Small molecule inhibition of the autophagy kinase ULK1 and identification of ULK1 substrates

Determination of the ULK1 kinase Consensus Phosphorylation Site

To identify additional substrates of ULK1 that may be important for the control of autophagy, we identified an optimal ULK1 phosphorylation consensus motif using arrayed degenerate peptide libraries, as we have previously performed for AMPK and AMPK-related kinases (Goodwin et al., 2014; Gwinn et al., 2008). To generate active ULK1 for these experiments, epitope-tagged ULK1 was co-expressed with its subunits FIP200 and Atg13 in HEK-293T cells and peptide eluted from affinity resin. The purified ULK1 complex exhibited robust kinase activity towards a known substrate, Atg13, in a dose-responsive fashion (Figure S1A). We used the purified ULK1 complex to screen a peptide library to determine its preferred sequence surrounding the phosphorylation site (Figure 1A). The results obtained with ULK1 correlate well with recent data on the peptide substrate specificity of the budding yeast ortholog of ULK1, Atg1 (Papinski et al., 2014). Unlike other Ser/Thr kinases (Miller et al., 2008; Turk, 2008) that phosphorylate sites near charged residues or proline, ULK1 had an unusual preference for hydrophobic residues at multiple positions surrounding the phosphorylation site. In particular, ULK1 strongly preferred a Leu or Met residue at position -3, while both aliphatic and aromatic hydrophobic residues were selected in the +1 and +2 positions. In addition, ULK1 strongly prefers Ser over Thr as the phosphoacceptor residue (Figure 1B). A consensus peptide substrate (ULKtide) that incorporated residues selected at each position flanking the phosphorylation site was efficiently phosphorylated by ULK1 in vitro. Starting with this peptide, we substituted key residues in ULKtide and measured the ability of our purified ULK1 complex to utilize these mutant peptides as substrates in an in vitro kinase assay. These experiments confirmed the importance of the -3, +1, and +2 positions for maximal phosphorylation efficiency (Figure 1C).

Identification of ULK1 substrates

A position-specific scoring matrix based on ULK1 peptide substrate specificity (Figure 1B) was used to bioinformatically search the human proteome for sites closely matching the ULK1 substrate consensus motif (Obenauer et al., 2003). We chose to focus first on those candidate substrates with highly conserved roles in autophagy, as a subset of core autophagy pathway components bears multiple ULK1 consensus sites. To define ULK1 phosphorylation sites in vivo, we took advantage of the fact that ULK1 is constitutively active when overexpressed in HEK-293T cells. We therefore compared global phosphorylation events on epitope-tagged candidate targets when co-expressed with wild-type (WT) ULK1 or a kinase-dead (KD) version. Using mass spectrometry, we identified phosphopeptides derived from each candidate protein under these conditions. These experiments revealed that in a subset of proteins bearing multiple ULK1 consensus sites, peptides from these candidate substrates were isolated that were highly phosphorylated in the presence of co-expressed WT but not KD ULK1. We first focused on the components of the ULK1 kinase complex itself, including FIP200, Atg13, and Atg101 (Figure 2 and Figure S2).

Atg101 was first identified in a mass spectrometry analysis of ULK1 interactors and found to encode a highly conserved integral component of the ULK1/FIP200/Atg13 complex in mammalian cell immunoprecipitations (Hosokawa et al., 2009; Mercer et al., 2009). Atg101 binds directly to Atg13 and is critical for Atg13 stabilization and its resultant stimulation of ULK1 kinase activity. To map potential ULK1-dependent phosphorylation events in Atg101, we co-expressed Flag-tagged Atg101 with WT or KD ULK1 and performed MS/MS analysis of total peptides in the Flag-Atg101 immunoprecipitates to map total phosphorylation sites in Atg101 under the two conditions. We observed that two specific serine sites (Ser11, Ser203) within human Atg101 were stoichiometrically phosphorylated in cells co-expressing WT but not KD ULK1 (Figure 2A). Notably. these two ULK1-dependent phosphorylation sites conform well to the optimal ULK1 substrate motif, in particular having hydrophobic residues at the -3 and +1 positions, which suggests that they may be direct ULK1 substrates in vivo. To further explore ULK1 phosphorylation of Atg101 in cells, we examined the migration pattern of Atg101 during SDS-PAGE in gels containing the Phos-Tag reagent, a phosphate binding dinuclear metal complex that specifically retards the mobility of phosphoproteins (Kinoshita et al., 2006). Comparing the mobility pattern of Atg101 on a Phos-Tag gel revealed a robust mobility change suggestive of phosphorylation (Figure 2B) when co-overexpressed in HEK-293T cells with WT ULK1. Mutation of Atg101 Ser11 abolished much of the mobility change, and this was further enhanced by additional mutation of Ser203 and nearby sites, which corroborated their identification as potential ULK1-dependent sites by mass spectrometry. We next performed a similar analysis of FIP200 and Atg13 phosphorylation events, and we discovered multiple serine sites in FIP200 and Atg13 bearing the ULK1 substrate consensus whose phosphorylation was induced by overexpressed WT ULK1 in cells (Figure S2A, S2B).

