Resource

Active Interaction Mapping Reveals the Hierarchical Organization of Autophagy

Graphical Abstract Authors Michael H. Kramer, Jean-Claude Farre´ , Koyel Mitra, ..., J. Michael Cherry, Suresh Subramani, Trey Ideker

Correspondence [email protected] (S.S.), [email protected] (T.I.)

In Brief Kramer et al. present a general procedure that guides molecular interaction mapping to assemble a hierarchical model of any biological system. Application to autophagy reveals the hierarchical organization of this process, including many new biological components and functions. This work provides an archetype for future studies in systems biology.

Highlights d Diverse -omics data integrated to assemble a unified hierarchical model of autophagy d Post hoc analysis prioritizes the most informative future data types d Consequently, 156,364 genetic interactions measured in autophagy-activating conditions d Multiple new functions involve Gyp1, Atg24, Atg26, Ssd1, Did4, Stp22, and others

Kramer et al., 2017, Molecular Cell 65, 761–774 February 16, 2017 ª 2017 Elsevier Inc. http://dx.doi.org/10.1016/j.molcel.2016.12.024 Molecular Cell Resource

Active Interaction Mapping Reveals the Hierarchical Organization of Autophagy

Michael H. Kramer,1,4 Jean-Claude Farre´ ,2,4 Koyel Mitra,1 Michael Ku Yu,1 Keiichiro Ono,1 Barry Demchak,1 Katherine Licon,1 Mitchell Flagg,1 Rama Balakrishnan,3 J. Michael Cherry,3 Suresh Subramani,2,* and Trey Ideker1,5,* 1Department of Medicine 2Section of Molecular Biology, Division of Biological Sciences University of California, San Diego, La Jolla, CA 92093, USA 3Department of Genetics, Stanford University, Stanford, CA 94304, USA 4Co-first author 5Lead Contact *Correspondence: [email protected] (S.S.), [email protected] (T.I.) http://dx.doi.org/10.1016/j.molcel.2016.12.024

SUMMARY 2004; Malleshaiah et al., 2010). Model predictions are system- atically compared to laboratory measurements, with discrep- We have developed a general progressive proced- ancies used to improve the model and design further ure, Active Interaction Mapping, to guide assembly experiments. of the hierarchy of functions encoding any biolo- Initially, formulating a model can be laborious, requires that gical system. Using this process, we assemble an much is already known about the system, and may be biased to- ontology of functions comprising autophagy, a cen- ward well-studied components. Alternatively, many studies have tral recycling process implicated in numerous dis- sought to construct models of biological systems directly from systematic datasets (Ergun€ et al., 2007; Ho et al., 2002). eases. A first-generation model, built from existing Saccharomyces Genome-scale data, including profiles of mRNAs, proteins or gene networks in , captures most metabolites, and gene and protein interactions, are analyzed to known autophagy components in broad relation to infer a network of genes describing the system (Behrends vesicle transport, cell cycle, and stress response. et al., 2010; Havugimana et al., 2012; Kim et al., 2011; Lefebvre Systematic analysis identifies synthetic-lethal inter- et al., 2012; Picotti et al., 2013). Such network models have been actions as most informative for further experiments; increasingly used to describe key biological systems, including consequently, we saturate the model with 156,364 tissue specificity (Greene et al., 2015), cell differentiation (Cahan such measurements across autophagy-activating et al., 2014), and diseases such as cancer (Carro et al., 2010; conditions. These targeted interactions provide Creixell et al., 2015; Leiserson et al., 2015). more information about autophagy than all previous At this juncture, we see two challenges. First, biological sys- datasets, producing a second-generation ontology tems do not merely consist of flat networks of genes; they are exquisitely hierarchical in nature. Models should seek to cap- of 220 functions. Approximately half are previously ture this hierarchy, which extends over multiple scales from unknown; we confirm roles for Gyp1 at the phago- nucleotides (<1 nm scale) to proteins (1–10 nm), protein com- phore-assembly site, Atg24 in cargo engulfment, plexes (10–100 nm), cellular processes (100 nm), organelles Atg26 in cytoplasm-to-vacuole targeting, and Ssd1, (1 mm), cells (1–10 mm), tissues (100 mm–100 mm), and com- Did4, and others in selective and non-selective auto- plex organisms (>1 m). Second, while the field has taken a phagy. The procedure and autophagy hierarchy are foundational step by enabling the construction of models at http://atgo.ucsd.edu/. directly from data rather than only literature, a next step will be to determine how to iteratively improve these models with the most informative new data. The ability to rationally INTRODUCTION motivate and design new -omics datasets is sorely needed in genomics; far too many large-scale experiments are per- A key promise of systems biology is to advance our understand- formed simply because one can, without justification or guid- ing of biological systems by combining genome-scale data with ance for why those experiments might lead to an improved model building in a virtuous cycle (Ideker et al., 2001a; Kitano, biological understanding. 2002). New genomics data improve the biological model and, Toward the first challenge, a hierarchical biological model in turn, analysis of the model informs collection of new datasets. already in widespread use is the Gene Ontology (GO) (Ashburner Systems biology typically surveys the literature to assemble et al., 2000). Through extensive literature curation efforts, GO knowledge of specific molecules, interactions, and reactions aims to factor the cell into a complete hierarchy of cellular com- involved in a biological system, then uses this knowledge to ponents and processes, each represented by a GO ‘‘term.’’ initialize a computational model (Covert et al., 2004; Davidson Recently, we showed that an ontology of the cell very much et al., 2002; Ideker et al., 2001b; Karr et al., 2012; King et al., like GO could be automatically constructed from diverse

Molecular Cell 65, 761–774, February 16, 2017 ª 2017 Elsevier Inc. 761 A Integrate data to best capture curated knowledge ii

Protein-protein interactions

Modeling i Genetic interactions Initialize Convert network to with public iii hierarchical model data Co-expression interactions

191 transcription factors, unknown relationship Regulators of casein kinase 2 bud neck proteins AtgO:220 autophagy membrane trafficking catalytic subunits and related processes promotion of condensin recruitment transcriptional membrane protein Rho1-dependent to replication fork 8 response to iron blockage site localization and polarized cell growth insertion / stress depletion salt tolerance kinases Golgi-localized, response ubiquitin related CAMK/CMGC kinases GET complex actin-regulating γ-adaptin ear palmitoylated AAA ATPases and process and kinase family spore morphogenesis vacuolar membrane cytoskeletal functions homology, Arf-binding specificity factors AtgO:185 vacuolar trafficking regulation proteins ankyrin proteins genes, unknown nucleosome staurosporine transport of repeat-containing, casein kinase I-like CDK9-related acetyltransferase of Atg19-receptor subgroup resistance pathway palmitoyl transferase MAP kinase and a proteins cyclin-dependent ARFs and ARF H4 components and cargos proteins MAPK inactivator (no plasma EGO complex kinases binding proteins actin related proteins Mapping membrane-associated subunits + Pib2, Pep7 ADP ribosylation piccolo NuA4 histone published casein kinase I-like EGO complex factors actin and nucleation acetyltransferase membrane trafficking, relationship) proteins subunits and Pib2 guanine nucleotide factor complex subunits cell growth and stress negative regulation of ADP ribosylation exchange factors for Sec7-Arf1 golgi responses I Hog1 signaling Gtr1-Gtr2 GTPase factors involved in ARFs vesicle transport complex Rho1 GTPase and Golgi vesicular regulation complex effector trafficking eIF2B subunits and chromosomal BET5 TRS23 regulator passenger complex regulation of TRS33 membrane trafficking, subunits polyamine uptake TRS130 TRS20 TRAPP Complex cell growth and stress eIF2B subunits BET3 responses II TRS31 TRS85 AtgO:138 late secretory GEA1 pathway (vesicle GYP1 SYS1 Ubp3-Bre5 ubiquitin GARP complex vacuolar H+-ATPase trafficking, docking ATG32 protease complex subunits GET3 GYP7 and endosome to and fusion) ATG22 CLG1 BRE5 Ypt31/Ypt32 family golgi traffic SEC18 VPS1 RAV1 UBP3 VPS51 VPS52 TLG2 PEP7 Rab GTPases SEC17 RIC1 VPS45 YPT32 membrane fusion conserved oligomeric 232 V-ATPase, V0 golgi (COG) complex PEP8 HOPS/CORVET AtgO:140 domain subunits complexes and COG4 VMA3 localization YPT53 YPT31 Vma12/Vma22 VMA11 VPS8 assembly complex YPT52 COG2 COG1 COG5 retromer complex VPS9 Align with GO, VMA22 VPH2 VPS3 COG8 cargo selection V-ATPase nucleotide COG3 COG6 COG7 subunits binding V1 subunits VTC1 vacuolar transporter HSV2 UTH1 VPS29 VMA2 VMA1 HOPS complex VTC4 chaperone complex VTI1 late secretory Q-snares VPS21 VPS33 VTC3 vacuolar transporter pathway / vacuolar VPS41 chaperone complex, membrane dynamics, PEP3 PKC1 REG1 SSK2 VTC2 VAM3 vacuolar docking and VAM3 cytosolic facing cell growth and stress PEP12 VAM7 VPS16 PCL1 responses fusion PEP5 CDC7 CDC28 Other ‘omics datasets componentsHierarchical model ATG32 GTR2 PDK1 homolog YPT7 BCK1 GCN3 Observe new VAM6 ATG20 ATG4 PKH1 Mon1-Ccz1 complex ATG18 KIN4 ATG29 kinases GEA1 HSL1 Intraluminal vesicle PTC6 KES1 PTC6 PKH2 ATG17 YAK1 ATG2 neck formation CCZ1 MON1 MEK1 DOA1 RAD53 DNA damage vesicle (especially ARF1 IRS4 RSP5 checkpoint kinase ATG31 ADR1 DOA4 SEC2 DID4 PI4P kinases CHK1 golgi) and organelle NPR1 autophagy inhibitors / DUN1 effectors CDC14 KSS1 ATG23 VPS4 traffic / cell cycle and CDC15 KIN28 MCK1 nutrient sensing STT4 general amino acid VTI1 TPK1 Ypt1/Sec4 family Rab stress response YPK1 BRO1 AtgO:247 engulfment of Atg14-Vps30 GCN4 control pathway PKH3 CDC5 LSB6 GTPases SEC4 KCC4 ATG24 selective autophagy cargos subcomplex of PI3P protein A kinases AtgO:96 GCN2 PDK1 homolog kinase complex I nucleus-vacuole junction YPT1 PKH2 ATG8 core machinery of TPK2 TPK3 VAC8 ATG12 Atg1-Atg13 complex kinases of autophagynuclear side of TRS85 (AtgO)AtgO:18 vesicle transport Ste11 kinase autophagy RIM15 VPS30 PKH1 biology, share nucleus-vacuole VPS30 ATG5 machinery of complex, MAPK ATG9 PI3P kinase complex junction ATG1 ATG5 YPT7 ATG9 ER-EE-MVB-vacuole ATG13 signaling ATG31 ATG14 II MAPK signaling and BEM1 NVJ1 and autophagy regulation of Msn2/4 Msn2/4-mediated ATG15 ATG18 SWH1 CLG1 nutrient-sensing ATG22 COG2 Atg8 activating 40 transcription yeast stress response ATG11 ATG16 VPS38 Ste11 kinase VPS4 ATG17 response PHO80 complex and regulation AtgO:240 VPS34 complex, MAPK ATG11 GYP7 Atg12 conjugation VPS15 GET3 PEP7 components of HOG Atg1-Atg13 complex signaling GYP1 ATG3 sensing pathway STE50 RIM15 system VPS1 ATG13 AtgO:246 183 AtgO:106 SCH9 VPS38 ATG7 STE50 COG6 kinases, unknown ATG1 Atg8 activating negative regulation of RIC1 ATG27 STE11 COG5 Ypt31/Ypt32 family relationship YGK3 NTH1 complex ATG12 ATG10 transcription from CLA4 PBS2 MSN4 YPK2 TLG2 Rab GTPases ATG3 RNA polymerase II TORC2 downstream golgi recruitment of YAK1 ATG7 cell cycle and Snf1 SKM1 calmodulin-binding RTG3 promoter effector kinases YPK1 Arl1 YPT31 STE11 TPS2 model signaling pathway YPT32 motor transport MARK homolog TPK2 STE20 glycogen synthase MSN2 kinases KIN28 PCL1 MAPK signaling and engulfment of MID2 MIG1 ARL1 ARL3 kinase 3 homologs MIG1 KIN1 TOS3 PCL5 TOR pathways regulators Mid2 subfamily of cell selective autophagy SHO1 KIN2 ADR1 KOG1 wall integrity sensors Ypt1/Sec4 family Rab SYS1 SAK1 septin-associated CDC14 MTL1 cargos SLT2 14-3-3 proteins GTPases membrane fusion ELM1 kinases KIC1 SSD1 MID2 Snf1 kinase subunits HSL1 Wsc family cell wall ATG24 ATG8 CBK1 sensors PTC1 MPS1 and regulator GIN4 CDC15 Hog signaling SEC4 YPT1 VTC1 ATG20 BMH1 REG1 KCC4 TOR complex RCK1 BMH2 SLG1 MAPKKKs CMK1 MCK1 cell cycle machinery catalytic subunits TOR1 MKK1 COG4 RIM11 Snf1 kinase complex, PCL6 WSC3 mitotic exit network PKC1 BCK1 Ubp3-Bre5 ubiquitin MYO2 MRK1 alpha and beta TOR2 Hog1 kinase and the SNF1 SWE1 kinases SSK22 protease complex subunits PBS2 SLT2 MKK2 SSK2 Rck2 kinase substrate CMK2 CMD1 CDC5 SIP2 PHO4 CAK1 HOG1 KSS1 DBF20 FUS3 UBP3 BRE5 regulation of G1/S DBF2 STE7 RCK2 AtgO:160 phase transition HOG1 iv SIC1 PHO85 Pho85-Pho80 CDC28 PHO80 CDK-cyclin comploex Analyze model to select new type of experiment. Add new data

BC

8 4 99 7 Core autophagy genes Relationships in GO 95 6 but not in network data 3 Other autophagy (GO) 5 All other genes 75 50 4 2 25

3 Density

2 5 1 Relationships in network 1 data but not in GO 1 0 0 12 Pairwise gene similarity score (GO) 0 1 2 3 4 5 6 7 Pairwise gene similarity score (integrated data) Similarity to core autophagy genes D Highly ICS2 similar IML1 Analysis of hierarchical organization NPR3 to infer gene ontology (CliXO) NPR2 PHO80 SIC1 Gene Biological Process KSP1 (Root) Term PCL5 PHO85 ICS2 UTR5 Regulation of ESA1 macroautophagy RPD3 PCL1 Regulation of Negative regulation Positive regulation autophagosome FAR11 of macroautophagy of macroautophagy assembly

Pairwise gene similarity score CLG1 NPR3 NPR2 RPD3 EPL1 SIC1 PHO80 PHO85 EPL1 ESA1 ATG23 IML1 Highly ATG23 PCL5 KSP1 FAR11 PCL1 CLG1 dissimilar UTR5 SIC1 IML1 ICS2 EPL1 PCL1 PCL5 ESA1 KSP1 UTR5 CLG1 RPD3 NPR2 NPR3 FAR11 ATG23 PHO85 PHO80

Figure 1. Program for Active Interaction Mapping (A) After initialization with public datasets (i), the approach proceeds by progressive iterations of data integration (ii), hierarchical model assembly (iii), and adaptive generation of new data (iv).

