Pex35 is a regulator of peroxisome abundance

Of the cellular organelles, peroxisomes are among the most poorly understood. Peroxisomes are equipped with only a modest number of proteins (∼100) and have a diameter in the range of the resolution of light microscopes (∼200 nm) but they are extremely dynamic in size, number and protein content (Smith and Aitchison, 2013). In addition, peroxisomes are intriguing and central organelles due to their involvement in various cellular metabolic pathways, such as those of lipids, reactive oxygen species (ROS), carbohydrates, polyamines and amino acids (Wanders and Waterham, 2006), as well as their contribution to an increasing number of non-metabolic functions, such as innate immunity (Dixit et al., 2010; Magalhães et al., 2016; Odendall et al., 2014).

It is not surprising therefore that malfunction of peroxisomes leads to peroxisome-specific diseases (Braverman et al., 2013; Weller et al., 2003) and contributes to the pathology of Alzheimer's and Parkinson's diseases, aging, cancer, type 2 diabetes and heart failure (Beach et al., 2012; Colasante et al., 2015; Fransen et al., 2013; Islinger et al., 2012; Trompier et al., 2014). A useful distinction divides peroxisomal diseases into two groups: those in which single enzymatic functions are defective, and those where peroxisome biogenesis is defective per se. In peroxisome biogenesis disorders (PBD), genes involved in peroxisome formation or maintenance are mutated. Such genes are termed PEX genes, and the proteins they encode are termed peroxins. In humans, 16 PEX genes are known (Thoms and Gärtner, 2012; Wanders and Waterham, 2006). Nearly all of the human PEX genes have orthologs in yeasts; however, yeasts have additional PEX genes without obvious homologues in mammals. PEX34 from the yeast Saccharomyces cerevisiae, a recent addition to the growing list of PEX genes, encodes a peroxisomal membrane protein that functions together with the Pex11 family to control peroxisome number (Tower et al., 2011). The Pex11 family in yeast encompasses the orthologous proteins Pex11, Pex25 and Pex27, and is involved in peroxisome division and proliferation (Thoms and Erdmann, 2005). Therefore, the Pex11 family is also referred to as PEX11-type of peroxisome proliferators (PPP) (Kettelhut and Thoms, 2014; Thoms and Gärtner, 2012). Despite the identification of new genes involved in division and proliferation our understanding of peroxisome division is still partial, and hence in-depth characterization of the functions of known effectors as well as discovery of more players in this process is forthcoming.

Classically, PEX genes have been identified by complementation screens using mammalian cells or yeast cells (Just and Kunau, 2014). The earlier yeast screens for PEX genes were based on the finding that β-oxidation in yeast is exclusively located in the peroxisome so that only peroxisome-competent cells were supposed to be able to utilize oleic acid as a sole carbon source (Just and Kunau, 2014). More recently, it was shown that oleic acid becomes toxic when some peroxisomal functions are compromised (Lockshon et al., 2007) and genes with these functions can therefore not be identified in traditional screen settings. Also, owing to the experimental limitations inherent in classical forward genetic screens, yeast screens for peroxisomal functions have identified mainly non-essential genes with strong phenotypes. Other, non-PEX genes that are required for peroxisome biogenesis or function have been identified by more dedicated approaches. These include the Sec61 ER translocon complex as an entry point into the endomembrane system for peroxisomal membrane proteins (Thoms et al., 2011a); or the ADP-ribosylation factors (ARFs), small GTPases involved in the control of many vesicular processes, that are also required for peroxisome formation (Anthonio et al., 2009; Lay et al., 2006).

