A new CRISPR screening approach for identifying novel autophagy-related factors and cytoplasm-to-lysosome trafficking routes

Selective autophagy comprises cytoplasm-to-lysosome trafficking routes that transport cargos using double-membrane vesicles (autophagosomes). Cargos are detected by receptor proteins, which typically also bind to lipid-conjugated LC3 proteins on autophagosome membranes. We dissected lysosomal delivery of four SQSTM1-like receptors by genome-wide CRISPR screening looking for novel autophagy-related (ATG) factors and trafficking routes. We uncovered new mammalian ATG factors including TMEM41B, an endoplasmic reticulum membrane protein required for autophagosome membrane expansion and/or closure. Furthermore, we found that certain receptors remain robustly targeted to the lysosome even in the absence of ATG7 or other LC3 conjugation factors. Lastly, we identified a unique genetic fingerprint behind receptor flux in ATG7KO cells, which includes factors implicated in nucleating autophagosome formation and vesicle trafficking factors not previously linked with autophagy. Our work uncovers new ATG factors, reveals a malleable network of autophagy receptor genetic interactions, and provides a valuable resource (http://35.196.197.20:3838/cscrispr/) for further mining of novel autophagy mechanisms.


Summary:
Selective autophagy comprises cytoplasm-to-lysosome trafficking routes that transport cargos using double-membrane vesicles (autophagosomes). Cargos are detected by receptor proteins, which typically also bind to lipid-conjugated LC3 proteins on autophagosome membranes. We dissected lysosomal delivery of four SQSTM1-like receptors by genome-wide CRISPR screening looking for novel autophagy-related (ATG) factors and trafficking routes. We uncovered new mammalian ATG factors including TMEM41B, an endoplasmic reticulum membrane protein required for autophagosome membrane expansion and/or closure. Furthermore, we found that certain receptors remain robustly targeted to the lysosome even in the absence of ATG7 or other LC3 conjugation factors. Lastly, we identified a unique genetic fingerprint behind receptor flux in ATG7 KO cells, which includes factors implicated in nucleating autophagosome formation and vesicle trafficking factors not previously linked with autophagy.

