Follicular lymphoma-associated mutations in the V-ATPase chaperone VMA21 activate autophagy creating a targetable dependency

ABSTRACT The discovery of recurrent mutations in subunits and regulators of the vacuolar-type H+-translocating ATPase (V-ATPase) in follicular lymphoma (FL) highlights a role for macroautophagy/autophagy, amino-acid, and nutrient-sensing pathways in the pathogenesis of this disease. Here, we report on novel mutations in the ER-resident chaperone VMA21, which is involved in V-ATPase assembly in 12% of FL. Mutations in a novel VMA21 hotspot (p.93X) result in the removal of a C-terminal non-canonical ER retrieval signal thus causing VMA21 mislocalization to lysosomes. The resulting impairment in V-ATPase activity prevents full lysosomal acidification and function, including impaired pH-dependent protein degradation as shown via lysosomal metabolomics and ultimately causes a degree of amino acid depletion in the cytoplasm. These deficiencies result in compensatory autophagy activation, as measured using multiple complementary assays in human and yeast cells. Of translational significance, the compensatory activation of autophagy creates a dependency for survival for VMA21-mutated primary human FL as shown using inhibitors to ULK1, the proximal autophagy-regulating kinase. Using high-throughput microscopy-based screening assays for autophagy-inhibiting compounds, we identify multiple clinical grade cyclin-dependent kinase inhibitors as promising drugs and thus provide new rationale for innovative clinical trials in FL harboring aberrant V-ATPase. Abbreviations: ALs: autolysosomes; APs: autophagosomes; ER: endoplasmic reticulum; FL: follicular lymphoma; GFP: green fluorescent protein; IP: immunoprecipitation; LE/LY: late endosomes/lysosomes; Lyso-IP: lysosomal immunoprecipitation; OST: oligosaccharide transferase; prApe1: precursor aminopeptidase I; SEP: super ecliptic pHluorin; V-ATPase: vacuolar-type H+-translocating ATPase


Introduction
Follicular lymphoma (FL) constitutes the most common indolent B-cell lymphoma, with an incidence and prevalence of ~15,000 and ~100,000 cases, respectively, in the US [1]. FL remains incurable with conventional therapies and most patients receive multiple treatment regimens over the course of their illness. While the overall clinical outlook for FL patients is improving, those with short remission durations following upfront chemoimmunotherapy, or patients transforming to large B cell lymphoma have substantially shortened overall survival [2]. A better understanding of FL growth and survival dependencies and the development of novel targeted FL therapies therefore remain a research priority [3][4][5].
The identification of MTOR-activating mutations in the small G-protein RRAGC, a component of the amino-acid signaling pathway to MTOR, has highlighted an opportunity to target this pathway in FL [6][7][8]. The RRAGC protein is part of a macromolecular complex consisting at a minimum of the protein and peptide hydrolases that degrade protein cargo delivered to lysosomes via autophagosomes, multivesicular bodies, microautophagy or macropinocytosis [22]. The mutations in ATP6V1B2 result in a hypomorphic V-ATPase with reduced ability for lysosomal acidification, impaired protein degradation and a resulting strong compensatory upregulation of autophagic flux. The upregulation of autophagy is important for FL B cell survival as the treatment of ATP6V1B2 mutant human FL B cells with inhibitors of ULK1 (unc-51 like autophagy activating kinase 1), the apical autophagy initiating kinase, results in FL cell death.
Various other genes related to V-ATPase function, including ATP6AP1 and ATP6AP2 are mutated in FL but the functional consequences of these mutations on FL biology and autophagy or MTOR regulation have not been elucidated [6]. The V-ATPase V 0 multiprotein transmembrane complex is assembled in the ER with the obligate involvement of a set of assembly chaperones [23][24][25][26]. These chaperones include ATP6AP1 and ATP6AP2 as well as the small protein VMA21. Here, we report on common mutations in VMA21 in FL that result in aberrant VMA21 cellular trafficking, and reduced V-ATPase holoenzyme function with an associated inability to fully acidify lysosomes. This lysosomal defect is associated with an aberrant lysosomal and cytosol metabolome and a strong compensatory activation of autophagy. These novel findings therefore unify the phenotypes of VMA21 and ATP6V1B2 mutant FL thus extending the pathobiological consequences of hypomorphic V-ATPase pathology to a combined total of 20-25% of FL.
We previously demonstrated a survival dependence of ATP6V1B2 mutant FL on autophagic flux, which motivated us to screen for drugs that inhibit VMA21-mutant induced autophagy. Our high-throughput screen of 4,577 compounds identified multiple compound classes with autophagy inhibitory properties, most prominently cyclin-dependent kinase inhibitors. The identification of compounds that are in clinical development or approved for clinical use in humans provides a direct path to novel therapeutic approaches in FL. Altogether, we report that a substantial subset of FL carry mutations that through various mechanisms result in hypofunctional V-ATPase, an elevated lysosomal pH and strong compensatory activation of autophagy, which lays the mechanistic foundation for innovative clinical trials repurposing various promising autophagy inhibitory drugs as therapeutics for V-ATPase-mutant FL.

The frequency and nature of VMA21 mutations in follicular lymphoma
Based on recurrent common mutations in genes associated with V-ATPase biology (ATP6V1B2, ATP6AP1, ATP6AP2), we re-sequenced the three coding exons of the V-ATPase chaperone gene VMA21 and adjacent intronic regions using Sanger sequencing in highly pure FL B cell DNA isolated from flow sorted FL B cells in a total of 151 FL grades 1-3A and 19 transformed FL cases [6,17,18]. We confirmed the somatic nature of all sequence variants using paired sorted lymph-node or blood-derived CD3 + T cell DNA. We identified 12% (18/151) of FL cases and 2/19 t-FL with non-synonymous VMA21 mutations, most of which were hemizygous ( Figure 1A-B).
The VMA21 gene encodes a small transmembrane ER/ Golgi chaperone protein of 101 amino acids. Topologically, the N and C termini are cytoplasmic and two transmembrane segments connect via a lumenal (ER, Golgi, lysosome) domain. The VMA21 gene is targeted in FL by a hotspot nonsense mutation (c.C277T; p.R93X) comprising 40% of all detected mutations resulting in the deletion of the C-terminal nine amino acids (amino acids 93-101). Additional nonsense mutations (p.W88X and p.W95X) also delete parts of the C terminus. Various missense mutations targeted other parts of the gene ( Figure 1A-C).
A review of gene mutations co-occurring in FL cases with mutant VMA21 demonstrated that mutations in other genes related to V-ATPase function and autophagy, including ATP6V1B2, ATP6AP1, ATP6AP2, and sporadic mutations in ATP6V0A4, ATP6V0C and ATP6V1A were absent in VMA21 mutant cases. Furthermore, the majority of VMA21 mutant cases carried mutations in KMT2D and CREBBP ( Figure S1).
Of the VMA21 mutations reported here, eleven were detected in males and one in a female while the sex of the patients with mutations detected in samples from the deidentified FL cohort II is unknown. Further, five VMA21 mutations were detected in untreated FL patients and seven in relapsed patients, while the clinical status of the patients with mutations detected in the de-identified samples is unknown.
A review of CRISPR-targeting data based on 1027 cell lines reported in the depmap portal (https://depmap.org/portal/) indicated that VMA21 disruption conferred a competitive disadvantage on cells (categorized as imparting reduced cell fitness). The effect, however, was less pronounced when compared with disruption of ATP6AP1 or ATP6AP2, which were essential for most of the cell lines tested.