Next, we examined components of the Beclin1/Vps34 complex, which lies directly downstream of the ULK1 complex in autophagy initiation (Kim and Guan, 2015). Similar to Atg101, mass spectrometry identified sites of phosphorylation in Beclin1 induced by co-expression of WT but not KD ULK1 in HEK-293T cells, and those sites most extensively phosphorylated in this context matched the ULK1 optimal substrate sequence, suggesting that Beclin1 may also be a direct ULK1 substrate (Figure 2C). Further analysis identified additional serine residues in Beclin1 that conform to the optimal ULK1 phosphorylation motif that were not observed by mass spectrometry analysis, and some of these contributed to Beclin1 mobility changes on Phos-Tag gels (Figure S2C). One of these sites, Ser15, was recently identified as a site of ULK1 phosphorylation and was reported to play a conserved role in autophagy induction (Russell et al., 2013). We found that mutation of at least three of these serine residues (Ser15, Ser30, Ser337) was required to prevent the observed mobility shift on Phos-Tag gels (Figure S2C). Examination of another component of the Beclin1 complex, Ambra1, also revealed multiple ULK1-dependent phosphorylation events in cells, suggesting that many components of the Beclin/Vps34 complex may be targeted by ULK1 (Figure 2D). Finally, we examined a known ULK1 interacting protein, Syntenin-1, which was also reported as an ULK1 substrate (Rajesh et al., 2011). Here, we found that the previously reported in vitro phosphorylation site, Ser6, along with a second site, Ser61, are jointly responsible for altered mobility of Syntenin-1 in the presence of ULK1 in vivo (Figure 2E). Collectively, the majority of the candidate ULK1 phosphorylation sites we identified across these substrates contain hydrophobic residues in the -3 position (Figure 2F), consistent with our identified optimal ULK1 substrate sequence (Figure 1B) and with another site recently reported in another ULK1 substrate, STING (Konno et al., 2013). In contrast with the preferred peptide sequence, hydrophobic residues were not over-represented at the +2 position, possibly because short sequence motifs bearing three hydrophobic residues in close proximity might be inaccessible to phosphorylation by kinases.

In contrast to the majority of the substrates we examined that contained between two to four ULK1-dependent phosphorylation sites, one protein, the lipid kinase Vps34, apparently contained only a single ULK1 phosphorylation site (Figure 2F). The highly conserved Ser249 of Vps34 was stoichiometrically phosphorylated in HEK-293T cells when co-expressed with WT but not KD ULK1 (Figure 3A, B). Mutation of Ser249 alone was sufficient to abolish phosphorylation of Vps34 by ULK1 in vitro (Figure 3C) and prevented a ULK1-induced mobility shift on a standard SDS-PAGE gel upon co-expression in cells (Figure 3D).

We explored the potential function of Vps34 phosphorylation by ULK1 by introducing nonphosphorylatable (Ser249Ala) or phospho-mimetic (Ser249Asp) mutants into conditional Vps34-floxed murine embryonic fibroblasts (MEFs) (Jaber et al., 2012). After first corroborating the requirement of Vps34 for proper autophagy and ultimate cell viability (Figure S3A, data not shown), we tested the effects of these mutants in five assays of Vps34 function in autophagy. However, Ser249 in Vps34 was not required under the conditions we examined (Figure S3B–E). Given that ULK1 appears to phosphorylate multiple sites in Beclin1 and Ambra1 at the same time that it is phosphorylating Ser249 of Vps34, the data suggest that the sum effects of ULK1 on the different Beclin1/Vps34 subcomplexes is a highly regulated series of events requiring further study.