So how can we identify more proteins required for peroxisomal function? In recent years this has been achieved by both genome-wide screens and dedicated proteomic analyses. For example, proteomics screens of peroxisomes were able to identify new peroxisomal proteins in fungi (Marelli et al., 2004; Schäfer et al., 2001; Thoms et al., 2008; Yi et al., 2002), plants (Eubel et al., 2008; Fukao et al., 2002; Reumann et al., 2007, 2009) and mammalian tissues, such as rat liver (Islinger et al., 2007, 2010; Kikuchi et al., 2004), mouse liver and kidney (Mi et al., 2007; Ofman et al., 2006; Wiese et al., 2007), and human liver (Gronemeyer et al., 2013). In addition, several genome-wide genetic screens in yeast have shed light on new proteins and their potential functions. For example, fitness and transcriptome responses to oleic and myristic acid have been used to define the networks underlying the metabolic responses to these fatty acids and to identify new proteins involved in peroxisome function (Lockshon et al., 2007; Smith et al., 2002, 2006). A collection of kinase and phosphatase gene deletion strains has also been evaluated for the expression of the peroxisomal matrix protein enzyme 3-ketoacyl-CoA thiolase (Pot1) (Saleem et al., 2008). Following a similar approach, a genome-wide screen for alterations in the expression of a GFP-labeled peroxisomal matrix protein led to the identification of 169 genes required for all the processes that lead to faithful biogenesis and propagation of peroxisomes (Saleem et al., 2010). Other systematic approaches including high-throughput experiments with quantitative analysis and modeling have been used to complete the peroxisomal proteome and to improve the understanding of peroxisome metabolism and biogenesis (Cohen et al., 2014; Wolinski et al., 2009). More recently, systematic high-content screens and bioinformatics analyses led to the identification of a new peroxisomal targeting protein, Pex9, and two new peroxisome matrix proteins, Pxp1 and Pxp2 (Effelsberg et al., 2016; Nötzel et al., 2016; Yifrach et al., 2016; Yofe et al., 2016). With each new study revealing additional new peroxisomal proteins it is clear that we have not yet saturated our understanding of the peroxisomal proteome and new peroxisomal proteins await discovery.

In this work, we report the results of a high-content screen in the yeast Saccharomyces cerevisiae, tailored to uncover mutants affecting peroxisome abundance. By creating a systematic library of gene mutants in yeast cells expressing a fluorescent marker for peroxisomes, followed by automated microscopy and computational image analysis, we identified a previously uncharacterized protein affecting peroxisome abundance. We show that this protein, which we name Pex35, is a peroxisomal membrane protein that controls peroxisome abundance.

In order to identify genes that regulate peroxisome number or size, we wanted to quantify these parameters in mutants of every yeast gene. To this end we created a query strain that enables visualization of peroxisomes and is compatible with the Synthetic Genetic Array (SGA) automated mating approach in yeast (Cohen and Schuldiner, 2011; Tong and Boone, 2006). The strain expressed the peroxisomal membrane protein Pex3 tagged with the red fluorescent protein mCherry (Pex3–mCherry), and a fluorescent marker for the endoplasmic reticulum (ER), Spf1–GFP, to enable automated cell segmentation. The query strain was then crossed into nearly 5000 strains of the deletion library (Giaever et al., 2002) and more than 1000 strains of the DAmP (Decreased Abundance by mRNA Perturbation) hypomorphic allele library for essential genes (Breslow et al., 2008). Following sporulation and selection of haploids, we obtained about 6000 unique haploid strains, each mutated in one yeast gene and marked with Pex3–mCherry and Spf1–GFP. This screening library was subjected to automated image acquisition and analysis during growth in glucose. Using an automated high-content microscopy setup specifically configured for this purpose (Breker et al., 2013), we recorded three images covering hundreds of cells per strain (Fig. 1A). We then developed a computational pipeline using the green fluorescent ER marker to find individual cells and counted the mean size of peroxisomes as well as their number per cell (Fig. S1). When analyzing peroxisome size (Fig. 1B; for complete data see Table S1) we found that mean cell peroxisome size varied through a range of approximately six-fold. Although some peroxisomal proteins and proteins associated with peroxisome formation were enriched at the edges of the distribution, there was no strong enrichment or near-complete coverage of mutants in peroxisomal proteins or peroxisome biogenesis factors in the strains with decreased or increased peroxisome size. Rather, genes controlling mitochondrial function or localization were dramatically enriched. Among the top 130 mutants with enlarged peroxisomes, 80 were mitochondrial proteins (five-fold enrichment, P-value=2×10⁻³¹) (Eden et al., 2009), confirming a tight connection between peroxisomes and mitochondria (Cohen et al., 2014; Mohanty and McBride, 2013; Schrader et al., 2013; Thoms et al., 2009). Indeed it is known that when mitochondrial metabolism is compromised, cells can compensate by upregulating peroxisome metabolism and size (Eisenberg-Bord and Schuldiner, 2016). Interestingly, a mutant causing some of the smallest peroxisomes was defective in carbamoyl phosphate synthase (Cpa1), which is important for the biosynthesis of the amino acid citrulline. These findings indicate that peroxisome size is not the best parameter to identify new peroxisomal proteins. However, the information regarding size may be used in the future to further understand the factors that create the functional link between mitochondria and peroxisomes.

