Cholesterol regulates insulin-induced mTORC1 signaling

The potential of a cell to integrate diverse intra- and extra-cellular cues is fundamental to growth and homeostasis. Aberrations in signal integration are hallmarks of various conditions like cancer, metabolic syndrome and aging (Laplante and Sabatini, 2012; Liu and Sabatini, 2020). In mammalian cells, mechanistic target of rapamycin complex 1 (mTORC1), a serine/threonine protein kinase complex, plays a key role in signal integration (Vézina et al., 1975; Brown et al., 1994). In response to nutrients, hormones and growth factors, mTORC1 gets activated resulting in phosphorylation of its downstream substrates to promote anabolic processes like protein translation, nucleotide synthesis and lipid synthesis, etc., as well as to suppress catabolic processes like macroautophagy (Keith and Schreiber, 1995; Dibble and Manning, 2013; Jewell and Guan, 2013). Under adverse conditions like starvation, mTORC1 remains inactive leading to the shutting down of cellular biosynthesis and to initiation of autophagy. The replenishment of nutrients via autophagy leads to the reactivation of mTORC1 and to the restoration of anabolism and growth (Yu et al., 2010; Russell et al., 2013; Yu and Long, 2015). Thus, by undergoing cycles of activation, inactivation and reactivation, mTORC1 coordinates cell growth and metabolism in a fluctuating environment (Liu and Sabatini, 2020).

Recent studies have proven beyond doubt that lysosomes play an important role in signal integration and activation of mTORC1 (Kim et al., 2008; Sancak et al., 2008). Two small GTPases of the Ras superfamily, Rags and Rheb, which reside on the limiting membrane of lysosomes act in mechanistically distinct ways to regulate mTORC1 (Menon et al., 2014; Perera and Zoncu, 2016). Rags are heterodimeric GTPases (comprising RagA–RagB or RagC–RagD; hereafter collectively denoted Rag) that are tethered to lysosomal membranes by virtue of their lysosomal-resident GEF, the Ragulator complex (Sancak et al., 2010). Rag GTPases undergo changes in the nucleotide loading states based on nutrient availability, leading to their interaction with the raptor component of mTORC1. Ultimately, this results in the recruitment of mTORC1 on lysosomal membrane to the near vicinity of Rheb (Bar-Peled et al., 2013). Rheb GTPase, on the other hand, allosterically stimulates mTORC1 by inducing a conformational change in the kinase complex via binding to a site away from its active site (Shams et al., 2021). Thus, the coordinated action of Rag and Rheb orchestrates complete activation of mTORC1 on lysosomes. It has been shown that Rag or Rheb mutants that are constitutively bound to GTP can cause the translocation of mTORC1 onto lysosomes and activate it irrespective of the metabolic status of the cell (Sancak et al., 2008; Bar-Peled et al., 2012). Thus, for signal integration to occur, a multitude of upstream cues must feed into mTORC1 via Rag or Rheb axis (Sabatini and Parsons, 2001; Mossmann et al., 2018).

Among many growth factors, insulin is a critical activator of mTORC1 (Manning et al., 2002; Potter et al., 2002). Within minutes of insulin stimulation, phosphorylation of mTORC1 substrates can be observed, suggesting rapid activation of mTORC1 by insulin (Inoki et al., 2002; Menon et al., 2014). The mechanism of insulin-stimulated mTORC1 signaling has also been elucidated. Insulin imposes its effects on mTORC1 via the Rheb axis through the phosphoinositide 3-kinase (PI3K)–Akt pathway (Manning et al., 2002; Garami et al., 2003; Menon et al., 2014). The Tuberous Sclerosis complex (TSC; comprising TSC1, TSC2 and TBC1D7) acts as a GAP for Rheb and is associated with lysosomes in the absence of insulin in a Rheb-dependent manner to inhibit Rheb function (Tee et al., 2003; Dibble et al., 2012). Through an Akt family protein kinase, insulin induces rapid phosphorylation of the TSC, which triggers dissociation of the TSC from lysosomes to release Rheb for mTORC1 activation (Menon et al., 2014). Elegant studies in mammalian cell lines have shown that input from amino acids is important to facilitate insulin-induced mTORC1 signaling (Menon et al., 2014). In the absence of amino acids, insulin fails to activate mTORC1 (Menon et al., 2014). In general, amino acids have been shown to act through the Rag axis for the recruitment of mTORC1 onto lysosomes thereby enabling Rheb-mediated activation of mTORC1 (Kim et al., 2008; Sancak et al., 2008). Nevertheless, some amino acids like arginine have been implicated in activating mTORC1 through the TSC–Rheb axis by relieving the allosteric inhibition of Rheb by TSC (Carroll et al., 2016). Although the role of amino acids is well established in insulin–mTORC1 axis operation, other factors regulating insulin-induced mTORC1 signaling remain poorly understood.