We next developed a phospho-specific antibody to Vps34 Ser249, whose signal was increased when ULK1 or ULK2, but not ULK3, was co-expressed with wild-type Vps34 but not a Ser249Ala Vps34 mutant in HEK-293T cells (Figure 3E).

Identification and characterization of a small molecule inhibitor of ULK1

To further examine how ULK1 regulates autophagy, we sought to identify small molecule inhibitors of the kinase. Starting with a focal adhesion kinase (FAK) inhibitor that showed potent cross-reactivity towards ULK1, we used a target-based reverse pharmacology approach to screen a focused library of pyrimidine analogues to identify small molecule ULK1 inhibitors. We performed in cellulo screening of the most promising 100 analogues from the in vitro kinase screening, testing their dose-dependent ability to inhibit the phosphorylation of Vps34 Ser249 by co-expressed ULK1 in HEK-293T cells. Structure-activity relationship (SAR) data were used to drive medicinal chemistry efforts that led to the ULK1 inhibitor tool compound SBI-0206965 (Figure 4A). Dose-response analysis of SBI-0206965 revealed an in vitro IC₅₀ of 108 nM for ULK1 kinase activity and 711 nM for the highly related kinase ULK2 (Figure 4B). We found SBI-0206965 (hereafter referred to as “6965”) inhibited Ser249 phosphorylation of overexpressed Vps34 when used at ~5 μM (Figure 4C). To ensure that 6965 was not specifically selective towards phosphorylation of Vps34 Ser249, we examined its ability to inhibit the phosphorylation of other ULK1 substrates we had identified. After validating a commercial antibody that could recognize a ULK1 phosphorylation site in Beclin1 (Ser15) when Beclin1 is overexpressed (Figure S4A), we found that 6965 inhibited Beclin1 Ser15 and Vps34 Ser249 phosphorylation to comparable extents in HEK-293T cells (Figure 4D). Furthermore, 6965 treatment collapsed the bandshift that Syntenin-1 and Atg13 undergo when co-expressed with wild-type ULK1 (Figure S4B, S4C). Unfortunately none of these existing phospho-specific antibodies against the ULK1 sites in Beclin1 or Vps34 are sensitive enough to recognize the endogenous proteins in a total cell lysate.

We next examined whether 6965 would inhibit endogenous ULK1 activity. To activate endogenous ULK1, we treated MEFs with either amino acid starvation media (EBSS) or the mTOR ATP-competitive inhibitors INK128 or AZD8055 (Chresta et al., 2010; Hsieh et al., 2012). In wild-type MEFs, we observed a mobility shift in endogenous Beclin1 and Atg13 in response to EBSS starvation media or the mTOR catalytic inhibitors, which was abolished in Ulk1/2-deficient MEFs (Figure 4E). The EBSS and mTOR-inhibitor induced mobility shift in Beclin1 and Atg13 was inhibited by 6965 co-treatment in wild-type MEFs (Figure 4E). Notably, neither Beclin1 nor Atg13 underwent a mobility shift upon treatment with either EBSS or the mTOR catalytic inhibitors in Ulk1/2-deficient MEFs, and no further decrease in their basal mobility was observed with 6965 co-treatment. We hypothesize that the mobility shifts observed in endogenous Beclin1 and Atg13 induced by mTOR inhibitors and starvation media reflect phosphorylation of endogenous Beclin1/Atg13 by endogenous ULK1/2 as they only occur in wild-type but not Ulk1/2-deficient MEFs.

SBI-0206965 is a highly selective kinase inhibitor

We next examined the specificity of 6965 using the DiscoveRx KINOMEscan panel of 456 purified human kinases in a competition-binding assay (Fabian et al., 2005; Karaman et al., 2008). By this analysis, 6965 was very selective, only inhibiting 10 out of 456 kinases >95% when tested at 10 μM (Figure 5A, 5B). The S(35) selectivity index of 6965 was 0.123, as measured by the percentage of the kinome inhibited below 35% of the control at this concentration (Figure S5A, S5B) [S(35) = (number of kinases with %Ctrl <35)/(number of kinases tested)]. This makes the selectivity profile of 6965 comparable to several kinase inhibitors currently used in clinical oncology, including Imatinib and Lapatinib (Figure S5B). Notably, by this assay, all the kinases bound by SBI-0206965 with an affinity comparable to ULK1 (FAK, Src, Abl, and Jak3) were tyrosine kinases (Figure 5C). Moreover, the only known functional ULK1 homolog in mammals, ULK2, was inhibited almost as well as ULK1 in this assay (Figure 5C), consistent with measurements obtained from in vitro kinase assays (Figure 4B).