(legend continued on next page)

762 Molecular Cell 65, 761–774, February 16, 2017 genome-scale datasets (Dutkowski et al., 2013; Kramer et al., RESULTS 2014). Using this method, we built a data-driven ontology of yeast, the Network Extracted Ontology (NeXO), which recapitu- Integrating Diverse -omics Data Constructs lated approximately 60% of known cellular components in GO. a First-Generation Model of Autophagy However, it was also clear that our hierarchical understanding To initialize the AI-MAP process, we sought to construct a ‘‘first- of cell biology is far from complete, leaving open the second generation’’ hierarchical model of autophagy, called the Auto- challenge: how to most efficiently improve cell function hierar- phagy Ontology (AtgO), based on publicly available molecular chies with high-value experiments. Thus far, methods have interaction networks in S. cerevisiae. Although not all datasets been devised for selecting the optimal next experiment consid- have been specifically designed to study autophagy, non-selec- ering a particular kind of model (e.g., a gene network), data tive or selective autophagy pathways have some activity in type (e.g., growth phenotyping), and a constrained repertoire nearly all conditions (Reggiori et al., 2012). In total, we obtained of treatments (e.g., single or double gene knockouts) (Atias a compendium of 78 networks of nine different data types, et al., 2014; Barrett and Palsson, 2006; Ideker et al., 2000; including protein interactions, genetic interactions, gene co- King et al., 2009, 2004). However, such questions of experi- expression, and gene-gene similarity based on shared protein mental design have not been approached for building general hi- sequence and structural information, as previously curated for erarchical models and, given the plethora of -omics data types YeastNet (Kim et al., 2014)(Figure 1Ai). that can now be generated, guidance is needed on which type To build an ontology from these heterogeneous datasets, we of data one should collect next to improve the models. first integrated all data into a single pairwise gene similarity Here, we describe a general program for elucidating the hier- network (Figure 1Aii) and then analyzed the hierarchical structure archy of functions underlying a cellular process, based on pro- of this network to infer a gene ontology (Figure 1Aiii). Pairwise gressive cycles of mapping and modeling that we call Active gene-gene similarities in the similarity network were determined Interaction Mapping, or AI-MAP (Figure 1A). We apply this using a statistical regression procedure that combines available approach to develop a hierarchical model of autophagy, the interaction datasets into a single quantitative similarity score. conserved process of ‘‘self-eating’’ by which cells respond to To learn how the different data types should be combined, starvation and other stress by degrading macromolecules and the regression was trained to model a ‘‘standard reference’’ recycling their constituent building blocks (Levine and Yuan, gene similarity network derived from GO (gene-gene similarity 2005). During autophagy, cellular contents are enclosed in a scores based on relatedness in GO of the two genes; see double-walled vesicle, the autophagosome, and delivered to STAR Methods). In this way, the data-derived similarity network the yeast vacuole or mammalian lysosome for degradation and captured known biological processes and components in GO recycling; such contents can include non-selective cytoplasmic when supported by data (Pearson r = 0.4 versus GO, p < contents (macroautophagy) or selective organelles, such as per- 10300, out-of-bag prediction); similar data combinations then oxisomes (pexophagy) or mitochondria (mitophagy). All non-se- identify new biological processes, components, and relation- lective and selective autophagy pathways share a core set of ships not previously documented (Figure 1B). molecular machinery (Jin and Klionsky, 2013), which has wide- For assembly of the ontology, we selected 492 candidate ranging effects on pathways such as tumor suppression (Liang genes with potential relation to autophagy based on literature et al., 1999), neurodegeneration (Rubinsztein et al., 2005), and or data (Figure 1C). A broad net was intentionally cast to allow aging (Rubinsztein et al., 2011). later steps of the AI-MAP process to dictate which of these In what follows, we (1) construct a first-generation hierarchical 492 genes are most related to core autophagy functions. The model of autophagy from publicly available interaction networks Clique Extracted Ontologies algorithm (CliXO; Kramer et al., for the budding yeast, Saccharomyces cerevisiae; (2) identify 2014) was then applied to analyze the hierarchical structure of synthetic-lethal genetic interactions as the most useful type of the data-derived gene similarity network among these 492 genes data for systematically improving the model; (3) generate static (Figure 1Aiii). By this method, nested communities of genes and differential genetic networks involving growth measure- apparent in the pairwise similarity data were identified, resulting ments of over 156,000 yeast genotypes across autophagy-rele- in a hierarchy of 218 terms and 310 term relations, which we call vant conditions; and (4) integrate these new datasets with previ- the Autophagy Ontology (AtgO 1.0; Figure 1D). To determine ous ones to arrive at a substantially improved understanding of which terms represent known biological functions and which autophagy. The active interaction mapping process creates a represent potential new biology, AtgO 1.0 was aligned with GO ‘‘living’’ ontology of cell function, which actively suggests new according to a previously described alignment procedure (Dut- data types for self-improvement and highlights new functions kowski et al., 2013). This process seeks a one-to-one mapping discovered as data are added. AI-MAP has general utility for between terms in AtgO and terms in GO; aligned terms, term any biological process. names, and descriptions are transferred from the GO reference.

(B) Pairwise gene similarity scores from curated knowledge (GO) versus the integrated data. The curves track percentiles in GO similarity scores at a given level of data similarity. (C) Selection of 492 autophagy-related genes. There are 20 genes (red) that are assigned to ‘‘core autophagy’’ by Jin and Klionsky (2013); another 102 genes (yellow) are annotated to GO:autophagy (GO:0006914). For all other genes (green), the average similarity score to core autophagy genes is calculated from the network data, and a similarity threshold (vertical line) is set to select 370 genes at least as similar to core genes as those in GO:autophagy. (D) Inference of the hierarchical model from (idealized) pairwise similarity scores using CliXO.

Molecular Cell 65, 761–774, February 16, 2017 763 A B YPK3 COG6 ATG26 PTC6 124 autophagy RAV1 VMA2 ATG4 protein 187 GYP1 localization to COG5 YAK1 GEA1 212 YGK3 nucleolar rDNA RIC1 repeats 97 VPS51 GARP complex 15 49 195 ADR1 MIG1 BEM1 protein TRAPPIII protein VPS52 retromer phosphorylation KIN28 complex, inner MCK1 127 complex SGV1 SAK1 1-phosphatidylinositol MRK1 chromosome 128 4-kinase activity shell regulation of cell ELM1 TOS3 PCL5 STT4 RIM11 passenger 248 VPS29 budding AMP-activated complex 243 protein kinase LSB6 PEP8 55 replication fork MAP kinase 194 SSD1 complex 118 protection kinase kinase REG1 Golgi to plasma 156 CDC15 28 activity 210 membrane 122 KIC1 AMP-activated eukaryotic 100 protein transport regulation of 128 protein kinase negative translation SYS1 nuclear-transcribed activity regulation of initiation factor ARL1 mRNA poly(A) ARL3 ubiquitin-protein 2B complex tail shortening 100 KIN2 SIP2 SNF1 235 ligase activity KIN1 involved in cytokinesis SKM1 RTG3 TPS2 RNA polymerase VPS45 mitotic cell cycle checkpoint PEP7 II CLA4 NTH1 MSN2 carboxy-terminal vesicle fusion TLG2 VTI1 41 MSN4 domain kinase SAM domain 74 SEC17 SEC18 activity negative binding regulation of 214 negative PHO4 STE50 ubiquitin-protein regulation of calmodulin ligase activity PCL6 PTC1 TAX4 macroautophagy binding involved in C ATG22 mitotic cell cycle YPT32 macroautophagy GIN4 PCL1 RCK1 114 CORVET complex ATG12 calmodulin-dependent protein kinase BMH1 BMH2 VAM3 MKK1 ribophagy HSL1 PHO80 91 IRS4 ATG3 166 activity Mon1-Ccz1 IRS4 186 UBP3 CBK1 VAM3 ATG27 VAC8 Golgi to plasma BRE5 complex ATG13 105 MYO2 YPT31 VPS3 SSD1 PHO80 membrane SEC4 CMK1 TOR1 ATG8 ATG20 RCK2 KOG1 VAM7 YPT52 protein transport 115 phosphatidylinositol VPS4 ATG26 TLG2 SIC1 SWE1 CDC14 CMD1 CMK2 TRS85 245 YPT53 TRS85 ARL3 YPT1 DBF20 3-kinase NPR1 VPS9 STT4 ATG24 CAK1 complex, class BRO1 VPS8 ATG22 CCZ1 ARL1 241 Golgi transport protein targeting SYS1 PHO85 DBF2 III, type II DOA1 PEP12 DOA4 complex VPS38 KES1 to vacuolar PEP12 226 cytokinesis 246 HOG1 VPS34 DOA4 membrane HOPS complex 109 KCC4 checkpoint STE11 MAP-kinase RSP5 scaffold activity VPS15 ARF3 205 152 ATG14 ATG31ATG17 MAP kinase CDC5 PBS2 SEC2 MON1 ATG4 YPT7 VAM6 kinase kinase CCZ1 130 ATG23 ATG29 COG3 COG2 COG4 STE20 MID2 SHO1 GEA2 ATG9 activity 138 PEP3 ARF2 ATG15 RIC1 COG5 COG7 COG1 MKK1 phosphatidylinositol VPS41 GEA1 200 ATG18 MTL1 3-kinase PEP5 ARF1 SEC7 COG8 COG6 SSK2 188 BCK1 VPS33 SSK22 complex, class 95 ATG11 127 VPS21 VPS16 ATG19 KSS1 FUS3 ATG15 III, type I VPS30 VAC8 CDC28 SIC1 TOR2 ATG24 STE7 MKK2 SLT2 163 YPK2 186 i. YPK1 29 nuclear outer RIM15 membrane SLG1 WSC3 162 i. late nucleophagy ATG14 SEC4 YPT1 ATG12 protein kinase A ATG27 TPK2 ATG10 ATG31 signaling ATG1/UKL1 TOR1 regulation of TRS31 ATG16 ATG3 signaling ATG17 BET5 SCH9 nuclear-transcribed ATG11 TPK1 NVJ1 D PKC1 TPK3 SWH1 YPT32 TRS33 autophagy ATG7 ATG5 ATG8 complex mRNA poly(A) TRS20 HSV2 ATG9 tail shortening ATG19 ATG22 ATG1 TRS23 ATG13 TAX4 PKH1 YPT31 PKH2 SEC16 BET3 SEC4 CUE5 SEC2 MON1 peroxisome IRS4 degradation protein targeting SEC17 to vacuole involved in ribophagy autophagy ATG34 microautophagy ATG36 MID2 PEP4 MKK2 BCK1 VTC1 BRE5 PKC1 VPS15 MEH1 NVJ1 RSP5 MKK1 WSC3 VTC3 UBP3 VPS34 ATG38 COG3 macroautophagy CMD1 SLG1 GEA1 VTC4 nucleophagy SLT2 GEA2 mitochondrion SLM4 VTC2 COG2 ARF1 degradation piecemeal TRS130 COG4 GENES (ovals) microautophagy YPT1 CDC48 ARF2 HSV2 of nucleus autophagosome Core autophagy genes regulation of VAM7 fusion autophagy ATG15 PTC6 VAM3 CCZ1 VAM6 ATG20 VPS41 WHI2 Other autophagy genes (GO) GTR2 negative VPS33 ATG33 YPT7 ATG21 RTG3 regulation of ICY2 UTH1 autophagy VPS30 Genes newly implicated by similarity GTR1 ATG16 AIM26 LMO1 PPM1 ATG10 ATG32 TOR1 ATG5 nucleus-vacuole late nucleophagy YIL165C positive ATG14 ATG7 autophagosome DCK1 TOR2 regulation of junction ATG13 TERMS (rectangles) ATG18 assembly YOR019W autophagy assembly ATG29 VAC8 i. FewSome Many regulation of ATG8 macroautophagy SEC18 ATG27 Genes annotated TRS85 SEC7 negative regulation of regulation of positive to (or below) term regulation of autophagosome macroautophagy assembly macroautophagy SHP1 ATG17 ATG23 NPR3 ATG2 SIC1 CLG1 ATG31 PHO85 PHO80 FAR11 IML1 ATG11 Unaligned Low High PCL5 RPD3 EPL1 KSP1 PCL1 ESA1 NPR2 Alignment score ATG3 ATG24 ATG9 AtgO 1.0 vs. GO ATG12 ATG4 ATG1

Figure 2. First-Generation Autophagy Ontology (A–C) AtgO 1.0 model of terms (rectangles), genes (ovals), and hierarchical interrelations among these (links). The rectangle size shows the number of genes annotated to that term; its color shows term alignment to GO: darker blue indicates higher similarity to GO; white terms/red outlines do not align. The oval color indicates the gene status. For compactness, the gene annotations are displayed only for terms AtgO:15 (B; no GO alignment) and AtgO:18 (C; aligned to GO: ‘‘macroautophagy’’). (D) Comparison to GO as curated from literature, showing GO terms and annotations for GO:0006914:autophagy and its descendants. The links of the same color are relations with the same parent term.

