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

Mechanistic target of rapamycin complex 1 (mTORC1) senses nutrient availability to appropriately regulate cellular anabolism and catabolism. During nutrient restriction, different organs in an animal do not respond equally, with vital organs being relatively spared. This raises the possibility that mTORC1 is differentially regulated in different cell types, yet little is known about this mechanistically. The Rag GTPases, RagA or RagB bound to RagC or RagD, tether mTORC1 in a nutrient-dependent manner to lysosomes where mTORC1 becomes activated. Although the RagA and B paralogues were assumed to be functionally equivalent, we find here that the RagB isoforms, which are highly expressed in neurons, impart mTORC1 with resistance to nutrient starvation by inhibiting the RagA/B GTPase-activating protein GATOR1. We further show that high expression of RagB isoforms is observed in some tumours, revealing an alternative strategy by which cancer cells can retain elevated mTORC1 upon low nutrient availability.

M echanistic target of rapamycin complex 1 (mTORC1) senses and integrates inputs from nutrients and growth factors to promote cell growth by stimulating cellular anabolic processes while inhibiting catabolic ones 1,2 . As a consequence, mTORC1 ensures that cell growth and metabolism are in line with the availability of biomolecular building blocks and the overall growth status of the organism. In metazoans, mTORC1 has different functions in different tissues. mTORC1 regulates synaptic transmission and myelination in the nervous system 3,4 , trophism in the skeletal muscle 5 and triglyceride synthesis and adipocyte differentiation in the adipose tissue 6 , to name a few. Such diversity of processes regulated by mTORC1 raises the question of whether the mTORC1 pathway is identically regulated in all tissues, or whether it can be tuned to the specific needs of a tissue or cell type. Consistent with the latter possibility, fluctuations of nutrient levels have different growth outcomes in different tissues. When nutrients become limiting to an organism, most tissues and organs enter a low-metabolic or catabolic state, while a few tissues and organs that provide vital functions to the organism, such as the brain or the heart, are notably spared. As an extreme example, severe nutrient restriction during intra-uterine growth causes the central nervous system to grow at the expense of other organs and results in small newborns with large heads, a process known as brain sparing 7 . Likewise, the brain is spared during acute hypoxia by re-routing cardiac output away from peripheral vascular beds 8 . As mTORC1 plays a central role in regulating cellular growth and metabolism, one could imagine that its activity is differentially modulated in response to nutrient limitation in different tissues and cell types. To our knowledge, such molecular mechanisms are not known.
A sophisticated molecular machinery has evolved that allows mTORC1 to sense nutrients. At the heart of the nutrient sensing capacity is a group of GTPases belonging to the Ras superfamily named the Rag GTPases in metazoa, and Gtr proteins in yeast [9][10][11][12] . Rag GTPases have two peculiar features that make them exquisitely dynamic, as expected for a system that has to respond rapidly to changing nutrient levels. First, unlike most other GTPases, the Rag GTPases do not associate with membranes directly through a lipid modification, but rather interact with a protein complex ('Ragulator') that is located on the lysosomal surface, the signalling hub of the mTORC1 pathway 13 . This allows the Rag GTPases to shuttle on and off the lysosome according to their nucleotide loading state 14 , which could regulate the amplitude of the response to nutrients. Second, unlike most other Ras-like GTPases, they function as obligate heterodimers, consisting in mammals of a RagA or RagB protein heterodimerizing with either RagC or RagD through complementary C-terminal roadblock domains (CRD). The two subunits of the heterodimer affect each other's GTP binding and hydrolysis rate 15 . Such inter-subunit crosstalk causes the dimer to quickly lock in one of two opposite configurations (GTP-GDP or GDP-GTP) according to nutrient levels, thus conferring rapidity and robustness to nutrient sensing.
As is often the case for small GTPases, the intrinsic GTP hydrolysis of the Rag GTPases is slow (k cat = 2.2 × 10 −4 -3.0 × 10 −3 min −1 ) (ref. 15 ). To allow swift responses as nutrient levels change, GTPase-activating proteins (GAPs) interact with the Rag GTPases to accelerate their hydrolysis rate, in an amino-acid-regulated fashion. The GATOR1 complex, composed of DEPDC5, Nprl2 and Nprl3, is the GAP for RagA/B 16  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. e-h, 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 (f-h). 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. k-n, 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. o-p, 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. 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   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 levels 30,31 (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 manner 32 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 missing 33 . 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.

Brain-enriched RagB isoforms regulate the dynamics of mTORC1 activity through GATOR1 inhibition
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 RagB short increased S6K1, 4EBP1 and TFEB phosphorylation to comparable levels, while these were lower in RagB long -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 RagB short -and RagB long -expressing cells, indicating persistent mTORC1 activity. This suggests that RagB short and RagB long 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 RagD 34 .
The effects of RagA, RagB short and RagB long 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 RagB short 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, RagB long failed to increase the lysosomal localization of mTOR in either immunofluorescence or lyso-IP experiments (Extended Data Fig. 4a-d), suggesting that RagB long interacts poorly with mTORC1. Indeed, RagB long co-immunoprecipitated substantially less Raptor than RagA or RagB short , 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 RagB short levels together with the appearance of RagB long , 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 RagB short and not at all by RagB long (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) RagB short and RagB long are more resistant to nutrient withdrawal compared with RagA, and (2) RagB long does not bind and recruit mTOR to lysosome as efficiently as the other isoforms.