When focusing on peroxisome number, we noticed that mutants with increased peroxisome number were not enriched for peroxisomal proteins and seemed to have defects in central cellular processes like translation, chromatin control or cytoskeletal dynamics (Fig. 1C; Table S1). However, the strains with the strongest reduction in peroxisome number were strikingly enriched in known peroxisomal membrane proteins (105-fold enrichment, P-value=10⁻²⁶) (Eden et al., 2009) (Fig. 1C). Among those are strains known to affect peroxisome morphology (Δpex19, Δpex27, Δinp1 and Δvps1). At the other end of the spectrum, loss of Pex17, a protein that is associated with the docking complex for peroxisome matrix protein import (Huhse et al., 1998), caused a drastic increase in peroxisome number. Importantly, several unknown proteins affected peroxisome number to a similar extent as mutations in known peroxisomal proteins [Fig. 1C (inset) and Table 1]. Of those, the most highly ranking uncharacterized open reading frame was YGR168C.

We decided to focus on YGR168C since a deletion of this previously uncharacterized gene caused a decrease in peroxisome abundance that is as dramatic as defects in the known peroxisomal biogenesis and inheritance proteins Pex12 and Vps1 (Fig. 2A). To ensure that this phenotype is independent of the peroxisomal marker (Pex3–mCherry) that we chose for the initial screen, we deleted YGR168C in four strains with GFP-tagged peroxisomal markers, including both matrix and membrane proteins. We found that peroxisome recognition of our image analysis program depended on the intensity of the marker used and the analysis settings. Therefore, peroxisome abundance values from measurement of fluorescent markers should be stated in arbitrary units that may only be used to compare between strains with the same marker. However, regardless of the marker used, loss of YGR168C caused a significant reduction in peroxisome number per cell when compared to a control strain (Fig. 2B).

Next, we manipulated Ygr168c protein expression by putting it under control of the strong TEF1 promoter (TEF1_prom). Overexpression of YGR168C also led to a decrease in peroxisomal fluorescent puncta either in glucose (Fig. 2C) or oleic acid as a sole carbon source (Fig. 2D). The fact that both deletion and overexpression caused similar phenotypes was puzzling to us and we decided to follow peroxisomal architecture in more detail.

To better characterize the effect on peroxisomes produced by either deletion or overexpression of YGR168C we analyzed peroxisome size and morphology by stimulated emission depletion (STED) microscopy. Strains were transformed with a plasmid expressing a peroxisome-targeted GFP variant (EYFP–SKL) and were analyzed by fluorescence nanoscopy using nanobodies directed against GFP.