Elevated LC3-II levels in VMA21 mutant (MUT) cells
Given the strong autophagy-activating phenotype of mutations in the V-ATPase subunit ATP6V1B2, we tested for effects of VMA21 mutations on autophagy using the steady state levels of the autophagosomal marker LC3-II [27]. We generated stable HEK293T cells carrying the doxycyclineinducible vector pCW57.1 alone or vector containing cDNAs for WT or MUT (p.93X) HA-tagged VMA21. Upon transgene induction, we detected strongly elevated levels of LC3-II in the mutant transfectants (Figure 2A, B). We also performed experiments with the V-ATPase and autophagy inhibitor bafilomycin A 1 and found that LC3-II levels in the mutant increased further, in agreement with elevated autophagy flux (Figure 2A, B).

Elevated numbers of autolysosomes in VMA21 MUT cells as measured by electron microscopy
One of the "gold standard" assays to quantify autophagic flux, here defined as the generation of autophagosomes (APs) and their subsequent maturation and fusion with lysosomes, is via the enumeration of APs, autolysosomes (ALs) and late endosomes/lysosomes (LE/LY) in cells based on electron microscopy (EM). Using the inducible HEK293T VMA21p.93X MUT or VMA21 WT cell lines we counted APs, ALs and LE/LYs in 70-90 EM images each per cell line. We found that the number of ALs was substantially and significantly elevated in the VMA21p.93X cell lines with a mean of 0.23 for vector, 0.25 for WT and 1.06 for MUT per cell, confirming pathological elevation of autophagic flux induced by the MUT VMA21 protein ( Figure 2C, D).

Elevated autophagic flux in yeast carrying the Vma21[Δ66-77] mutation in the genome
To further investigate the effect of the VMA21p.93X mutation on autophagy and autophagic flux, we proceeded to test its effects in a yeast model. Sequence alignment showed that the VMA21p.93X mutation corresponds to yeast Vma21[∆66-77]. Therefore, we generated a Vma21[∆66-77] mutation in the genome. As a readout of autophagy activity, we measured green fluorescent protein (GFP)-Atg8 processing in S. cerevisiae cells expressing either full-length Vma21 (WT) or the truncated version Vma21[∆66-77]. Atg8 binds to both sides of the phagophore [28], and a portion of Atg8, or GFPtagged Atg8, is delivered to the vacuole via autophagy; GFP is relatively resistant to vacuolar hydrolases and the free form can be resolved and detected by western blot. The Vma21 [∆66-77] mutant showed an increased level of free GFP under nitrogen-starvation conditions, suggesting elevated autophagic flux in the mutant cells ( Figure 2E, F).
We extended our measurement of autophagy to the quantitative Pho8∆60 assay. Upon deletion of the first 60 amino acids at the N terminus, Pho8 phosphatase can only be  -77] were analyzed for autophagy. The cells were grown in YPD (-N, 0 min) to mid-log phase, and shifted to nitrogen starvation (SD-N) (E) The GFP-Atg8 processing assay. Pgk1 was used as a loading control. (F) The ratio of free GFP to Pgk1 after 45 min SD-N treatment was quantified. Mean ± SD of n = 4 independent experiments. Unpaired, 2-tailed t-test; *: p < 0.05. (G) Pho8∆60 enzymatic activity was measured under growing and nitrogen-starvation conditions. Mean ± SD of n = 5 independent experiments. Two-way ANOVA; ***: p < 0.001. (H) The prApe1 processing assay. Dpm1 was used as a loading control. (I) The ratio of Ape1 to prApe1 after 15 min SD-N treatment was quantified. Mean ± SD of n = 3 independent experiments. Unpaired, 2-tailed t-test; *: p < 0.05. (J) The Atg8 lipidation assay. Pgk1 was used as a loading control. (K) The ratio of Atg8-PE to Pgk1 after 45 min of nitrogen starvation was quantified. Mean ± SD of n = 4 independent experiments. Two-way ANOVA; *: p < 0.05, **: p < 0.01. delivered to the vacuole nonselectively through autophagy [29]. Subsequently, the Pho8∆60 propeptide gets cleaved by vacuolar hydrolases, thereby activating its phosphatase activity. We observed a significant increase in autophagydependent Pho8∆60 phosphatase activity under nitrogenstarvation conditions in the Vma21[∆66-77] mutant cells ( Figure 2G).
Next, we performed one additional assay that monitors a type of selective autophagy. Ape1 is a vacuolar hydrolase that is originally synthesized as a cytosolic precursor form (prApe1). The prApe1 can be delivered to the vacuole through either macroautophagy or the cytoplasm-to-vacuole targeting/ Cvt pathway [30]. In the vacuole, prApe1 is processed to a mature form with lower molecular mass that can be easily detected by western blot. The assay was performed in cells with a vac8∆ background, which prevents vacuolar delivery of prApe1 under growing conditions; however, the precursor is efficiently delivered via autophagy under nitrogen-starvation conditions. Unlike the previous two assays, import of prApe1 depends on selectivity factors including the receptor Atg19 and the scaffold Atg11. We observed a similar autophagy increase in the Vma21[∆66-77] mutant cells under starvation conditions ( Figure 2H, I).
To analyze an earlier aspect of autophagy, we examined the induction step, following the endogenous Atg8 protein through a lipidation assay. The lipidated form, Atg8-PE, displays a faster electrophoretic mobility relative to nonlipidated Atg8 that can be detected by western blot. The cells expressing Vma21[∆66-77] showed a substantial increase in the levels of Atg8-PE under both growing and nitrogen-starvation conditions, indicating upregulated autophagy induction ( Figure 2J, K).

Generation of quantitative autophagic flux reporter cell lines expressing inducible VMA21 WT or MUT
We generated HEK293T cell lines stably expressing the wellstudied cleavable autophagy reporter fusion protein GFP-LC3 -(ATG4 proteolytic cleavage site)-RFP together with inducible HA-VMA21 WT or MUT ( Figure 3A) [31]. Upon autophagy induction, the GFP-LC3-RFP fusion protein is cleaved into GFP-LC3 and RFP, followed by lipid conjugation of LC3 and association of GFP-LC3-II with APs and ALs. We employed these cells for a variety of autophagy assays and for chemical autophagy inhibitor screens detailed below.