RagB short and RagB long 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 lysosome 16 . One possible explanation why mTORC1 activity 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. f-h, 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. i-k, 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. 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   -T389-S6K1 P in different nucleotide loading states and the mTORC1 subunit Raptor. GTP-locked RagA and RagB short interacted comparably with Raptor, while GTP-locked RagB long bound more weakly (Extended Data Fig. 5a,b), consistent with the reduced ability of wild-type RagB long 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 RagB short and RagB long should not be retaining mTORC1 on the lysosome. Consistent with these results, strong overexpression of GDP-locked RagA, RagB short or RagB long 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 RagB short -expressing cells, while high levels of GATOR1 were required to cause an appreciable drop in S6K1 phosphorylation in RagB long -expressing cells ( Fig. 2a,b, Extended Data Fig. 6a,b). Together, these data indicate that the RagB isoforms are comparatively resistant to GATOR1.
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, RagB short , RagB long or metap2 as a control. Indeed, cells expressing RagA or RagB short 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 RagB short . Consistent with the low binding of Raptor to RagB long , mTORC1 activity in RagB long -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 RagB long > RagB short > RagA and that this different resistance to GATOR1 is probably the main functional distinction between RagB short and RagA.
RagB short 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 counterpart 17 , 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 molecules 35 .
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 interface 17,18 , 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. RagB short interacted less than RagA with the inhibitory interface, but interacted similar to RagA with the GAP interface. RagB long interacted even less with the GATOR1 inhibitory interface as compared with RagA and RagB short , 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 RagB short . To compare how strongly RagA versus RagB short 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 interface 17 . 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 RagB short -expressing cells caused a reduction in mTORC1 activity, because this reconstitutes the GATOR1 complex (lanes 1-4 in Fig. 3a-d). In RagB short -expressing cells, DEPDC5 Y775A expression led to an even stronger inhibition of mTORC1 compared with wild-type DEPDC5, because the DEPDC5 Y775A mutant cannot bind RagB short protein and thus cannot be inhibited by it 17 (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 RagB short on GATOR1 through DEPDC5 binding. In RagA-expressing cells, however, DEPDC5 Y775A 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 RagB short binds DEPDC5.  If this is the case, then the difference between RagA and RagB short 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 RagB short -expressing cells (lane 3 versus lane 7, Fig. 3e,f), consistent with RagB short being comparatively resistant to GATOR1, while this difference was gone in cells expressing DEPDC5 Y775A (Fig. 3g,h). These results indicate that RagB short is resistant to GATOR1 activity not because it is less sensitive to the GAP activity (indeed, RagB short binds equally well to the GAP side of GATOR1 as RagA, Fig. 2i-  Ragb short or Ragb long 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 Mg 2+ 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) Ragb long . 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 Ragb long or a non-GTp-bound mutant of Ragb long (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 Ragb longT54N and Ragb shortT54N proteins purified from RagAbKO HeK293T cells as dimers with FLAG-tagged RagC Q120L (1 μg dimer per lane). The experiment was repeated once. i,j, Malachite-green GTpase assay with 1 μM of RagA•RagC S75N either alone or mixed with an equimolar amount of Ragb longT54N •RagC Q120L (i) or Ragb shortT54N •RagC Q120L (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 Ragb long 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.    because RagB short , but not RagA, is able to inhibit GATOR1 through DEPDC5 binding (Fig. 3i). RagB short 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 RagB short , we generated a mutant of RagB short lacking the N-terminal extension (ΔN mutant) or a mutant of RagB short 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•RagC S75N and RagB short •RagC S75N 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 above 35 . Using this set-up, we found that also in vitro GATOR1 had reduced activity towards RagB short compared with RagA (Extended Data Fig. 7d) and this difference was reduced when using GATOR1 containing DEPDC5 Y775A (Extended Data Fig. 7e), consistent with the in vivo results (Fig. 3a-h).
In sum, these results suggest that the N-terminal extension of RagB short enables it to inhibit GATOR1 via DEPDC5. RagB long has low affinity for GTP. We next turned our attention to RagB long . RagB long 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 RagB long has reduced affinity for GTP. Indeed, the 28-amino-acid insertion in RagB long resides within the switch I region, which forms part of the GTP-binding pocket. A GTP pull-down assay revealed that RagB long interacts much less with GTP than RagA or RagB short , almost at background levels ( Fig. 4a-c). These results are in line with a previous report in which a radiolabelled-GTP binding assay was employed 11 . 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 RagB short with both FLCN-FNIP2 and p18, while mutation of the catalytic glutamine causing RagA or RagB short to lock into GTP binding abrogated or reduced FLCN-FNIP2 and p18 interaction (Extended Data Fig. 8a-d). In contrast, the analogous mutant of RagB long that cannot hydrolyse GTP retained substantial interaction with FLCN-FNIP2 and p18 (Extended Data Fig. 8a-d), consistent with RagB long having impaired GTP binding. Likewise, wild-type RagB long bound p18 more strongly than RagA or RagB short (Extended Data Fig. 8e,f). In sum, all these data indicate that RagB long binds GTP less well than RagA or RagB short .