Introduction:
Macroautophagy (hereafter autophagy) is an umbrella term for a series of vesicle formation and transport mechanisms that deliver cytoplasmic material to the lysosome for degradation (Bento et al., 2016;Mizushima et al., 2011;Nakatogawa et al., 2009). During bulk autophagy, regions of cytoplasm are nonspecifically encapsulated into de novo-generated, double-membrane vesicles.
These vesicles (autophagosomes) are then trafficked to lysosomes where, upon autophagosomelysosome fusion, the inner autophagosomal membrane and its contents are degraded by lysosomal hydrolases.
There are approximately 40 autophagy-related (ATG) factors that operate at distinct stages of the autophagosome biogenesis process (Mizushima et al., 2011). Nucleation of the cup-shaped, autophagosome precursor membrane (phagophore) is thought to be mediated by the RB1CC1/ATG101/ATG13 scaffolding complex and post-Golgi vesicles containing ATG9A (Itakura and Mizushima, 2010;Kishi-Itakura et al., 2014;Koyama-Honda et al., 2013). This is followed in short succession by phagophore membrane recruitment of a phosphoinositide-3 kinase (PI3K) complex and phosphatidylinositol-3 phosphate effectors such as WIPI2 (Dooley et al., 2014). Subsequent expansion of the phagophore membrane and vesicle closure is associated with lipid conjugation of LC3-family proteins (Atg8 in yeast) via a ubiquitin-like conjugation cascade dependent on ATG7 (E1-like), ATG3 (E2-like) and ATG5/ATG12/ATG16L1 (E3-like) (Geng and Klionsky, 2008;Ichimura et al., 2000). In addition to their role in bulk autophagy, these ATG factors work together with a diverse group of proteins called autophagy receptors to enable capture of specific cytoplasmic targets, such as damaged organelles, intracellular pathogens, and protein aggregates, during autophagosome formation (Zaffagnini and Martens, 2016). In the majority of known selective autophagy cases, receptor-bound targets become tethered to growing phagophores via receptor-mediated interactions with lipidated LC3 (Stolz et al., 2014).
Defining the precise role of lipidated LC3 during either bulk or selective autophagosome formation has been historically challenging for several reasons. First, the six mammalian LC3family proteins have similar sequences and perform related functions. A recent study overcame this issue of LC3 redundancy as it pertains to Parkin-dependent mitophagy by generating a sixtuple LC3 KO (LC3 NULL ) cell line (Nguyen et al., 2016). Envelopment of receptor-bound mitochondria by the phagophore membrane persisted in LC3 NULL cells but the resulting autophagosomes were defective for fusion with the lysosome. Second, there has been a shortage of tools for visualizing autophagosome formation without relying on microscopic reporters of LC3 conjugation. To get around this problem, another recent work monitored bulk autophagy in cells lacking LC3 conjugation factors using localization of Syntaxin 17 (STX17), a SNARE protein that is recruited to autophagosomes following vesicle closure (Itakura et al., 2012). Autophagosome-like structures (LC3 negative, Syntaxin 17 positive) formed and fused with the lysosome but failed to undergo inner autophagosomal membrane degradation, potentially because of a vesicle closure defect (Tsuboyama et al., 2016).
There are also cytoplasm-to-lysosome pathways that appear to operate normally in the absence of LC3 conjugation while still relying on other ATG factors to varying degrees. One pioneering study found conditions during which bulk protein degradation was mediated by doublemembrane autophagosomes formed by an ATG5/7-independent mechanism (Nishida et al., 2009).
Yet another alternative route to the lysosome was revealed in a very recent study of how the NCOA4 autophagy receptor enables iron starvation to induce degradation of ferritin (a cytoplasmic, iron-binding protein aggregate) (Goodwin et al., 2017). With the notable exception of ATG9, ferritin destruction persisted in the absence of virtually all other known ATG factors, including RB1CC1 (alternatively named FIP200), a scaffold protein essential for normal phagophore nucleation (Goodwin et al., 2017). In sum, these discoveries are beginning to re-shape the monolithic view of how ATG factors and autophagy receptors function into a more complex web of overlapping pathways for robust cytoplasm-to-lysosome delivery.
Here, we further explored the possible existence of new ATG factors and alternative routes to the lysosome by focusing on SQSTM1 and SQSTM1-like family of autophagy receptors (NDP52, TAX1BP1, and NBR1) (Birgisdottir et al., 2013). Inspired by the pHsensitive, tandem-fluorescent reporter for monitoring LC3 lysosomal delivery (tfLC3) (Kimura et al., 2007;Pankiv et al., 2007), we used tfReceptors as phenotypic reporters for genome-wide CRISPR screening. This approach recovered virtually all known ATG factors and identified several novel candidates as screen hits. Further follow-up by quantitative microscopy and biochemical analysis revealed that TMEM41B is a previously uncharacterized endoplasmic reticulum (ER) membrane protein with a critical role in phagophore expansion and/or vesicle closure. In addition, we found that lysosomal targeting of NBR1 (and to lesser extent SQSTM1 and TAX1BP1) persisted in cells lacking LC3 conjugation factors. A subsequent screen performed using ATG7 KO tfNBR1 cells enabled us to compare ATG7-dependent and ATG7independent genetic modifiers across the entire genome. Notably, all forms of NBR1 lysosomal delivery were absolutely dependent on ATG factors implicated in phagophore nucleation (RB1CC1, ATG101, and ATG9A) but ATG7-independent delivery had a unique requirement for additional factors with known roles in alternative forms of vesicle trafficking.

RESULTS:
Validation of tfReceptors as phenotypic reporters of autophagy SQSTM1 and members of the SQSTM1-like family of proteins are among the most widely studied autophagy receptors in mammalian cells (Zaffagnini and Martens, 2016). We constructed six homologous gene cassettes encoding the tandem-fluorescent (tf) protein RFP-GFP by itself (tfEmpty) or as an N-terminal fusion to either LC3 (isoform B) or SQSTM1 and three SQSTM1-like family members: NBR1, TAX1BP1, and NDP52 ( Fig S1A). Following cassette integration at the AAVS1 locus, we measured tf flux to the lysosome as the ratio of RFP fluorescence to GFP fluorescence, the latter of which is selectively quenched when exposed to the low pH of the lysosomal lumen ( Fig 1A) (Kimura et al., 2007;Pankiv et al., 2007). As expected, inducing autophagosome formation by inhibiting mTOR with torin preferentially increased lysosomal delivery of tfSQSTM1 and tfLC3 over tfEmpty in HEK293T cells ( Fig 1B).
Notably, we confirmed that the torin-induced increase in the Red:Green ratio was due specifically to a decrease in GFP fluorescence as expected for this reporter system ( Fig S1B).
Finally, we validated that both basal and torin-induced lysosomal targeting of tfSQSTM1 was dependent on two phagophore nucleation factors, RB1CC1 and ATG13 ( Fig 1C, Fig S1C).