Elevated levels of free GFP in VMA21 MUT cells carrying GFP-LC3-RFP autophagy reporters
Upon association of GFP-LC3-II with phagophores and subsequent fusion of APs with lysosomes, the GFP-LC3-II protein is subjected to partial proteolysis by lysosomal proteases resulting in free GFP. We therefore measured free GFP in HEK293T cells and OCI-LY1 lymphoma cells inducibly expressing WT or MUT VMA21 or WT or MUT ATP6V1B2 (p.371C and/or p.400Q) as controls. We detected elevated levels of endogenous LC3-II and elevated levels of free (lysosomal) GFP in all MUT cell lines ( Figure 3B, C and quantification in Figure 3 E, F).

Elevated numbers of autolysosomes in VMA21 MUT cells and inhibition by ULK1 or PIK3C3/VPS34 inhibitors
Following induction of VMA21 MUT expression in the reporter cells, we detected many green cytoplasmic vesicles (APs and ALs) and much fewer such vesicles in VMA21 WT cells indicative of VMA21-MUT induced autophagy ( Figure 3D). The GFP fluorescence signal is usually quenched in acidic lysosomes over time. However, as detailed below, the lysosomes in VMA21 MUT cells are not as acidic as in WT cells thus allowing for persistence of GFP protein and the green ALs as shown in Figure 3D.
Next, we tested the effects of two well-studied small molecule inhibitors of the proximal autophagy initiator proteins ULK1 (MRT68921) and PIK3C3/VPS34 (SAR405) on VMA21-MUT induced autophagic flux [32,33]. As can be seen in Figures 3D-G, both compounds substantially inhibited the appearance of APs and ALs in VMA21 MUT cells, thus demonstrating involvement of canonical macroautophagy initiation complexes in VMA21-MUT activated autophagy.

Results from VMA21 cellular localization studies using confocal microscopy
Yeast Vma21 is one of five proteins known to be necessary for successful V-ATPase V 0 complex assembly in the ER [23,34]. Vma21 is an assembly factor and a chaperone for the V 0 complex, accompanying it as it traffics from the ER toward the Golgi, although mechanistic details on how this is accomplished are lacking. Upon arrival at the Golgi, Vma21 dissociates from the V 0 complex and Vma21 is shuttled back to the ER via a C-terminal KKXX (KKED in yeast Vma21) retrieval sequence [23].
To study the functional consequences of the deletion of the C terminus (amino acids 93-101) in human MUT VMA21, we employed confocal microscopy to localize HA-tagged WT and MUT VMA21 in cells. We found that WT VMA21 predominantly localized to the ER/Golgi, whereas MUT VMA21 was strongly enriched in lysosomes as evidenced by enhanced colocalization of MUT VMA21 with the lysosomal protein LAMP1 ( Figure 4A, B).
How does human WT VMA21 cycle back to the ER? The human VMA21 protein lacks a typical COPI complex (a vesicle coating protein complex comprising seven subunits) recognition signal at the C terminus (KKXX or KXKXX), which is needed to transport proteins back from the cis-Golgi to the ER, but instead carries the sequence R 93 QWREGKQD 101 . We proceeded to test if the sequence KQD was sufficient to serve as an ER retrieval signal and if its deletion could mimic the phenotype of the VMA21p.93X mutant. We generated the VMA21 MUTs p. K99A and p.delKQD(99-101) and analyzed their intracellular localization. Like VMA21 p.93X, both p.K99A and p. delKQD(99-101) MUTs enriched in lysosomes, thus providing novel evidence that the VMA21 C-terminal KQD sequence is functional as an atypical ER retrieval signal ( Figure S2A, C). Both mutants also activated autophagy as detected in the GFP-LC3-RFP reporter cell lines ( Figure  S2B, D).

Results from VMA21 localization studies using lysosomal immunoprecipitation (lyso-IP)
Given the central roles of the lysosome in V-ATPase biology, MTOR and autophagy regulation and its emerging role in MUT VMA21 biology, we employed whole-lysosome immunoprecipitation by following the methods published by Abu-Remaileh et al. [35]. We generated HEK293T cells stably expressing the triple HA-tagged lysosomal membrane protein TMEM192 (3xHA-TMEM192) as a bait together with inducible WT or MUT VMA21. Following induction of VMA21 expression, we immunoprecipitated lysosomes with anti-HAbeads added in limiting amounts.
Next, we measured the amount of selected organelle indicator proteins (CANX [calnexin], ER; GOLGA2/GM130, Golgi; ACTB, cytosol; LAMP1, lysosomes) as well as ATP6V0, ATP6V1 and VMA21 in the post-IP supernatant and in the lysosomal preparations via immunoblotting ( Figure 4C, D). We confirmed that the lyso-IP method enriched for lysosomes and, importantly, that MUT VMA21 (which runs faster than WT on SDS-PAGE gels) was enriched in lysosomes. Of additional interest, the amount of ATP6V0 or ATP6V1 proteins in lysosomes was not altered.
To determine if the Vma21[∆66-77] mutation affects V-ATPase assembly in yeast, we measured vacuolar protein levels of V-ATPase subunits Vma2 and Vma4, which are homologous to human ATP6V1B1 and ATP6V1E1, respectively. Vacuoles were purified from logarithmically grown yeast cells by dextran lysis and flotation, followed by western

An aberrant VMA21-MUT associated proteome defined via immunoprecipitation (IP)-mass spectrometry
We conducted two IP-mass spectrometry experiments to define the VMA21 WT-and MUT-protein associated interactome. Using OCI-LY1 lymphoma cells expressing inducible HA-tagged VMA21 WT or MUT or cells carrying vector alone, we immunoprecipitated out of detergent cell lysates using anti-HA beads and proceeded with quantitative tandem mass spectrometry and peptide identification. Following data normalization incorporating the normalized spectral absorbance factor/NSAF and the relative amounts of immunoprecipitated VMA21 WT or MUT proteins, we detected a greater amount of almost all V-ATPase protein subunits (V 0 + V 1 ) in the VMA21 MUT IPs compared with VMA21 WT ( Figure 5A, B). The only exception was ATP6V0C, the central c subunit of the membrane-spanning V-ATPase V 0 complex, of which slightly less precipitated with MUT VMA21 when compared with WT VMA21. The latter finding is likely a reflection of the VMA21 MUT protein's inability to cycle back to the ER, resulting in reduced binding and assembly of additional V-ATPase V 0 complexes and thus a net decrease in the sum of ATP6V0C bound to VMA21 complexes existing in the ER, Golgi, and post-Golgi compartments.
In addition, the V-ATPase accessory subunits ATP6AP1 and ATP6AP2, which are required for V-ATPase V 0 assembly, also co-precipitated with VMA21 WT and MUT. The proteins ATP6AP1 and ATP6AP2, which are both mutated in FL, are components of the V-ATPase V 0 complexes and are located centrally within the V 0 membrane-spanning ring formed by nine ATP6V0C/V 0 -c subunits ( Figure 5A) [20,21,26].
We also inspected the VMA21 WT-and MUT-associated proteome for other interactions that could shed light on their functions. Importantly, we detected the COPI complex (comprising the subunits COPA, COPB1, COPB2, ARCN1/COPD, COPE, COPG1, COPG2) in the IPs further supporting the involvement of this trafficking complex in VMA21 and V-ATPase biology (please note that only small or no amounts of COPE were recovered and thus are omitted from the display in Figure 5C). Of note, less COPI subunit proteins (five out of six) precipitated with MUT than WT VMA21 supporting prior conclusions of impaired COPI interaction with VMA21 MUT proteins ( Figure 5C), although effects of additional proteins remain possible.
Multiple other proteins with interesting functions coprecipitated with VMA21 and could be the subject for future studies (Tables S1-2). For instance, we detected subunits (DDOST, RPN1, RPN2, DAD1, and STT3A) of the ER resident oligosaccharide transferase complex (OST), most of which precipitated strongly and highly preferentially with the VMA21 MUT protein ( Figure 5D). Given the essential role that OST serves in the protein complex that adds complex high mannose oligosaccharides to nascent ER-resident proteins and the observed glycosylation defects in patients carrying germline mutations in VMA21, ATP6AP1 and ATP6AP2, it is possible that the enhanced VMA21 MUT-OST interaction impairs such glycosylation possibly through aberrant OST sequestration [37].