We subsequently measured the selectivity of SBI-0206965 against its top binding kinases using radiolabel kinase assays in vitro at varying inhibitor concentrations. Here, we tested the ten kinases whose binding was most suppressed by 6965 by the DiscoverRx ATP binding assay (from Figure 5C). Of these kinases, only FAK and FLT3 showed an IC₅₀ comparable to ULK1 when measured by an in vitro kinase assay (Figure 5D), and 6965 inhibited Src, Jak3, and ULK2 five- to ten-fold less potently than ULK1 and FAK. Consistent with its efficacy of FAK inhibition in vitro, we found that at 10 μM, 6965 reduced endogenous FAK signaling in HEK293T cells (as judged by FAK and Paxillin phosphorylation) to an extent comparable to its inhibition of ULK1 signaling, albeit when ULK1 was overexpressed (Figure S5C). To examine the selectivity of 6965 more rigorously, we examined its impact on endogenous ULK1 and FAK signaling, as well as a number of other critical growth pathways, in wild-type MEFs that were starved or treated with mTOR inhibitors to endogenously activate ULK1 (Figure 4E, S5D). Notably, in MEFs, 6965 did not impair endogenous FAK, AMPK, mTOR, Akt, or Erk signaling at 10 μM, a dose at which it is capable of preventing a mobility shift in Atg13 that occurs in response to mTOR inhibition (Figure 4E, S5D). The suppression of endogenous ULK1 signaling was also monitored by a very recently developed phospho-specific antibody to one of the ULK1 sites in Atg13 (Ser318, which is Ser355 in human Atg13 isoform1) (Figure S5D). In fact, at 10 μM 6965 only modestly reduced FAK and Src signaling (as judged by FAK and Paxillin phosphorylation) and increased both AMPK and Erk-dependent signaling with increasing doses (Figure S5D). Thus, 6965 appears quite selective for ULK1 in vitro and in vivo, with a secondary selectivity towards FAK.

SBI-0206965 suppresses autophagy induced by mTOR inhibition, and this is phenocopied by ULK1 siRNA

To test the ability of 6965 to block autophagy and cell survival, initial studies were performed in A549 lung cancer cells, which are highly sensitive to mTOR inhibition (Sun et al., 2005). We observed that the catalytic ATP-competitive mTOR kinase inhibitor AZD8055 induced robust autophagy as visualized by accumulation of the Cyto-ID autophagy dye (Chan et al., 2012), and this effect was strongly suppressed by treatment with 5 μM 6965 (Figure 5E). Next, we genetically assessed the requirement for ULK1 versus other kinases inhibited by 6965 to induce autophagy after pharmacological mTOR inhibition. A robust high-throughput microscopy method for quantifying GFP-LC3 puncta was established using the PC3 prostate-cancer cell line stably expressing a GFP-LC3 construct (Degtyarev et al., 2008). Using this assay, we performed a focused RNAi analysis of the top 20 kinases identified in the DiscovRx screen as best binding to 6965. Quantitative measurement on wells of cells transfected with control siRNAs revealed a consistent 2-fold induction in GFP-LC3 puncta formation after treatment with either of the mTOR catalytic inhibitors INK128 or AZD8055 (Figure 5F,G). Strikingly, of the 18 kinases tested, only one kinase, ULK1, nearly fully abolished the LC3 puncta induced by the mTOR inhibitors (Figure 5F; Figure S5E). The ability of ULK1 siRNA to nearly fully ablate the autophagic response induced by mTOR inhibition suggests that, in this cell line at least, ULK1 is essential for stimulating autophagy in response to mTOR suppression.