First, using STED microscopy we determined the size of peroxisomes with unprecedented accuracy. We used images of 71 peroxisomes to determine that the diameter of wild-type peroxisomes is 174 nm (±8 nm, s.e.m.), which is below the resolution of conventional light microscopy (Fig. 3A,D; Fig. S2A). Second, these images, as opposed to regular diffraction-limited analysis, revealed a clue as to the basis of the function of YGR168C. While the number of peroxisomes was reduced in Δygr168c, their size was not (Fig. 3B,D; Fig. S2A). However, peroxisomes in the overexpressing strain, which appeared to be enlarged in confocal microscopy analysis, showed a striking multi-lobular morphology when analyzed by STED microscopy (Fig. 3C,D; Fig. S2A). This phenotype may be due to hyper-fission of the organelles into tiny peroxisomes, which clump together giving the false impression of a single big peroxisome when visualizing cells by fluorescence microscopy; alternatively the multi-lobular phenotype may be due to the initiation of a proliferation process that is aborted at an early stage. Both scenarios are compatible with a role of Ygr168c in controlling peroxisome fission. Loss of YGR168C may therefore cause the opposite effect – reduction in fission – causing the observed reduction in peroxisome number. Taken together our results demonstrate that Ygr168c is a bona fide regulator of peroxisome number and size in yeast.

Since Ygr168c has such a strong effect on peroxisome number and size, we wanted to better characterize this protein. The C-terminally tagged strain could not be visualized above background autofluorescence when expressed under its own promoter in cells growing in glucose medium, indicating that expression of the protein in glucose medium is very low. When we expressed the protein with either a C- or N-terminal fluorescent tag under control of the medium strength TEF1 promoter, Ygr168c tagged on either terminus was localized to punctate structures that completely colocalized with the peroxisomal markers Pex3–mCherry or Pex3–GFP (Fig. 4A). The Ygr168c protein is predicted by TMHMM (Krogh et al., 2001) and TOPCONS (Bernsel et al., 2008) to be a membrane protein with five transmembrane domains and a large C-terminal domain facing the cytosol (Fig. 4B). We examined the expression and localization of GFP–Ygr168c in the absence of Pex19, a farnesylated membrane protein biogenesis factor (Rucktäschel et al., 2009). In Δpex19 cells, the residual GFP signal is distributed in the cytosol and the ER (Fig. 4C). These findings further support that Ygr168c is a peroxisomal protein that is withheld in or mislocalized to the ER in the absence of Pex19. Similar results were obtained in the absence of PEX3. GFP–Ygr168c expression was reduced and the residual GFP signal was distributed in the cytosol and the ER (Fig. S2B), in agreement with Ygr168c being a peroxisomal protein. To confirm that Ygr168c is a bona fide peroxisomal protein, we purified peroxisomes from oleate-induced wild-type cells with a GFP-marked Ygr168c. Ygr168c was recovered from the peroxisome fraction of the Nycodenz density gradient (24%–35% interface) and was absent in the cytosolic and the mitochondrial fractions (Fig. 4D).

The N-terminal domain of Ygr168c is predicted to contain a mitochondrial-targeting signal (MTS) (Emanuelsson et al., 2007). As some peroxisomal membrane proteins can be dually targeted to mitochondria, we analyzed the intracellular localization of an N-terminal fragment of Ygr168c comprising the first and second predicted transmembrane domain (TMD; amino acids 1 to 107) that contain the predicted MTS as a fusion protein with mCherry. Mitochondrial targeting of the truncated protein was not observed, suggesting that the N-terminal domain does not function as an MTS in this context or under these conditions. Interestingly, this fusion protein now localized to both the ER and peroxisomes (Fig. 4E), in agreement with the proposed passage of peroxisomal membrane proteins through the ER (van der Zand et al., 2010). Our finding also suggests that peroxisome-specific targeting information is in the C-terminal domain of Ygr168c.

The hallmark of many strains with a defect in maintaining normal peroxisome abundance is a growth defect when grown on oleic acid as the sole carbon source. Indeed, the deletion of YGR168C did not affect growth on glucose (Fig. S2C), yet showed reduced growth and reduced formation of halos on oleic acid medium, demonstrating the lack of oleic acid consumption (Fig. S2D). Interestingly, the overexpression did not give a similar phenotype (Fig. S2E). However, overexpression of a GFP-tagged isoform (TEF1_prom:GFP-YGR168C) did cause a phenotype suggesting that it somehow aggravates the effect of overexpression or creates a dominant negative (Fig. S2E). Growth curves of these strains on oleate as a sole carbon source confirm this finding: the growth of Δygr168 and TEF1_prom:GFP-YGR168C is slower than a control strain although is better than Δpex3 (Fig. 4F). Quantitative assessment of oleate consumption at the end point of the growth assay showed that mutant cells indeed utilized less of the available oleic acid (Fig. 4G).

