and to the inhibitory interface of GATOR1 (Fig. 2f–h). As both interactions are enhanced when RagA/B are loaded with GTP, one possible explanation is that RagBlong has reduced affinity for GTP. Indeed, the 28-amino-acid insertion in RagBlong resides within the switch I region, which forms part of the GTP-binding pocket. A GTP pull-down assay revealed that RagBlong interacts much less with GTP than RagA or RagBshort, almost at background levels (Fig. 4a–c). These results are in line with a previous report in which a radiolabelled-GTP binding assay was employed11. As additional confirmation, we also assessed the interaction of the RagA and RagB isoforms with the p18 subunit of Ragulator and the folliculin complex (FLCN–FNIP2), two complexes that sense the nucleotide loading state of the Rag GTPases but do not interact directly with switch I (refs. 24,25,36,37). As expected, mutations that disrupt GTP binding led to strong interaction of RagA or RagBshort with both FLCN–FNIP2 and p18, while mutation of the catalytic glutamine causing RagA or RagBshort to lock into GTP binding abrogated or reduced FLCN–FNIP2 and p18 interaction (Extended Data Fig. 8a–d). In contrast, the analogous mutant of RagBlong that cannot hydrolyse GTP retained substantial interaction with FLCN–FNIP2 and p18 (Extended Data Fig. 8a–d), consistent with RagBlong having impaired GTP binding. Likewise, wild-type RagBlong bound p18 more strongly than RagA or RagBshort (Extended Data Fig. 8e,f). In sum, all these data indicate that RagBlong binds GTP less well than RagA or RagBshort.

Fig. 4: RagBlong acts as a ‘sponge’ for the GAP interface of GATOR1.figure 4

a, Schematic representation of the GTP pull-down assay. b,c, GTP pull-down of RagA, RagBshort or RagBlong in the presence of magnesium or with addition of 20 mM EDTA as negative control: representative example (b) and quantification of three independent experiments, with RagA Mg2+ set to 1 (c). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. Two-way ANOVA and Sidak’s post-hoc test. d,e, Co-IP of Nprl2 and Nprl3 with Rag dimers consisting of wild-type RagC and GTP-locked RagA (Q66L) expressed with or without non-GTP-bound (T54N) RagBlong. The experiment was performed using RagA/B and DEPDC5 triple-knockout cells to assess specifically the binding to the GAP interface of GATOR1: representative example (d) and quantification of three independent experiments, with the Nprl2/3 condition set to 1 (e). Bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. One-way ANOVA and Tukey’s post-hoc test. f,g, S6K1 phosphorylation in RagABKO cells transiently transfected with RagA either alone or together with wild-type RagBlong or a non-GTP-bound mutant of RagBlong (T54N) during amino-acid starvation (amino-acid-free DMEM + 10% dFBS): representative example (f) and quantification of six independent experiments, with RagA-expressing cells at timepoint 0 set to 1 (g). Circle/square indicates average, and error bars represent standard deviation; n = 6 biological replicates. Two-way ANOVA and Tukey’s post-hoc test. h, Coomassie staining of RagBlongT54N and RagBshortT54N proteins purified from RagABKO HEK293T cells as dimers with FLAG-tagged RagCQ120L (1 μg dimer per lane). The experiment was repeated once. i,j, Malachite-green GTPase assay with 1 μM of RagA•RagCS75N either alone or mixed with an equimolar amount of RagBlongT54N•RagCQ120L (i) or RagBshortT54N•RagCQ120L (j) in solution in the presence of the indicated amounts of GATOR1. Quantification of four (i) or three (j) independent experiments. Circle indicates average, and error bars represent standard deviation; n = 3 (j) and 4 (i) replicates. Two-way ANOVA and Sidak’s post-hoc test. k, Schematic representation of the mechanism whereby non-GTP-bound RagBlong titrates away the GAP interface of GATOR1. Exact P values are shown in the graphs. Source numerical data and unprocessed blots are available in source data.

Source data

To understand why RagBlong has reduced GTP binding, we mutagenized individually all the 28 amino acids encoded by exon 4 of Rragb. Among all the mutants screened, we found that only L94A could increase GTP binding, albeit only mildly (Extended Data Fig. 8g,h and data not shown). This suggests that the low affinity of RagBlong for GTP depends on complex structural features rather than on the identity of one specific residue. Consistent with increased GTP binding, addition of the L94A mutation to the GTP-locking mutant of RagBlong decreased its affinity for FLCN–FNIP2, but not the neighbouring D96A mutation that does not increase GTP binding (Extended Data Fig. 8i,j). Finally, we tested to what extent the weak binding of RagBlong to GATOR1 via DEPDC5 depends on its particular switch I sequence or its N-terminal extension. Consistent with the additional 28-amino-acid loop in the switch I sequence of RagBlong imposing a large constraint, removal of the N-terminal extension of RagBlong did not visibly improve GATOR1 binding via DEPDC5 (Extended Data Fig. 8k,l). In sum, the 28-amino-acid insertion in RagBlong impairs both GTP binding and GATOR1 binding via DEPDC5.