Mutant VMA21 confers a defect on the V-ATPase holoenzyme proton pump resulting in elevated lysosomal pH
We previously showed that MUT ATP6V1B2 results in a hypomorphic V-ATPase enzyme and higher lysosomal pH. We therefore proceeded with measurements of lysosomal pH in stably transduced HEK293T cells expressing WT or MUT VMA21. Cells were loaded with a pH-sensitive dual reporter dye conjugated to dextran and the cells were subsequently analyzed using a fluorescence-activated cell sorter. As controls, cells in parallel were briefly incubated with buffers at a defined pH of 3.5, 4.5 or 5.5, or were treated with bafilomycin A 1 to completely inhibit V-ATPase function ( Figure 6A). Whereas treatment with bafilomycin A 1 resulted in an elevated lysosomal pH as expected from its ability to inhibit the V-ATPase, we detected an acidic lysosomal pH in the range of 4.2 in VMA21 WT cells. We measured a less acidic lysosomal pH in the range of >4.5-5.0 in the VMA21 MUT cell lines and in cells expressing a VMA21-targeted shRNA but not a scrambled shRNA. Therefore, the VMA21 MUT caused a partial lysosomal acidification defect resulting in a modestly elevated lysosomal pH ( Figure 6A, B).
We proceeded with complimentary measurements of the ability of the V-ATPase to acidify the vacuolar lumen in WT or mutant yeast cells. Pho8 was modified by fusing a pHsensitive super ecliptic pHluorin (SEP) fluorophore protein on its C terminus; the fluorescent pH-biosenser Pho8-SEP displays higher fluorescence intensity at higher pH levels [38]. The fluorescence intensity corresponding to Pho8-SEP was highest in vma21∆ mutant cells, with WT cells displaying the lowest signal ( Figure 6C, D). The Vma21[∆66-77] mutant displayed an intermediate signal indicating an increased vacuolar pH compared with WT cells (Figure 6C, D).

The assembly of V-ATPase holoenzyme on lysosomes is preserved in VMA21 MUT cells but the activity of V-ATPase is reduced
Next, we measured V-ATPase activity in lysosomes isolated via lyso-IP from cells carrying WT or MUT VMA21. We also measured V-ATPase activity is cells treated with the V-ATPase inhibitor bafilomycin A 1 as well as in cells treated with an shRNA targeting VMA21. We found that the activity of V-ATPase in lysosomes of VMA21 MUT cells treated with bafilomycin A 1 was fully inhibited and that targeting VMA21 with an shRNA also substantially lowered V-ATPase activity. The V-ATPase activity in the VMA21 MUT was ~50% of WT providing a mechanistic explanation for the observed higher lysosomal pH ( Figure 6E).
To determine whether the elevated vacuolar pH in Vma21 [∆66-77] mutant yeast cells is caused by compromised V-ATPase activity, we first examined the growth phenotypes of WT and Vma21[∆66-77] mutant cells on plates buffered to pH 7.5. Yeast mutants lacking V-ATPase activity exhibit a distinctive set of growth defects including reduced growth at elevated pH [39]. The Vma21[∆66-77] mutant showed a diminished ability to grow at pH 7.5 relative to the WT, and, in agreement with the intermediate vacuolar pH as assessed with Pho8-SEP, its phenotype was not as severe as that seen with the vma21 null mutant ( Figure 6F).
We also directly measured V-ATPase activity in isolated vacuoles through detecting the hydrolysis of ATP in a bafilomycin A 1 -sensitive manner. Consistent with the growth phenotypes, the Vma21[∆66-77] mutant showed reduced V-ATPase activity equivalent to approximately 41% of the WT strain, whereas the V-ATPase activity dropped to approximately 9% of the WT in vma21 null cells ( Figure 6G).

The aberrant metabolome of VMA21 MUT lysosomes defined via lyso-IP and lysosomal metabolomics
To identify the lysosomal metabolome changes imparted by MUT VMA21 on cells, we measured the relative amounts of small molecules in lysosomes isolated from cells carrying MUT or WT VMA21 or parental cells treated with bafilomycin A 1 . We isolated lysosomes with lyso-IP using limiting amounts of HA-beads, to precipitate equal amounts of intact lysosomes and subjected the lysosomes to untargeted small molecule metabolomics (Tables S3-4).
As an initial QC step, we correlated the relative amounts of metabolites in bafilomycin A 1 -treated cells with published data by Abu-Remaileh et al. [35], using similar conditions and detected a high statistical correlation (R = 0.82). Next, we performed correlation analyses on bafilomycin A 1 -treated samples compared with VMA21 MUT samples and detected a high correlation (R = 0.89) for proteinogenic amino acids confirming a similarity in the lysosomal metabolome of cells with elevated lysosomal pH ( Figure 7A).
Subsequently, we focused on di/tripeptides. We found that a variety of lysosomal di/tripeptides were much reduced in lysosomes isolated from VMA21 MUT cells compared with WT cells, suggesting decreased generation due to impaired proteolysis by pH-dependent hydrolases and peptidases ( Figure 7B). Surprisingly, however, we found that the lysosomal concentrations of proteinogenic amino acids was not uniformly lower in VMA21 MUT lysosomes but instead comprised amino acids that were lower (e.g., histidine, asparagine, lysine, methionine, tryptophan) or higher (e.g., proline, alanine, glutamic acid, serine) than in VMA21 WT cells ( Figure 7C, E). These data strongly suggested that the lysosomal amino acid concentration is determined not only by the rate of protein and peptide hydrolysis but also by pHdependent changes in lysosomal amino acid handling and transport into the cytosol.
We also measured the relative amounts of amino acids in the post IP supernatant fraction and found that most amino acids were modestly lower in MUT than in WT cells ( Figure 7D, E and Tables S5-6). The lowered cytoplasmic amino acid concentration likely activates the signal that increases autophagic flux in VMA21 MUT cells.