Term AtgO:9, which aligned to GO:0006914 (autophagy), encap- Analysis of Model Reveals Genetic Interactions as Most sulated 234 genes organized into 82 subterms (Figures 2A–2C), Effective New Data including 19 of the 20 core autophagy genes and 67 other genes Next, we determined which new network data would be most annotated to GO:‘‘autophagy’’ (Figure 2D). Also included were capable of improving the AtgO model in further experiments (Fig- 148 genes newly associated with autophagy; within these, we ure 1Aiv). For this purpose, we removed all data of a given type observed a significant enrichment for functions in cell cycle (51 and evaluated the resulting decrease in model performance, genes, p < 1012), cellular response to stress (42 genes, p < measured as the ability to capture the GO reference (Figure 3A). 1012), and vesicle transport (47 genes, p < 1024), placing these The largest contribution was from protein-protein interactions, functions in broad relation to autophagy. closely followed by genetic interactions; that is, recovery of GO This AI-MAP protocol is available in a Jupyter Notebook at depends most strongly on data of these two types, making these http://atgo.ucsd.edu; notebook execution constructs the AtgO more desirable. Beyond this absolute importance measure, we model described herein; however, general ontology models were also interested in the importance of data relative to the pre- can be constructed by running the notebook with new data sup- sent time, i.e., the extent to which adding another dataset of a plied by the user. given type is expected to improve the model. Therefore, we

764 Molecular Cell 65, 761–774, February 16, 2017 measured the decrease in model performance as individual New Targeted Interactions Markedly Improve the studies of a particular type were sequentially removed (Fig- Hierarchical Model ure 3B). The unit of one ‘‘study’’ was used to normalize for Next, we evaluated the extent to which the new interactions different temporal and financial costs of generating each type improved the AtgO ontology. When used as the sole source of of data; it roughly estimates the amount of data one -omics lab data, networks from each of the static conditions showed can produce in one study on average (i.e., data per lab study). some ability to reconstruct GO (r = 0.13–0.18; Figure 3G). The dif- Removing single studies of genetic interactions caused the ferential networks showed better performance (r = 0.23–0.31), steepest instantaneous decrease in performance, suggesting consistent with previous findings on the utility of differential that such studies may continue to provide the most information. interaction mapping (Bandyopadhyay et al., 2010; Ideker and A similar result was obtained when individual interactions rather Krogan, 2012). Integrating all static and differential conditions than whole studies were progressively removed (Figure S1, into a single network yielded the best correspondence with GO available online). Interestingly, gene expression data provided (r = 0.42), suggesting that multiple conditions reveal different as- very little information not captured by other data types, either pects of cell biology. Remarkably, this performance was better when removed as a group (Figure 3A) or one study at a time than all previous data combined (r = 0.30; Figure 3H). Integrating (Figure 3B). previous with new data performed best of all (r = 0.48). We desig- nate the ‘‘second-generation’’ ontology of 220 terms resulting Conditional Genetic Interaction Networks Incorporate from these integrated data as AtgO 2.0 (Figure 4). New Autophagy Genes Beyond its improved ability to reconstruct GO, we found that Guided by the above analysis, we designed a systematic screen the hierarchical structure of this model captured many potential for genetic interactions targeted at genes and conditions rele- new biological functions and relationships. In particular, analysis vant to autophagy (Figures 3C and 3D). Synthetic genetic array of AtgO 2.0 revealed that the majority of terms (56%) involved technology (SGA) (Tong and Boone, 2006) was used to query previously unknown biological findings, which we categorized 52 autophagy-related genes for genetic interactions against an into four broad types: (1) terms representing previously unknown array of 3,007 genes, covering approximately two-thirds of the subfunctions of autophagy (previously unknown groupings of non-essential yeast genome. SGA uses high-throughput robotic primarily known autophagy genes), (2) terms representing previ- colony pinning on agar to create and score growth of many dou- ously unknown subfunctions related to autophagy (groupings of ble gene deletion strains in parallel, here yielding 52 3 3,007 = genes previously attributed to diverse functions), (3) terms repre- 156,364 tests for gene-gene interaction (Table S1). These SGA senting superprocesses that integrate known processes, and (4) networks were created in three conditions: rapamycin, which terms representing a known process, but expanded by adding pharmacologically induces autophagy; amino-acid starvation, genes (Figure 5A; Table S2). In what follows, we survey brief, which metabolically induces autophagy; and an untreated con- but suggestive, experimental findings from several of these trol. An established computational workflow (Bean et al., 2014) types. We have constructed a web portal to AtgO 2.0 (http:// was used to assign quantitative S-scores to all gene pairs, with atgo.ucsd.edu/), which permits exploration of the autophagy hi- positive S-scores indicating faster than expected growth erarchy with visualization of the network data supporting (epistatic or suppressive interaction) and negative S-scores indi- each term. cating slower than expected growth (synthetic-sick or lethal interaction). SGA networks were also computed in ‘‘differential’’ A Key Regulator of Vesicular Trafficking, Gyp1, Is configurations (Bandyopadhyay et al., 2010), based on the Required for Autophagy-Related Pathways differences in S-scores between (1) rapamycin and untreated, AtgO 2.0 suggested a broader involvement of vesicle trafficking, (2) nitrogen starvation and untreated, and (3) nitrogen starvation docking, and fusion pathways in autophagy (Figure 4C): for and rapamycin. example, we noted AtgO:18 (‘‘vesicle transport machinery of Several array genes displayed strong differential interactions ER-EE-MVB-vacuole and autophagy’’; 37 genes), which inte- with core autophagy genes, most specifically upon rapamycin grated core autophagy or pexophagy genes with many Rab treatment (Figure 3E). These included SSD1, encoding an GTPases and GAPs known primarily for their function in Golgi mRNA-binding protein that represses translation (Jansen et al., transport (GYP1, YPT1, YPT31-32, and SEC4)(Jean and Kiger, 2009); DID4 and STP22, encoding subunits of the ESCRT com- 2012)(Figure 4Ci). GYP1, the GAP of this pathway, displayed sig- plexes, which are required for autophagy in humans, but had nificant genetic interactions with core autophagy genes under not yet been examined in yeast (Rusten and Stenmark, 2009); starvation and rapamycin treatment, including positive interac- GYP1, a GTPase-activating protein (GAP) (Du and Novick, tions between Gyp1 and the Atg9-recycling system (Atg9 and 2001); IRA2, an inhibitor of RAS-cAMP (Rødkaer and Faerge- Atg18), as well as with the two core Ubiquitin-like conjugation man, 2014); PIB2, a phosphatidylinositol(3)-phosphate binding systems (Atg3, Atg5, and Atg7) (Figure 3E); these interactions protein of unknown function (Shin et al., 2001); and YPL247C, were not observed in untreated conditions (Table S1). Gyp1 an uncharacterized open reading frame. To investigate whether had been previously localized to the Golgi (Du and Novick, these genes play a direct role in macroautophagy, we scored 2001); however, we observed that GFP-Gyp1 is also localized their null mutants by the Pho8D60 assay, which provides a quan- to the phagophore assembly site (PAS), the location of the auto- titative marker of autophagic flow into the vacuole (Noda et al., phagosome-generating machinery (Figure 6A). Functional ex- 1995). With the exception of YPL247C, autophagy showed a periments with a gyp1D strain revealed a requirement for Gyp1 clear dependence on all of these genes (Figure 3F). in PAS formation, as deletion of GYP1 caused mislocalization

Molecular Cell 65, 761–774, February 16, 2017 765 AB1.00 10 0.99

0.98 8 0.97

6 0.96 0.95 4 0.94

0.93 2 Genetic interactions 0.92

Performance decrease (%) Protein-protein interactions 0 0.91 Co-expression Fraction of maximum performance 0.90 Genetic Domain Genomiccontext 0 1 2 3 4 5 6 7 8 9 10 11 12 50 similarity Phylogenetic Protein-proteininteractions interactions Co-expression Predicted3D structure from co-occurence Studies removed Interaction data type removed Rapamycin C Untreated D 52 Starvation

Rapamycin - untreated

Starvation - untreated query genes Autophagy related Starvation - rapamycin query genes Autophagy related

0 0 Array genes 3007 Genetic interaction S-score Array genes 3007 ≤-3 -2 -1 0 1 2 ≥3 EF DID4 IRA2 GYP1 PIB2 SSD1 STP22 YPL247C 100 ATG16 0h ATG12 ATG10 80 4h ATG7 ATG5 ATG4 60 ATG3 ATG8 ATG14 60 activity (%) 60 activity 40

ATG18 Δ ATG9 ATG31

Pho8 20 ATG17 ATG13 ATG1 0 WT atg8Δ did4Δ ira2Δ gyp1Δ pib2Δ ssd1Δ stp22Δ ypl247cΔ GH 0.5 Static 0.5 Differential 0.4 0.4 Combined

0.3 0.3

0.2 0.2 Performance at Performance at reconstructing GO 0.1 reconstructing GO 0.1

0.0 0.0

ffled tion - ted tion -cin va a va y ffled reated re m ombined, Untreated Starvation Rapamycin apamycin - Star R Star ll conditions ffled GO unt unt rapa A combined All prior data This screen All priorthis data screen + All prior data + All prior data + this This screen shu this screen shu All conditions c screen, shu

Figure 3. Active Analysis of Data Types Leads to Conditional Genetic Interaction Maps (A) Performance decrease in AtgO 1.0 when all data of one type are excluded from the model. The performance is measured as the Pearson correlation between pairwise similarity scores derived from integrated data versus GO, focusing on gene pairs within the 492 autophagy-related genes (see main text). (B) Performance degradation as single studies of the indicated type (color) is cumulatively excluded. The mean (points) and SD (error bars) are calculated over 50 random sets of removed studies. See also Figure S1. (C and D) New genetic interaction maps between 52 autophagy query genes and 3,007 non-essential array genes in untreated conditions (rich media; C) as well as rapamycin and starvation conditions (D). The differential maps between each pair of conditions are also displayed. See also Table S1. (E) Differential genetic interactions (rapamycin, untreated) between core autophagy genes and implicated array genes. (legend continued on next page)