Cholesterol has long been known as the essential building block of cell membranes and as an important substrate for the biosynthesis of steroids and vitamin D. Recent invaluable studies from Zoncu lab have shown that, similar to nutrients, lysosomal cholesterol can activate mTORC1 in a Ragulator–Rag GTPase-dependent manner (Castellano et al., 2017; Lim et al., 2019; Davis et al., 2021). Several mechanisms have been proposed to relay lysosomal cholesterol sufficiency to enable mTORC1 activation. SLC38A9, a lysosomal-resident amino acid transporter that normally senses and relays arginine content was found to be required for cholesterol sensing (Castellano et al., 2017). This function was distinct from the amino acid-sensing function of SLC38A9 and was dependent on the cholesterol-responsive motifs found on the transporter (Castellano et al., 2017). More recently, a GPCR-like lysosomal cholesterol-sensing protein LYCHOS (also known as GPR155), which also contains cholesterol-responsive motifs was identified as functioning as a facilitator of cholesterol-assisted lysosomal recruitment of mTORC1 by sequestering GATOR1, a GAP for Rag GTPase (Shin et al., 2022). Studies on individuals with Niemann-Pick type C (NPC) disease further substantiated the link between cholesterol and mTORC1 (Infante et al., 2008). It was demonstrated that Niemann-Pick type C1 (NPC1), the lysosomal cholesterol efflux transporter, binds to SLC38A9 and negatively regulates mTORC1 function (Castellano et al., 2017). Accordingly, individuals with pathogenic variants in NPC1 that render the transporter dysfunctional have an accumulation of cholesterol in lysosomes (Vanier et al., 1996; Carstea et al., 1997), and exhibit hyperactive mTORC1 and aberrant growth signaling (Lim et al., 2019; Davis et al., 2021). In addition, continuous delivery of cholesterol across ER–lysosome contacts via oxysterol-binding protein (OSBP) and its anchors VAPA and VAPB at endoplasmic reticulum (ER) also aggravated the mTORC1 phenotype driving mitochondrial dysfunction and neurodegeneration in individuals with NPC (Lim et al., 2019; Davis et al., 2021). However, the requirement of cholesterol for the functioning of insulin-induced mTORC1 signaling in various contexts remains unclear.

In this study, we investigate the role of cholesterol in the progression of the insulin–mTORC1 signaling axis. Specifically, we test how cholesterol and insulin integrate spatiotemporally to enable mTORC1 activation and autophagy-dependent mTORC1 reactivation as well as how this signal integration goes awry in Smith–Lemli–Opitz syndrome (SLOS), a cholesterol biosynthesis disorder (Smith et al., 1965). We demonstrate that the rapid progression of insulin-induced mTORC1 signaling is hampered in the absence of cholesterol and is restored in the presence of cholesterol. We also report that supplementation of cholesterol is important to restore the activity and insulin sensitivity of mTORC1 during prolonged amino acid starvation. Genetic inhibition of autophagy abolished the restorative effect of cholesterol, suggesting that reactivation and insulin responsiveness of mTORC1 is autophagy dependent. In all these contexts, cholesterol supports the insulin–mTORC1 axis by promoting the recruitment of mTORC1 to the lysosomal surface without perturbing the insulin-mediated spatial arrangement of TSC. Using fibroblasts from individuals with SLOS and a pharmacologically generated cell model, we demonstrate that the insulin–mTORC1 signaling axis is perturbed in SLOS. Finally, we show that the insulin–mTORC1 axis defect in SLOS can be corrected by supplying exogenous cholesterol or by expressing a constitutively active form of Rag GTPase in the presence of insulin, demonstrating the fundamental importance of the signal integration mechanism described herein.

HeLa cells, which are sensitive to both insulin and have a cholesterol response, were used for the experiments (Sancak et al., 2008; Menon et al., 2014; Castellano et al., 2017). We first confirmed the independent effects of cholesterol and insulin on mTORC1 function in HeLa cells. The mTORC1 signaling function was assessed by determining the steady-state levels of the ribosomal p70S6 kinase (also known as RPS6KB1) phosphorylated at T389 [denoted pS6K (T389)] and eukaryotic translation initiation factor 4E-binding protein 1 (also known as EIF4EBP1) phosphorylated at S65 [denoted p4EBP1 (S65)], two direct substrates of mTORC1. Acute sterol depletion using methyl β-cyclodextrin (MCD) showed a dose- and time-dependent decline in pS6K (T389) levels (Fig. S1A–B′) suggesting mTORC1 dysfunction (Castellano et al., 2017). The effect on mTORC1 was specific to cholesterol deprivation, as the addition of cholesterol restored pS6K (T389) levels in a dose- and time-dependent manner (Fig. S1C–D′). From these experiments, 1% MCD treatment for 2 h was found to be sufficient to inactivate mTORC1 (Fig. S1A–B′), and the addition of 20 µg/ml cholesterol for 2 h to significantly restore mTORC1 function (Fig. S1C–D′). Similarly, serum starvation led to a loss of pS6K (T389) levels, which recovered upon the addition of 1 µM insulin for 30 min (Fig. 1A,A′). These results validated earlier findings showing independent effects of insulin and cholesterol on mTORC1 (Menon et al., 2014; Castellano et al., 2017).

To understand the role of cholesterol in insulin-induced mTORC1 signaling, serum-starved HeLa cells were treated with MCD. As expected, this led to undetectable levels of pS6K (T389) indicating complete inactivation of mTORC1 (Fig. 1A,A′). Interestingly, the pS6K (T389) levels remained undetectable when either insulin or cholesterol was added back (Fig. 1A,A′). However, the addition of both insulin and cholesterol resulted in a marked increase in pS6K (T389) levels suggesting that integration of cholesterol and insulin signals is required for the proper functioning of mTORC1 (Fig. 1A,A′). Similar results were obtained with human skin fibroblasts (Fig. 1B,B′). Next, to check whether cholesterol is required for the rapid progression of mTORC1 signaling upon insulin stimulation, we followed the temporal dynamics of mTORC1 signaling. In insulin-stimulated cells, mTORC1 signaling initiated early, with pS6K (T389) levels visible at ∼10 min of insulin addition, reaching a maximum by 30 min and sustained thereafter for at least an hour (Fig. 1C,C′, left panel). However, in the absence of sterols, mTORC1 signaling initiated much later, showing detectable levels of pS6K (T389) only after 60 min of insulin addition (Fig. 1C,C′, middle panel). Importantly, this delay in mTORC1 activation was rectified when cholesterol was replenished (Fig. 1C,C′, right panel). Together, this indicates a key role for cholesterol in the rapid progression of insulin-induced mTORC1 signaling.