Primary human follicular lymphoma B cells carrying mutant VMA21 are sensitive to inhibition of autophagic flux
A recent report described the discovery of the novel potent ULK1 kinase inhibitor MRT68921 [33,40], and we had previously confirmed the ability of MRT68921 to inhibit the phosphorylation of the direct ULK1 substrate ATG13 [18]. Here, we treated primary purified human FL B cells isolated from lymph node biopsies that carried WT or MUT VMA21 with escalating doses of MRT68921 and measured the cell fraction alive after 72 h using CellTiter-glow®-based viability. Given the preciousness of the FL biopsy material we combined the new FL WT data (N = 4; Figure 8A; black curves) with data from ten similarly treated WT FL cases ( Figure 8A; gray curves) previously published, as findings were very similar [18]. We found that FL B cells carrying MUT VMA21 were substantially more sensitive to ULK1-inhibitor-induced cell death than VMA21 WT lymphoma cells which were not sensitive (IC50 values calculated based on the Hill equation of 1 µM and 4.5 µM, respectively) ( Figure 8A-B; red curves. L34 was a bone marrow aspirate containing a mixture of FL B cells and normal B cells). Of additional interest, the two FL cases carrying the VMA21 p.93X mutation were most sensitive to MRT68921 treatment. We also reviewed our published findings on MUT ATP6V1B2 ( Figure 8C; blue curves), which showed that all V-ATPase-impaired FL are sensitive to ULK1 inhibition. Finally, treatment of VMA21 MUT and WT FL B cells with SAR405 (a PIK3C3/VPS34 inhibitor) also showed heightened sensitivity of VMA21 MUT FL ( Figure S4) [41].
In summary, these important findings based on primary human FL B cells uncover a survival addiction to activated autophagic flux in VMA21 MUT FL B cells, providing support for novel therapeutic approaches targeting autophagy in FL.

High-throughput chemical screens identify highly potent inhibitors of autophagy induced by MUT VMA21
We conducted a high-throughput microscopy-based screen of a compound library containing 4,577 clinically approved drugs or tool compounds that could be repurposed, aiming at identification of small molecules that inhibited VMA21-MUT induced autophagy (Table S7). The rationale for these screens were two-fold: i) to further elucidate mechanisms of VMA21-MUT induced autophagy, and ii) to identify pathways and molecular targets suitable for therapeutic development in FL patients. We seeded HEK293T reporter cells carrying GFP-LC3-RFP fusion proteins into 384-well plates and left them untreated or induced VMA21 MUT protein with doxycycline and treated cells with experimental compounds at 4 µM. As controls we chose the ULK1 inhibitor MRT68921 and the PIK3C3/VPS34 inhibitor SAR405. To activate autophagy, we used torin 1.
The accumulation of GFP-stained green puncta following autophagy activation was quantitatively scored via highthroughput microscopy ( Figure 9A-B) and the percent inhibition of green puncta formation by compounds calculated from normalization against untreated cells (Tables S8-9). As can be seen in Figure 9C, the ULK1 inhibitor MRT68921 (black dots) resulted in ~90% inhibition across many replicates, while untreated cells (in blue) showed no inhibition. Many compounds (gray) were inactive while some in fact stimulated autophagy. We selected 112 primary screens hits (red), which demonstrated autophagy inhibition in the primary screen between 50% to > 95%, for triplicate assay repeats, and based on these results selected 32 compounds for further study ( Figure 9B).
The review of these autophagy inhibitory compounds and their molecular targets gave interesting results. Multiple autophagy inhibitory drugs belonged to the class of cyclin- dependent kinase inhibitors, which hitherto had not received dedicated attention as autophagy inhibitors [42]. A subsequent enrichment analysis comparing the fraction of autophagy inhibitory compounds within compound classes (e.g., CDK inhibitors or PIK3C3/VPS34 inhibitors) uncovered strong enrichment for CDK targets, autophagy components, phosphoinositide 3-kinase, and a few other pathways ( Figure 9D).
Next, we derived dose-response curves for each of the 32 compounds ( Figure 9E and Figure S3). In these assays, MRT68921 displayed an IC50 of 428 nM and SAR405 of 104 nM. However, neither one of these compounds is in active clinical development. The clinical grade multi-targeted CDK inhibitor dinaciclib potently inhibited VMA21-MUT induced autophagy at an IC50 of 16 nM. We identified multiple other CDK inhibitors with potent autophagy-inhibiting properties including SNS-032, AT7519, AT9283, PHA-793887, AZD5438, R547, PHA-767491, abemaciclib, and LY2857785 ( Figure 9E and Figure S3). These compounds variably target CDKs 1, 2, 3, 5, 7, 8, and 9 with most of them targeting a CDK core set including CDK2, CDK7 and CDK9. Additional CDKs of interest that are targeted by these compounds are CDK16, CDK17 and CDK18 (PCTAIRE kinase family), which have recently been implicated in autophagy regulation [43]. The clinically used drug ribociclib, targeting CDK4 and CDK6 was not active and the specific CDK7 inhibitor THZ1 was also only modestly active. Furthermore, RO-3306 with relative specificity for CDK1 and CDK2 was inactive. Together, these results narrow the CDK profile of interest to CDK9 and CDK16 to CDK18 but relevant targeting of other CDKs or non-CDKs remains possible.
Multiple other compounds with attractive properties, including the clinically used drugs flavopiridol (a pan CDK inhibitor; IC50 = 242 nM) and homoharringtonine (a ribosome and protein synthesis inhibitor; IC50 = 20 nM) approved for CML treatment were identified and confirmed ( Figure 9E).
Next, we treated primary purified human FL WT or MUT for VMA21 as above with dinaciclib, flavopiridol or homoharringtonine ( Figure S4). We found that all three drugs killed FL WT or MUT for VMA21 at low concentrations and thus could be considered for future FL clinical trials.