To understand why RagB long 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 RagB long 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 RagB long 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 RagB long 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 RagB long imposing a large constraint, removal of the N-terminal extension of RagB long did not visibly improve GATOR1 binding via DEPDC5 (Extended Data Fig. 8k,l). In sum, the 28-amino-acid insertion in RagB long 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 RagB long purified from HEK293T cells (the yield of RagB long 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).
RagB long titrates away the GAP interface of GATOR1. The low affinity of RagB long for GTP and its strong interaction with the GAP Two-way ANOVA and Tukey's post-hoc test. g,h, Ragb but not RagA knockdown blunts bicuculline-stimulated dendritogenesis of hippocampal neurons grown in 10% MeM amino-acid concentration (methods). bar height indicates average, and error bars represent standard deviation; n = 3 biological replicates. One-way ANOVA and Dunnett's post-hoc test. i-k, Violin plot of of Rraga/Rragb mRNA levels and their ratio in normal (GTeX) or cancer (TCGA) tissues. n GTeX = 7,825 biological replicates, n TCGA = 10,534 biological replicates. l, Violin plot of the Rragb/Rraga ratio in normal brain (GTeX brain) or brain cancers (TCGA brain), compared with all other samples (GTeX non-brain and TCGA non-brain). n GTeX non-brain = 6,688 biological replicates, n TCGA non-brain = 9,840 biological replicates, n GTeX brain = 1,136 biological replicates, n TCGA brain = 694 biological replicates.   (Fig. 2i-k) made us reason that a substantial pool of non-GTP-bound RagB long could act as a 'sponge' that binds GATOR1 on the GAP interface and titrates it away from RagA, RagB short or the small pool of GTP-bound RagB long in a cell. In this way, RagB long 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 RagB long harbouring the T54N mutation outcompetes GTP-locked RagA for Nprl2/3 binding (Fig. 4d,e). By acting as a GATOR1 inhibitor, RagB long should confer increased resistance to nutrient starvation when co-expressed with the other RagA/B isoforms. Indeed, co-expression of either wild-type RagB long or the RagB longT54N mutant together with RagA caused less inactivation of mTORC1 upon amino-acid starvation compared with expression of RagA alone (Fig. 4f,g). As the RagB longT54N mutant cannot bind mTORC1, this is consistent with RagB long acting via GATOR1 inhibition rather than direct mTORC1 activation.
We next tested if RagB long can inhibit GATOR1 GAP activity towards RagA also in vitro through this mechanism. We first confirmed that also in vitro RagB long binds the GAP interface of GATOR1 more strongly than RagA or RagB short using purified GDP-loaded Rag dimers and a GATOR1 complex containing DEPDC5 Y775A to disrupt the inhibitory interface (Extended Data Figs. 7b and 9e,f). Next, to test if RagB long can inhibit GATOR1 specifically via the GAP interface, we purified RagB longT54N , which interacts strongly with the GAP interface of GATOR1 but not its inhibitory interface ( Fig. 2f-k), as a dimer with GTP-locked RagC Q120L (Fig. 4h) to mimic the Rag conformation probably occurring in vivo owing to inter-subunit crosstalk 15 . Consistent with direct GATOR1 inhibition, the GAP activity of GATOR1 towards RagA was lower in the presence of equimolar amounts of RagB longT54N (Fig. 4h,i) but not of the analogous RagB shortT54N (Fig. 4j). This indicates that inhibition of GATOR1 via GAP-interface binding is a specific feature of RagB long , and that RagB short instead only inhibits GATOR1 via the inhibitory interface ( Fig. 3 and Extended Data Fig. 7).
In sum, we find that RagB long inhibits GATOR1 by acting as a 'sponge' for its GAP interface to potentiate signalling through the other Rag isoforms (Fig. 4k).
RagB short and RagB long additively inhibit GATOR1. Since RagB short and RagB long inhibit GATOR1 activity through distinct mechanisms-RagB short via DEPDC5 binding and RagB long 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 + RagB short , or RagA + RagB short + RagB long . 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 + RagB short (Fig. 5a-i). Analogously, in a time course of complete amino-acid starvation, cells expressing all three RagA/B isoforms inactivated mTORC1 more slowly than cells expressing only RagA (Extended Data Fig. 9h,i).
To test if similar effects can also be observed in a cell line that endogenously expresses all three RagA/B isoforms, we turned to the mouse neuroblastoma Neuro-2a cell line. As expected, knockout of RagA, which leaves only the more GATOR1-resistant RagB isoforms in the cells, caused mTORC1 activity to be lower and largely insensitive to amino-acid removal (Extended Data Fig. 9j,k), similar to what was observed in HEK293T cells (Fig. 1k,l). In contrast, knockout of the RagB isoforms, which leaves only RagA in the cells, reduced mTORC1 activity but at the same time caused the cells to remain amino-acid sensitive (Extended Data Fig. 9l,m), paralleling the differences seen between HEK293T cells expressing all three RagA/B isoforms versus only RagA (Fig. 5i).
In sum, these results indicate that RagB short and RagB long inhibit GATOR1 additively to render mTORC1 more resistant to nutrient depletion, according to the scheme in Fig. 5j. This may cause different tissues in the body, which express different ratios of RagA, RagB short and RagB long , to regulate mTORC1 activity differently upon amino-acid restriction.