Given that the recruitment of mTORC1 to the lysosomal surface is a key step in mTORC1 activation (Bar-Peled et al., 2012), we next investigated whether cholesterol affects this step during insulin stimulation. In serum-starved, sterol-depleted cells, the mTOR staining was found predominantly in the cytosol with negligible colocalization with the lysosomal marker LAMP-2 (Fig. 2A,B). The addition of 1 µM insulin for 30 min did not affect mTOR localization in these cells, whereas the addition of cholesterol triggered the recruitment of mTOR onto lysosomes (Fig. 2A,B).

To confirm whether cholesterol-mediated lysosomal recruitment of the kinase complex leads to mTORC1 signaling during insulin stimulation, we replenished cholesterol in a time-dependent manner in insulin-stimulated cells and followed both recruitment and activation of mTORC1. In the absence of cholesterol, even after 30 min of insulin addition, the mTOR–LAMP-2 colocalization (Fig. 3A,B) and increase in pS6K (T389) levels were not apparent (Fig. 3C,C′). However, the efficiency of kinase recruitment increased over time with significant mTOR–LAMP-2 colocalization observed after 1 h of cholesterol addition (Fig. 3A,B). This was consistent with the time taken for cholesterol to reach lysosomes (Hölttä-Vuori et al., 2008; Castellano et al., 2017). The kinase recruitment coincided with an increase in pS6K (T389) levels in insulin-stimulated cells (Fig. 3C,C′) implying that cholesterol-assisted mTORC1 recruitment supports insulin-induced mTORC1 signaling.

Phosphorylation leading to the dissociation of TSC from lysosomes to cytosol upon insulin stimulation is a crucial step in insulin-induced mTORC1 signaling. Hence, we checked whether cholesterol has a role in insulin-mediated phosphorylation and spatial arrangement of TSC. TSC phosphorylation was comparable in sterol-depleted HeLa cells with or without cholesterol restimulation, as revealed by immunoblotting (Fig. 4A). However, TSC phosphorylation was negligible when cells were serum-starved, and insulin restimulation for 30 min induced an increase in TSC phosphorylation irrespective of cholesterol status (Fig. 4A). Together, these results imply that cholesterol does not interfere with insulin-mediated TSC phosphorylation. A pool of TSC was always found to be associated with lysosomes in serum-starved cells showing significant colocalization with LAMP-2 (Fig. S2A–A″,C). Insulin stimulation for 30 min caused the relocation of TSC into the cytosol as revealed by a significant reduction in TSC–LAMP-2 colocalization (Fig. S2B–B″,C) confirming the role of insulin in the spatial arrangement of TSC (Menon et al., 2014). In these conditions, sterol depletion or cholesterol restimulation did not induce any significant change in the localization of TSC implying that cholesterol might not be involved in the spatial arrangement of TSC (Fig. 4B,C).

It has been shown that prolonged amino acid starvation leads to an initial decline in mTORC1 signaling and restoration in the later phase, a process termed mTORC1 reactivation. This process is crucial for sustaining cellular metabolic homeostasis during adverse conditions. To assess the role of cholesterol in mTORC1 reactivation, we first investigated the temporal dynamics of mTORC1 signaling in response to amino acid starvation with or without supplementation of cholesterol. As glutamine has been shown to play a key role in mTORC1 reactivation, cells were starved of all amino acids except glutamine (Tan et al., 2017). In sterol-depleted cells, amino acid starvation led to a progressive loss in the phosphorylation levels of S6K (Fig. 5A,A′). The marked loss in mTORC1 signaling persisted throughout the incubation period (Fig. 5A,A′). In the cholesterol-supplemented cells, amino acid starvation also led to an initial loss of mTORC1 signaling. However, the cholesterol induced a restoration of mTORC1 signaling in the later phase of amino acid starvation, as shown by the recovery of pS6K (T389) levels (Fig. 5B,B′). To verify that the restorative effect of cholesterol on pS6K (T389) levels was indeed mediated by mTORC1, we treated cells with Torin 1, a catalytic inhibitor of mTOR. Whereas cholesterol restored pS6K (T389) levels during amino acid starvation, the effect was abolished by inhibition of mTORC1 (Fig. 5C–D′).

We next investigated whether cholesterol-assisted mTORC1 reactivation involves lysosomal localization of the kinase complex. We observed significant colocalization of mTOR with LAMP-2 after 8 h of amino acid starvation (Fig. 5E,E′) suggesting that mTORC1 is recruited on lysosomes. In the absence of cholesterol, colocalization of mTOR with LAMP-2 was minimal even after 8 h of amino acid starvation, as shown by a diffused cytoplasmic pattern of mTOR, which marginally overlapped with LAMP-2 (Fig. 5E,E′). This correlated well with diminished mTORC1 signaling after 8 h of amino acid starvation (Fig. 5A,A′). However, cholesterol supplementation restored colocalization of mTOR and LAMP-2 at 8 h post-amino acid starvation (Fig. 5E,E′), which coincided with reactivation of mTORC1 signaling (Fig. 5B,B′). Collectively, the data show that cholesterol is essential to restore mTORC1 lysosomal localization and signaling during prolonged amino acid starvation.

Notably, the effect of cholesterol on mTORC1 reactivation occurred at a later phase of amino acid starvation (between 4–6 h) (Fig. 6A,A′), a period that correlates well with the effective induction of autophagy. These results compelled us to check whether cholesterol-assisted mTORC1 reactivation is autophagy dependent in HeLa cells. Consistent with this notion, genetic ablation of Atg5 in HeLa cells (Atg5 KO) abolished the restorative effect of cholesterol on mTORC1 signaling, as demonstrated by a persistent loss of pS6K (T389) (Fig. 6B,B′).