Discussion
In this study, we report on the identification of common mutations in VMA21, a small ER-resident chaperone protein for the vacuolar-type H + -translocating ATPase (V-ATPase) in 12% of follicular lymphoma (FL). The subsequent in-depth functional characterization uncovered that the FL-associated VMA21 hotspot nonsense mutation p.R93X, comprising 40% of all VMA21 mutations as well as p.88X and p.95X result in the deletion of a C-terminal non-canonical ER retrieval sequence, which causes mutant VMA21 protein to mislocalize to lysosomes. Consequently, there is a reduction in total lysosomal V-ATPase activity and a partial inability to fully acidify the lysosomes. The pathological elevation in lysosomal pH caused a defect in the degradation of lysosomal cargo proteins as measured through lysosomal metabolomics and a reduction in the concentration of cytosolic amino acids. Overall, this aberrant lysosomal phenotype associated with a substantial compensatory increase in autophagic flux, which we here demonstrate constitutes an Achilles heel survival dependency for FL. Targeting the apical autophagy regulating ULK1 kinase complex with a small molecule inhibitor resulted in the death of VMA21 MUT FL B cells, while VMA21 WT cells were largely resistant to such treatment. Therefore, we are identifying the compensatory activation of autophagic flux as a survival dependency and its inhibition as a novel therapeutic opportunity in FL with reduced V-ATPase function.
FL harbors mutations in ATP6V1B2, VMA21, ATP6AP1 and ATP6AP2, and sporadic mutations in other V-ATPase proteins, which together affect one third of all FL. As such, the V-ATPase emerges as a commonly targeted enzyme in FL. In prior work, we have shown that mutations in the V-ATPase V 1 subunit ATP6V1B2 create a catalytically hypomorphic V-ATPase that results in elevated lysosomal pH and activated autophagy and a dependency for survival. The work presented here unifies the phenotypes of FL with mutations in ATP6V1B2 and VMA21 and converges on a novel pathobiological mechanism driving FL. Last, how mutated ATP6AP1 and ATP6AP2 affect these processes and FL biology is unknown and awaits ongoing detailed study.
The recently reported cryo-EM structures of V-ATPase holoenzymes purified from brain and other sources placed ATP6AP1 and ATP6AP2 proteins centrally within the membrane-spanning ring-like structures formed by nine V-ATPase ATP6V0C subunits and both proteins were identified as central components of the mature V-ATPase. Of interest, the VMA21 protein was absent in all reported structures implying that it interacts only transiently with the V-ATPase holoenzyme or was lost during enzyme purification. In the VMA21 IP-mass spec experiments reported here we detected the entire V-ATPase holoenzyme, including V 0 , V 1 and ATP6AP1-ATP6AP2 in both VMA21 WT and MUT immunoprecipitates but measured increased amounts of coimmunoprecipitating proteins with the mutants. The increased co-immunoprecipitation of MUT VMA21 and V-ATPase holoenzyme together with results from confocal microscopy studies supports the idea that MUT VMA21 remains aberrantly associated with V-ATPase in the lysosomes. This aberrant association results in only a partially functional or possibly immature V-ATPase enzyme or in the direct inhibition of V-ATPase activity by aberrantly associated MUT VMA21 protein.
We previously demonstrated that purified human primary FL B cells carrying MUT ATP6V1B2 acquire the ability to survive in lowered or no leucine-containing media. Considering the lowered cytoplasmic leucine and amino acid concentration that we measured in VMA21 MUT cells together with these findings seem paradoxical at first. Leucine signals to MTORC1 via different mechanisms including its metabolite acetyl-CoA [44,45], the latter being generated via multiple metabolic pathways. One possible explanation relies on metabolic adaptation by MUT FL B cells allowing these cells to preferentially use alternative fuel sources to generate acetyl-CoA and, in this context, it is relevant that we found that dichloroacetate/DCA, known to elevate intracellular acetyl-CoA levels, abolishes aberrant activation of autophagy in VMA21 MUT cells [46].
In addition to resistance to nutrient starvation as likely exists in pathologically enlarged lymph nodes, why do FL frequently select for mutations that cause lysosomal dysfunction and compensatory elevations in autophagy flux? Considering the important role of the lysosome in MHC peptide loading and B cell antigen presentation, is the elevated lysosomal pH permitting immunological escape from T cells? Is the elevated autophagy needed to overcome impairments to baseline autophagy as part of the high BCL2 protein level in FL and the effects on the BCL2-BECN1-PIK3C3/VPS34 complexes [47], and could BCL2 inhibitors be used in combination with autophagy inhibitors as therapy for FL?
To exploit novel therapeutic opportunities derived from the survival addiction of VMA21 MUT FL on intact autophagic flux, we conducted a screen for compounds that could inhibit MUT-VMA21 induced autophagy. Through these efforts we identified several cyclin-dependent kinase inhibitors (e.g., dinaciclib, flavopiridol, and multiple others) that strongly abolished autophagy activation and demonstrated their ability to kill purified FL cells at low concentrations. Of course, these drugs have other known effects, which would need to be considered when planning further study and possible development as FL therapeutics. These results also motivate future studies into the role of specific CDKs in autophagy regulation in lymphoma, and recent findings of involvement of CDK16 in autophagy activation could serve as a mechanistic lead [43].
In summary, our studies elucidate novel mechanisms of action of FL-associated mutations in chaperones and components of the V-ATPase and provide a framework for future biological or therapeutic studies of hypomorphic or hypofunctional V-ATPase in FL. Our findings open up novel avenues for the treatment of FL, possibly through intermittent inhibition of autophagy [41,48], and we identify multiple hitherto unsuspected compounds as attractive in this setting [33,40,49].

Patient characteristics and study approvals
Results from the Sanger-based resequencing of exons 1-3 and adjacent intronic sequences of VMA21 using template genomic DNA isolated from 151 flow-sorted FL and 19 transformed FL (t-FL) are reported here [50]. Lymphoma patients were enrolled into two lymphoma repositories at the University of Michigan Comprehensive Cancer Center (IRBMED #HUM00007985 and IRBMED #HUM00017055) after signing informed consent documents approved by the University of Michigan Institutional Review Board (IRBMED). All investigations were performed in accordance with ethical guidelines outlined in the declaration of Helsinki. Genomic research on all specimens was approved through IRBMED #HUM00005467. We furthermore analyzed by Sanger sequencing DNA from 49 flow-sorted FL cases collected between 1990 and 2005 as previously described that were derived from de-identified leftover clinical material [51].

Exon resequencing of VMA21 in FL
Primers to amplify and sequence VMA21 were designed using the primer 3 program (https://bioinfo.ut.ee/primer3-0.4.0/) and sequences generated using direct sequencing. The sequence traces were analyzed using Mutation Surveyor® software (Softgenetics®) and by visual inspection. Mutations were confirmed to be somatically acquired using unamplified lymphomacell derived DNA and paired CD3-cell derived DNA isolated from flow-sorted LN or blood T cells or from buccal DNA.

VMA21 cDNA mutagenesis and lentiviral vector generation, cell transfection and cell transduction to generate stable cell lines
Cell lines: The OCI-LY1 lymphoma cell line was obtained from the Ontario Cancer Institute, Toronto, Canada. Cell line authentication was performed through sequence analysis of known gene mutations. The HEK293T cell line was purchased from Takara (632180).

Generation of stable doxycycline-inducible lymphoma cell and 293T cell lines
The methods followed our previously published detailed protocols [18] and cell pools were used for experiments described herein.

Measurement of autophagic flux using electron microscopy
Stable HEK293T cell lines overexpressing WT or MUT HA-VMA21 were induced with doxycycline as above. Cells were then fixed with 2% glutaraldehyde (EM grade) in 0.2 M HEPES, pH 7.4 at room temperature for 2 h. During fixation, the cells were scraped off the culture dish and pelleted. The fixed cell pellets were kept at +4°C until sample preparation for transmission electron microscopy. Images were taken at 3000 x primary magnification, using the principle of uniform random sampling, and analyzed using ImageJ. Each image was zoomed in on the screen for counting of profiles. The cytoplasmic area was estimated using point counting. The number of autophagosomes, autolysosomes and endolysosomes was counted for ~70 to 90 images from each group.