Genome-wide screens for tfReceptor modifiers detect known ATG factors to saturation
To get an unbiased view of autophagy pathways, we expressed our tf proteins in K562 cells (a myelogenous leukemia line), which can be cultured in suspension to facilitate genetic screening. Cells expressing either tfLC3 or tfEmpty served as receptor-agnostic controls.
Bafilomycin A1 (BafA1), a known inhibitor of autophagosome-lysosome fusion, blocked basal targeting of tfLC3 and tfReceptors to the lysosome (Fig 2A). Conversely, lysosomal targeting of tfLC3 and tfReceptors was enhanced following treatment with torin (Fig 2A). Notably, tfEmpty did not respond to these treatments, implying that nonselective engulfment of tfReceptors into autophagosomes is negligible under these conditions (Fig 2A).
Next, we performed genome-wide, pooled CRISPR knock-out screens in K562 cells coexpressing Cas9 with either tfLC3 or tfReceptors (Fig 2B). Following transduction with a library of lentivirus-encoded single guide RNAs (sgRNA; 19,114 genes), we used FACS to collect the top and bottom ~30% of cells ranked on the basis of their Red:Green ratios. Read counts of sgRNAs derived from each cell fraction were obtained by an established Illumina sequencing approach before being processed and analyzed using MAGeCK (Li et al., 2014b(Li et al., , 2015 (Table   S1). The resulting output for each gene includes a beta score, a proxy for its enrichment as a phenotypic suppressor or enhancer ( Fig 2C, Table S2). An interactive interface for these data is also available online at http: //35.196.197.20:3838/cs_crispr/ . Inspection of deletions that decrease tfLC3 lysosomal targeting revealed that we successfully recovered most known autophagy-related factors, thereby validating our experimental setup (Fig 2C). Comparing genetic modifiers across tfLC3 and tfReceptors revealed a strong phenotypic overlap comprising most well-established ATG factors, as well as positive and negative regulators of mTORC1. In addition, our analysis distinguished between related protein complex pairs based on their subunit ablation phenotypes. For example, in the case of the homotypic fusion and vacuole protein sorting (HOPS) and class C core vacuole/endosome tethering (CORVET) tethering complexes, we found evidence that only HOPS was required for autophagy. Similarly, we found that the BLOC-one-related complex (BORC) facilitated autophagy while the biogenesis of lysosome-related organelles complex (BLOC-1) was dispensable. Lastly, our analysis identified factors whose precise functions in autophagy have not been firmly defined (e.g. VPS37A, C18orf8), as well those that are completely uncharacterized (e.g. TMEM41B).
To validate the top screen hits, we transduced our panel of K562 cell lines with individual sgRNAs and measured phenotypes of single gene knockouts (Fig 3A,B, Fig S3A). This approach allowed us to eliminate rare screen hits (e.g. sgUROD) that were spurious because they modified the Red:Green ratio by changing RFP rather than GFP fluorescence ( Fig S3B). The vast majority of tested sgRNAs confirmed that they were true screen hits whose phenotypic strengths as single gene deletions largely mirrored their relative beta score rankings (Fig 3B and 2C). Lastly, we controlled for potential tf-specific artifacts by detecting accumulation of endogenous LC3 and SQSTM1 in cells ablated for certain novel factors among the screen hits ( Fig S3C). In sum, our genetic screening strategy robustly identified known ATG factors and enabled us to pursue new factors whose important roles in autophagy might have been overlooked.