We next assessed whether the crosstalk between cholesterol and autophagy is necessary to sustain insulin-induced mTORC1 signaling during prolonged amino acid starvation. When Atg5 wild-type (WT) cells were stimulated with insulin after 6 h of amino acid starvation, there was an increase in pS6K (T389) levels suggesting that the reactivated mTORC1 is responsive to insulin (Fig. 6C,C′). While, acute sterol depletion abolished this effect on mTORC1, supplementation of cholesterol restored insulin-sensitive mTORC1 signaling (Fig. 6C,C′). Thus, the restorative effect of cholesterol on mTORC1 signaling during amino acid starvation is sufficient to support the insulin input. This is in accordance with the ability of cholesterol to restore proper lysosomal localization of mTORC1 (Fig. 5E,E′). Although cholesterol was able to sustain insulin-responsive mTORC1 signaling during amino acid starvation in Atg5 WT cells, we found that the effects were abolished in Atg5 KO cells (Fig. 6C,C′). To evaluate whether autophagy contributes by supplying an amino acid pool as reported earlier (Yu et al., 2010), we supplemented amino acids to Atg5 KO cells. A partial but evident restoration of mTORC1 signaling was detected in the presence of amino acids suggesting autophagy might contribute in part, by supplying an intracellular amino acid pool (Fig. 6C,C′).

Our results so far depicted the requirement of cholesterol in sustaining the insulin-mTORC1 axis under starvation conditions. Based on these data, we hypothesized that the natural cholesterol insufficiency in human disorders of cholesterol biosynthesis might lead to defects in the insulin-mTORC1 axis. To test this idea, we analyzed the insulin–mTORC1 axis in fibroblasts derived from individuals with SLOS. These individuals had pathogenic variants in the DHCR7 enzyme, which functions to generate cholesterol by reducing 7-dehydrocholesterol (7DHC) in the final step of the Kandutsch–Russell biosynthetic pathway (Irons et al., 1993; Tint et al., 1994). Skin fibroblasts from individual #1 (SLOS#1: GM05788) had G138V and H405Y variants, whereas those from individual #2 (SLOS#2: GM03044) had translocations in chromosomes 13 and 14. Under basal conditions, the pS6K (T389) levels were reduced in both SLOS #1 and SLOS #2 as compared to their wild-type counterparts (WT #1:GM05565C and WT #2:GM05399C) in the presence of insulin (Fig. 7A–A′). Based on the genotype–phenotype correlation, SLOS#1 is classified as severe and SLOS#2 as a moderate category. Consistent with this, the pS6K (T389) levels were reduced to a greater extent in SLOS#1 as compared to SLOS#2, WT #1 and WT #2 cells. The reduction in pS6K (T389) levels was more apparent in cells grown in lipoprotein-deficient serum (LPDS) suggesting that mTORC1 signaling is sensitive to cholesterol availability (Fig. 7A,A′). To confirm this observation, we compared temporal dynamics of mTORC1 signaling in SLOS and WT cells cultured initially on exogenous sterols (FBS) and then switched to exogenous sterol-deficient conditions (LPDS). Both WT and SLOS cells showed an initial decrease in pS6K (T389) levels when shifted from FBS to LPDS (Fig. 7B,B′). However, WT cells sustained mTORC1 signaling throughout the LPDS incubation period, whereas SLOS cells demonstrated a progressive decline as indicated by a reduction in pS6K (T389) levels (Fig. 7B,B′). Overall, this data indicates defective mTORC1 signaling in SLOS patients.

The differential response of mTORC1 observed in WT and SLOS cells might be due to the difference between exogenous and de novo synthesis of sterols to control mTORC1 activity. To test this, we blocked and reactivated exogenous and endogenous sterol pathways in both WT and SLOS cells. The pS6K (T389) levels were sustained in WT despite blocking exogenous sterol supply (LPDS) (Fig. 7C,C′). The residual pS6K (T389) levels further diminished when cells were treated with statin, a reversible inhibitor of HMG-CoA reductase enzyme involved in de novo sterol synthesis. The pS6K (T389) levels were restored upon removal of statin (washout) and increased several fold when exogenous sterols were supplemented via FBS (Fig. 7C,C′). By contrast, the pS6K (T389) levels were diminished at steady state in SLOS fibroblasts (Fig. 7C,C′). The levels were undetectable in LPDS, statin or washout conditions. In addition, pS6K (T389) levels increased in SLOS upon supplementation of exogenous sterols via FBS, but not to the extent that we saw in WT (Fig. 7C,C′). Together, it suggests that the mTORC1 signaling is dynamically regulated by availability of sterols and that in the absence of exogenous sterols; mTORC1 signaling is supported by de novo sterol synthesis in WT but not in SLOS patients.

Next, we investigated whether the insulin–mTORC1 axis is altered in individuals with SLOS. Although mTORC1 responded to insulin in a dose-dependent manner in both WT and SLOS, the extent of mTORC1 activation was comparably lower in SLOS even at the highest insulin concentration tested suggesting a defect in the insulin–mTORC1 axis in SLOS (Fig. 7D,D′). To confirm whether this defect in SLOS is due to cholesterol insufficiency, we fed SLOS cells with increasing concentrations of MCD-conjugated cholesterol. Insulin stimulation in the absence of exogenous cholesterol showed negligible mTORC1 signaling as revealed by diminished pS6K (T389) levels (Fig. 7E,E′). However, there was a dramatic increase in pS6K (T389) levels upon the addition of MCD-conjugated cholesterol (Fig. 7E,E′). Collectively, these data indicate that the insulin-mTORC1 axis is perturbed in SLOS due to cholesterol insufficiency and that the exogenous supplementation of cholesterol can rescue insulin–mTORC1 axis defects in SLOS.