Finally, we assayed the accumulation of peroxisomes in real time. The GAL1 promoter was introduced to regulate GFP–Ygr168c in a strain containing Pex3–RFP as a peroxisomal marker. Time-lapse microscopy following a transfer of this strain from glucose- to galactose-containing medium showed that GFP–Ygr168c accumulated in Pex3-marked peroxisomes. Furthermore, while cells that did not show GFP–Ygr168c expression had a normal peroxisome abundance, cells with GFP–Ygr168c expression displayed an abnormal phenotype of one enlarged peroxisome per cell over time (Fig. 4H).

Taken together, these data show that Ygr168c is a new peroxisomal membrane protein in yeast. Manipulation of the expression level of Ygr168c led to changes in peroxisome number and size, as well as to growth defects on oleate. As Ygr168c bears all signs of a peroxin, we named it Pex35.

Having identified a new peroxisomal membrane protein, we were eager to functionally characterize it and understand how it exerts its effect on peroxisome number and size. Since peroxisomes have an important role in lipid metabolism, we first performed a detailed lipidomic analysis but could find no significant changes in whole-cell lipid composition compared to a control strain (Table S2). This result suggests that Pex35 does not influence gross membrane lipid composition. Coupled with the relatively small growth defect of Δpex35 strains grown on oleic acid our results suggest that Pex35 is not directly involved in lipid metabolism.

We then performed a synthetic growth screen to try and uncover proteins or pathways that become important when cells are lacking Pex35. In this screen, a Δpex35 strain was crossed with the libraries comprising the ∼6000 deletion and hypomorphic strains that we had previously used for the visual screen. After mating, sporulation and selection of haploids, growth of the double mutants was compared to growth of the individual mutant strain (Table 2; Table S3). Focusing on synthetic lethal interactions where the double mutants are dead while each single mutant is alive, we saw that redox homeostasis becomes important in the absence of Pex35. For example, deletion of the glutathione (GSH) synthetase (GSH2) required for the synthesis of GSH from γ-glutamyl cysteine and glycine was synthetic lethal with Δpex35. This suggests that the mutant strongly depends on a reducing environment for survival. To test this, we measured the glutathione redox potential in the peroxisomal matrix using a newly described peroxisomal reporter (Elbaz-Alon et al., 2014). Notably, overexpression of Pex35 had already caused a substantial pro-oxidative shift under steady state conditions. Moreover, following treatment with a bolus of hydrogen peroxide, the probe remained more reduced in the deletion strain whereas it became more oxidized upon overexpression of Pex35 (Fig. S2F).

Other pathways that were over-represented were those of polyamine, cysteine, lysine and arginine biosynthesis (Table 2). To test how amino acid levels are altered in the absence of Pex35, we extracted metabolites from wild-type, Δpex35, TEF_prom:GFP-PEX35 and Δpex3 strains. We used high-performance liquid chromatography (HPLC)-based metabolomics to characterize amino acid levels in each strain and found that amino acid levels were indeed significantly affected (Fig. S2G). There was a remarkable 70% reduction in the level of citrulline in Δpex35, which was matched by a more than two-fold increase of citrulline in TEF_prom:GFP-PEX35. This was in agreement with the strong lethality of this strain in combination with mutants in the arginine biosynthetic pathway. These findings reinforce the notion that peroxisomes in yeast are important in regulating or executing aspects of amino acid metabolism (Natarajan et al., 2001), and highlights how defects in mitochondrial function (such as in arginine biosynthesis) can activate the retrograde response to enhance peroxisomal functions (Chelstowska and Butow, 1995).