TMEM41B is a novel, ER membrane protein required for phagophore maturation
Despite being a complex membrane-mediated process, there is only a handful of integral membrane proteins known to be directly involved in autophagosome formation. For this reason, we focused on TMEM41B, a largely uncharacterized factor predicted to encode a multi-pass membrane protein with a C-terminal ER retention signal (Jackson et al., 1990). A global, subcellular-localization approach previously assigned TMEM41B's residence to the ER (Itzhak et al., 2016). To confirm this finding and also to facilitate subsequent microscopy analysis of autophagosome formation in TMEM41B KO cells, we turned to adherent HEK293T cells. Using a split-fluorescent protein approach (Kamiyama et al., 2016), we tagged the N-terminus of endogenous TMEM41B (predicted to by cytosolic) with the 11 th beta strand of GFP (GFP11) and co-expressed a complementary GFP fragment (beta strands 1 through 10 [GFP1-10]). This resulted in the expected reticular GFP fluorescence signal that colocalized with the ER marker calnexin ( Fig S4A).
Next, we made a stable TMEM41B KO cell line ( Fig S4B) and established that this neither elicited an ER stress response nor diminished its induction by tunicamycin, an inhibitor of protein glycosylation in the ER (Fig S4C). Several lines of evidence argue that autophagy is severely blocked in the absence of TMEM41B. First, TMEM41B KO cells accumulated LC3 in its lipidated form (LC3-II), a phenotype that was rescued by transgenic expression of TMEM41B ( Fig S4D). In addition, we found no evidence for further LC3 accumulation in TMEM41 KO cells following addition of BafA1 ( Fig 4A). Second, we also detected accumulation of autophagy receptors (e.g. SQSTM1, etc) in TMEM41B KO cells ( Fig 4B). Third, autophagy flux was abolished in TMEM41B KO cells as measured using tfLC3 and tfSQSTM1, which phenocopied the ATG7 KO control (Fig S4E, S4F).
To define which step in autophagy requires TMEM41B, we first monitored a signal induction pathway for autophagosome initiation. mTOR kinase activity normally represses autophagy initiation by inhibiting the ULK1 kinase (Alers et al., 2012). Active ULK1 phosphorylates ATG13, a component of the RB1CC1 scaffolding complex required for phagophore nucleation (Ganley et al., 2009;Hosokawa et al., 2009;Jung et al., 2009). We found that torin treatment stimulated ATG13 phosphorylation at Ser318, a known ULK1 site, comparably in wildtype and TMEM41B KO cells ( Fig 4C). For comparison, this signaling assay robustly detected loss of ULK1 activity in cells lacking RB1CC1, a known ULK1 kinasecoactivator ( Fig 4C). These data argue that TMEM41B is required for a step downstream of phagophore nucleation.
To monitor phagophore elongation in TMEM41B KO cells, we visualized several markers of this process by cell microscopy. Semi-automated and custom-written scripts for image processing and segmentation allowed us to make quantitative phenotypic measurements. We found that TMEM41B KO cells displayed ~3-fold more LC3+/ SQSTM1+ punctae than wildtype cells under basal conditions (Fig 5A, 5B). To further test if these structures correspond to intermediates in autophagosome formation, we analyzed localization of the early autophagy marker WIPI2 in TMEM41B KO cells under basal and torin-induced conditions. WIPI2 transiently associates with nascent phagophores before it dissociates prior to vesicle closure (Dooley et al., 2014;Polson et al., 2010). We found that torin treatment did not further increase the level of LC3+ punctae in TMEM41B KO cells and that the majority of LC3+ structures were positive for the early autophagosome marker WIPI2 (Fig 5C, 5D). For comparison, torin induced a strong increase in LC3+ punctae formation in wildtype cells and these LC3+ structures had relatively little overlap with WIPI2 ( Fig 5D). Notably, in TMEM41B KO cells both LC3+ and WIPI2+ structures were dependent on the upstream autophagy factor RB1CC1 for formation ( As another way of visualizing terminal steps in autophagosome formation, we turned to STX17, a SNARE protein that is recruited to mature autophagosomes following vesicle closure of elongated phagophores (Itakura et al., 2012). A recent study showed that STX17+ structures accumulate in the absence of ATG7 and we confirmed this finding (Fig S5B, S5C) (Tsuboyama et al., 2016). By contrast, we found no appreciable amounts of STX17 on LC3+ structures that accumulate in TMEM41B KO cells (Fig S5B, S5C). As a final test of whether LC3+ structures in TMEM41B KO cells represent closed autophagosomes defective for fusion with the lysosome, we carried out an LC3 protease protection assay (Fig 6A-C). To maximally accumulate any sealed autophagosomes, we inhibited autophagosome-lysosome fusion with BafA1 for 18 hours before gently lysing cells by mechanical disruption. Trypsin treatment of control wildtype cell extracts revealed that lipidated LC3 (LC3-II) was protease-resistant in the absence but not presence of Triton X-100, a non-ionic detergent that solubilizes membranes. By contrast, LC3-II was protease-sensitive in TMEM41B KO and control RB1CC1 KO cells (Fig 6B, 6C). Taken together, our microscopy and biochemical evidence support the view that TMEM41B is required for a late step in phagophore maturation into fully-sealed autophagosomes.