RagB maintains high mTORC1 in neurons during starvation.
RagB short and RagB long are abundantly expressed in the mammalian brain (Fig. 1a,b). Interrogation of a transcriptomic dataset of the main cell types of the brain 39 revealed that RagB is expressed in neurons, astrocytes and oligodendroglial lineage cells, but not in microglia or endothelial cells (Fig. 6a-c). Moreover, the ratio of RagB to RagA is highest in neurons and oligodendroglial lineage cells. We confirmed that RagB short and RagB long are highly expressed in neurons also at the protein level as compared with mouse embryonic fibroblasts (MEFs) isolated in parallel from the same mouse embryos (Fig. 6d). In contrast, Rheb and its GAP TSC2, relaying growth factor signalling to mTORC1, did not differ strongly between neurons and MEFs (Extended Data Fig. 10a).
On the basis of the mechanism outlined above, expression of all three RagA/B isoforms should lead to relative resistance of mTORC1 activity to nutrient deprivation, analogous to the RagABKO cells reconstituted to express all three RagA/B isoforms (Fig. 5a-i). Indeed, we found that mTORC1 activity in neurons is rather resistant to amino-acid removal. A time course of amino-acid starvation at day in vitro (DIV) 5 showed slower dynamics of mTORC1 inactivation in neurons as compared with MEFs, with S6K phosphorylation being largely retained after 30 min of amino-acid starvation (Extended Data Fig. 10b,c). In more mature neurons (DIV 10), mTORC1 activity became largely resistant to amino-acid removal despite complete and prolonged starvation for 24 h and simultaneous stimulation of synaptic activity with bicuculline to increase their metabolic rate (Fig. 6e,f and Extended Data Fig. 10d,e).
We then tested the contribution of the RagA/B isoforms to this resistance. Knockdown for RagB, but not RagA, caused a significant drop in S6K phosphorylation in response to combined amino-acid starvation and bicuculline treatment (Fig. 6e,f and Extended Data Fig. 10f,g), suggesting that high RagB levels in neurons are at least partially responsible for their resistance to nutrient depletion and that this effect depends on neuronal activity, possibly because it increases protein synthesis and thus amino-acid usage 40 . Growth of dendritic arbours, key for neuronal function, is controlled by extracellular cues such as synaptic activity and by intracellular signalling cascades, including mTORC1, which in this context has been mostly studied downstream of growth factor signalling [41][42][43][44][45] . We therefore asked whether nutrient sensing through the RagA/B proteins could also be involved in this process by assessing dendrite growth in a standardized medium containing a low amino-acid concentration (10% of MEM) 46,47 and upon synaptic stimulation with bicuculline. Consistent with the changes in mTORC1 activity detected by western blot, knockdown of RagB, but not RagA, using two independent sets of short hairpin RNAs (shRNAs) blunted the synaptic-activity-dependent stimulation of dendritic growth (Fig. 6g,h).
RagB maintains high mTORC1 in tumour cells during starvation. Somatic inactivating mutations in all GATOR1 components have been reported in small subsets of glioblastomas and ovarian cancers 16 . Additionally, 17% of follicular lymphomas display RagC mutations that cause partial mTORC1 resistance to low nutrients 48 , indicating that disruption of the nutrient-sensing mechanism of mTORC1 is a recurring event in various cancer types. Analogously, elevated expression of the more GATOR1-resistant RagB isoforms could be an alternative route to maintaining mTORC1 activity under the low nutrient conditions experienced by cancer cells. Consistent with this hypothesis, a comparison of RagA and RagB (short and long) transcript levels in normal (GTEX) and cancer samples (TCGA) revealed that, although variable when considered separately (Fig. 6i,j), the ratio RagB/RagA was substantially increased in a subset of cancer samples (Fig. 6k). This increase did not depend on a potential overrepresentation of brain samples in the TCGA collection, since an identical trend could be appreciated also when considering only non-brain cancers (Fig. 6l). Among the samples that showed a RagB/RagA ratio greater than 0.5, we found disparate cancer primary sites and, interestingly, an enrichment in acute myeloid leukaemia samples as compared with the whole TCGA collection (Extended Data Fig. 10h,i). Interestingly, cancer samples with low RagB/RagA ratios also had lower expression of each of the three GATOR1 subunits as compared with samples with RagB/RagA ratios greater than 0.5 (Extended Data Fig. 10j-l), consistent with the hypothesis that high RagB or low GATOR1 expression could be alternative mechanisms to confer nutrient stress resistance to cancer cells. To validate these findings experimentally, we interrogated the Cancer Cell Line Encyclopedia for cell lines with high expression of the RagB isoforms and found the EFO21 ovarian cancer cell line as the top-scoring one. EFO21 cells have high levels of RagB short protein compared with RagA in comparison with other normal or cancer cell lines, together with low but detectable RagB long expression (Fig. 6m). We then tested the functional relevance of this profile of RagA/B expression by knocking down RagA or the RagB isoforms. As expected, depletion of the RagB isoforms lowered mTORC1 activity both under nutrient-rich conditions and during amino-acid starvation (Fig. 6p,q), while depletion of RagA (leaving only RagB in the cell) caused more persistent mTORC1 activity (Fig. 6n,o). Consistent with these results, knockdown of RagB in EFO21 cells caused lower protein synthesis rate, detected via incorporation of fluorescently labelled O-propargyl-puromycin (OPP), especially when exposed to lower amino-acid concentrations (Fig. 6r).