So far, our data suggested that Pex35 is a peroxisome membrane protein affecting peroxisome number and size, as well as redox and amino acid homeostasis. To try and understand the mechanism by which Pex35 functions, we performed a systematic complementation assay using the two parts of a specially adapted dihydrofolate reductase (DHFR) enzyme (Tarassov et al., 2008). In this assay, one half of the DHFR enzyme is fused to Pex35 (‘bait’), which is expressed in ∼5700 yeast strains expressing fusions of each individual proteins with the other half of DHFR (‘prey’). When the Pex35 bait is in proximity to one of the library prey proteins, methotrexate-resistant DHFR activity is reconstituted, and cells can grow in the presence of methotrexate, an inhibitor of the endogenous essential DHFR activity (Fig. 5A).

When seeking for growth with an N-terminally tagged Pex35, we could not find any high-confidence complementing proteins, suggesting that the N-terminus is essential for the interaction of Pex35 with effector proteins localized at the peroxisome (data not shown) and reinforcing the previous notion that the overexpression of an N-terminally tagged GFP–Pex35 is more toxic than regular overexpression. However, when we performed the assay with a C-terminally tagged strain, either under the native Pex35 promoter or the strong TEF1 promoter, we identified 11 proteins previously known to be involved in peroxisomal functions among the top 16 hits (Fig. 5B,C; Table S4). These included Pex19, the peroxisomal membrane protein chaperone, supporting the notion that Pex35 is a membrane protein. Other proteins that were in the vicinity of Pex35 were peroxisomal membrane proteins or membrane-associated proteins (enrichment factor 70, P=1.6×10⁻¹⁷) (Eden et al., 2009). Especially intriguing was the identification of the Pex11 family proteins, Pex11 and Pex25, and Arf1, because these proteins have previously been shown to interact with each other, and all of them have a role in peroxisome fission, a process that would affect peroxisome number and size (Fig. 5B,C).

To examine whether the physical proximity with Arf1 can explain the effect of Pex35 on peroxisome abundance, we first tested whether the abundance of Arf1 or the localization of Arf1–mCherry were affected by deleting or overexpressing Pex35. We found that both deletion and overexpression of PEX35 dramatically increased Arf1 levels (Fig. S3A). Moreover, PEX35 also affected Arf1–mCherry distribution. In control strains, Arf1 localizes mainly in punctate structures, in agreement with the Golgi localization of Arf1 (Stearns et al., 1990). However, when PEX35 was deleted or the tagged form was overexpressed, Arf1 redistributed to various internal membranes (Fig. 6A). It is well known that, while the C-terminal tag of Arf1 creates a dysfunctional form, it does report accurately on its subcellular localization (Huh et al., 2003; Jian et al., 2010).

Arf1 has previously been implicated in peroxisome formation (Anthonio et al., 2009; Lay et al., 2005; Passreiter et al., 1998). We therefore investigated whether Pex35 was mediating its effect on peroxisome abundance through Arf1 by looking at the effect of Δpex35 deletion or overexpression on the background of loss of Arf1 (Δarf1). Deletion of ARF1 in the Δpex35 knockout lead to a marked increase in peroxisome size and aggravated the effect of Δpex35 on peroxisome number (Fig. 6B,C). Even more striking, the deletion of ARF1 in the Pex35-overexpressing strain rescued the peroxisome-enlarging effect of Pex35 overexpression and normalized the number of peroxisomes to control levels (Fig. 6B,C). These findings suggest that Pex35 exerts its function through a mechanism that requires Arf1.

Next, we wanted to analyze Arf1 at the organellar level. We purified peroxisomes by cell fractionation from strains that contained both GFP-tagged Pex3 as a peroxisome marker and RFP-tagged Arf1, either in a wild-type or in a Δpex35 background (Fig. S3B). The absence of the peroxisome marker in the cytosol or the mitochondria fraction showed that other compartments were free from peroxisomes. Arf1–RFP was clearly enriched in the peroxisome and mitochondria fractions, in agreement with the results of the quantitative whole-cell analysis (Fig. S3A). The distinction of peroxisomal and mitochondrial pools of Arf1, however, was not possible in this experiment, because the ‘light-mitochondrial’ faction co-fractionates with peroxisomes in the step gradient.