To test whether the insulin–mTORC1 axis defects in SLOS is due to altered spatial arrangement of mTOR and TSC; we analyzed localization of mTOR and TSC in WT and SLOS fibroblasts. WT cells grown in FBS or LPDS showed considerable colocalization of mTOR and LAMP-2, suggesting sustained mTOR recruitment on lysosomes (Fig. S3A,B,E). However, consistent with the immunoblotting results, SLOS cells grown in FBS or LPDS exhibited reduced colocalization between mTOR and LAMP-2 indicating mTOR localization to lysosomes is altered in SLOS patient fibroblasts (Fig. S3A,B,E). Surprisingly, TSC localization was not altered in the presence or absence of insulin in both WT and SLOS cells (Fig. S3C,D,F) (Demetriades et al., 2016).

Given that mTORC1 is a key determinant of cell growth in mammals, we surmised that the altered insulin– mTORC1 axis could lead to growth defects in SLOS cells. Consistently, cell size defects were evident in SLOS fibroblasts cultured in LPDS. Compared to the WT counterpart, a relative decrease in the average cell area was seen in SLOS cells (Fig. S3H). To eliminate the effect of cell spreading, we assessed the size of trypsinized fibroblasts using flow cytometry and observed that the average cell diameter of SLOS fibroblasts is lower than that of WT fibroblasts (Fig. S3G,I).

In response to nutrient levels, heterodimeric Rag GTPases facilitate the recruitment of mTORC1 complex onto lysosomes for activation. Also, mutations that render Rag GTPases in constitutively active conformations have been shown to bypass the requirement for nutrients to translocate and activate mTORC1 on lysosomes (Sancack et al., 2008). To understand the molecular mechanisms underlying cholesterol-mediated rescue of insulin–mTORC1 axis defects in SLOS cells, we analyzed mTORC1 translocation and activation in cells expressing a constitutively active Rag GTPase mutant (GFP–RagB*Q99L). For the ease of transgene expression, we switched to a SLOS model generated in HeLa cells using AY9944 dihydrochloride, a pharmacological inhibitor of DHCR7 (Xu et al., 2011). AY9944 has been used extensively to recapitulate the biochemical hallmarks of SLOS both in vitro and in vivo (Fliesler et al., 2004, 2007). Consistent with those results, HeLa cells treated with AY9944 showed reduced colocalization of mTOR and LAMP-2 (Fig. S4A,A′,B,C) as well as diminished mTORC1 activation (Fig. 8A,A′) at steady state and upon insulin stimulation, confirming the insulin–mTORC1 axis defects observed in individuals with SLOS. However, the TSC axis remained functional and unaltered in the HeLa-SLOS model (Fig. S4D,E). In addition, similar to SLOS patient fibroblasts, the mTORC1 defects in the HeLa-SLOS model were also rescued upon supply of exogenous MCD-conjugated cholesterol (Fig. 8B,B′). Together, these results imply that the HeLa-SLOS model used in our study recapitulated the insulin-mTORC1 axis defects in fibroblasts from individuals with SLOS.

The HeLa-SLOS model expressing GFP alone showed significantly reduced colocalization of mTOR with LAMP2 (Fig. 8C,C′). Accordingly, mTORC1 signaling was markedly reduced in GFP-transfected cells even upon insulin stimulation as demonstrated by loss of pS6K (T389) levels (Fig. 8D,D′). Importantly, expression of GFP–RagB*Q99L restored mTORC1 translocation to lysosomes as revealed by a significant colocalization of mTOR and LAMP2 (Fig. 8C,C′). However, these cells failed to restore mTORC1 signaling in the absence of insulin. An increase in mTORC1 signaling was observed in GFP–RagB*Q99L-expressing cells upon insulin stimulation reinforcing the importance of integration of the Rag and Rheb axis for the complete activation of mTORC1 (Fig. 8D,D′). Based on these findings, insulin–mTORC1 axis defects in SLOS cells can be corrected in a cholesterol and Rag GTPase-dependent manner.

In mammals, extracellular cues like insulin (and other growth factors) act as triggers for mTORC1-dependent growth signaling. However, growth happens only if this insulin–mTORC1 axis is supported by a sufficient intracellular pool of nutrients. Studies in model organisms have revealed a genetic link between sterols and the insulin–mTORC1 axis for growth. In Caenorhabditis elegans and Drosophila, which are sterol auxotrophs (Shamsuzzama et al., 2020; Carvalho et al., 2010), it has been shown that the external supply of sterols is very important to sustain growth, larval development and longevity (Merris et al., 2003; Zanco et al., 2021). Moreover, mutants for genes encoding the SREBP pathway in Drosophila (Seegmiller et al., 2002) have shown reduced cell, organ and body size, which is reminiscent of the growth defects seen where there are defects in insulin or insulin growth factor, Akt, TSC and TOR signaling (Porstmann et al., 2008; Oldham et al., 2002; Zhang et al., 2000). Similarly, SREBP-2- deficient and hypomorphic mice also showed attenuated growth rates and reduced body weight and size (Vergnes et al., 2016). However, direct evidence supporting the role of cholesterol in insulin-induced mTORC1 signaling and adaptation is lacking.