Consistent with these in vivo data, in vitro GAP assays revealed that GATOR1 was hardly able to stimulate GTP hydrolysis by RagBlong purified from HEK293T cells (the yield of RagBlong purified from bacteria was too low) (Extended Data Fig. 9a–c), whereas it efficiently stimulated GTP hydrolysis by RagA purified from either bacteria or HEK293T cells (Extended Data Fig. 9c,d).

RagBlong titrates away the GAP interface of GATOR1

The low affinity of RagBlong for GTP and its strong interaction with the GAP interface of GATOR1 when non-GTP bound (Fig. 2i–k) made us reason that a substantial pool of non-GTP-bound RagBlong could act as a ‘sponge’ that binds GATOR1 on the GAP interface and titrates it away from RagA, RagBshort or the small pool of GTP-bound RagBlong in a cell. In this way, RagBlong would act as a GATOR1 inhibitor. This possibility is consistent with the RagA/B proteins being stoichiometrically in great excess compared with the GATOR1 subunits (RagA/B, 58,467.8 protein copies per cell; DEPDC5, 649.6; Nprl2, 11,268.6; Nprl3, 9,860.6; data from HeLa cells)38. Indeed, we found that non-GTP-bound RagBlong harbouring the T54N mutation outcompetes GTP-locked RagA for Nprl2/3 binding (Fig. 4d,e). By acting as a GATOR1 inhibitor, RagBlong should confer increased resistance to nutrient starvation when co-expressed with the other RagA/B isoforms. Indeed, co-expression of either wild-type RagBlong or the RagBlongT54N mutant together with RagA caused less inactivation of mTORC1 upon amino-acid starvation compared with expression of RagA alone (Fig. 4f,g). As the RagBlongT54N mutant cannot bind mTORC1, this is consistent with RagBlong acting via GATOR1 inhibition rather than direct mTORC1 activation.

We next tested if RagBlong can inhibit GATOR1 GAP activity towards RagA also in vitro through this mechanism. We first confirmed that also in vitro RagBlong binds the GAP interface of GATOR1 more strongly than RagA or RagBshort using purified GDP-loaded Rag dimers and a GATOR1 complex containing DEPDC5Y775A to disrupt the inhibitory interface (Extended Data Figs. 7b and 9e,f). Next, to test if RagBlong can inhibit GATOR1 specifically via the GAP interface, we purified RagBlongT54N, which interacts strongly with the GAP interface of GATOR1 but not its inhibitory interface (Fig. 2f–k), as a dimer with GTP-locked RagCQ120L (Fig. 4h) to mimic the Rag conformation probably occurring in vivo owing to inter-subunit crosstalk15. Consistent with direct GATOR1 inhibition, the GAP activity of GATOR1 towards RagA was lower in the presence of equimolar amounts of RagBlongT54N (Fig. 4h,i) but not of the analogous RagBshortT54N (Fig. 4j). This indicates that inhibition of GATOR1 via GAP-interface binding is a specific feature of RagBlong, and that RagBshort instead only inhibits GATOR1 via the inhibitory interface (Fig. 3 and Extended Data Fig. 7).

In sum, we find that RagBlong inhibits GATOR1 by acting as a ‘sponge’ for its GAP interface to potentiate signalling through the other Rag isoforms (Fig. 4k).

RagBshort and RagBlong additively inhibit GATOR1

Since RagBshort and RagBlong inhibit GATOR1 activity through distinct mechanisms—RagBshort via DEPDC5 binding and RagBlong via GAP-interface binding—we predicted that co-expression of both isoforms together with RagA, as physiologically observed in the brain, would have additive effects. To this end, we transfected RagABKO cells with equal total amounts of RagA, RagA + RagBshort, or RagA + RagBshort + RagBlong. Cells transfected with all three isoforms had total Rag protein levels similar to cells transfected with only RagA (Extended Data Fig. 9g), thus allowing us to see differences in RagA versus RagB function rather than levels. In line with our prediction, we observed that, although complete and prolonged amino-acid starvation caused comparable mTORC1 inhibition in all cases, co-expression of all three RagA/B isoforms enabled cells to maintain elevated mTORC1 activity when amino-acid levels were reduced to 50%, 25% or 10% of normal cell culture levels, compared with cells expressing only RagA or RagA + RagBshort (Fig. 5a–i). Analogously, in a time course of complete amino-acid

留言 (0)

沒有登入
gif