SBI-0206965 prevents ULK1-dependent cell survival following nutrient deprivation

A well-established function of autophagy is to promote cell survival under conditions of nutrient deprivation. For example, ATG5-deficient immortal, nontumorigenic baby mouse kidney epithelial cells display no growth defects when cultured under normal media conditions; however when such cells are placed into starvation media, they undergo apoptosis at a greatly accelerated rate compared to controls (Guo et al., 2011). Similarly, we previously demonstrated that RNAi against ULK1 and ULK2 phenocopied RNAi against ATG5 with regards to the loss of cell viability under nutrient deprived conditions (Egan et al., 2011b). To examine whether our small molecule ULK1 inhibitor would similarly affect cell survival under nutrient deprived conditions, we treated MEFs with 6965 in the context of normal media, amino acid-deprived media, or glucose-deprived media. After 18 hours of amino acid deprivation, 15% of the vehicle-treated MEFs were positive for both 7-AAD and AnnexinV, indicative of end-stage apoptosis (Figure 6A), versus 42% of the 6965-treated cells. Similar effects were also seen in glucose-deprived MEFs, where 6965 also promoted enhanced cell death (Figure S6A). An immunoblot timecourse analysis of amino-acid starved cells revealed that active cleaved caspase-3 and the cleavage of its target PARP was only robustly observed in starved, 6965 co-treated cells (Figure 6B). Interestingly, the immunoblot analysis revealed that 6965 treatment induced the loss of ULK1 and Atg13 protein, but only in nutrient-deprived and not nutrient-replete conditions. Perhaps only in this context when ULK1 is activated does the direct binding of 6965 stimulate ULK1 turnover. We also directly compared 6965 to chloroquine (CQ), a lysosomotropic agent that acts as a general autophagy inhibitor (Kroemer, 2015), in their ability to drive apoptosis under starvation conditions. While both 6965 and CQ induced cell death only under starvation conditions, it was notable that 10 μM 6965 was more effective than 20 μM CQ at rapidly inducing apoptosis in this context (Figure S6B). Control immunoblotting revealed that 20 μM CQ was effective at blocking autophagic flux, preventing the turnover of the Atg8 family member LC3 as well as the autophagy cargo receptor p62/SQSTM1 (Figure S6C, 6B). The degradation of both ULK1 and Atg13 was again specifically observed under starvation conditions in cell treated with 6965 (Figure S6C). Of note, administration of CQ or bafilomycin A1, but not the proteasome inhibitor MG132, abrogated the degradation of both ULK1 and Atg13 indicating that their own turnover is lysosomal in nature (Figure S6D).

Beyond the general role of autophagy that occurs in many cell types as a response mechanism to nutrient deprivation, there has also been considerable interest in the role of autophagy as a survival mechanism for tumor cells faced with nutrient limitations due to poor vascularization as well as metabolic stress resulting from chemotherapies or targeted therapeutics (White, 2015). We examined whether 6965 would selectively promote apoptosis under conditions in which autophagy is actively engaged in tumor cells similar to what we observed in MEFs. Indeed, 6965 promoted apoptosis (AnnexinV+ cells) particularly in the nutrient-starved state in human U87MG glioblastoma cells and in murine lung carcinoma cells derived from a genetically engineered mouse model of lung cancer driven by mutant Kras and loss of p53 (Figure 6C,6D). Notably, these murine lung cancer cells have been previously shown to be particularly dependent on autophagy (Guo et al., 2013).

The small molecule ULK1 inhibitor SBI-0206965 converts the cytostatic response to catalytic mTOR inhibitors into a cytotoxic response

Given that mTOR inhibition induces ULK1 activation (Figure 4E and S5C) and ULK1 siRNA or 6965 suppresses autophagy induced by mTOR inhibitors (Figure 5E–G), we examined whether 6965 would reduce cell survival in the face of mTOR inhibition. A549 cells were treated with a constant dose of AZD8055 (1 μM), which induces cytostatic growth arrest, and escalating doses of 6965. At a 5 μM dose, 6965 triggered apoptosis in 23% of cells when combined with AZD8055 versus 10% of cells in the absence of AZD8055 and 7% of cells treated with AZD8055 alone (Figure 7A). The induction of AnnexinV+ apoptotic cells was heightened at 10 μM dosing of 6965. As observed in MEFs with nutrient deprivation combined with 6965, immunoblot analysis revealed that only the combination of 6965 with mTOR inhibitors triggered caspase activation in A549 cells, paralleling the FACS analysis of cell death (Figure S7A). Moreover, degradation of total ULK1 and Atg13 was observed as before (Figure 6B, S6C): only in the presence of the autophagy activating stimulus (AZD8055) and 6965 (Figure S7A).