766 Molecular Cell 65, 761–774, February 16, 2017 191 transcription factors, A unknown relationship Regulators of casein kinase 2 bud neck proteins AtgO:220 autophagy membrane trafficking catalytic subunits and related processes promotion of condensin recruitment transcriptional membrane protein Rho1-dependent to replication fork 8 response to iron blockage site localization and polarized cell growth insertion / stress depletion salt tolerance kinases Golgi-localized, GENES (ovals) response ubiquitin related CAMK/CMGC kinases GET complex actin-regulating γ-adaptin ear palmitoylated AAA ATPases and process and kinase family spore morphogenesis vacuolar membrane cytoskeletal functions homology, Arf-binding Core autophagy genes specificity factors AtgO:185 vacuolar trafficking regulation proteins i. ankyrin proteins genes, unknown nucleosome staurosporine transport of repeat-containing, Other autophagy genes (previously in GO) casein kinase I-like CDK9-related acetyltransferase of Atg19-receptor subgroup resistance pathway palmitoyl transferase MAP kinase and a proteins cyclin-dependent ARFs and ARF H4 components and cargos proteins MAPK inactivator (no New genes implicated by similarity plasma EGO complex kinases binding proteins actin related proteins membrane-associated subunits + Pib2, Pep7 ADP ribosylation piccolo NuA4 histone published casein kinase I-like EGO complex factors actin and nucleation acetyltransferase membrane trafficking, relationship) proteins subunits and Pib2 guanine nucleotide factor complex subunits cell growth and stress negative regulation of ADP ribosylation exchange factors for Sec7-Arf1 golgi responses I Hog1 signaling TERMS (rectangles) Gtr1-Gtr2 GTPase factors involved in ARFs vesicle transport B complex Rho1 GTPase and Golgi vesicular regulation complex effector FewSome Many trafficking eIF2B subunits and chromosomal BET5 TRS23 regulator passenger complex regulation of TRS33 membrane trafficking, Genes annotated subunits polyamine uptake TRS130 TRS20 TRAPP Complex cell growth and stress eIF2B subunits to (or below) term BET3 responses II TRS31 TRS85 AtgO:138 late secretory GEA1 pathway (vesicle GYP1 SYS1 Ubp3-Bre5 ubiquitin GARP complex vacuolar H+-ATPase trafficking, docking New subprocess (autophagy) ATG32 protease complex subunits GET3 GYP7 and endosome to and fusion) ATG22 CLG1 BRE5 Ypt31/Ypt32 family golgi traffic SEC18 New subprocess (autophagy-related) VPS1 RAV1 UBP3 VPS51 VPS52 TLG2 PEP7 Rab GTPases SEC17 RIC1 VPS45 New superprocess (integrating known processes) ii. YPT32 membrane fusion conserved oligomeric 232 V-ATPase, V0 golgi (COG) complex PEP8 HOPS/CORVET Expanded process, new gene(s) added AtgO:140 domain subunits complexes and COG4 VMA3 localization YPT53 YPT31 Vma12/Vma22 VMA11 VPS8 Known process, not previously in GO assembly complex YPT52 COG2 COG1 COG5 VPS9 retromer complex VMA22 VPH2 VPS3 Known process, in GO COG8 cargo selection V-ATPase nucleotide COG3 COG6 COG7 subunits v. binding V1 subunits VTC1 vacuolar transporter HSV2 UTH1 VPS29 VMA2 VMA1 HOPS complex VTC4 chaperone complex VTI1 late secretory Q-snares VPS21 VPS33 VTC3 vacuolar transporter pathway / vacuolar VPS41 chaperone complex, membrane dynamics, PEP3 PKC1 REG1 SSK2 VTC2 VAM3 vacuolar docking and VAM3 cytosolic facing cell growth and stress PEP12 VAM7 VPS16 PCL1 components responses fusion PEP5 CDC7 CDC28 ATG32 GTR2 PDK1 homolog YPT7 VAM6 BCK1 GCN3 ATG20 ATG4 PKH1 Mon1-Ccz1 complex ATG18 KIN4 ATG29 kinases GEA1 HSL1 Intraluminal vesicle PTC6 KES1 PTC6 PKH2 C ATG17 YAK1 ATG2 neck formation CCZ1 MON1 MEK1 DOA1 RAD53 DNA damage vesicle (especially ARF1 IRS4 RSP5 checkpoint kinase ATG31 ADR1 DOA4 SEC2 DID4 PI4P kinases CHK1 golgi) and organelle NPR1 autophagy inhibitors / DUN1 effectors CDC14 KSS1 ATG23 VPS4 traffic / cell cycle and CDC15 KIN28 MCK1 nutrient sensing STT4 general amino acid VTI1 i. TPK1 Ypt1/Sec4 family Rab stress response YPK1 BRO1 AtgO:247 engulfment of Atg14-Vps30 GCN4 control pathway PKH3 CDC5 viii. LSB6 GTPases SEC4 KCC4 ATG24 selective autophagy cargos subcomplex of PI3P protein A kinases AtgO:96 GCN2 PDK1 homolog kinase complex I nucleus-vacuole junction YPT1 PKH2 ATG8 core machinery of TPK2 TPK3 VAC8 ATG12 Atg1-Atg13 complex kinases nuclear side of TRS85 AtgO:18 vesicle transport Ste11 kinase PKH1 autophagy RIM15 nucleus-vacuole VPS30 VPS30 ATG5 machinery of complex, MAPK ATG9 PI3P kinase complex junction ATG1 ATG5 YPT7 ATG9 ER-EE-MVB-vacuole ATG13 signaling ATG31 ATG14 II MAPK signaling and BEM1 NVJ1 and autophagy regulation of Msn2/4 Msn2/4-mediated ATG15 ATG18 SWH1 CLG1 nutrient-sensing ATG22 COG2 Atg8 activating 40 transcription yeast stress response ATG11 ATG16 vii. VPS38 Ste11 kinase VPS4 ATG17 response PHO80 complex and regulation AtgO:240 VPS34 complex, MAPK ATG11 GYP7 Atg12 conjugation VPS15 GET3 PEP7 components of HOG Atg1-Atg13 complex signaling GYP1 ATG3 sensing pathway STE50 RIM15 system VPS1 ATG13 AtgO:246 183 AtgO:106 iv. SCH9 VPS38 ATG7 STE50 COG6 kinases, unknown ATG1 Atg8 activating negative regulation of RIC1 ATG27 STE11 i. COG5 Ypt31/Ypt32 family relationship YGK3 NTH1 complex ATG12 ATG10 transcription from CLA4 PBS2 MSN4 YPK2 TLG2 Rab GTPases ATG3 RNA polymerase II TORC2 downstream golgi recruitment of YAK1 ATG7 cell cycle and Snf1 SKM1 calmodulin-binding RTG3 promoter effector kinases YPK1 Arl1 YPT31 STE11 TPS2 signaling pathway YPT32 motor transport MARK homolog TPK2 STE20 glycogen synthase MSN2 kinases vi. KIN28 PCL1 MAPK signaling and engulfment of MID2 MIG1 ARL1 ARL3 kinase 3 homologs MIG1 KIN1 TOS3 PCL5 TOR pathways regulators Mid2 subfamily of cell selective autophagy SHO1 KIN2 ADR1 KOG1 wall integrity sensors Ypt1/Sec4 family Rab SYS1 SAK1 septin-associated CDC14 MTL1 cargos SLT2 14-3-3 proteins GTPases membrane fusion ELM1 kinases KIC1 SSD1 MID2 Snf1 kinase subunits HSL1 Wsc family cell wall ATG24 ATG8 CBK1 sensors PTC1 MPS1 and regulator GIN4 CDC15 Hog signaling SEC4 YPT1 VTC1 ATG20 BMH1 REG1 KCC4 TOR complex RCK1 BMH2 SLG1 MAPKKKs CMK1 MCK1 cell cycle machinery catalytic subunits TOR1 MKK1 COG4 RIM11 Snf1 kinase complex, PCL6 WSC3 mitotic exit network PKC1 BCK1 Ubp3-Bre5 ubiquitin MYO2 MRK1 alpha and beta TOR2 Hog1 kinase and the SNF1 SWE1 kinases SSK22 protease complex subunits PBS2 SLT2 MKK2 SSK2 Rck2 kinase substrate CMK2 CMD1 CDC5 SIP2 PHO4 CAK1 HOG1 KSS1 DBF20 FUS3 UBP3 BRE5 iii. regulation of G1/S DBF2 STE7 RCK2 AtgO:160 phase transition HOG1 PHO85 Pho85-Pho80 SIC1 CDC28 PHO80 CDK-cyclin comploex

Figure 4. Second-Generation Ontology (A–C) AtgO 2.0 hierarchical model. AtgO:220:‘‘autophagy and related processes’’ and its descendants are displayed. The gene annotations are shown for branches beneath AtgO:11:‘‘membrane trafficking, cell growth, and stress responses II’’ (B) and AtgO:14:‘‘vesicle (especially golgi) and organelle traffic/cell cycle and stress response’’ (C). The term names have been manually curated by our team, and thus may differ from previous versions of AtgO or GO. See also Tables S2 and S3 and Figure S2. of GFP-Atg8 (Figure 6B) and an Atg8-processing defect (Fig- formation and rescues the autophagy defect of a Ypt1 GEF ure 6C), reduced maturation of prApe1 into mApe1 via selective mutant (Trs85, upstream of Ypt1), but not of a Ypt1 effector autophagy (Figure 6D), and impaired macroautophagy under ni- mutant (Atg11, downstream of Ypt1) (Lipatova et al., 2012). trogen starvation as measured by the Pho8D60 assay (Figure 3F). When Ypt1 was overexpressed in a gyp1D strain, the gyp1D These assays confirm a general requirement of Gyp1 in auto- defect in PAS formation was not rescued (Figure 6E). This indi- phagy-related pathways. cates that Gyp1 acts downstream of Ypt1 during autophagy, The Rab GTPase Ypt1 has also been implicated in autophagy similar to its function as the GAP for Ypt1 in ER-to-Golgi protein (Wang et al., 2015); Gyp1 may act as the heretofore unknown trafficking (Figure 6F). GAP on Ypt1 at the PAS. To test this possibility, we used a Finally, some of the same autophagy-deficient phenotypes Ypt1 overexpression system, which leads to increased PAS were observed for a vps1D strain (Figures 6C and 6D). VPS1

(F) Pho8D60 enzymatic activity measurement of gene deletions in (E). The samples were collected from growing cells (0 hr; mid-log phase in YPD) and after nitrogen starvation (4 hr SD-N). All activity measurements were normalized to Pho8D60 activity in the wild-type 4 hr sample (100%). The error bars indicate SD of three replicates. (G and H) Performance of reconstructing GO, using new data only (G) or integrating new with prior data (H). The performance is measured over gene pairs within the set of 52 autophagy query genes, in comparison to randomized data or GO (mean of 100 trials).

Molecular Cell 65, 761–774, February 16, 2017 767 AtgO:185 transport of Atg19-receptor cargos Figure 5. Characterization of AtgO 2.0 A AtgO:18 vesicle transport machinery of (Atg26) (A) Summary of AtgO terms by types of biological ER-EE-MVB-vacuole and autophagy findings. (Gyp1) 3% (B) Pairwise gene similarity matrix and resulting AtgO 2.0 Terms AtgO 2.0 hierarchy for a subset of core auto- 14% New subprocess (autophagy) 23% phagy genes. New subprocess (autophagy-related) New superprocess (integrating known processes) Expanded process, new gene(s) added understand the Atg8-Atg24 relationship, Known process, not previously in GO 34% 21% we have provisionally named AtgO:247 Known process, in GO 5% ‘‘engulfment of selective auto- phagy cargo.’’

AtgO:247 (Atg8/Atg24) Atg26 Is Required for Processing of Large Aggregates in the Cvt B ARF1 Analysis of hierarchical organization Pathway 7.0 to infer gene ontology (CliXO) ATG27 Term AtgO:185 expanded from two ATG10 genes, Atg11 and Atg19 (Figures 2Ci BCK1 To root ATG12 ARF1 and 7A), to include Atg26 and Atg27 (Fig- AtgO:219 ATG7 ures 4Ai and 7B), based on strong genetic engulfment of core machinery 5.0 selective of autophagy ATG3 autophagy cargo interaction profile similarity across the AtgO:240 183 ATG13 ATG24 ATG8 Atg12 new conditions. Three of these genes conjugation system (Atg11, Atg19, and Atg27; Yen et al., ATG1 AtgO:246 ATG10 ATG27 Atg8 activating 2007; Yorimitsu and Klionsky, 2005) ATG8 complex ATG12

Pairwise gene similarity score Other core autophagy have known roles in the cytoplasm-to- ATG24 ATG7 genes ATG3 vacuole targeting (Cvt) pathway, by which 2.5 BCK1 Atg1-Atg13 complex aggregates of the aminopeptidase pre- ATG1 ATG13 cursor prApe1 are transported to and pro- TG24 TG13 TG12 TG10 TG27 ATG8 ATG1 ATG3 ATG7 ARF1 BCK1 A A A A A cessed in the vacuole to mature Ape1. AtgO:185 implied involvement of the re- is also annotated to AtgO:18 and had been formerly impli- maining gene, Atg26, in transport of prApe1 aggregates to the cated in pexophagy (Mao et al., 2014); these results indicate vacuole. In the yeast Pichia pastoris, the Atg26 ortholog partici- it may have a more general role in autophagy-related pates in pexophagy, where it is required for the degradation of pathways. large peroxisomes; hence, its designation as an ‘‘Atg’’ gene (Nazarko et al., 2009; Oku et al., 2003). Engulfment of Selective Autophagy Cargos Involves Although previous work in S. cerevisiae had found no involve- Atg8 and Atg24 ment of Atg26 in standard autophagy assays (Cao and Klionsky, AtgO:247 was a new term in AtgO 2.0 that grouped ATG8, en- 2007), we observed that Atg26 co-localizes in distinct, overlap- coding a ubiquitin-like protein that localizes to isolation mem- ping puncta with Atg19, the known receptor for prApe1 aggre- branes and mature autophagosomes (Noda et al., 2010), with gates (Figure 7C). We then monitored processing of prApe1 in ATG24, a regulator of membrane fusion (Ano et al., 2005; Kanki wild-type and atg26D strains, as well as in atg1D cells, as a pos- and Klionsky, 2008), based on a combination of strong genetic itive control for disruption of prApe1 processing (Cheong et al., interaction profile similarity and co-expression (Figures 4Bi and 2008). Normal maturation of prApe1 was observed in wild- 5B). Working with the yeast Pichia pastoris, where the very large type, but not atg1D, cells (Figure 7D). The atg26D mutant also peroxisomes serve as a model cargo for imaging of autophagy showed processing defects, especially in cells overexpressing

(Oku and Sakai, 2008), we followed Atg24-GFP in both wild- prApe1, which produce larger aggregates (50 mM CuSO4; Fig- type and atg8D cells. We noted a change in Atg24 localization ure 7D). Furthermore, we found that atg26D cells contained specific to the atg8D genotype, in which Atg24 occupies a significantly more large prApe1 aggregates than did wild-type large perimeter of the peroxisome cargo, as opposed to only a cells (p = 0.03; Figures 7E and 7F); overexpression exacerbated few punctate structures in wild-type cells (Figure 6G). Atg8 this defect (5 mM CuSO4, p = 0.02; 50 mM CuSO4, p = 0.0003). function has been thought to depend on Atg7, an E1-like enzyme Using rapamycin to induce macroautophagy, wild-type cells that conjugates Atg8 to phosphatidylethanolamine (PE), a lipid. successfully processed all large prApe1 aggregates even when

Therefore, we also examined the localization of Atg24 in an they were greatly overexpressed (50 mM CuSO4; Figure 7F). In atg7D mutant, predicting that it would have the same phenotype contrast, many aggregates remained in the atg26D mutant after as that in atg8D cells. Surprisingly, Atg24 localization was rapamycin treatment (Figure 7G). Finally, in addition to prApe1, not affected in atg7D cells; thus, the new term appears to repre- the Cvt pathway transports the vacuolar hydrolases aminopepti- sent a separate activity of Atg8, independent of Atg8 conjugation dase 4 (Ape4) and the Ty1 transposon to the vacuole (Suzuki to PE. Although further investigations will be needed to fully et al., 2011; Yuga et al., 2011). We found that atg26D strains

768 Molecular Cell 65, 761–774, February 16, 2017 ABD

DIC GFP-Gyp1 BFP-Ape1 Merge DIC GFP-Atg8 Δ Δ Δ WT atg6 gyp1 WT vps1 WT

SD prApe1 Ape1 Actin

gyp1Δ SD + Rapamycin (1h) DIC GFP-Atg8 C E WT WT atg6Δ gyp1Δ vps1Δ Minutes 0 20 40 60 120 0 20 40 60 120 0 20 40 60 120 0204060120 in SD-N

Ypt1 GFP-Atg8 overexpression

Free GFP

Ypt1 overexpression Actin + gyp1Δ

F G Vacuole Atg24 Peroxisome DIC (FM4-64, Red) (GFP, Green) (BFP-SKL, Blue) Merge Ypt1 RAB-GDP Inactive GDP WT Pi GTP

Autophagy Gyp1 TRAPPIII GAP and GEF complex

GDP atg8Δ

GTP RAB-GTP Active Ypt1 Autophagy COG atg7Δ Effectors Atg11 complex

Figure 6. Analysis of Gyp1 and Atg8/Atg24 (A) Co-localization (yellow arrows) of Gyp1 and the PAS marker Ape1 in wild-type cells under nominal conditions (SD) or after rapamycin (SD + rapamycin) treatment, using fluorescence microscopy. (B) Atg8 localization in wild-type and gyp1D cells. (C and D) Atg8 cleavage (C) and prApe1 maturation assay (D) in wild-type, atg6D (positive control), gyp1D, and vps1D strains. (E) Atg8 localization in context of wild-type, Ypt1 overexpression, and gyp1D with Ypt1 overexpression after rapamycin treatment for 1 hr. (F) Model proposing the conversion of GDP-bound Ypt1 into the GTP-bound, active form, catalyzed by guanidine nucleotide exchange factor TRAPPIII during autophagy. Ypt1-GTP is recognized by at least two autophagy effectors, Atg11 and the COG complex, and is converted back to the GDP-bound, inactive formby the GTPase-activating protein Gyp1. (G) Localization of Atg24 in wild-type, atg7D, and atg8D strains of P. pastoris under pexophagy conditions. The scale bar represents 5 mm. had reduced processing for each of these cargos (Figures 7H but which had not yet been curated by GO (Figure 5A; repre- and 7I). These results support a role for Atg26 in the transport senting 23% of all terms). For example, AtgO:138 recovered the of several cargos that use Atg19 as their receptor, including large Ubp3p-Bre5p ubiquitin protease complex implicated in ribophagy prApe1-containing aggregates, Ty1, and Ape4; hence, we gave (Figure 4Bii) (Kraft et al., 2008); AtgO:160 recovered the Pho85- AtgO:185 the name ‘‘transport of Atg19-receptor cargos.’’ Pho80 CDK-cyclin complex, which regulates autophagy (Fig- ure 4Biii) (Yang et al., 2010); AtgO:106 recovered Skm1, Ste20, AtgO as an Automated Literature-Curation Device and Cla4, part of a complex that downregulates sterol uptake (Fig- By incorporating evidence from all available -omics studies, ure 4Biv) (Lin et al., 2009); and AtgO:140 recovered the Vma12- AtgO 2.0 recovered many functions already reported in literature, Vma22 assembly complex (Figure 4Bv) (Graham et al., 1998).