In sum, these results indicate that high expression of the RagB isoforms, as observed physiologically in neurons and pathologically in some cancers, confers partial resistance to low nutrients and thus contributes to determining cell-specific dynamics of mTORC1 responses.

Discussion
On the basis of their high sequence similarity, RagA and RagB have been assumed to be functionally equivalent. The findings presented here, together with those in an accompanying study 34 , show that this is not the case. RagA and the two RagB isoforms have distinct profiles of interaction with Raptor and GATOR1, which correspond to different levels of mTORC1 activity and result in different responses to nutrient deprivation. First, RagB long interacts only weakly with Raptor as compared with RagA and RagB short and thus cannot sustain strong lysosomal accumulation and mTORC1 activity in nutrient-rich conditions, when expressed alone. Second, RagB short and RagB long , but not RagA, are able to counteract the activation of GATOR1 upon nutrient removal by inhibiting this complex through two distinct mechanisms, which cause persistent mTORC1 activity. When all isoforms are co-expressed, the two inhibitory mechanisms provided by the RagB isoforms potentiate mTORC1 activity and slow down the dynamics of its inactivation when nutrients are scarce. High expression of the two RagB isoforms, together with RagA, is observed physiologically in neurons, where it helps maintain mTORC1 activity during amino-acid starvation. We finally show that cancer cells can use this mechanism to their advantage by upregulating the RagB isoforms.
The inhibition of GATOR1 by RagB short depends on the interaction between this Rag isoform and the GATOR1 subunit DEPDC5. Previously, an inhibitory interaction between RagA/B and GATOR1 via DEPDC5 was described in cells endogenously expressing both RagA and RagB short and named 'inhibitory mode' to distinguish it from the interaction via Nprl2/3 that executes the GAP activity 17 .
Our results indicate that the inhibitory mode is probably a specific feature of RagB short binding to DEPDC5 and that, while RagA also interacts with DEPDC5, this interaction is compatible with normal GATOR1 activity. Thus, the inhibitory effect detected in cells expressing both RagA and RagB short might actually reflect just the contribution of RagB short -mediated inhibition of GATOR1. This finding is rather unexpected since RagA and RagB short are likely to have very similar interfaces for DEPDC5 binding. Although no structure of RagB short in complex with GATOR1 nor of RagB short alone is yet available, the N-terminal extension of RagB short is expected to be in the vicinity of the DEPDC5-switch I interface on the basis of a published structure of GATOR1 in complex with RagA 17 . Hence it is possible that this N-terminal extension of RagB short acts on DEPDC5. Consistent with this, we observed that removal of the N-terminal extension restores the same affinity for GATOR1 as RagA and results in complete mTORC1 inactivation upon nutrient starvation as in RagA-expressing cells (Extended Data Fig. 3g-j). The N-terminal extension of RagB short might then force an orientation of the GATOR1 complex where the Nprl2/3 subunits are less favourably placed for the execution of the GAP activity on the nearby RagA/B GTPases (Fig. 3i).
RagB long had been previously described as a brain-specific splicing isoform of RagB of unknown function 11 . Here we show that RagB long has very peculiar features that distinguish it from the other two RagA/B isoforms and from the other Rag GTPases altogether: (1) it has low affinity for GTP; (2) it interacts poorly with Raptor; (3) it interacts strongly with the Nprl2/3 subunits of GATOR1 when not bound to GTP. On this basis, we propose that its primary function is to act as a 'sponge' that obstructs the GAP-competent interface of GATOR1 from the other RagA/B isoforms, thereby enhancing mTORC1 activity. The low affinity of RagB long for GTP is therefore a key aspect of its function, as it enables RagB long to interact strongly with Nprl2/3 while the other RagA/B isoforms are GTP bound, and it allows RagB long to be insensitive to GATOR1 without hyperactivating mTORC1. RagB long has close parallels to the so-called pseudo-GTPases, a group of atypical GTPases defined by their inability to bind GTP or, alternatively, to hydrolyse it 49 . While the low GTP binding of pseudo-GTPases usually depends on mutations in the G1 motif interacting with the phosphates or in the G4 and G5 motifs interacting with the guanosine ring, in the case of RagB long this is probably achieved through insertion of the sequence encoded by exon 4 of Rragb in its switch I region, which might alter the GTP binding pocket, together with providing a high-affinity interface for Nprl2/3 binding. Future studies will be needed to better understand which consequences this sequence imposes on the structure of RagB long and to correlate such structural features with the functional peculiarities of this RagA/B isoform.