Our findings reported here provide direct evidence for the coordination of cholesterol and insulin in driving mTORC1-dependent growth signaling at the cellular level. In this study, we show that cholesterol is required for the temporal progression of insulin-induced mTORC1 signaling. Consistent with the earlier reports, cholesterol supports the insulin–mTORC1 axis by recruiting mTORC1 on the lysosomal membrane without disturbing TSC (Castellano et al., 2017). Although amino acids have been shown to recruit mTORC1 onto lysosomes and support insulin-stimulated mTORC1 signaling, it appears that this is perhaps insufficient to completely activate mTORC1 within the required timeframe if cholesterol is not present on the lysosomal membrane (Dibble and Manning, 2013; Laplante and Sabatini, 2012; Castellano et al., 2017). Mammalian cells acquire cholesterol mainly from two sources – de novo biosynthesis in the ER regulated by the SREBP pathway or exogenous uptake of low-density lipoprotein (LDL) through clathrin-mediated endocytosis followed by its conversion into free cholesterol inside lysosomes (Brown and Goldstein, 1974, 1975; Anderson et al., 1976). Also, organellar crosstalk via membrane contact sites and sterol transfer proteins to relay the local availability of cholesterol might facilitate cholesterol exchange between compartments (Ikonen and Olkkonen, 2023). All these mechanisms are stringently regulated via feedback loops to maintain cholesterol homeostasis at the cellular and systemic levels (Anderson et al., 1977; Infante et al., 2008). These mechanisms independently or collectively might ensure cholesterol sufficiency on the lysosomal membrane to enhance the efficiency of insulin-induced mTORC1 signaling (Ikonen and Olkkonen, 2023). Although we have tested only HeLa cells and human skin fibroblasts in our study, the requirement of lysosomal cholesterol for insulin–mTORC1 function might be more universal. Recent studies in Ldlr^−/− mice and individuals showing reduced T-cell activation due to inefficient mTORC1 signaling arising from altered cholesterol routing to lysosomes owing to familial hypercholesterolemia further support this notion (Bonacina et al., 2022).

During starvation, mTORC1 is suppressed to activate autophagy to restore amino acid balance (Yu et al., 2010; Russell et al., 2013; Yu and Long, 2015). The amino acids replenished by autophagy together with glutamine metabolism reactivate mTORC1 and restore mTORC1 activity (Tan et al., 2017). In our investigations, cholesterol was found to be essential for lysosomal recruitment, reactivation and insulin-responsive signaling of mTORC1 during amino acid starvation despite the presence of functional autophagy and glutamine. Although from our results, it is difficult to pinpoint which of these steps require lysosomal cholesterol flux, the presence of cholesterol throughout the starvation period appears supportive. The suppression of mTORC1 and induction of autophagy to reactivate mTORC1 are shown to be of utmost physiological importance. Knock-in mice expressing a GTP-bound form of Rag GTPase (constitutively active) and Atg5-knockout mice died within one day of birth due to abnormally low levels of amino acids and glucose in plasma and tissues (Efeyan et al., 2013; Kuma et al., 2004). Currently, the role and mechanism of action for cholesterol in neonatal autophagy remain poorly understood (Kuma et al., 2004). Along these lines, it would be very interesting to see whether, in neonatal mice, reducing the levels of cholesterol soon after birth when the maternal cholesterol supply stops would cause the same effect. Also, during starvation, the restored mTORC1 will suppress autophagy and induces lysosome reformation to complete the feedback loop (Yu et al., 2010). The requirement of cholesterol in this context is also worth investigating.

Our findings in fibroblasts from individuals with SLOS and the HeLa-SLOS model showing reduced cell growth associated with defective basal insulin–mTORC1 axis function are physiologically relevant. It has been reported that the birth size and growth of the biochemically more severely affected individuals with SLOS is substantially less, likely due to early defects in insulin–mTORC1 axis (Donnai et al., 1986; Curry et al., 1987). In addition, infants with SLOS are small for gestational age and most continue to grow below the third centile (Ryan et al., 1998). Importantly, individuals with SLOS exhibit several metabolic defects, some of which could be associated with insulin signaling (Cunniff et al., 1997). For instance, cognitive impairment, renal hypoplasia, pancreatic islet cell hyperplasia, adrenal hyperplasia, electrolyte abnormalities, hypoglycemia and hypertension etc., which are usually found in individuals with insulin resistance and type-2 diabetes, can also be seen in individuals with SLOS (Pauli et al., 1997; Tint et al., 1994).

It is interesting to note that the basal mTORC1 signaling was reduced in SLOS cells despite the supply of serum containing LDL-cholesterol. Recent studies have reported that sterol insufficiency leads to defective membrane curvature and fission thereby reducing the efficiency of clathrin-mediated endocytosis (CME) in SLOS. In addition, sterol supplementation failed to completely rescue CME function, especially in individuals with severe SLOS, suggesting a default rewiring of CME to a slower mode in those individuals (Anderson et al., 2021). Importantly, other endo-lysosomal trafficking defects have also been reported in SLOS models (Kumar et al., 2021). Thus, the reduction in basal mTORC1 signaling in SLOS that we observe could be very well due to defective CME-mediated uptake or trafficking of LDL. Elevated autophagy and mitochondrial dysfunction are reported in SLOS (Chang et al., 2014). This could be explained by cholesterol insufficiency suppressing the insulin–mTORC1 axis and blocking the possibilities of mTORC1 reactivation via autophagy (Yu et al., 2010).

It has been reported that 7DHC being able to compensate for certain functions of cholesterol and accumulation of 7DHC is a hallmark of SLOS (Wassif et al., 2002). However, we did not observe a visible compensatory effect of 7DHC in the operation of insulin–mTORC1 axis in SLOS, at least in our experimental framework. This could be due to several possibilities. Firstly, 7DHC might not be able to compensate for cholesterol in the regulation of the insulin–mTORC1 axis, as the machinery that senses and relays cholesterol sufficiency on lysosomes (such as SLC38A9, NPC1, LYCHOS, etc.) might not be able to efficiently sense 7DHC (Castellano et al., 2017; Lim et al., 2019; Davis et al., 2021; Shin et al., 2022). Although we cannot neglect this possibility, it seems unlikely to be the case because of the structural similarities in cholesterol and 7DHC (Chang et al., 2014). Secondly, our analysis was focused mainly on the first 48 h of LDL deprivation where the contribution of 7DHC accumulation in supporting the insulin–mTORC1 axis and cell growth might be negligible. Thirdly, although 7DHC accumulation has been reported in SLOS, the total sterol concentration might not be sufficient to reach the threshold or optimal level required for the insulin–mTORC1 axis to proceed normally (Anderson et al., 2021). Our rescue assays with increasing concentrations of cholesterol also support this idea.