Molecular Cell 65, 761–774, February 16, 2017 769 ABPairwise gene similarity score Figure 7. Analysis of Atg26 Involvement in 5.6 ATG11 ATG11 the Transport of Atg19-Receptor Cargos

4.8 To root (A and B) The refinement of AtgO 1.0 (A) to AtgO ATG19 ATG19 2.0 (B), resulting in expansion of AtgO:185, which

ATG26 ATG26 AtgO:185 we name ‘‘transport of Atg19-receptor cargos.’’ 3.6 prApe1 aggregate processing (C) Co-localization of dTomato-Atg26 and GFP- ATG27 ATG27 2.8 Atg19 in wild-type cells before and after 1 hr of ATG11 ATG27 ATG19ATG26 rapamycin treatment. ATG11 ATG11 ATG27 ATG26 ATG19 ATG27 ATG26 ATG19 (D) Western blot of prApe1 and Ape1 across ge- D D C dTomato- D WT atg1Δ atg26Δ netic backgrounds (wild-type, atg1 , and atg26 ) DICAtg26 GFP-Atg19 Merge Hours in 0 3 6120 3 612 03612 with prApe1 expressed from a copper-inducible rapamycin promoter and incubated 16 hr in SD-Cu media plus

prApe1 copper (CuSO4), followed by removal from CuSO4 Ape1 0 and treatment with rapamycin for indicated times. (E) Fluorescence micrographs of cells expressing

Actin endogenous prApe1, prApe1-BFP from the endogenous promoter, and prApe1 from the 16 hrs pretreatment) +Rapamycin -Rapamycin

4, copper-inducible promoter. The cells incubated E Vacuole 16 hr in SD-Cu + 50 mm CuSO4, followed by 14 hr DIC(FM4-64)( ) Ape1-BFPp Mergeg prApe1 in SD-Cu media + rapamycin. M CuSO Ape1

μ 5 (F) Analysis of cells with constructs as in (E). The WT fraction of cells with observed prApe1 aggregates Actin (light blue) and large prApe1 aggregates >1 mM

∆ (red) after 16 hr incubation in SD-Cu media +

atg1 indicated CuSO4, followed by 14 hr incubation in prApe1 SD-Cu + rapamycin are shown. There were >250 Ape1 50 prApe1 over-expression (

∆ cells that were analyzed. The asterisk indicates significant difference from wild-type under the atg26 Actin same condition as judged by Fisher’s exact test. F G (G) Same as (F), but pretreatment in CuSO4, fol- 0.6 0.6 w/prApe1 aggregates lowed by 14 hr incubation in SD-Cu + rapamycin. w/prApe1 aggregates * * w/prApe1 aggregates >1 μm (H and I) Quantitative analysis of processed GFP 0.5 w/prApe1 aggregates >1 μm 0.5 for GFP-Ape4 and GFP-GAG Ty processing * * * 0.4 0.4 assay. The error bars represent SD of three repli- * 0.3 0.3 cates. The asterisks indicate significant difference * compared to wild-type as judged by the one-sided 0.2 0.2 Fraction of cells Fraction of cells t test (p < 0.05). * 0.1 * * 0.1 * * * * * * * 0.0 0.0 ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ WT WT WT WT WT WT atg1 atg1 atg1 atg1 atg1 atg1 atg26 atg26 atg26 atg26 atg26 atg26 nucleus (Dawaliby and Mayer, 2010; 0550 0550 Kvam and Goldfarb, 2004). Two of these prApe1 over-expression prApe1 over-expression genes (NVJ1 and SWH1) were not anno- (μM CuSO4, 16 hrs pretreatment) (μM CuSO4, 16 hrs pretreatment) HI tated as such in GO and thus at our request 100 100 were annotated to this term (GO:0034727). 80 * * * 80 As another example, we found that eight 60 60 * * proteins annotated to autophagy by 40 40 * * (% of WT) (% of WT) * AtgO, but not GO, already had literature 20 20 support for autophagy phenotypes: Hog1 GFP-Ape4 processing 0 0 WT atg11Δ atg19Δ atg26Δ atg27Δ Gag processing GFP-Ty1 WT atg11Δ atg19Δ atg26Δ atg27Δ (Prick et al., 2006), Pep12 (Kanki et al., 2009), Snf1 (Wang et al., 2001), Stt4 We submitted each of these terms to GO curators and, given the (Wang et al., 2012), Tpk1 (Budovskaya et al., 2004), Vps4 (Nebauer ample prior evidence, all were accepted for inclusion in the et al., 2007), Vps51 (Reggiori et al., 2003), and Vps52 (Reggiori Gene Ontology (GO:1990861, GO:1990860, GO:1990872, and et al., 2003). Here too, these new gene annotations were accepted GO:1990871, respectively). Other terms missing from GO, but for inclusion in GO. Thus, AI-MAP can automatically mine, struc- supported by literature, include AtgO:246, which captured an ture, and unify knowledge of a cellular process to create a central E1/E2 enzyme complex that activates Atg8 in a ubiquitin-like research resource that is complementary to literature-curated cascade (Figures 4Bvi and 5B), and AtgO:240, a second ubiqui- models. tin-like cascade activating Atg12 (Figures 4Bvii and 5B) (Kaiser et al., 2012). Numerous new AtgO gene annotations were also Application to Human Autophagy supported by prior studies: for instance, term AtgO:96 was as- To test the feasibility of the Active Interaction Mapping approach signed Vac8, Nvj1, and Swh1, which form the nucleus-vacuole in the study of human biology, we applied the same procedure junction (Figure 4Bviii) during piecemeal microautophagy of the described in yeast to a compendium of human data including

770 Molecular Cell 65, 761–774, February 16, 2017 gene expression profiling, protein-protein interactions, genetic as well as what type of information is most needed next (Fig- interactions, co-localization, and sequence similarity. As in ure 3B). After data generation, the value of the new dataset is yeast, we began with a seed set of genes (33) designated as systematically evaluated in the context of all existing data (Fig- having core functions common to non-selective and selective ures 3G and 3H). Here, such assessments pointed us to genetic autophagy (Jin and Klionsky, 2013). This process resulted in a interactions, leading us to generate a new targeted dataset, human autophagy ontology (hAtgO 1.0) of 1,452 genes and which was more powerful at recovering knowledge about auto- 1,664 terms, 173 of which align significantly to existing terms phagy in GO than all previous public datasets combined (Fig- in the human GO, and the rest of which represent potential ure 3H). In this regard, the significant information gained by new autophagy subprocesses and components (Figure S2A; including multiple conditions (Figure 3G) argues that condition- Table S3). Compared to yeast, a smaller percentage of the specific and differential interaction maps have provided a parti- ontology (10% versus 35%) aligned to existing GO terms, which cularly worthwhile bolus of data. may be due to better GO annotation of yeast than human, more In the future, many research communities, each focused on a complete data in yeast than human, or both. As in yeast, much of cellular process of interest, may find it useful to organize around the core machinery of autophagy was grouped together shared hierarchical models as a means for representing cell bio- (hAtgO:2962), which contains 14 core and 8 other genes (Fig- logical knowledge, communicating new findings, and designing ure S2A). Also as before, our confidence in the model is targeted datasets. Once a baseline model of a process is increased by the presence of recently discovered autophagy created, the value of any new dataset can be evaluated by its biology that is not yet captured by GO; for example, the Hunting- ability to improve this baseline. The value of new interaction tin protein, Htt, is placed among core autophagy proteins, mapping efforts can thus be rigorously evaluated rather than consistent with the recent discovery that this important disease assumed, and the design of these experiments can be guided gene acts physiologically as a scaffold for selective autophagy rationally, based on current knowledge and data. (Rui et al., 2015). Turning to the data most valuable for construct- ing hAtgO, we see a striking difference as compared to yeast, as STAR+METHODS the Active Interaction Mapping process relies primarily on gene expression profiling to construct the model of the human auto- Detailed methods are provided in the online version of this paper phagy system (Figure S2B). This result suggests the generality and include the following: of Active Interaction Mapping, as it can quickly be adapted to a new species (human) and can rely on the data most available d KEY RESOURCES TABLE and useful for that species and system. d CONTACT FOR REAGENT AND RESOURCE SHARING d EXPERIMENTAL MODEL AND SUBJECT DETAILS DISCUSSION B Yeast Strains d METHOD DETAILS This study provides a roadmap for how to progressively eluci- B Public Data for S. cerevisiae date the ontology of cellular functions underlying any biological B Data Integration process. Studying a new process is straightforward, requiring B Construction of Autophagy Ontology (AtgO) 1.0 starting knowledge of some of the genes involved and access B Genetic Interaction Mapping to public -omics datasets. At all points of the study, virtually B Construction of Autophagy Ontology (AtgO) 2.0 any genomic data can be analyzed since almost any data linked B Biochemical Studies to genes or proteins can inform gene-gene similarity networks. B Fluorescence Microscopy Studies The current AtgO 2.0 model, assembled entirely from network B Other Methods data, is larger and has differences in content and structure from B Human Autophagy Ontology (hAtgO) the literature-curated GO. Many of these differences imply new d QUANTIFICATION AND STATISTICAL ANALYSIS biological findings, as reported above. Other differences be- d DATA AND SOFTWARE AVAILABILITY tween the models may arise due to differences in policy between the AI-MAP process and human curation. For example, AtgO SUPPLEMENTAL INFORMATION groups core autophagy genes under the single term ‘‘core machinery of autophagy,’’ whereas GO repetitively annotates Supplemental Information includes two figures and three tables and can be each core gene to all non-selective and selective autophagy found with this article online at http://dx.doi.org/10.1016/j.molcel.2016. pathways, including ‘‘macroautophagy,’’ ‘‘microautophagy,’’ 12.024. and ‘‘nucleophagy.’’ Furthermore, GO generally does not group paralogous genes of similar function, whereas AtgO creates a AUTHOR CONTRIBUTIONS term for functionally similar paralogs, the same as for any func- tional gene set (e.g., ‘‘PDK homolog kinases’’ and ‘‘ADP ribosy- M.H.K., M.K.Y., and T.I. designed the study and developed conceptual ideas. lation factors’’). M.H.K. and M.F. constructed AtgO. K.M., J.-C.F., and K.L. performed the ge- netic interaction screens. M.H.K. and J.-C.F. curated AtgO 2.0 term names. An important aspect of Active Interaction Mapping is that het- R.B. and J.M.C. curated new GO terms and annotations. K.O. and B.D. erogeneous new and existing data are interwoven in the same created the http://atgo.ucsd.edu web portal. J.-C.F., M.H.K., and S.S. de- modeling framework. This enables a comprehensive assess- signed and performed biochemical experiments. M.H.K., J.-C.F., S.S., and ment of the current support for a biological model (Figure 3A), T.I. wrote the manuscript with input from other authors.

Molecular Cell 65, 761–774, February 16, 2017 771 ACKNOWLEDGMENTS Cheong, H., Nair, U., Geng, J., and Klionsky, D.J. (2008). The Atg1 kinase com- plex is involved in the regulation of protein recruitment to initiate sequestering This work was supported by grants from the NIH to T.I. (R01-ES014811, R01- vesicle formation for nonspecific autophagy in Saccharomyces cerevisiae. GM084279, and P50-GM085764), S.S. (R01-DK41737 and a Chancellor’s As- Mol. Biol. Cell 19, 668–681. sociates Chair), and J.M.C. (U41-HG001315 and U41-HG002273). Additional Collins, S.R., Schuldiner, M., Krogan, N.J., and Weissman, J.S. (2006). A strat- funding in support of trainees was provided to M.H.K. (F30-HG007618 and egy for extracting and analyzing large-scale quantitative epistatic interaction T32-GM007198) and M.K.Y. (T32-GM008806). We wish to thank Dr. Claudine data. Genome Biol. 7, R63. Kraft, Dr. Maho Niwa, Dr. Yoshinori Ohsumi, Dr. Scott Emr, and Dr. Takahiro Covert, M.W., Knight, E.M., Reed, J.L., Herrgard, M.J., and Palsson, B.O. Shintani for providing reagents. We acknowledge Zlatan Hodzic for tech- (2004). Integrating high-throughput and computational data elucidates bacte- nical work. rial networks. Nature 429, 92–96.