To examine whether 6965 would broadly synergize with other mTOR inhibitors, we examined the ability of rapamycin and a second mTOR catalytic inhibitor, INK128 (MLN0128), to synergize with 6965 (Figure 7B). Similar to AZD8055, both rapamycin and INK128 exhibited a mild apoptotic response on their own in A549 cells but significantly enhanced this apoptotic response when combined with 6965 (Figure 7B). Notably, the extent of AnnexinV+ apoptotic cells induced by the three different mTOR inhibitors was mirrored by the proportion of Atg13 protein loss and the extent to which Beclin1 mobility was impaired (Figure 7C). This also appeared to reflect the potency of mTOR inhibition, as assessed by 4EBP1 mobility shift and loss of the direct mTOR phosphorylation of ULK1 on Ser757 (Figure 7C). A model for the selectivity of different mTOR substrates for inhibition by rapamycin or catalytic mTOR inhibitors proposed that substrate quality may dictate how easily a substrate phosphorylation site is blocked by mTOR inhibition (Kang et al., 2013). From this analysis, ULK1 was identified an excellent mTOR substrate in vitro, similar to 4EBP1, and required full mTOR catalytic inhibition to reduce its phosphorylation. Hence, stronger mTOR inhibition, provided by catalytic mTOR inhibitors, is expected to be superior to rapamycin with regards to blocking phosphorylation of ULK1 at the Ser757 mTOR inhibitory site (Kang et al., 2013). Indeed, the extent of apoptosis induced by mTOR inhibitors combined with 6965 thus parallels the ability of these three mTOR drugs to inhibit ULK1 phosphorylation and to induce degradation of the ULK1/Atg13 complex in the presence of 6965. Furthermore, equimolar 6965 was more potent than chloroquine at inducing apoptosis in combination with all three mTOR drugs (Figure 7B, C).

Discussion

Given that the most upstream component of the conserved autophagy cascade encodes the only serine/threonine kinase in the cascade, it has been assumed that ULK1-dependent phosphorylation of other components of the pathway must instruct and provide proper temporal and spatial cues for autophagy initiation. Here, we shed additional light on the complexities involved in the autophagic process by identifying multiple ULK1-dependent phosphorylation sites in Beclin1, Ambra1, Atg101, Syntenin-1, and Vps34. Recent reports that AMPK phosphorylates Beclin1 and Vps34 on distinct sites from these ULK1 sites (Kim et al., 2013) and that the stress-induced MK2/MK3 kinases also target additional sites in Beclin1 (Wei et al., 2015) indicates that much work remains to fully dissect the context-specific, temporal, and spatial differences in the regulation of Vps34/Beclin1 complexes. The induction of ULK1 activity following catalytic mTOR inhibition alone (Figure 4E and S5C) is consistent with amino acid deprivation induction of ULK1 kinase activity, which does not appear to involve AMPK. mTORC1 phosphorylates and inhibits ULK1 via at least one well-established serine site in ULK1, Ser757 (Kang et al., 2013; Kim et al., 2011) and the extent to which different mTOR inhibitors blocked phosphorylation of Ser757 correlated with the ability of 6965 to promote apoptosis. Given that ULK1 is inhibited by mTORC1, further delineation of ULK1 substrate phosphorylation sites may yield valuable biomarkers for mTOR inhibitors, as the attendant increase in ULK1-dependent substrate phosphorylation would be predicted to reflect the relative intensity and effective duration of mTOR inhibition. This appears to be the case as revealed by endogenous phospho-Atg13 levels in MEFs following starvation or mTOR inhibitor treatment (Figure S5D).

The fact that ULK1 is the only conserved serine/threonine kinase in the autophagy cascade makes it a very attractive target for therapeutic development. Our finding that ULK1 and its binding partner Atg13 are selectively degraded by combining 6965 with starvation or with mTOR inhibitors but not by any of these treatments alone indicates that specifically the active pool of ULK1 kinase complexes may be sensitive to 6965-induced lysosomal degradation. These results provide additional biomarkers for ULK1 inhibition in vivo, as they suggest that ULK1 or Atg13 will be degraded when effectively inhibited by on-target small molecule inhibitors in those cells that rely on ULK1 for survival. Notably, when comparing three mTOR inhibitors for their ability to synergize with 6965 to induce cell death in A549 cells, the extent of apoptosis was paralleled best by loss of Atg13 protein, which in turn mirrored the level of caspase-cleaved PARP. Given the number of ongoing clinical trials combining CQ and CQ analogs with existing cancer therapies, which include mTOR inhibitors (Rangwala et al., 2014), it also worth noting that equimolar 6965 induced more apoptosis than CQ when combined with mTOR inhibitors (Figure 7B,C).