Generation of GFP-LC3-RFP reporter cell lines
HEK293T cells were transfected with pMRX-IP-GFP-LC3-RFP (Addgene, 84573; deposited by Dr. Noboru Mizushima) and packaging plasmids and virus-containing supernatant media collected 48 h after transfection. The virus-containing media were used to transduce HEK293T cells and OCI-LY1 lymphoma cells. After viral transduction, cells were treated with 2 µg/ml of puromycin for 2 days. After 3-5 days of culture in full medium, cells were than infected with pCW57blast lentiviruses carrying WT or MUT HA-VMA21 cDNAs followed by selection with 5 µg/ml blasticidin S (Cayman Chemical, 14499). The expression of GFP-LC3-RFP was verified by fluorescence microscopy and immunoblotting and the inducible expression of VMA21 was verified by immunoblotting.

Measurement of GFP-positive autophagosomes and autolysosomes
Doxycycline-inducible stable 293T expressing GFP-LC3-RFP and VMA21 WT or MUT cell lines were cultured on chambered cover glasses and induced with doxycycline as above. Cell Images were taken using a Leica SP5 confocal imaging system.

easurement of free GFP
Doxycycline-inducible stable 293T cells or OCI-LY1 cells expressing GFP-LC3-RFP and VMA21 WT or MUT were cultured in DMEM medium supplemented with 10% FBS and induced with doxycycline as above. Cells were harvested and prepared for immunoblotting. Band intensities were analyzed by densitometry (ImageJ, version 2.0.0, NIH).

Intracellular localization of WT and MUT HA-VMA21 in HEK293T cells using confocal microscopy
Cells were plated on pre-sterilized chambered cover glasses. The expression of HA-VMA21 WT and MUT was induced with doxycycline as above. Cells were fixed and permeabilized with 4% paraformaldehyde for 10 min at room temperature (RT) and subsequently in 0.1% Triton X-100 (Sigma, T8532) for 5 min at 4°C. Samples were blocked with 4% bovine serum albumin (BSA; Dot Scientific, DSA30075-100) in phosphatebuffered saline (PBS; Lonza, 17-516 F) for 1 h at RT. Staining was performed using primary antibodies (Table 1) diluted in 4% BSA in PBS. Species-specific conjugated secondary antibodies were diluted 1:1000 in 4% BSA in PBS and applied for 1 h at RT. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI; Molecular Probes/Fisher, D1306). The coverslips were mounted onto glass slides with ProLong® Gold Antifade Mountant reagent (Thermo Fisher, P36930). Images were taken using a Leica SP5 confocal imaging system. Pearson's correlation coefficient analysis was used to determine colocalization and was performed using ImageJ v2.0.0 software.

Cell line generation and lysosome isolation
The lysosome isolation was done using the rapid Lyso-IP method developed by Abu-Remaileh et al, 2017 [35]. Briefly, HEK293T cells stably expressing the lysosomal bait protein 3xHA-TMEM192 were generated using the lentiviral plasmid pLJC5-Tmem192-3xHA (Addgene, 102930; deposited by Dr. David Sabatini) as above. The control HEK293T cells were generated expressing 2xFlag-TMEM192 using pLJC5-Tmem192-2xFlag (Addgene, 102929; deposited by Dr. David Sabatini). After puromycin selection the HEK293T cells were transduced with pCW57(blast) lentiviruses carrying cDNA encoding non-tagged WT or MUT VMA21, respectively, followed by selection using 5 µg/ml blasticidin S. To perform isolation of intact lysosomes via immunoprecipitation (Lyso-IP), cells were lysed using a Dounce homogenizer and the lysates were centrifuged at 1000 x g for 2 min at 4°C. The supernatant containing the cellular organelles including lysosomes was subsequently immunoprecipitated using anti-HA-conjugated magnetic beads (Thermo Scientific, 88836), followed by three quick washings using ice cold Tris-buffered saline (TBS; Bio-Rad, 1706435). After immunopurification, the bead-bound lysosomes were snap frozen in liquid nitrogen.

Measurement of V-ATPase components in lysosomes
The 3xHA-TMEM192-expressing 293T cells carrying inducible VMA21 WT or MUT were induced with doxycycline as above. Lysosomes were isolated from 1.5 × 10 7 cells and using a limiting amount (25 µL) of anti-HA beads and the beadbound lysosomes subsequently lysed using standard lysis buffer for immunoblotting as described above. In parallel, 10 6 cells were harvested and lysed directly using the same buffer to obtain whole-cell lysates. Immunoblottings for various V-ATPase components were performed on protein from lysosomal and whole cell lysates.

Untargeted lysosomal and cytoplasmic metabolomics assay
The 3xHA-TMEM192-expressing 293T cells (10 8 cells/construct and condition for lysosomal metabolomics and 10 7 cells for cytoplasmic metabolomics) carrying inducible VMA21 WT or MUT were induced with doxycycline as above. Lysosomes were isolated using anti-HA beads. For lysosomal metabolomics, the IP-enriched lysosomes and the post-IP supernatants were kept and snap frozen with liquid nitrogen and subsequently processed in the Metabolomics Core at the University of Michigan. Small molecules were extracted, and untargeted metabolomics profiling was conducted using an HPLC/Q-TOF Mass Spec system. Manual peak integration was combined with automated peak calling to identify metabolites. Data reduction was performed with the software Binner (https://doi.org/10.1093/ bioinformatics/ btz798). The accurate mass and retention time was then used for the database search against RefMet (MetabolomicsWorkbench.org). Features of interest from the database search were individually confirmed by tandem mass spectrometry (MS/MS) analysis.

Immunoprecipitation followed by quantitative tandem mass spectrometry
A total of 5 × 10 7 OCI-LY1 cells inducibly overexpressing HA-VMA21 WT or MUT were induced with doxycycline as above. Cells were pelleted, washed, and lysed on ice for 20 min in modified lysis buffer containing 0.2% NP-40 Surfact-Amps Detergent Solution (Thermo Fisher Scientific, 85124) as above. The detergent-soluble fraction of the cell lysates was obtained by centrifugation at 20,817 x g for 10 min. A total of 200 μL of protein A agarose beads (Cell Signaling Technology, 9863) were blocked with 20 μL total mouse IgG-UNLB (Southern Biotech, 0107-01) for 2 h. The blocked protein A agarose beads were washed and resuspended in lysis buffer. To preclear the protein samples, 50 μL conjugated protein A agarose beads were added to the protein lysate for 2 h at 4°C. Next, the anti-HA-loaded beads (Sigma-Aldrich, A2095) were blocked with 5% BSA in TBS, 0.05% Tween-20 (Sigma, P7949) TBST for 2 h, followed by washing with lysis buffer. A total of 40 μL of anti-HAconjugated beads were added to precleared cell lysates and incubated with rotation for 5 h at 4°C. The beads were washed four times with lysis buffer containing 150 mmol/L NaCl and pelleted by centrifugation at 2,000 × g for 5 min.
The methods for protein SDS-PAGE and mass spectrometry using nano LC/MS-MS with a Waters M-Class highperformance liquid chromatography (HPLC) system interfaced to a Thermo Fisher Fusion Lumos mass spectrometer and data analysis using Mascot (Matrix Science) were as recently published [52].