Receptor-specific engagements with ATG7-independent autophagy
Returning to other results from our genetic screens, we noted that NBR1 stood out because of its unusually weak dependence on ATG7 and other LC3 lipidation factors ( Fig 7A). A recent study showed that loss of RB1CC1 induces an alternative lysosome trafficking route for NCOA4 and TAX1BP1 (Goodwin et al., 2017). To examine if NBR1 can also access this route, we analyzed RB1CC1 KO tfNBR1 cells but found that no evidence supporting this possibility, either under basal conditions or following torin treatment ( Fig 7B, Fig S7A-C). By contrast, in ATG7 KO cells lysosomal delivery of tfNBR1 was strongly stimulated by torin and inhibited by BafA1 ( Fig S7C). To estimate the extent of NBR1's engagement with ATG7-independent autophagy in K562 cells, we treated ATG9A deletions as true autophagy nulls and calculated that 65% of tfNBR1 autophagy flux was maintained in an ATG7-independent context (Fig 7C).
Similarly, 30% of TAX1BP1 and 15% of SQSTM1 was delivered to the lysosome in the absence of ATG7, while delivery was negligible for NDP52 and LC3.

Forward genetic screen for modifiers of ATG7-independent autophagy
To directly probe the genetics of ATG7-independent autophagy by an unbiased approach, we performed a forward genetic screen in K562 ATG7 KO cells co-expressing Cas9 and tfNBR1 ( Fig 7D). From our list of screen hits and subsequent hit validation by individual sgRNA experiments, we found that lysosomal delivery of NBR1 in the absence of LC3 lipidation is absolutely dependent on several canonical ATG factors, including ATG9A, RB1CC1 and ATG101, as well as components of the late endocytic pathway including RAB7A and HOPS ( Fig 7E). Next, we compared modifiers of tfNBR1 in wildtype and ATG7 KO backgrounds to look for interesting genetic interactions. This analysis revealed that loss of ATG7 induced complete buffering to several ATG factors, including ATG3 and ATG10. Furthermore, we detected synthetic genetic interactions between ATG7 and several factors with poorly-defined roles in autophagy (e.g. EI24), as well as TMEM41B and the autophagy receptor TAX1BP1. Lastly, we found evidence that NBR1 autophagy flux in the absence of LC3 lipidation is dependent on certain vesicle trafficking (e.g. RAB1A and TRAPP complex components).