Although it has been established that cholesterol functions in a Rag-GTPase-dependent manner to activate mTORC1, its utilization to correct insulin–mTORC1 axis defects in the context of a cholesterol biosynthesis disorder like SLOS remains largely unexplored (Castellano et al., 2017). Currently, dietary cholesterol supplementation is the most followed global treatment strategy for SLOS (Linck et al., 2000). However, the outcome has been modest probably due to CME defects preventing efficient dietary cholesterol absorption or endo-lysosomal trafficking defects or other unexplored possibilities (Svoboda et al., 2012; Sikora et al., 2004). Our rescue experiments using constitutively active Rag GTPase might suggest a new dimension to correct the insulin–mTORC1 axis defects in individuals with SLOS.

In conclusion, we have shown that cholesterol sufficiency is key to driving insulin–mTORC1 growth signaling under basal, starvation and disease conditions. At this point, our investigations are limited to cultured cells and human skin fibroblasts. Future investigations in animal models and individuals with SLOS would help us derive stronger conclusions regarding the requirement of cholesterol in the functioning of the insulin–mTORC1 axis and its variations at the level of cell, tissue, organ and organism. Importantly, other than SLOS, it is known that there are many rare diseases with no effective treatment wherein mutations in genes encoding cholesterol biosynthetic enzymes or other regulators of cholesterol homeostasis are present (Porter and Herman, 2011; Platt et al., 2014). It would be tempting to speculate that there are insulin–mTORC1 axis defects in those diseases as well. However, the extent of cholesterol insufficiency, impairment of insulin–mTORC1axis function and its contribution to disease pathogenesis and clinical phenotypes observed in these diseases will require thorough study.

For immunoblotting, antibodies against p70 S6K (49D7) (#2708), phospho-p70S6K (Thr389) (#9234), 4EBP1 (53H11) (#9644), p4EBP1 (Ser65) (#9451), TSC2 (D93F12) (#4308), phospho-TSC (Thr1462) (#3617), Atg5 D5F5U (#12994) from Cell Signaling Technology and anti GFP antibody (#600-401-215) from Rockland antibodies were used at 1:1000 dilution. Anti-β-actin, (clone AC-15) (#A5441) was from Sigma-Aldrich and was used at 1:10,000 dilution. HRP-conjugated goat anti-rabbit-IgG (H+L) (#1721019) and goat anti-mouse-IgG (H+L) (#1721011) antibodies from Bio-Rad were used at 1:5000. For immunofluorescence, antibodies against mTOR (7C10) (#2938) from Cell Signaling Technology, and LAMP2 (H4B4) (#sc-18822) from Santa Cruz Biotechnology were used at 1:200 dilution. Donkey anti-rabbit-IgG (H+L) Alexa Fluor 488 conjugate (#A-21206) and donkey anti-mouse-IgG (H+L)- Alexa Fluor 568 conjugate (#A-10037) were from Invitrogen, Thermo Fisher Scientific and were used at 1:500 dilution. Recombinant DNA plasmids used are pEGFP-N1 (Clontech), pRK5 GFP RagB*Q99L (Addgene #112749).

HeLa cell line, an adenocarcinoma from the human cervix (ATCC) was procured from the National Cell Repository, National Centre for Cell Science, Pune, India. HeLa WT and HeLa Atg5 KO cells were kind gifts from Prof. Richard Youle (NINDS, NIH, USA). These cells were maintained in DMEM (GE Hyclone, SH30249.01) supplemented with 10% fetal bovine serum (GIBCO, #10270106) and 2 mM L-glutamine (Thermo Fisher Scientific, #25030081) (complete medium) at 37°C in an atmosphere of 5% CO₂. Human wild-type skin fibroblasts WT #1 (GM05565C), WT #2 (GM05399C) and skin fibroblasts from individuals with SLOS SLOS#1 (GM03044) and SLOS#2 (GM05788) were obtained from Coriell Institute for Medical Research, USA and were cultured and maintained in DMEM supplemented with 15% fetal bovine serum, 100 U/ml penicillin, and 100 µg/ml streptomycin (Thermo Fisher Scientific, #15140122) and 2 mM L-glutamine (complete medium) at 37°C in an atmosphere of 5% CO₂. All cell lines were authenticated and tested for contamination.

For all experiments, cells were grown in 10 cm dishes unless otherwise specified. To perform sterol depletion, sub-confluent HeLa cells in culture dishes were rinsed twice with serum-free media and incubated in DMEM containing 0.5–2.0% methyl β-cyclodextrin (MCD) (Sigma, #C4555) supplemented with 0.5% lipoprotein-deficient serum (LPDS) (Capricorn scientific, # FBS-DL-12B) for 2 h. Human skin fibroblast cells were treated with DMEM containing 7.5% LPDS for 12- 48 h. (Anderson et al., 2021). For cholesterol stimulation, these cells were treated with DMEM supplemented with 0.5% LPDS and 0.1% MCD with 2–50 μg/ml cholesterol (Sigma, #C3045) pre-heated at 37°C in a water bath for 2 h with frequent vortexing (resulting in an MCD and cholesterol mix with a 1:1 molar ratio) (Castellano et al., 2017). For insulin stimulation, cells were serum starved in DMEM (16 h) and stimulated with 50 nM–1 μM insulin (Sigma, #19278) for 1–60 min (Menon et al., 2014). In insulin and cholesterol combination experiments, sterol depletion, cholesterol and insulin restimulations were performed after 16 h of serum starvation. For amino acid starvation, after rinsing the cells twice with DMEM without amino acids (HIMEDIA, # AL967) cells were incubated for 0.5–10 h (based on the experiment) in DMEM lacking all amino acids except L-glutamine. For amino acid stimulation, cells were replaced with either standard DMEM containing all amino acids or NEAA plus EAA mixture (Thermo Fisher Scientific, #11130051, #11140050) for 10–15 min. For reactivation experiments, sterol depletion and cholesterol restimulation was done prior to amino acid starvation.