Both RagB isoforms are highly expressed in neurons. We find here that mTORC1 activity in neurons is exceptionally resistant to nutrient depletion and that this effect depends at least in part on their high RagB levels. What is the physiological function of the resistance of neuronal mTORC1 to nutrient starvation? At the organismal level, the various organs are metabolically wired to privilege those fulfilling vital functions such as the brain. Thus, starvation induces autophagy in most organs within 24 h, including liver, pancreas, kidney, skeletal muscle and heart, but not in the brain 50 . Autophagy-derived substrates, such as nucleosides, amino acids and lipids are then metabolized in the liver to produce glucose and ketone bodies necessary to feed the brain 51 . As mTORC1 is a potent inhibitor of autophagy, resistance of neuronal mTORC1 to nutrient starvation could then support the preferential upregulation of autophagy in the peripheral tissues and the subsequent re-routing of nutrients to the brain. Additionally, mTORC1 is involved in a variety of crucial neuron-specific functions, including dendritic growth [43][44][45] , synaptic plasticity 4 , axon guidance and both developmental and adult neurogenesis 4,52,53 . High expression of the RagB isoforms in neurons could therefore lend more robustness to these critical processes, by ensuring that oscillations in nutrient levels do not immediately result in corresponding changes in mTORC1 activity. Consistent with this interpretation, we find that the stimulation of dendritic growth by synaptic activity is blunted upon knockdown of the RagB isoforms when neurons are exposed to low amino-acid concentrations.
Germline monoallelic loss-of-function mutations in all three GATOR1 subunits have been recognized as the most frequent genetic cause of focal epilepsies, accounting for almost 10% of the cases 54 . Moreover, bi-allelic loss-of-function mutations in some subunits of the KICSTOR complex, anchoring GATOR1 to the lysosomal surface 55,56 , have been associated with a broad range of neurodevelopmental disorders, including mental retardation and developmental and epileptic encephalopathy [57][58][59][60][61][62][63][64] . The propensity of such whole-body mutations to specifically give rise to neurological manifestations seems at odds with the ubiquitous importance of the mTORC1 signalling axis. As a comparison, mutations in the TSC complex lead to skin lesions such as facial angiofibromas, kidney angiomyolipomas and pulmonary lymphangiomyomatosis, alongside neurologic disorders 65 . The inhibition of GATOR1 by the RagB isoforms that we describe here helps explain this discrepancy, as a decrease in GATOR1 function is expected to have the most dramatic consequences in cells expressing higher amounts of these two isoforms. Genetic or pharmacological inhibition of the RagB isoforms might then be a particularly appealing therapeutic intervention, as it could partially restore GATOR1 function selectively in the brain.
Components of the nutrient-sensing machinery are often mutated in cancer. A priori, it is unclear whether one would expect it to be beneficial or deleterious for cancer cells to properly sense their nutrient status. For instance, one could imagine that the ability to sense the lack of nutrients and thereby downregulate anabolic processes might be important for cancer cells to avoid metabolic catastrophe and ensure viability. By contrast, one might imagine that having constitutively high activation of mTORC1 might give a growth and proliferation advantage to cancer cells. This logic may depend on the type of tumour in consideration, and on the location of cells within the tumour: leukaemia cells are likely to have constitutively high nutrient availability and may therefore benefit more from constitutively active nutrient sensing, whereas cells at the centre of a solid tumour that is poorly perfused might survive better with an intact nutrient stress response. Hence, it is interesting to look at what has been found from cancer sequencing. DEPDC5, a component of the GATOR1 complex, is frequently mutated in gastrointestinal stromal tumours, leading to constitutively high mTORC1 activity 66 . All three components of the GATOR1 complex are mutated in a variety of carcinomas, including hepatocellular carcinoma, bladder carcinoma and breast carcinoma 16 . Likewise, recurrent gain-of-function mutations in RagC that increase mTORC1 binding and activation were found in follicular lymphoma 48 . Together, these data indicate that constitutive activation of mTORC1 is advantageous for some cancer cells despite the cost of impaired nutrient sensing. The data we present here fit this general trend. We find that many tumour cells shift their relative balance of RagA/B expression towards RagB and away from RagA. Our data indicate that this leads to stronger and more constitutive activation of mTORC1, thereby providing cancer cells with a growth advantage.
Altogether, we provide here evidence that the functions of RagA and RagB have substantially diverged during evolution, with RagB assuming the role of a GATOR1 decoy substrate and inhibitor that enables elevated mTORC1 activity in a tissue-specific manner.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41556-022-00977-x. For amino-acid starvation of cell lines, homemade DMEM containing 25 mM glucose was prepared according to the formulation of the commercial medium but omitting all amino acids, and completed by addition of 10% dialysed FBS (dFBS) (Life Technologies #A33820-01) (−aa medium). For amino-acid stimulation, the medium was supplemented with 1× MEM non-essential amino acids (Life Technologies #11140035), 1× MEM amino acids (Sigma-Aldrich #M5550) and 2 mM l-glutamine (Life Technologies #25030024) (+aa medium). Before any amino-acid treatment, cells were first incubated in +aa medium for 1 h.

Cell lysis, immunoblots and immunoprecipitations.
For immunoblot experiments, subconfluent cells in six-well plates were washed once with ice-cold DPBS and lysed in 100 μl lysis buffer (40 mM HEPES pH 7.4, 1% Triton X-100, 10 mM β-glycerophosphate, 10 mM sodium pyrophosphate, 30 mM NaF, 2× Roche Complete protease inhibitor cocktail with EDTA, 1× Roche PhosSTOP phosphatase inhibitors). Lysates were cleared by max speed centrifugation at 4 °C in a tabletop centrifuge and processed immediately for immunoblots or stored at −80 °C until needed. The same amount of cell lysate was loaded on multiple membranes, one for each protein of interest, including loading controls.