Received: February 19, 2016 Creixell, P., Schoof, E.M., Simpson, C.D., Longden, J., Miller, C.J., Lou, H.J., Revised: November 21, 2016 Perryman, L., Cox, T.R., Zivanovic, N., Palmeri, A., et al. (2015). Kinome-wide Accepted: December 22, 2016 decoding of network-attacking mutations rewiring cancer signaling. Cell 163, Published: January 26, 2017 202–217. Davidson, E.H., Rast, J.P., Oliveri, P., Ransick, A., Calestani, C., Yuh, C.H., REFERENCES Minokawa, T., Amore, G., Hinman, V., Arenas-Mena, C., et al. (2002). A genomic regulatory network for development. Science 295, 1669–1678. Ano, Y., Hattori, T., Oku, M., Mukaiyama, H., Baba, M., Ohsumi, Y., Kato, N., Dawaliby, R., and Mayer, A. (2010). Microautophagy of the nucleus coincides and Sakai, Y. (2005). A sorting nexin PpAtg24 regulates vacuolar membrane with a vacuolar diffusion barrier at nuclear-vacuolar junctions. Mol. Biol. Cell dynamics during pexophagy via binding to phosphatidylinositol-3-phosphate. 21, 4173–4183. Mol. Biol. Cell 16, 446–457. Du, L.L., and Novick, P. (2001). Yeast rab GTPase-activating protein Gyp1p Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., localizes to the Golgi apparatus and is a negative regulator of Ypt1p. Mol. Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al.; The Gene Ontology Biol. Cell 12, 1215–1226. Consortium (2000). Gene ontology: tool for the unification of biology. Nat. Genet. 25, 25–29. Dutkowski, J., Kramer, M., Surma, M.A., Balakrishnan, R., Cherry, J.M., Krogan, N.J., and Ideker, T. (2013). A gene ontology inferred from molecular Atias, N., Gershenzon, M., Labazin, K., and Sharan, R. (2014). Experimental networks. Nat. Biotechnol. 31, 38–45. design schemes for learning Boolean network models. Bioinformatics 30, € i445–i452. Ergun, A., Lawrence, C.A., Kohanski, M.A., Brennan, T.A., and Collins, J.J. (2007). A network biology approach to prostate cancer. Mol. Syst. Biol. 3,82. Bacon, R.A., Salminen, A., Ruohola, H., Novick, P., and Ferro-Novick, S. (1989). The GTP-binding protein Ypt1 is required for transport in vitro: the Graham, L.A., Hill, K.J., and Stevens, T.H. (1998). Assembly of the yeast vacu- Golgi apparatus is defective in ypt1 mutants. J. Cell Biol. 109, 1015–1022. olar H+-ATPase occurs in the endoplasmic reticulum and requires a Vma12p/ Vma22p assembly complex. J. Cell Biol. 142, 39–49. Bandyopadhyay, S., Mehta, M., Kuo, D., Sung, M.K., Chuang, R., Jaehnig, E.J., Bodenmiller, B., Licon, K., Copeland, W., Shales, M., et al. (2010). Greene, C.S., Krishnan, A., Wong, A.K., Ricciotti, E., Zelaya, R.A., Himmelstein, Rewiring of genetic networks in response to DNA damage. Science 330, D.S., Zhang, R., Hartmann, B.M., Zaslavsky, E., Sealfon, S.C., et al. (2015). 1385–1389. Understanding multicellular function and disease with human tissue-specific networks. Nat. Genet. 47, 569–576. Barrett, C.L., and Palsson, B.O. (2006). Iterative reconstruction of transcrip- tional regulatory networks: an algorithmic approach. PLoS Comput. Biol. Havugimana, P.C., Hart, G.T., Nepusz, T., Yang, H., Turinsky, A.L., Li, Z., 2, e52. Wang, P.I., Boutz, D.R., Fong, V., Phanse, S., et al. (2012). A census of human soluble protein complexes. Cell 150, 1068–1081. Barrett, T., Wilhite, S.E., Ledoux, P., Evangelista, C., Kim, I.F., Tomashevsky, M., Marshall, K.A., Phillippy, K.H., Sherman, P.M., Holko, M., et al. (2013). Ho, Y., Gruhler, A., Heilbut, A., Bader, G.D., Moore, L., Adams, S.L., Millar, A., NCBI GEO: archive for functional genomics data sets—update. Nucleic Taylor, P., Bennett, K., Boutilier, K., et al. (2002). Systematic identification of Acids Res. 41, D991–D995. protein complexes in Saccharomyces cerevisiae by mass spectrometry. Bean, G.J., Jaeger, P.A., Bahr, S., and Ideker, T. (2014). Development of ultra- Nature 415, 180–183. high-density screening tools for microbial ‘‘omics’’. PLoS ONE 9, e85177. Ideker, T., and Krogan, N.J. (2012). Differential network biology. Mol. Syst. Behrends, C., Sowa, M.E., Gygi, S.P., and Harper, J.W. (2010). Network orga- Biol. 8, 565. nization of the human autophagy system. Nature 466, 68–76. Ideker, T.E., Thorsson, V., and Karp, R.M. (2000). Discovery of regulatory inter- Bicknell, A.A., Tourtellotte, J., and Niwa, M. (2010). Late phase of the endo- actions through perturbation: inference and experimental design. Pac. Symp. plasmic reticulum stress response pathway is regulated by Hog1 MAP kinase. Biocomput. 5, 305–316. J. Biol. Chem. 285, 17545–17555. Ideker, T., Galitski, T., and Hood, L. (2001a). A new approach to decoding life: Breiman, L. (2001). Random forests. Machine Learning 45, 5–32. systems biology. Annu. Rev. Genomics Hum. Genet. 2, 343–372. Budovskaya, Y.V., Stephan, J.S., Reggiori, F., Klionsky, D.J., and Herman, Ideker, T., Thorsson, V., Ranish, J.A., Christmas, R., Buhler, J., Eng, J.K., P.K. (2004). The Ras/cAMP-dependent protein kinase signaling pathway reg- Bumgarner, R., Goodlett, D.R., Aebersold, R., and Hood, L. (2001b). Integrated ulates an early step of the autophagy process in Saccharomyces cerevisiae. genomic and proteomic analyses of a systematically perturbed metabolic J. Biol. Chem. 279, 20663–20671. network. Science 292, 929–934. Cahan, P., Li, H., Morris, S.A., Lummertz da Rocha, E., Daley, G.Q., and Jansen, J.M., Wanless, A.G., Seidel, C.W., and Weiss, E.L. (2009). Cbk1 regu- Collins, J.J. (2014). CellNet: network biology applied to stem cell engineering. lation of the RNA-binding protein Ssd1 integrates cell fate with translational Cell 158, 903–915. control. Curr. Biol. 19, 2114–2120. Cao, Y., and Klionsky, D.J. (2007). Atg26 is not involved in autophagy-related Jean, S., and Kiger, A.A. (2012). Coordination between RAB GTPase and pathways in Saccharomyces cerevisiae. Autophagy 3, 17–20. phosphoinositide regulation and functions. Nat. Rev. Mol. Cell Biol. 13, Carro, M.S., Lim, W.K., Alvarez, M.J., Bollo, R.J., Zhao, X., Snyder, E.Y., 463–470. Sulman, E.P., Anne, S.L., Doetsch, F., Colman, H., et al. (2010). The transcrip- Jin, M., and Klionsky, D.J. (2013). The core molecular machinery of autopha- tional network for mesenchymal transformation of brain tumours. Nature 463, gosome formation. In Autophagy and Cancer, H.G. Wang, ed. (New York: 318–325. Springer), pp. 25–45.

772 Molecular Cell 65, 761–774, February 16, 2017 Kaiser, S.E., Mao, K., Taherbhoy, A.M., Yu, S., Olszewski, J.L., Duda, D.M., Mao, K., Liu, X., Feng, Y., and Klionsky, D.J. (2014). The progression of Kurinov, I., Deng, A., Fenn, T.D., Klionsky, D.J., and Schulman, B.A. (2012). peroxisomal degradation through autophagy requires peroxisomal division. Noncanonical E2 recruitment by the autophagy E1 revealed by Atg7-Atg3 Autophagy 10, 652–661. and Atg7-Atg10 structures. Nat. Struct. Mol. Biol. 19, 1242–1249. Nazarko, T.Y., Farre´ , J.C., and Subramani, S. (2009). Peroxisome size provides Kanki, T., and Klionsky, D.J. (2008). Mitophagy in yeast occurs through a se- insights into the function of autophagy-related proteins. Mol. Biol. Cell 20, lective mechanism. J. Biol. Chem. 283, 32386–32393. 3828–3839. Kanki, T., Wang, K., Baba, M., Bartholomew, C.R., Lynch-Day, M.A., Du, Z., Nebauer, R., Rosenberger, S., and Daum, G. (2007). Phosphatidylethanolamine, Geng, J., Mao, K., Yang, Z., Yen, W.L., and Klionsky, D.J. (2009). A genomic a limiting factor of autophagy in yeast strains bearing a defect in the carboxypep- screen for yeast mutants defective in selective mitochondria autophagy. tidase Y pathway of vacuolar targeting. J. Biol. Chem. 282, 16736–16743. Mol. Biol. Cell 20, 4730–4738. Noda, T., and Klionsky, D.J. (2008). The quantitative Pho8Delta60 assay of Karr, J.R., Sanghvi, J.C., Macklin, D.N., Gutschow, M.V., Jacobs, J.M., Bolival, nonspecific autophagy. Methods Enzymol. 451, 33–42. B., Jr., Assad-Garcia, N., Glass, J.I., and Covert, M.W. (2012). A whole-cell Noda, T., Matsuura, A., Wada, Y., and Ohsumi, Y. (1995). Novel system for computational model predicts phenotype from genotype. Cell 150, 389–401. monitoring autophagy in the yeast Saccharomyces cerevisiae. Biochem. Kim, Y.A., Wuchty, S., and Przytycka, T.M. (2011). Identifying causal genes Biophys. Res. Commun. 210, 126–132. and dysregulated pathways in complex diseases. PLoS Comput. Biol. 7, Noda, N.N., Ohsumi, Y., and Inagaki, F. (2010). Atg8-family interacting motif e1001095. crucial for selective autophagy. FEBS Lett. 584, 1379–1385. Kim, H., Shin, J., Kim, E., Kim, H., Hwang, S., Shim, J.E., and Lee, I. (2014). Oku, M., and Sakai, Y. (2008). Pexophagy in Pichia pastoris. Methods YeastNet v3: a public database of data-specific and integrated functional Enzymol. 451, 217–228. gene networks for Saccharomyces cerevisiae. Nucleic Acids Res. 42, Oku, M., Warnecke, D., Noda, T., Muller,€ F., Heinz, E., Mukaiyama, H., Kato, D731–D736. N., and Sakai, Y. (2003). Peroxisome degradation requires catalytically active King, R.D., Whelan, K.E., Jones, F.M., Reiser, P.G., Bryant, C.H., Muggleton, sterol glucosyltransferase with a GRAM domain. EMBO J. 22, 3231–3241. S.H., Kell, D.B., and Oliver, S.G. (2004). Functional genomic hypothesis gener- Papinski, D., Schuschnig, M., Reiter, W., Wilhelm, L., Barnes, C.A., Maiolica, ation and experimentation by a robot scientist. Nature 427, 247–252. A., Hansmann, I., Pfaffenwimmer, T., Kijanska, M., Stoffel, I., et al. (2014). King, R.D., Rowland, J., Oliver, S.G., Young, M., Aubrey, W., Byrne, E., Liakata, Early steps in autophagy depend on direct phosphorylation of Atg9 by the M., Markham, M., Pir, P., Soldatova, L.N., et al. (2009). The automation of sci- Atg1 kinase. Mol. Cell 53, 471–483. ence. Science 324, 85–89. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Kitano, H. (2002). Systems biology: a brief overview. Science 295, 1662–1664. Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., et al. (2011). Scikit-learn: Klionsky, D.J., Cueva, R., and Yaver, D.S. (1992). Aminopeptidase I of machine learning in python. J. Mach. Learn. Res. 12, 2825–2830. Saccharomyces cerevisiae is localized to the vacuole independent of the Pfaffenwimmer, T., Reiter, W., Brach, T., Nogellova, V., Papinski, D., secretory pathway. J. Cell Biol. 119, 287–299. Schuschnig, M., Abert, C., Ammerer, G., Martens, S., and Kraft, C. (2014). Kraft, C., Deplazes, A., Sohrmann, M., and Peter, M. (2008). Mature ribosomes Hrr25 kinase promotes selective autophagy by phosphorylating the cargo re- are selectively degraded upon starvation by an autophagy pathway requiring ceptor Atg19. EMBO Rep. 15, 862–870. the Ubp3p/Bre5p ubiquitin protease. Nat. Cell Biol. 10, 602–610. Picotti, P., Cle´ ment-Ziza, M., Lam, H., Campbell, D.S., Schmidt, A., Deutsch, Kramer, M., Dutkowski, J., Yu, M., Bafna, V., and Ideker, T. (2014). Inferring E.W., Ro¨ st, H., Sun, Z., Rinner, O., Reiter, L., et al. (2013). A complete mass- gene ontologies from pairwise similarity data. Bioinformatics 30, i34–i42. spectrometric map of the yeast proteome applied to quantitative trait analysis. Kvam, E., and Goldfarb, D.S. (2004). Nvj1p is the outer-nuclear-membrane Nature 494, 266–270. receptor for oxysterol-binding protein homolog Osh1p in Saccharomyces Pierson, E., Koller, D., Battle, A., Mostafavi, S., Ardlie, K.G., Getz, G., Wright, cerevisiae. J. Cell Sci. 117, 4959–4968. F.A., Kellis, M., Volpi, S., and Dermitzakis, E.T.; GTEx Consortium (2015). Lefebvre, C., Rieckhof, G., and Califano, A. (2012). Reverse-engineering hu- Sharing and specificity of co-expression networks across 35 human tissues. man regulatory networks. Wiley Interdiscip. Rev. Syst. Biol. Med. 4, 311–325. PLoS Comput. Biol. 11, e1004220. Leiserson, M.D., Vandin, F., Wu, H.T., Dobson, J.R., Eldridge, J.V., Thomas, Prick, T., Thumm, M., Haussinger,€ D., and Vom Dahl, S. (2006). Deletion of J.L., Papoutsaki, A., Kim, Y., Niu, B., McLellan, M., et al. (2015). Pan-cancer HOG1 leads to Osmosensitivity in starvation-induced, but not rapamycin- network analysis identifies combinations of rare somatic mutations across dependent Atg8 degradation and proteolysis: further evidence for different pathways and protein complexes. Nat. Genet. 47, 106–114. regulatory mechanisms in yeast autophagy. Autophagy 2, 241–243. Levine, B., and Yuan, J. (2005). Autophagy in cell death: an innocent convict? Reggiori, F., Wang, C.W., Stromhaug, P.E., Shintani, T., and Klionsky, D.J. J. Clin. Invest. 115, 2679–2688. (2003). Vps51 is part of the yeast Vps fifty-three tethering complex essential Liang, X.H., Jackson, S., Seaman, M., Brown, K., Kempkes, B., Hibshoosh, H., for retrograde traffic from the early endosome and Cvt vesicle completion. and Levine, B. (1999). Induction of autophagy and inhibition of tumorigenesis J. Biol. Chem. 278, 5009–5020. by beclin 1. Nature 402, 672–676. Reggiori, F., Komatsu, M., Finley, K., and Simonsen, A. (2012). Autophagy: Lin, M., Unden, H., Jacquier, N., Schneiter, R., Just, U., and Ho¨ fken, T. (2009). more than a nonselective pathway. Int. J. Cell Biol. 2012, 219625. The Cdc42 effectors Ste20, Cla4, and Skm1 down-regulate the expression of Resnik, P. (1995). Using information content to evaluate semantic similarity in a genes involved in sterol uptake by a mitogen-activated protein kinase-inde- taxonomy. Proceedings of the 14th International Joint Conference on Artificial pendent pathway. Mol. Biol. Cell 20, 4826–4837. Intelligence 1, 448–453. Lipatova, Z., Belogortseva, N., Zhang, X.Q., Kim, J., Taussig, D., and Segev, N. Rivera-Molina, F.E., and Novick, P.J. (2009). A Rab GAP cascade defines the (2012). Regulation of selective autophagy onset by a Ypt/Rab GTPase module. boundary between two Rab GTPases on the secretory pathway. Proc. Natl. Proc. Natl. Acad. Sci. USA 109, 6981–6986. Acad. Sci. USA 106, 14408–14413. Malleshaiah, M.K., Shahrezaei, V., Swain, P.S., and Michnick, S.W. (2010). The Rødkaer, S.V., and Faergeman, N.J. (2014). Glucose- and nitrogen sensing scaffold protein Ste5 directly controls a switch-like mating decision in yeast. and regulatory mechanisms in Saccharomyces cerevisiae. FEMS Yeast Res. Nature 465, 101–105. 14, 683–696. Manjithaya, R., Jain, S., Farre´ , J.C., and Subramani, S. (2010). A yeast MAPK Rubinsztein, D.C., DiFiglia, M., Heintz, N., Nixon, R.A., Qin, Z.H., Ravikumar, cascade regulates pexophagy but not other autophagy pathways. J. Cell Biol. B., Stefanis, L., and Tolkovsky, A. (2005). Autophagy and its possible roles in 189, 303–310. nervous system diseases, damage and repair. Autophagy 1, 11–22.