Discussion:
The power of forward genetics in yeast is the foundation on which the field of autophagy research firmly stands (Harding et al., 1995;Thumm et al., 1994;Tsukada and Ohsumi, 1993).
Yeast genetic screens were instrumental in identifying the first autophagy receptor and the ~30 core ATG factors required for autophagosome biogenesis in yeast (Scott et al., 2001). Despite deep conservation between yeast and mammalian ATG factors, there are many mammalian-specific forms of autophagy that remain poorly understood. The recent emergence of new CRISPR/Cas9based technologies has begun facilitating efforts to define novel mammalian autophagy factors and pathways by forward genetic screening.
Here, we combined the power of CRISPR screening with that of tandem fluorescent reporters for monitoring cytoplasm-to-lysosome targeting of several known autophagic factors. To facilitate discovery, these data are also presented online in an interactive format at http: //35.196.197.20:3838/cs_crispr/ . Among our screen hits, we found virtually all known ATG factors, plus autophagy modulators that only became discovered while this work was in preparation (Jia et al., 2017;Vaites et al., 2017), as well as fully uncharacterized proteins. Within this latter category, we identified TMEM41B as an ER membrane protein important for a late step in phagophore elongation and/or vesicle closure. Going forward, it will be important to test potential connections between TMEM41B and other ER-associated autophagy factors (e.g. VMP1). Another promising extension of our work will be to apply CRISPR screening to additional tf-based reporters of cytoplasm-to-lysosome targeting, including autophagy targets themselves.
Following primary screening, our focus was drawn to one receptor, NBR1, because of its unusually persistent autophagy flux in the absence of many ATG factors, including those essential for LC3 lipidation. This initial finding was surprising to us because of the well-established paradigm that NBR1 and most other autophagy receptors tether their targets to lipidated LC3 via their LC3-interacting regions (LIRs) (Birgisdottir et al., 2013). To get a better handle on this unexpected result, we carried out a secondary screen in ATG7 KO tfNBR1 cells. Our specific choice for this LC3-lipidation null background was driven by two considerations. Historically, ATG7independent autophagosomes were among the first forms of alternative autophagy to be reported (Nishida et al., 2009). Practically, ATG7 KO cells are widely used in autophagy research as an autophagy-null control.
We found that genetic ablation of ATG9A, RB1CC1, and ATG101 consistently gave rise to the most severe loss-of-function phenotypes for both ATG7-dependent and ATG7independent autophagy. In the absence of ATG7, tfNBR1 autophagy flux became increasingly dependent on certain ATG factors (e.g. ATG2B, WDR45) and less on others, including certain late endosomal factors (e.g. EPG5 and VPS37A). This type of buffering genetic interaction places LC3, EPG5, and VPS37A in the same genetic pathway. Consistently, both LC3-family proteins and EPG5 have recently been defined as vesicle/target-membrane specificity determinants in autophagosome-lysosome fusion (Nguyen et al., 2016;Wang et al., 2016). Thus, our working model is that autophagosomes formed during ATG7-independent autophagy are trafficked to lysosomes by an alternative trafficking mechanism mediated by factors that we found were selectively required for ATG7-independent but not ATG7-dependent autophagy (e.g. RAB1A, GDI2, TRAPPC2L) . A mutually non-exclusive hypothesis emerges from our observation that NBR1 trafficking becomes co-dependent on TAX1BP1 in the absence of ATG7.
This autophagy receptor is known to interact with myosin IV and was previously postulated to control vesicle trafficking by the actin/myosin network (Tumbarello et al., 2015). Intriguingly, a recent study showed how activation of TBK1 kinase enables TAX1BP1 to direct lysosomal targeting of ferritin by a novel ESCRT-mediated process (Goodwin et al., 2017). Our work implicates TAX1BP1 in another noncanonical cytoplasm-to-lysosome pathway and should help raise the interest level in this understudied autophagy receptor.
Mechanistically, it remains unclear how NBR1, as well as TAX1BP1 and SQSTM1 to lesser extents, engages the autophagosome biogenesis process in the absence of LC3 lipidation. In yeast, several receptors are known to locally induce autophagosome formation by binding to a Cterminal region of Atg11 (Kamber et al., 2015), which has weak homology to a C-terminal sequence of RB1CC1 (Li et al., 2014a;Yorimitsu and Klionsky, 2005). Analogous mechanisms could explain our observation that NBR1's lysosomal delivery remains selective in the absence of ATG7 albeit in a manner still dependent on RB1CC1 and ATG101. Consistently, there is evidence that mammalian autophagy receptors (both mitophagy receptors (Lazarou et al., 2015) and TRIMs (Kimura et al., 2016)) can recruit an RB1CC1-containing complex (RB1CC1/ATG101/ATG13/ULK1) to locally induce autophagosome formation.
LC3 positive and negative double-membrane vesicles have previously been shown to coexist in certain cell types (Nishida et al., 2009). To what extent do receptors then engage alternative autophagy pathways in cells permissive for LC3 lipidation? Answering this important question will necessitate development of receptor-based reporters that differentiate flux through related pathways comprising many overlapping components. At one extreme, it is possible that LIRcontaining receptors normally engage only LC3-marked autophagosomes. This would then imply that ablation of ATG7 induces compensatory mechanisms, which bypass the need for receptor-LC3 interactions. It also remains unclear what distinguishes receptors that engage ATG7independent autophagy from those that don't (e.g. NBR1 vs NDP52). Since NBR1 is known to interact with a diverse set of autophagy targets (Deosaran et al., 2013;Kenific et al., 2016;Kirkin et al., 2009;Lamark et al., 2009), one intriguing possibility is that its engagement with this alternative autophagy pathway is target dependent.
Our systematic, genome-wide analysis of NBR1 autophagy flux casts a new light on the roles of ATG factors within the autophagy cannon. We found that some are absolutely essential (e.g. RB1CC1 and ATG9A) whereas others (e.g. WIPI2 and ATG14) only become critical in the absence of ATG7. At a molecular level, RB1CC1 and ATG9A are among the first ATG factors to be recruited to ER-associated sites at which LC3-marked autophagosomes eventually form (Koyama-Honda et al., 2013). RB1CC1 doesn't have a strong yeast homolog by sequence analysis, but is very likely structurally and functionally equivalent to Atg17 and Atg11 in yeast (Li et al., 2014a). Recent studies have proposed that Atg17 is sufficient to remodel Atg9 vesicles into a cupshaped structure (Bahrami et al., 2017;Ragusa et al., 2012). Our work makes it further imperative to establish if RB1CC1 scaffolding of ATG9A vesicles is a conserved phagophore nucleation mechanism. Even more speculatively, our data suggests that autophagosomes that are produced by a common phagophore nucleation process can then engage a variable set of mechanisms involved in downstream autophagy steps. Further mapping of pairwise genetic interactions between ATG factors is a promising approach for teasing apart these complexities inherent in the mammalian ATG network. See also, Figure S1.   See also, Figure S4. See also, Figure S5.  respectively. The median Red:Green ratio measured in sgATG7 cells was then used to calculate the fraction of tfReporter delivered to the lysosome according to the following formula: (ratio sgATG7 -ratio sgATG9A )/(ratio sgControl 1 -ratio sgATG9A ). P-values were determined using a student's t-test. ***, p <0.001; **, p <0.01; ns, not significant.
(D) Gene correlation plot (19,117 genes) of average beta scores for tfNBR1 in wildtype and ATG7 KO cells. Notable autophagy-related genes are highlighted in black. Also highlighted (but without their gene names) are components of the TRAPP (blue) complex that scored as hits.
Dashed lines, top 1% of beta scores.
(E) Wildtype and ATG7 KO K562 cells co-expressing Cas9 and tfNBR1 were transduced with individual sgRNAs for indicated genes. The median Red:Green ratio of each population was used to calculate the fold-repression according to the following formula: (ratio sgGeneratio sgATG9A )/(ratio sgControl -ratio sgATG9A ), with the assumption that sgATG9 yields a true autophagy-null phenotype. Deeper shades of red indicate stronger suppressor phenotypes. Genes were clustered on the basis of their patterns of genetic interactions.   Figure 3A. These data are represented as part of the heatmap in Figure 3B.       List of 20-nucleotide oligos (plus overhangs) used to clone individual sgRNAs.