Immunoblots were quantified using Image Lab (Biorad). The band intensity of each sample was normalized to the sum of all samples in one experiment, to allow comparisons between different experiments. In immunoprecipitation experiments, quantifications represent the ratio between co-immunoprecipitating and immunoprecipitating protein in the IP after normalization for the co-immunoprecipitating protein in the whole-cell lysate (normalized to α-tubulin).
Lyso-IP. Lyso-IP experiments were performed following previously described methods 69 . Briefly, 12 × 10 6 control or knockout HEK293T cells were seeded in 15-cm dishes and transiently transfected 24 h later with 6 μg of plasmid coding for TMEM192-3xHA. Two days after transfection, the medium was exchanged with fresh medium 1 h before cell collection. Cells were scraped in 1 ml ice-cold PBS containing protease and phosphatase inhibitors, centrifuged at 1,000g for 2 min at 4 °C in a tabletop centrifuge, resuspended in 950 μl PBS containing protease and phosphatase inhibitors, and gently homogenized with 20 strokes in a 5-ml douncer. Then, 25 μl cell suspension was transferred to a new tube before homogenization and lysed with 70 μl 1% Triton lysis buffer to generate whole-cell lysate samples. Intact cells were pelleted at 1,000g for 2 min, and the supernatant was incubated with 150 μl pre-washed anti-HA magnetic beads (ThermoFisher #88837) for 3 min with gentle rotation. Immunoprecipitated lysosomes were washed three times with PBS containing protease and phosphatase inhibitors and subsequently lysed in 100 μl 1% Triton lysis buffer.

Immunofluorescence.
A total of 1.2 × 10 5 HEK293T cells were seeded on fibronectin-coated 12-mm glass coverslips. One day later, cells were washed once in DPBS, fixed for 15 min in 4% formaldehyde in PBS (ThermoFisher #28908), permeabilized for 10 min with 0.2% Triton X-100 in PBS, blocked 1 h with 0.25% BSA in PBS, incubated overnight with primary antibodies, incubated 1 h with fluorophore-conjugated secondary antibodies, counterstained with 5 µg ml −1 DAPI (Applichem #A1001) and then mounted on glass slides using Vectashield (Vector Labs #H-1000). Coverslips were imaged with a confocal microscope (Leica TCS SP8) using a 63× objective. A total of 16 random fields of view from two coverslips per condition were acquired. mTOR co-localization analysis was performed using CellProfiler. The mTOR staining was segmented using an Otsu-based adaptive thresholding algorithm to separate the bright spots of lysosomal mTOR from the cytosolic one. Similarly, the LAMP2 staining was segmented to identify LAMP2-positive objects. Finally, the percentage of the area of the LAMP2-positive objects covered by the lysosomal mTOR was calculated.
RNA extraction and reverse-transcription qPCR. Total RNA was extracted from cultured cells with TRIzol (Life Technologies #15596018) as per the manufacturer's instructions. Two micrograms of RNA per sample was reverse transcribed with Maxima H Minus reverse transcriptase (ThermoFisher #EP0753) and an oligo-dT primer. Expression of RagA, RagB, RagC and RagD was determined by qPCR with isoform-specific primers (Supplementary Table 1) using a SYBRGreen-based master mix (primaQUANT CYBR 2x qPCR SYBRGreen Master Mix with LOW ROX, Streinbrenner, #SL-9913). The absolute number of copies in each sample in Extended Data Fig. 1g was extrapolated from a standard curve created using a serial dilution of defined amounts of plasmids containing the coding sequences of each Rag GTPase. The relative abundance of RagA and RagB in Extended Data Fig. 10f,g was determined with the 2 −ΔΔCt method using Rpl13a as loading control.

Purification of RagA/B GTPases and GATOR1 from HEK293T cells.
pcDNA3.1-based plasmids coding for untagged RagA, RagB shortT54N or RagB long (wild type or T54N) and FLAG-tagged RagC (S75N or Q120L) were co-transfected at a 1:1 ratio in RagABKO HEK293T cells previously seeded in 20 15-cm dishes using PEI. Forty-eight hours after transfection, cells were scraped in 60 ml ice-cold lysis buffer (40 mM HEPES pH 7.4, 100 mM NaCl, 5 mM MgCl 2 , 1% Triton X-100, 100 μM ATP and 1× Roche Complete protease inhibitor cocktail without EDTA) and homogenized with 20 strokes of a Dounce homogenizer on ice. After clearing through centrifugation for 1 h at 23,000g at 4 °C, the lysate was incubated with 400 μl (packed volume) of anti-FLAG M2 affinity gel beads shaking for 3 h, followed by three washes with lysis buffer supplemented with 300 mM NaCl. After exchanging buffer to buffer E (50 mM HEPES pH 7.4, 100 mM KOAc and 2 mM DTT), the bound proteins were eluted overnight at 4 °C by adding 250 μg ml −1 FLAG peptide (Sigma-Aldrich #F3290). The eluate was then concentrated in 30 kDa MWCO PES columns, supplemented with glycerol (20% final concentration), aliquoted, snap-frozen and stored at −80 °C. The GATOR1 complex was purified following a similar procedure. pRK5-based plasmids coding for FLAG-DEPDC5 (wild type or Y775A), HA-Nprl2 and HA-Nprl3 were co-transfected at a 1:2:2 ratio, respectively, in control HEK293T cells. After immunoprecipitation as above, the anti-FLAG beads were buffer-exchanged to CHAPS buffer (50 mM HEPES pH 7.4, 150 mM NaCl, 2 mM MgCl 2 and 0.1% CHAPS) and the GATOR1 complex was eluted overnight at 4 °C by adding 250 μg ml −1 FLAG peptide. The eluate was then concentrated in 100 kDa MWCO PES columns, supplemented with glycerol (10% final concentration) and 2 mM DTT, aliquoted, snap-frozen and stored at −80 °C.