Molecular Cell 65, 761–774, February 16, 2017 773 Rubinsztein, D.C., Marin˜ o, G., and Kroemer, G. (2011). Autophagy and aging. AMP-activated protein kinase, and the cyclin-dependent kinase Pho85p. Mol. Cell 146, 682–695. Cell. Biol. 21, 5742–5752. Rui, Y.N., Xu, Z., Patel, B., Chen, Z., Chen, D., Tito, A., David, G., Sun, Y., Wang, K., Yang, Z., Liu, X., Mao, K., Nair, U., and Klionsky, D.J. (2012). Stimming, E.F., Bellen, H.J., et al. (2015). Huntingtin functions as a scaffold Phosphatidylinositol 4-kinases are required for autophagic membrane traf- for selective macroautophagy. Nat. Cell Biol. 17, 262–275. ficking. J. Biol. Chem. 287, 37964–37972. Rusten, T.E., and Stenmark, H. (2009). How do ESCRT proteins control auto- Wang, J., Davis, S., Menon, S., Zhang, J., Ding, J., Cervantes, S., Miller, E., phagy? J. Cell Sci. 122, 2179–2183. Jiang, Y., and Ferro-Novick, S. (2015). Ypt1/Rab1 regulates Hrr25/CK1d ki- nase activity in ER-Golgi traffic and macroautophagy. J. Cell Biol. 210, Schuldiner, M., Collins, S.R., Weissman, J.S., and Krogan, N.J. (2006). 273–285. Quantitative genetic analysis in Saccharomyces cerevisiae using epistatic miniarray profiles (E-MAPs) and its application to chromatin functions. Warde-Farley, D., Donaldson, S.L., Comes, O., Zuberi, K., Badrawi, R., Chao, Methods 40, 344–352. P., Franz, M., Grouios, C., Kazi, F., Lopes, C.T., et al. (2010). The GeneMANIA prediction server: biological network integration for gene prioritization and pre- Shin, M.E., Ogburn, K.D., Varban, O.A., Gilbert, P.M., and Burd, C.G. (2001). dicting gene function. Nucleic Acids Res. 38, W214–W220. FYVE domain targets Pib1p ubiquitin ligase to endosome and vacuolar mem- Yang, Z., Geng, J., Yen, W.L., Wang, K., and Klionsky, D.J. (2010). Positive or branes. J. Biol. Chem. 276, 41388–41393. negative roles of different cyclin-dependent kinase Pho85-cyclin complexes Suzuki, K., Morimoto, M., Kondo, C., and Ohsumi, Y. (2011). Selective auto- orchestrate induction of autophagy in Saccharomyces cerevisiae. Mol. Cell phagy regulates insertional mutagenesis by the Ty1 retrotransposon in 38, 250–264. Saccharomyces cerevisiae. Dev. Cell 21, 358–365. Yen, W.L., Legakis, J.E., Nair, U., and Klionsky, D.J. (2007). Atg27 is required Tong, A.H., and Boone, C. (2006). Synthetic genetic array analysis in for autophagy-dependent cycling of Atg9. Mol. Biol. Cell 18, 581–593. Saccharomyces cerevisiae. Methods Mol. Biol. 313, 171–192. Yorimitsu, T., and Klionsky, D.J. (2005). Atg11 links cargo to the vesicle-form- Tong, A.H., Lesage, G., Bader, G.D., Ding, H., Xu, H., Xin, X., Young, J., Berriz, ing machinery in the cytoplasm to vacuole targeting pathway. Mol. Biol. Cell G.F., Brost, R.L., Chang, M., et al. (2004). Global mapping of the yeast genetic 16, 1593–1605. interaction network. Science 303, 808–813. Yuga, M., Gomi, K., Klionsky, D.J., and Shintani, T. (2011). Aspartyl aminopep- Wang, Z., Wilson, W.A., Fujino, M.A., and Roach, P.J. (2001). Antagonistic con- tidase is imported from the cytoplasm to the vacuole by selective autophagy in trols of autophagy and glycogen accumulation by Snf1p, the yeast homolog of Saccharomyces cerevisiae. J. Biol. Chem. 286, 13704–13713.

774 Molecular Cell 65, 761–774, February 16, 2017 STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Anti-vacuolar aminopeptidase 1 (Ape1) Klionsky et al., 1992 N/A Anti-b-actin (C4) Santa Cruz Biotechnology Cat# sc-47778; RRID: AB_626632 Anti-GFP (JL8) Clontech Cat# 632381; RRID: AB_2313808 Anti-Mouse IgG-HRP BioRad Cat#170-6516; RRID: AB_11125547 Anti-Rabbit IgG-HRP BioRad Cat#170-1019 Biological Samples Saccharomyces cerevisiae (BY4741) knockout ThermoFisher Cat# 95400.BY4741 library Chemicals, Peptides, and Recombinant Proteins Rapamycin Sigma Cat# R0395 Nourseothricin (clonNAT) Jena Bioscience Cat# AB-102L Geneticin selective antibiotic (G418 sulfate) ThermoFisher Cat# 11811031 Phenylmethanesulfonyl fluoride (PMSF) Sigma Cat# P7626 r-nitrophenyl phosphate (pNPP) Sigma Cat# N9389 r-nitrophenol Sigma Cat# N7660 Pierce BCA Protein Assay Kit ThermoFisher Cat# 23225 Deposited Data YeastNet v3 Kim et al., 2014 http://www.inetbio.org/yeastnet/ GeneMania Warde-Farley et al., 2010 http://pages.genemania.org/data/ Gene Expression Omnibus (GEO) Barrett et al., 2013 https://www.ncbi.nlm.nih.gov/geo/ Gene Network Analysis Tool (GNAT) Pierson et al., 2015 http://mostafavilab.stat.ubc.ca/gnat/ Experimental Models: Organisms/Strains MATa leu2–3,112, trp1, ura3–52, Bicknell et al., 2010 MNY1121; WT Pho8D60 pho8::pho8D60, pho13::URA3 MATa leu2–3,112, trp1, ura3–52, Bicknell et al., 2010 MNY1123; atg8D Pho8D60 pho8::pho8D60, pho13::URA3 atg8D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-186; ypl247cD Pho8D60 pho8::pho8D60, pho13::URA3 ypl247cD::kanMX MATa, leu2–3,112, trp1, ura3–52, This study sAtg0-187; gyp1D Pho8D60 pho8::pho8D60, pho13::URA3 gyp1D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-188; pib2D Pho8D60 pho8::pho8D60, pho13::URA3 pib2D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-189; ira2D Pho8D60 pho8::pho8D60, pho13::URA3 ira2D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-190; ssd1D Pho8D60 pho8::pho8D60, pho13::URA3 ssd1D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-191; did4D Pho8D60 pho8::pho8D60, pho13::URA3 did4D::kanMX MATa leu2–3,112, trp1, ura3–52, This study sAtg0-192; stp22D Pho8D60 pho8::pho8D60, pho13::URA3 stp22D::kanMX MATa his3D1 leu2D0lys2D0 ura3D0, pDP103 This study sAtg0-520; WT+BFP-Ape1+GFP-Gyp1 (BFP-Ape1), NRB1322 (GFP-Gyp1) MATa his3D1 leu2D0lys2D0 ura3D0, This study sAtg0-1; WT+GFP-Atg8 pRS315-GFP-Atg8 (Continued on next page)

Molecular Cell 65, 761–774.e1–e5, February 16, 2017 e1 Continued REAGENT or RESOURCE SOURCE IDENTIFIER MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-12; atg6D+GFP-Atg8 atg6D::kanMX, pRS315-GFP-Atg8 MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-39; gyp1D+GFP-Atg8 gyp1D::kanMX, pRS315-GFP-Atg8 MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-48; vps1D+GFP-Atg8 vps1D::kanMX, pRS315-GFP-Atg8 MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-150; gyp1D+GFP-Atg8+OverYpt1 gyp1D::kanMX, pRS315-GFP-Atg8, NRB167 (Ypt1 Overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0, pRS315- This study sAtg0-152; WT+GFP-Atg8+OverYpt1 GFP-Atg8, NRB167 (Ypt1 Overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0, pTS551 This study sAtg0-79; WT+GFP-Ape4 (GFP-Ape4) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-80; atg11D+GFP-Ape4 atg11D::kanMX, pTS551 (GFP-Ape4) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-81; atg19D+GFP-Ape4 atg19D::kanMX, pTS551 (GFP-Ape4) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-82; atg26D+GFP-Ape4 atg26D::kanMX, pTS551 (GFP-Ape4) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-83; atg27D+GFP-Ape4 atg27D::kanMX, pTS551 (GFP-Ape4) MATa his3D1 leu2D0 lys2D0 ura3D0, pYEX-BX This study sAtg0-111; WT+GFP-Ty1 Gag (GFP–Ty1 Gag) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-112; atg11D+GFP-Ty1 Gag atg11D::kanMX, pYEX-BX (GFP–Ty1 Gag) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-113; atg19D+GFP-Ty1 Gag atg19D::kanMX, pYEX-BX (GFP–Ty1 Gag) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-114; atg26D+GFP-Ty1 Gag atg26D::kanMX, pYEX-BX (GFP–Ty1 Gag) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-115; atg27D+GFP-Ty1 Gag atg27D::kanMX, pYEX-BX (GFP–Ty1 Gag) MATa his3D1 leu2D0 lys2D0 ura3D0, pCK770 This study sAtg0-194; WT+GFP-Atg19+tdTomato-Atg26 (pRS415-GFP-Atg19), pCU416- tdTomato-Atg26 MATa his3D1 leu2D0 lys2D0 ura3D0, pCK782 This study sAtg0-3; WT+OverApe1 (Ape1 overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0, pCK782 This study sAtg0-4; WT+OverApe1+BFP-Ape1 (Ape1 overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-28; atg26D+OverApe1 atg26D::kanMX, pCK782 (Ape1 overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-29; atg26D+OverApe1+BFP-Ape1 atg26D::kanMX, pDP105 (Ape1 overexpression+BFP-Ape1) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-195; atg1D+OverApe1 atg1D::kanMX, pCK782 (Ape1 overexpression) MATa his3D1 leu2D0 lys2D0 ura3D0 This study sAtg0-196; atg1D+OverApe1+BFP-Ape1 atg1D::kanMX, pDP105 (Ape1 overexpression+BFP-Ape1) PPY12 his4 arg4 pPIC6a-BFP- This study sJCF322; WT+BFP-SKL+Atg24-GFP SKL::Blasticidinr, pJCF210(Atg24-GFP)::HIS4 atg7D::ScARG4 his4 arg4 pPICz-BFP- This study sJCF351; atg7D+ BFP-SKL+Atg24-GFP SKL::Zeocinr, pJCF210(Atg24-GFP)::HIS4 (Continued on next page)