Table S4. Other oligos
List of oligos used in T7 endonuclease assays to confirm TMEM41B deletion and oligos used for endogenous labeling of TMEM41B with the 11 th beta strand of GFP.

Table S5. List of primers for cloning tfReporters
List of primers used to clone LC3B and receptors into the KpnI site of tfEmpty to generate each tfReporter.    Table S5. To generate blasticidin resistant versions of each cassette, the puromycin resistance gene was excised from pCS418 with an XhoI/SpeI digest and replaced with an analogous geneblock fragment (oVD5075, Integrated DNA technologies) encoding a blasticidin resistance gene (BSD).
Tissue Culture: All cells were grown in a standard water-jacketed incubator with 5% CO2. (1:500) was used as a common additive. All cells were passaged less than 25 times. For passaging, cells were trypsinized with Trypsin-EDTA (25300-054, Life Technologies).
Puromycin (2µg/ml), blasticidin (5µg/ml) and zeocin (50µg/ml) were added when necessary for selection. Database was used to verify that the identified alleles matched those of the expected cell type.

Transient transfection and nucleofection: Prior to transfection, HEK293T cells were seeded in
OptiMEM Reduced Serum media (51985-034, ThermoFisher). At 90% confluency, cells were transfected overnight using Lipofectamine 3000 reagent according to the manufacturer's recommendations. The following morning media was exchanged and cells were passaged for another 24 hours prior to drug treatment. K562 cells were nucleofected using a Nucleofector™ 2b Device (Lonza) using nucleofector kit T (VACA-1002, Lonza) and protocol T-016.