For Torin1 (Cell Signaling Technology, #14379) treatment, cells were grown in complete medium and treated with 100 nM Torin1 for 24 h. For inhibition of de novo cholesterol synthesis, cells were rinsed twice in serum-free DMEM and cultured under cholesterol-depleted conditions in 7.5% LPDS for 48 h with 5 μM AY9944 dihydrochloride (Tocris Bioscience, #1639) or 5 μM Simvastatin (Sigma-Aldrich, #S6196). DMSO-treated cells were used as a control in these experiments. For transfections, HeLa cells grown in six-well plates were transfected with specified plasmids using Lipofectamine 2000 (Invitrogen, #2423710) followed by 5 μM AY9944 treatment and insulin stimulation.

For cell lysate preparation, cells were washed with 1× PBS, lysed, and sonicated in lysis buffer (1% Triton X-100, 1× assay buffer (25 mM HEPES-KOH, pH7.2, 125 mM potassium acetate, 5 mM magnesium acetate, 2 mM EDTA, 2 mM EGTA), 1× protease inhibitor cocktail (Sigma, #4693159001) and phosphatase inhibitor cocktail (Sigma, #4906845001). Lysates were centrifuged at 20,000 g for 30 min at 4°C and the supernatant was taken for analysis. The samples were heated in SDS-sample buffer containing dye at 95°C for 5 min. Normalized lysates of equal volume were used for SDS-PAGE and transferred to nitrocellulose membranes and blocked overnight in TBS with 0.1% Tween 20 (TBST) containing 5% skimmed milk powder or 5% BSA. The membranes are incubated with specific primary antibodies diluted in 1% milk or BSA made in TBST either at room temperature for 1 h or at 4°C overnight. After washes with TBST, the membranes are incubated with horseradish peroxidase-conjugated secondary antibodies and developed via Amersham ECL (GE Life Sciences) or Clarity ECL (Bio-Rad) western blotting substrates using autoradiography.

To measure the cell size, human WT and SLOS skin fibroblasts were grown in complete medium and treated with LPDS for 48 h. Cells were then washed, trypsinized and analyzed for FACS using a CytoFLEX S (Beckman Coulter) flow cytometer analyzer with CytExpert software. To measure the cell area, phase contrast images from WT and SLOS fibroblasts were taken using IX73 inverted microscope (Olympus) using a 10× objective. The cell area was manually measured using the freehand tool in Image J (n=3, 100 cells in total).

HeLa cells were plated on glass coverslips in 35 mm tissue culture dishes. The dishes were rinsed with PBS once and coverslips were fixed for 10 min with 4% paraformaldehyde (Electron Microscopy Science, # 15710) in PBS at room temperature. For mTOR and LAMP2 staining, dishes were rinsed twice with PBS and cells were permeabilized with 0.5% Triton X-100 in PBS for 10 min. After rinsing twice with PBS, the cells were blocked in blocking buffer (3.75% BSA and 0.05% saponin in PBS) and rinsed three times with PBS. The incubation with primary antibodies (1:200) was performed in blocking buffer for 1 h at room temperature, rinsed three times with PBS, and incubated with secondary antibodies diluted in blocking buffer (1:500) for 1 h at room temperature in the dark, washed three times with PBS. DNA was stained with Hoechst 33342 DNA dye (1:2000 in PBS; Thermo Fisher Scientific, #H3570). For TSC and LAMP2 staining, cells were permeabilized with PBT solution (1× PBS and 0.1% Tween-20) for 10 min, and blocked with BBT solution (1× PBS, 0.1% Tween-20 and 0.1% BSA) for 45 min. The incubation with primary antibodies diluted (1:200) in BBT was undertaken for 2 h, cells were washed four times with BBT solution and incubated for 1 h with secondary fluorescent-conjugated antibodies diluted (1:500) in BBT solution. After two washes with PBT, the nuclei were stained with Hoechst 33342 DNA dye (1:2000 in PBT). Coverslips were mounted on slides using Fluromount-G (Electron Microscopy Science, #17984-25).

Images were acquired using a Leica SP8 WLL Laser Scanning Confocal Microscope equipped with a Plan Apo 63× oil objective and LAS X software. A 405 nm diode laser for Hoechst, a 488 nm argon laser for Alexa Fluor 488, and a 470–670 nm white light laser for Alexa Fluor 568 and Alexa Fluor 647 were used to capture fluorophores. Z-stacks were obtained by sequential scans of 0.3 µm step size between each optical section and were merged to obtain max-Z projection when required and were used as images after 2D deconvolution. Colocalization analysis of fluorescence confocal images were undertaken using the JACoP plugin embedded in Image J/Fiji software. Graphs represent mean±s.d. P values were calculated using a two-tailed unpaired Student's t-test and the graphs were prepared in GraphPad Prism 8.0.

We are grateful to Prof. Chandrabhas Narayana, Director, Rajiv Gandhi Centre for Biotechnology (RGCB) for providing the necessary infrastructure, facilities and advice to support this work. We are thankful to Dr T. R. Santhoshkumar for the advice and support throughout this study. We thank Dr Mahak Sharma and Dr Harikumar for their valuable suggestions. We thank Prof. Richard Youle and Dr Ravi Manjithaya for providing the Atg5 WT and Atg5 KO HeLa cells. We acknowledge the technical assistance of Yedukrishnan, Reshmi, Nishada and Krishnendu in the early stages of this work and Lekshmi during revision of the manuscript. We thank the Central Bioimaging facility, RGCB for help with microscopy and FACS.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261402.reviewer-comments.pdf.