Our findings on the effects of Pex35 on Arf1 made us wonder whether altering PEX35 could exert an effect on the secretory pathway functions of Arf1 (Fig. 6A). In order to test whether Arf1-dependent anterograde traffic was affected, we tested secretion of the intra-ER chaperone Kar2 (a BiP homolog). The slow leak of Kar2 from the ER, through the Golgi and its secretion into the medium is a good way to observe dysregulated secretion (Copic et al., 2009). Indeed, overexpressing GFP–Pex35 (which led to a drastic increase in Arf1 levels) decreased secretion to 50% or less, similar to what is seen upon directly overexpressing Arf1 (Fig. 6D). Interestingly, the deletion of PEX35 (although also leading to Arf1 overexpression) could compensate for the effect of Arf1 overexpression, suggesting that in normal physiology PEX35 restrains Arf1 function (Fig. 6D). Equally striking was the finding that the deletion of PEX3 alone exerted a secretion defect in the same range, like strong activation of the unfolded protein response (UPR; seen by deletion of the ER protein SPF1), or the overexpression of Pex35 or Arf1 (Fig. 6D). To follow up this finding, we tested the localization of Arf1–RFP in a Δpex3 background. The deletion of PEX3 led to a surprising loss of focal Arf1 localization (Fig. 6E). At this point it is not clear whether this is due to a direct association of Arf1 with Pex3 or Pex35, or if it is due to the upregulation of Arf1 or the large proportion of Arf1 associated with peroxisomes that is altered upon Pex3 loss, but in all these cases, these data point to a strong role for peroxisomes in the function of the secretory pathway.

One of the current challenges in molecular cell biology is to integrate multiple levels of understanding spanning holistic omics approaches and detailed analysis of single molecular entities. In this study, we go through such an analysis, starting from a large-scale genetic screen with the complete deletion and DAmP collection, and end up with the identification of a new peroxisomal protein and insights into the mechanism by which it may regulate peroxisome abundance.

The initial screen relied on three automated stages: generation of a screening library comprising deletions and functional knockdowns of nearly all yeast genes, automated image acquisition, and a computational pipeline for image segregation and quantification. This approach complements the classical forward genetic screen based on growth on oleate, with the advantage of increased sensitivity gained by statistical analysis of quantitative data. Our screen was stringent as well as exhaustive – most other genes previously implicated in many aspects of peroxisome formation, inheritance and maintenance clustered on the top of our scoring list. We found that, although some strains defective in peroxisome proliferation have a strongly altered (reduced) peroxisome size, reduction in peroxisome number is a more exhaustive criterion to cover genes required for peroxisome formation and maintenance.

Of the genes with the strongest effect on peroxisome number, we decided to focus on YGR168C, which we termed PEX35. We show that Pex35 is a peroxisomal membrane protein that had not been identified previously, probably due to its low expression level in cells grown on glucose and its moderate, yet quantifiable, effect on peroxisome number and size. Deletion of PEX35 leads to a reduction in peroxisome number, and overexpression of PEX35 to an increase in hyper-fissioned aggregated tiny peroxisomes or to an early block in the proliferation of peroxisomes. These results, together with the oleate growth phenotypes, classify Pex35 as a bona fide peroxin.

We found that C-terminally tagged Pex35 localizes in the vicinity of the Pex11-type of peroxisome proliferators Pex11, Pex25 and Pex34, and the small GTPase Arf1. All of these proteins are known to influence peroxisome abundance (Lay et al., 2006; Thoms and Erdmann, 2005; Tower et al., 2011), so it is attractive to speculate the Pex35 is part of a regulatory unit that controls peroxisome fission, and therefore numbers, in the cell. The effect on Arf1 was particularly intriguing, because ARFs have previously been implicated in peroxisome formation, but their contribution has been poorly established. The mammalian ARF proteins Arf1 and Arf6 have been shown to be present on rat liver peroxisomes, and to bind to peroxisomes in a GTP-specific manner. Interference with either Arf1 or Arf6 changes peroxisome morphology (Anthonio et al., 2009; Lay et al., 2005; Passreiter et al., 1998). In yeast, the ARF proteins Arf1 and Arf3 are required for peroxisome fission albeit in opposing manners (Anthonio et al., 2009; Lay et al., 2005). Additional support of the involvement of ARFs in peroxisome function comes from the finding that the expression of a dominant-negative Arf1 mutant, Arf1Q71L, blocks protein transport from the ER to the peroxisome in plant cells (McCartney et al., 2005). However, whether ARFs are actively recruited to peroxisomes, how ARFs affect peroxisomes and how this affects general cellular trafficking fidelity is still a mystery.