Generation of gene knock-out cell lines using CRISPR-Cas9 gene editing:
For the generation of stable knock-outs, HEK293T and K562 cells were transfected or nucleofected, respectively, as described above. Oligo sequences for sgRNAs were generated by CHOPCHOP or extracted from the Brunello library and cloned into the indicated vectors as outlined above under 'sgRNA oligonucleotide (oligo) ligation protocol' (Doench et al., 2016;Labun et al., 2016). Oligos are listed under Table S4. Cells were diluted by limiting dilution or cell sorting into 96 well plates and clonally expanded. Expanded cells were confirmed for knock-out by western blot or PCR/T7 endonuclease testing.
Cells were grown overnight in Opti-MEM media (5% FBS, no antibiotics) (31985062, ThermoFisher) to 90% confluency. Cells were than transfected with pVSV-G, pSPAX2 and packaging constructs at a 1:3:4 ratio. Transfection proceeded for 6-8 hour before media was refreshed. Virus was collected and pooled at 24 and 48 hours post-transfection. Virus was pelleted at 1000g 2X 10 min, aliquoted and frozen in single-use aliquots. For retrovirus production, all methods were the same except pUMVC was exchanged for pSPAX2.

Viral transduction:
Cells were incubated in appropriate media containing 8µg/ml polybrene and lacking penicillin/streptomycin. Cells were transduced overnight. Media was exchanged for media lacking polybrene for 24 hours prior to antibiotic selection.
Protease protection assay: Our protocol for mammalian protease protection assay was based on Zhao et al. (Zhao et al., 2013). Cells were seeded so they would be 70% confluent at 5pm. Normal Goat Serum (NGS) in PBS) and washed once in PBS. Primary antibody was diluted in 5% NGS at the dilutions described elsewhere in these methods. 75ul of antibody mixture was spotted on parafilm in a humidified chamber and inverted coverslips were incubated with antibody overnight at 4C. After incubation, coverslips were washed 3X 10 min in PBS.
Image analysis Fluorescence microscopy images were processed using a newly developed Python analysis pipeline built around the pyto_segmenter analysis package (Weir et al., 2017). First, regions of images containing cells were identified. To do so we first fit a Gaussian distribution to the fluorescence intensity distribution for a smoothed green channel (488 nm excitation) z-stack from an empty field. Using this Gaussian fit we predicted the probability that each pixel in the smoothed green channel z-stack for each field corresponded to background (non-cell) or foreground (cell). We assigned each pixel with a p(background) < 10 -5 to the cells, thus creating a "cell mask". After removing small specks (<100,000 pixels volume) to eliminate debris, we removed out-of-focus planes from the cell mask using a Support Vector Machine (SVM) classifier as described previously (Hsu and Chen, 2008). Next, nuclei were segmented from the blue (DAPI) channel by slice-by-slice relative thresholding followed by watershed segmentation using the pyto_segmenter package. Cells were segmented using watershed segmentation from nuclei seeds.
Cell edges were eroded to eliminate blurred edge excess included during the p-value transformation. Cells contacting the edge of the field were removed from analysis. Next, punctae were segmented in the green and red (561 nm) channels using the pyto_segmenter package with empirically determined Canny edge detection thresholds. The number of total punctae and punctae overlapping with objects in the other fluorescence channel were counted and tabulated data were saved in .csv format. Plotting was performed using R and the ggplot2 package. See the image analysis package for details.
Cells were manually scraped off plates and a GenElute Megaprep kit (Sigma, NA0600-1KT) was used to purify plasmid DNA.

NGS data analysis:
The 5' end of NGS reads were trimmed to 5'-CACCG-3' using Cutadapt. The count function of MAGeCK (version 0.5.3) was used to extract read counts for each sgRNA. The mle function was used to compare read counts from cells displaying increased and decreased Red:Green ratios.
The output included both beta scores and false discovery rates. Beta scores for each sgRNA for each tfReporter were averaged across 2-4 experiments. Across all experiments, average read counts were 200-400 per sgRNA. To generate heatmaps for each reporter (e.g. Fig 2), the beta scores for each gene were normalized by the beta score for ATG9A.

Flow cytometry:
All samples were pelleted, washed 1X in cold PBS, and filtered through strainer cap tubes (VWR, 21008-948) prior to analysis. Flow cytometry data were collected on an LSRII flow cytometer. Data were analyzed in FlowJo (FlowJo, LLC) and R. The biomodality of populations was determined in an unbiased manner using the BifurGate tool in FlowJo. At least 5000 cells were collected for all samples.