e2 Molecular Cell 65, 761–774.e1–e5, February 16, 2017 Continued REAGENT or RESOURCE SOURCE IDENTIFIER atg8D::kanMX his4 arg4 pPICz-BFP- This study sJCF353; atg8D+ BFP-SKL+Atg24-GFP SKL::Zeocinr, pJCF210(Atg24-GFP)::HIS4 MATa can1D::MFA1pr-HIS3::MFa1pr-LEU2 Tong et al., 2004 Y3656, query strain ura3D0 leu2D0 his3D1 lys2D0) Recombinant DNA BFP-Ape1 Pfaffenwimmer et al., 2014 pDP103 pRS306 pADH1-GFP-HA-GYP1-CYC1term Rivera-Molina and Novick, 2009 NRB1322 pRS315-GFP-Atg8 This study pRS315-GFP-Atg8 Ypt1 Overexpression Bacon et al., 1989 pNB167 GFP-Ape4 Yuga et al., 2011 pTS551 pRS415-GFP-Atg19 Pfaffenwimmer et al., 2014 pCK770 GFP–Ty1 Gag Suzuki et al., 2011 pYEX-BX pCU416-tdTomato-Atg26 This study pCU416-tdTomato-Atg26 Ape1 overexpression Pfaffenwimmer et al., 2014 pCK782 Ape1 overexpression+BFP-Ape1 Pfaffenwimmer et al., 2014 pDP105 Atg24-GFP This study pJCF210 pPIC6a-BFP-SKL This study pPIC6a-BFP-SKL pPICz-BFP-SKL This study pPICz-BFP-SKL Software and Algorithms Active Interaction Mapping This Study http://atgo.ucsd.edu/download.html CliXO v0.3 Kramer et al., 2014 https://mhk7.github.io/clixo_0.3/ Ontology Alignment Algorithm Dutkowski et al., 2013 https://mhk7.github.io/alignOntology/ ImageJ NIH https://imagej.nih.gov/ij/ Bean Colony Analyzer Toolkit Bean et al., 2014 https://github.com/brazilbean/ Matlab-Colony-Analyzer-Toolkit R R Project http://www.R-project.org. Jupyter Jupyter Project http://jupyter.org/ Python https://www.python.org/ Scikitlearn http://scikit-learn.org/stable/

CONTACT FOR REAGENT AND RESOURCE SHARING

As Lead Contact, Trey Ideker is responsible for all reagent and resource requests. Please contact Trey Ideker at [email protected] with requests and inquiries.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Yeast Strains Saccharomyces cerevisiae and Pichia pastoris laboratory strains as described in Key Resources Table with growth conditions as appropriate for experiments as described in Method Details.

METHOD DETAILS

Public Data for S. cerevisiae Data from YeastNet v3 (Kim et al., 2014) were obtained from http://www.inetbio.org/yeastnet/, consisting of the following numbers of datasets classified by type: 50 co-expression, 1 domain co-occurrence, 1 genomic neighbor, 10 genetic interaction, 12 high- throughput protein-protein interaction, 1 phylogenetic profile, 1 protein network tertiary structure, 1 low-throughput protein-protein interaction, and 1 co-citation.

Molecular Cell 65, 761–774.e1–e5, February 16, 2017 e3 Data Integration Input datasets were provided as features to a random forest regression system for prediction of gene-gene pairwise similarity, trained using pairwise gene similarities in GO as a bronze standard. The network of predicted similarities we call the ‘data-derived gene sim- ilarity network.’ We aimed to use the Gene Ontology (GO) as a standard reference for learning to integrate data in a supervised fashion. Previously, we showed that a weighted network of gene-gene similarities derived from a gene ontology can be analyzed to reconstruct the full hierarchy of terms and term relations with near perfect precision and recall (Kramer et al., 2014). Because the network and the ontology can be interconverted, both contain the same information. Based on this equivalence, we guided our integration of data using a gene similarity network derived from GO instead of using GO directly. We derived this network by calculating the Resnik semantic similarity for each pair of genes (Resnik, 1995). Semantic similarities are calculated across the Bio- logical Process and Cellular Component branches of GO, downloaded on 6/2/2015 from http://www.geneontology.org. The 78 input datasets were integrated into a single network by using them as features in a supervised learning of the GO gene similarity network. Learning was performed using random forest regression (Breiman, 2001) from the Python scikit-learn package (Pedregosa et al., 2011). Predictions were made ‘‘out of bag,’’ i.e., the similarity of a gene pair was predicted based on information learned from other gene pairs. In effect, the random forest learns patterns in the networks which recapitulate information in GO. Hence, only relations in GO that can be systematically explained from data are included, any relations not justified by the data are excluded, and new relations not in GO are added when the network data support them (Figure 1B).

Construction of Autophagy Ontology (AtgO) 1.0 We analyzed the data-derived gene similarity network using the Clique Extracted Ontology (CliXO) algorithm, version 0.3 (http:// mhk7.github.io/clixo_0.3/; Kramer et al., 2014), with parameters a = 0.1 and b = 0.5. We identified significantly aligned terms between AtgO and GO using a previously described ontology alignment procedure with FDR threshold of 10% and minimum alignment score of 0.24 (http://mhk7.github.io/alignOntology/; Dutkowski et al., 2013). Alignment was performed first against the sub-branch of GO rooted at GO:‘‘autophagy’’ and then the entire GO, prepared as in Dutkowski et al. (2013); significant alignments to the autophagy sub-branch were given first priority.

Genetic Interaction Mapping Strain construction, plating of mutants, mutant selection, and scoring of genetic interactions in each condition were performed using a previously defined protocol (Collins et al., 2006; Schuldiner et al., 2006). Using a replica pinning robot, haploid double mutants rep- resenting crosses of 52 autophagy-related query genes and 3007 array mutants were grown on agar plates that were either un- treated, rapamycin treated or nitrogen deficient. Plates were photographed and colony sizes normalized, spatially corrected, and quantified using the Colony Analyzer Toolkit (Bean et al., 2014). For the three replicates per double mutant, the resulting experimental data were used to assign a quantitative S-score based on a modified t-test that compares the observed double mutant growth rate to that expected assuming no interaction exists, again using the Colony Analyzer Toolkit. Differential interactions were calculated by subtracting the S-scores for the same double mutant pair across conditions. Resulting S-scores available in Table S1.

Construction of Autophagy Ontology (AtgO) 2.0 The data-derived gene similarity network was recalculated incorporating both prior data and the new genetic interaction screen. The CliXO algorithm was applied with previous parameters. Term names in AtgO 2.0 were manually curated by our team, considering the names of aligned terms in GO, when available, as well as an extensive literature review. Reasoning and citations for term names are included in Table S2.

Biochemical Studies The prApe1 processing assay The prApe1 processing assays for Figure 6D: cells were grown for 16 hr in YPD medium (1% yeast extract, 2% peptone and 2% glucose) without exceeding exponential phase (OD600 above 1), 1 OD600 equivalents of cells were collected and Trichloroacetic acid (TCA) precip- itated using a final concentration of 12.5% TCA and incubated at least for 30 min at 80C. Next TCA-treated cells were pelleted by centrifugation (10 min at 21,000 g, room temperature), washed twice with ice-cold 80% acetone, and air-dried. After dissolving the pellets in 100 ml of 1% SDS/0.1 N sodium hydroxide, 20 mlof63 SDS sample buffer was added. Samples were boiled for 5 min. For SDS-polyacrylamide gel electrophoresis, 10 ml of each sample was used per lane and blotted onto nitrocellulose membranes. Western blots were blocked with 5% dry skim milk in Tris-buffered saline with 0.1% tween-20 (TTBS) and probed with anti-Ape1 (1:5000; rabbit; a gift from Dr. Daniel Klionsky) and anti-b-actin antibodies (1:5000; mouse; C4, Santa Cruz Biotechnology) diluted in TTBS with 5% dry skim milk and incubated overnight at 4C. Membranes were probed with anti-rabbit or anti-mouse IgG HRP (1:5000; BioRad). All blots were detected with HyGLO Quick Spray (Denville Scientific Inc.) on a Medical film processor SRX-101A (Konica). The prApe1 processing assays for Figure 7D: cells harboring the plasmid overexpressing prApe1 (pCK782, a gift from Dr. Claudine Kraft; Papinski et al., 2014) were grown for 16 hr in SD-Cu medium (0.67% yeast nitrogen base [YNB] without amino acids and copper, 2% glucose plus any required amino acid, nucleotide, or vitamin supplement) without exceeding exponential phase

(OD600 above 1) and transferred to SD-Cu medium with 200 nM Rapamycin and the indicated CuSO4 concentration at an OD600 of 1. One ml of cells was collected at different times, TCA precipitated and analyzed as described above. e4 Molecular Cell 65, 761–774.e1–e5, February 16, 2017 Pho8D60 assay The Pho8D60 assay (WT and atg8D strains were a gift from Dr. Maho Niwa) was performed as described previously (Manjithaya et al.,

2010; Noda and Klionsky, 2008). For the alkaline phosphatase assay, five OD600 equivalents of yeast cells were harvested, washed once with cold water and once with wash buffer (0.85% NaCl and 1mM PMSF) and resuspended in 500 ml lysis buffer (20 mM Pipes, pH 7.0, 0.5% Triton X-100, 50 mM KCl, 100 mM potassium acetate, 10 mM MgSO4, 10 mM ZnSO4, and 1 mM PMSF). The cells were lysed by vortexing at full speed 10 times with 250 ml equivalents of glass beads for 1 min and incubated for 1 min on ice in between. The lysate was centrifuged at 14,000 g for 5 min at 4C and the supernatant was recovered without disturbing the pellet. 100 ml of this supernatant was added to 400 ml reaction buffer (250 mM Tris-HCl, pH 8.5, 0.4% Triton X-100, 10 mM MgSO4, and 1.25 mM p-nitro- phenyl phosphate [pNNP]), and samples were incubated for 10-15 min at 30C before terminating the reaction by adding 500 mlof stop buffer (2 M glycine, pH 11). Production of pnitrophenol was monitored by measuring the absorbance at 400 nm (A400) using a spectrophotometer (DU-730; Beckman Coulter), and the concentration in nmol of pnitrophenol in the samples was calculated by graphing the adjusted A400 values relative to a standard curve of commercial pnitrophenol (0 to 100 nmol). Protein concentration in the extracts was measured with the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific), and the specific activity was calculated as nmol pnitrophenol/min/mg protein. The GFP-Atg8, GFP-Ape4 and GFP-Ty1 Gag processing assays For the GFP-Atg8, GFP-Ape4 and GFP-Ty1 Gag processing assays, yeast strains harboring the plasmids GFP-Atg8 (pRS315-GFP- Atg8, in house plasmid), GFP-Ape4 (pTS551, a gift from Dr. Takahiro Shintani; Yuga et al., 2011) and GFP-Ty1 Gag (pYEX-BX[GFP– Ty1 Gag], a gift from Dr. Yoshinori Ohsumi; Suzuki et al., 2011) were grown to mid log phase in SD medium (0.67% YNB without amino acids, 2% glucose plus any required amino acid, nucleotide, or vitamin supplement) lacking auxotrophic amino acids and then shifted to SD-N medium (0.17% YNB without amino acids and ammonium sulfate, and 2% glucose) at an OD600 of 1 for the indicated time. At each indicated time point, 1 mL of cell culture was removed and TCA precipitated as described above. The protein extracts were resolved by SDS-PAGE and proteins were revealed by western blotting with anti-GFP (1:2500; mouse; JL8, Clontech) and anti- b-actin (1:5000; mouse; C4, Santa Cruz Biotechnology) antibodies. Densitometry was performed using ImageJ software.

Fluorescence Microscopy Studies Cell culture conditions for microscopy were performed as described for biochemical studies. Fluorescence microscopy images were acquired at indicated times using a motorized fluorescence microscope (Axioskop 2 MOT, Carl Zeiss MicroImaging, Thornwood, NY) coupled to a monochrome digital camera (AxioCam MRm, Carl Zeiss MicroImaging) and processed using the AxioVision 4.8.2 soft- ware. To quantify prApe1 aggregates images, the fluorescence microscopy parameters such as exposition time, gain, and binning were kept constant. Pexophagy condition used for Pichia pastoris in Figure 6G: cells were grown in methanol medium (0.67% YNB without amino acids and 1% methanol plus any required amino acid) starting an OD600 of 0.2 for 15-16 hr and shifted to SD for 1 hr, then images were acquired.

Other Methods Cloning, gene deletion and yeast transformation were performed using standard methods.

Human Autophagy Ontology (hAtgO) Datasets utilized are listed in Table S3. All quantitative interaction/co-expression data were linearly transformed into 8 bit integers prior to data integration. Ontology constructed using available Active Interaction Mapping Jupyter notebook downloadable at atgo.ucsd.edu/download.html.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical tests performed using R and python. Significance of pearson correlation calculated using pearsonr function in python’s scipy package. Statistical GO gene set enrichments in the main text calculated using the hypergeometric test in R. For Figures 7H and 7I, statistical comparison of processed GFP for GFP-Ape4 and GFP-GAG Ty processing assay were calculated by comparing 3 replicates of each strain to 3 replicates of wild-type using the one sided t test in R.

DATA AND SOFTWARE AVAILABILITY

Genetic interaction S-scores as measured for 52 autophagy-related query genes and 3007 array genes under 3 static and 3 differ- ential conditions: Table S1. Active Interaction Mapping (AIM) software and data necessary to perform AIM in S. cerevisiae and H. sapiens are available for download as a Jupyter notebook at atgo.ucsd.edu/download.html. The AtgO 2.0 model is also available for browsing and exploration at atgo.ucsd.edu.

Molecular Cell 65, 761–774.e1–e5, February 16, 2017 e5