Recently, the crystal structure of ATP-competitive kinase inhibitors bound to ULK1 was reported, marking the first report of the ULK1 atomic structure (Lazarus et al., 2015). However, the small molecule inhibitors developed in this study were non-selective, inhibiting tens of kinases >95% which made them impossible to utilize for cell culture studies (Lazarus et al., 2015). Our study here on 6965 therefore reports one of the first tool compounds that inhibits ULK1 with exquisite selectivity and is suitable for cell studies for the autophagy field. Future efforts will explore the use of this compound and its derivatives in animal models of cancer and other diseases.

mTOR inhibitors are being widely tested in clinical trials for oncology, and rapamycin analogues are the approved standard of care for advanced renal cell carcinoma and other solid tumors (Benjamin et al., 2011). From a therapeutic perspective, an exciting prediction is that ULK1 inhibition will convert the cytostatic response to mTOR inhibition to a cytotoxic response due to loss of autophagic maintenance of cell survival (Figure 7D). We directly tested this hypothesis and found that it holds true in A549 non-small cell lung cancer cells treated with three different mTOR inhibitors (Figure 7). An important step moving forward will be to delineate those tumors that are most likely to show benefit from ULK1 inhibitors like 6965. In hamartoma syndromes, such as tuberous sclerosis complex and Peutz-Jeghers syndrome, inactivating mutations in the tumor suppressors TSC1, TSC2, PTEN, and LKB1/STK11 result in mTOR hyperactivation, and these mutations are targetable by mTOR inhibitors (Bissler et al., 2008). One would hypothesize that the tumors most sensitive to mTOR inhibitors and hence those perhaps most likely to benefit from dual mTOR/ULK1 inhibition will be those tumors bearing mutations in genes encoding components of the core mTOR signaling machinery itself (Grabiner et al., 2014). Another possibility is that combining a ULK1 inhibitor such as 6965 with mTOR inhibitors may allow for lower dosing schedules of catalytic mTOR inhibitors. ULK1 inhibition therefore represents an exciting new modality to avoid therapeutic resistance in the many patients currently treated with mTOR inhibitors. The additionally identified ULK1-dependent phosphorylation sites, functional assays, and ULK1 inhibitor tool compound identified in this study will help catalyze further exploration of the roles of this highly conserved autophagy kinase in cell biology and in therapeutic possibilities for a wide variety of diseases.

Antibodies and reagents

Cell Signaling Antibodies used: Atg13 (#6940), Beclin1 (#3495), PARP (#9542), 4E-BP1 (#9452), phospho-ULK1 S757 (#6888), Ulk1 (#8054), p70 S6K (#9202), phospho-S6 S235/236 (#4858), Myc tag (#2278), LC3B (#3868), Vps34 (#4263), phospho-AMPK T172 (#2535), phospho-ACC S79 (#3661), phospho-Paxillin Y118 (#2541), and PathScan Multiplex Western Cocktail I (#5301). Phospho-Vps34 Ser249 antibody was developed in collaboration with Gary Kasof at Cell Signaling Technology. Sigma antibodies used: ULK1 (A7481), β-actin (A5541), and Flag tag polyclonal (F7425). Gabarap antibody from Abgent (PM037). Phospho-FAK Y397 from Abcam (ab4803). FAK from Epitomics (2146-1). p62 SQSTM1 antibody from Progen (03-GPP62-C). Phospho-Beclin1 S15 from Abbiotec (254515). Phospho-Atg13 S318 (human S355) from Rockland (600-401-C49S). EBSS (14155-063) and glucose-free media (11966-025) from Gibco/Life Technologies. Chloroquine (C6628) and bafilomycin A1 (B1793) from Sigma. AZD-8055 (A-1008) from Active Biochem. AnnexinV-PE Apoptosis Detection Kit from BD Biosciences. Phos-tag™ AAL-107 from NARD (#304-93521). Ad5-CMV-Cre was purchased from the University of Iowa viral vector core.