Measurement of lysosomal pH in HEK293T cells expressing WT or MUT HA-VMA21
HEK293T cells inducibly expressing WT or MUT HA-VMA21 were grown in 24-well plates and induced with doxycycline as above. A knockdown using lentiviruses carrying VMA21 shRNA (Millipore Sigma, TRCN0000122433) or scrambled shRNA was performed in parallel. LysoSensor Yellow/Blue dextran (Thermo Fisher Scientific, L22460) was added at 500 μg/mL and cells incubated for 16 h, followed by a 1-h chase in dextran/dye-free medium. The methods otherwise follow our published protocols [18].

Measurements of lysosomal V-ATPase activity
HEK293T cells stably expressing TMEM192-3xHA or TMEM192-2xFlag and carrying inducible non-tagged VMA21 WT or MUT were induced with doxycycline as above. Lysosomes were subsequently isolated using Lyso-IP as described above. The V-ATPase activity was measured in the isolated lysosomes using the ATPase/GTPase Activity Assay Kit (Sigma-Aldrich, MAK113), and 4 mM ATP (Sigma-Aldrich, A7699). The absorbance was read at 620 nM using a plate reader. Cells treated with 100 nM bafilomycin A 1 for 2 h were used as positive control for complete inhibition of V-ATPase.

High-throughput screening for compounds inhibiting MUT-VMA21 induced autophagy
The autophagy reporter 293T cells expressing GFP-LC3-RFP and inducible VMA21 WT or MUT were plated into black 384well tissue culture plates (PerkinElmer, 6005182) at a density of 5000 cells/well using a multidrop device (Thermo Fisher Scientific, Waltham, MA). Phenol-red free DMEM (Gibco, 21063029) was used for cell culture. Doxycycline was added at a concentration of 1 µg/ml. The compound library used for drug screening (Drug Repurposing Set) was created by the Center for Chemical Genomics at the University of Michigan. The Drug Repurposing Set is a collection of FDA-approved compounds and selected clinical drug candidates (Table S7). The compounds were added using a liquid handler (Sciclone Advanced Liquid Handling 3000 V&P pintool) in 0.2-µL volume (final DMSO concentration = 0.4%). All compounds were tested as singletons at 4 µM. The ULK1 inhibitor MRT68921 (Cayman Chemical, 1190379) at 1 µM was added to one column of 16 wells on each plate as a control for autophagy inhibition (internal plate and assay control). Cells without doxycycline treatment (uninduced cells) were also plated into one column on each plate as an assay baseline control. Following the addition of doxycycline and compounds the plates were incubated for 24 h at 37°C in 5% CO 2 . The 384wells plates were than imaged and analyzed using an ImageXpress Micro Widefield High Content Screening System at the Center for Chemical Genomics, Life Sciences Institute, University of Michigan. Next, the number of GFPpositive puncta per cell were quantified. Cells with more than or equal to five GFP-positive puncta were categorized as autophagy-activated cells and the total number of these cells were divided by the total number of diffusely GFP + cells in the same high-power image. A hit was defined as a compound able to inhibit the number of MUT-VMA21 associated autophagyactivated cells to a level ≤50% of the doxycycline-induced VMA21-MUT expression control group. The average inhibition of MRT68921-positive controls was 86% with a Z' value > 0.4. Hits from the primary screen and selected compounds targeting CDKs were further tested in secondary screenings at 4 µM in triplicates. Compounds validated in the secondary confirmatory screen were individually subjected to doseresponse assays performed in duplicate at concentrations from 8.67 nM to 10 µM.

Drug treatment and cell viability assays of purified FL B cells
Purified FL B cells were plated in RPMI 1640 medium (Hyclone, SH30027.01) supplemented with 20% heat inactivated FBS at a density of 10 6 cells/ml and cultured at 37°C with 5% CO 2 . Torin 1 (10997) and SAR405 (16979) were procured from Cayman Chemical. Apilimod (SML2974) was procured from Millipore Sigma. Flavopiridol (S1230), pitavastatin (S1759), homoharringtonine (S9015) and dinaciclib (S2768) were purchased from Selleck Chemicals. Stock solutions of 100 mM were made with DSMO and were diluted serially in medium. DMSOonly was used as vehicle control. After 72 h of incubation, cell viability was measured with Celltiter-Glo® (Promega, G7572) and quantified on a microplate luminescence reader.

Strains, media, and growth conditions
Yeast strains (SEY6210 and WLY176) were used to generate the Vma21[∆66-77] mutation and vma21∆ in the genome using standard methods [53,54]. Cells were cultured in rich medium (YPD; 1% [w:v] yeast extract, 2% [w:v] peptone], and 2% [w:v] glucose) as nutrient-rich conditions. Autophagy was induced through nitrogen starvation by shifting cells in mid-log phase from YPD to SD-N (0.17% yeast nitrogen base without ammonium sulfate or amino acids [ForMedium, CYN0501], and 2% [w:v] glucose) for the indicated times, referred to as starvation conditions. For growth assays, cells were grown in YPD, pH 5 to mid-log phase and then serially diluted and transferred to YPD plates buffered to pH 5 or pH 7.5. Growth phenotypes were then determined after 36 h at 30°C incubation.

Vacuole assays
Vacuole purification was performed as previously described [58]. Vacuole lysates were prepared and subjected to SDS-PAGE. Vma2 and Vma4 were detected by western blot. Antisera to Vma2 [59] was used at 1:5,000 dilution, and anti-Vma4 [60] was used at 1:3,000 dilution. V-ATPase activity (ATP hydrolysis rate) was quantified colorimetrically as the phosphate released based on a previous method [61]. Bafilomycin A 1 inhibition was measured by adding bafilomycin A 1 directly to the assay mixture to a final concentration of 100 nM. Specific V-ATPase activity represents the rate of bafilomycin A 1 -sensitive ATP hydrolysis.

Statistical Methods
Statistical comparisons between two conditions were carried out using unpaired two-sample t-tests or the Mann-Whitney U-test (where indicated). Bonferroni adjustments were used when a positive result from any one of several tests was used to support a single hypothesis [62]. For analysis of metabolism data, unpaired two-sample t-tests and false discovery rate correction was calculated using Scipy and Statsmodels Python packages [63,64]. Volcano plots were drawn using the Matplotlib Python package [65]. For concentrationresponse curves in chemical screening experiments and FL kill assays, sigmoidal dose-response models with variable slope for inhibition (based on the Hill equation) were used to estimate the IC50 using commercial software (GraphPad Prism). All bar plots, line plots and heatmaps were draw using GraphPad Prism.