Malachite-green GTPase assay. The Rag GTPase activity was assayed using a colourimetric malachite-green based reaction that detects the phosphate released upon GTP hydrolysis, according to previously published protocols 72 . Briefly, aliquots of the Rag dimers and GATOR1 were thawed on ice, diluted in dilution buffer (30 mM Tris-HCl pH 7.4, 0.1% CHAPS and 100 mM NaCl) to 5× the final concentrations in the assay, and then mixed together in 1× GTPase buffer (10 mM, 2 mM MgCl 2 , 0.05% CHAPS, 10 mM NaCl, 100 μM DTT and 300 μM GTP) at the final concentrations of 1 μM for the Rags and 50 nM to 2 μM for GATOR1 in a 20 μl volume. For the experiments in Extended Data Fig. 7d,e, the Rag heterodimers were first immobilized on 50 μg magnetic beads (Dynabeads His-Tag, Life Technologies #10103D). The reaction was started by incubating the samples at 30 °C. After 3 h, the reaction was terminated with 5 μl 0.5 M EDTA. The colourimetric signal was generated by adding 75 μl of the malachite-green solution (1 mg ml −1 malachite-green oxalate and 10 mg ml −1 ammonium molybdate tetrahydrate, in 1 M HCl), incubated 5 min at room temperature, and then immediately read with a microplate reader (SPECTROstar Omega, BMG LABTECH) as 660 nm absorbance. Samples without Rags nor GATOR1, with Rags but no GATOR1, and with GATOR1 but no Rags were analysed in parallel to allow for background subtraction and to ensure that the GTPase activity detected was not due to contaminating GTPases. The amount of phosphate per sample was then calculated using a standard curve of NaH 2 PO 4 •H2O in the range 0-50 μM. The amount of phosphate generated by GATOR1-stimulated GTP hydrolysis was then calculated by subtracting the background signal of the reaction with only the Rag heterodimers.
In vitro GATOR1-Rag interaction. Six micrograms of purified GATOR1 was bound on anti-HA magnetic beads (ThermoFisher #88837) for 2 h in CHAPS buffer (40 mM HEPES pH 7.4, 150 mM NaCl, 2 mM MgCl 2 and 0.1% CHAPS) followed by three washes with CHAPS buffer supplemented with 500 mM NaCl. Three micrograms of each Rag heterodimer was then added and incubated overnight with the GATOR1-containing beads in CHAPS buffer supplemented with 200 μM GDP. The same amount of Rag dimers was incubated in parallel with empty beads as background control. On the following day, the beads were washed three times with CHAPS buffer supplemented with 500 mM NaCl and the GATOR1-Rag complexes were eluted in 2× Laemmli buffer.
Extraction of mouse tissues and preparation of lysates, isolation of fibroblasts (MEFs) and primary cortical neurons from mouse embryos and amino-acid treatments. Mice were housed in pathogen-free conditions, 12-h day/night cycle, at 21 °C temperature and 50-60% relative humidity. Food and water were available ad libitum. For mouse tissue analysis, C57B6/J control female postnatal-day 60-90 adult mice were killed with isofluorane, and the tissues of interest were dissected, snap-frozen in liquid nitrogen and stored at −80 °C until needed. To prepare tissue lysates, the frozen tissues were ground on dry ice in Eppendorf tubes using plastic pestles and lysed by adding 500 μl lysis buffer (40 mM HEPES pH 7.4, 1% Triton X-100, 10 mM β-glycerophosphate, 10 mM sodium pyrophosphate, 30 mM NaF, 2× Roche Complete protease inhibitor cocktail with EDTA and 1× Roche PhosSTOP phosphatase inhibitors) and then processed as described above for cell lysates. MEFs were prepared from six to eight embryonic-day 15 mouse embryos from time-mated C57B6/J mice. After removal of the heads, Extended Data Fig. 2 | Response of mTORC1 targets to amino-acid removal. (a, b) phosphorylation of the mTORC1 targets S6K1, TFeb, 4ebp1, ULK1, and the S6K targets S6 in control cells upon amino-acid starvation (amino-acid free DMeM + 10% dFbS) for the indicated time points. (a) Representative example. (b) Quantification of 3 independent experiments, with unstarved cells set to 1. Circle=average, error bars=standard deviation, n = 3 biological replicates. Source numerical data and unprocessed blots are available in source data. (d,f) Quantification of 3 independent experiments, with unstarved HA-rap2a-transfected control cells set to 1. bar Height=average, error bars=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 1x amino acids. exact p values are shown in the graphs. Source numerical data and unprocessed blots are available in source data.