Initially, Pex11 was suggested to be the receptor of ARFs and coatomer on peroxisomes (Passreiter et al., 1998). This notion was questioned when it was found that mutations that prevent coatomer binding to Pex11 had no effect on peroxisome function in yeast or trypanosomes (Maier et al., 2000). In addition, the COPI inhibitor, brefeldin A, does not inhibit peroxisome biogenesis (South et al., 2000; Voorn-Brouwer et al., 2001). However, today there is new evidence for the involvement of membrane coats in peroxisome function (David et al., 2013; Dimitrov et al., 2013; Lay et al., 2006) and it is becoming clear that peroxisome biogenesis may be distinct from peroxisome maintenance by fission. Moreover, as we saw in our own work, fluorescence microscopy alone may not be a sensitive enough readout for such changes, as sub-diffraction images can give more detailed information that may be easier to reconcile with genetic data.

In recent years, reports on non-vesicular functions of ARFs are becoming more common (Jackson and Bouvet, 2014). We found that either deletion or overexpression of Pex35 drives Arf1 out of its physiological localization, mirroring the phenotype of an increased GAP function or a reduced GEF function, because the active GTP-bound form of Arf1 is recruited to the membrane, whereas the inactive GDP-bound form is released from the membrane (Donaldson and Jackson, 2011). Of note, the Arf1-GEF Sec7 was also found in the proximity of Pex35 in our DFHR screen (Table S4, rank 29), although the significance of that interaction remains to be established. While deletion of Pex35 elevates Arf1 protein levels (Fig. S3A), we do not understand its mechanistic connection with Arf1. It may very well be that the upregulation of Arf1 is a consequence of reduced cellular function and cellular compensation mechanisms. In support of this, the ability of overexpression of Arf1 to reduce leakage of Kar2 is abolished in the absence of Pex35 (Fig. 6D).

Our findings illuminate both the role of Arf1 as a regulator of peroxisome abundance and its dependency on healthy peroxisome biogenesis. Pex35 may be the crucial link in these processes: loss of focal localization of Arf1 in a peroxisome-defective Δpex3 strain may be due to the loss of peroxisomal recruitment of Arf1 in this strain.

Intriguingly, Pex35 has domains that show homology with three human proteins (Fig. S4), two of which have membrane-shaping properties. One region of Pex35 is homologous to human Rtn1, an ER reticulon, and another with Epn4, an epsin-related protein that binds the clathrin coat. Both proteins shape membrane curvature, which is essential for the fission process. In addition, potential homology was also found to Pxmp4, a human peroxisomal protein. These areas of homology suggest that Pex35 functions have been conserved in humans and they are suggestive of possible mechanism of Pex35 function.

Our work suggests that regulation of peroxisome size and number is linked tightly with secretory pathway functions. Physiological secretion is dependent on the presence of peroxisomes. On the other hand, peroxisome formation itself requires a functional secretory pathway. In this way, the metabolic status of the cell could be coupled to secretion and growth in a simple and elegant mechanism. Such a cross-talk between these two systems could have dramatic effects on cellular homeostasis and hence would impact peroxisomes in health and disease.

All yeast strains in this study are based on the BY4741/2 laboratory strains (Brachmann et al., 1998). Strains created in this study are listed in Table S5. All genetic manipulations were performed using the lithium acetate, polyethylene glycol (PEG) and single-stranded DNA (ssDNA) method for transforming yeast strains (Gietz and Woods, 2006) using plasmids previously described (Janke et al., 2004; Longtine et al., 1998), as listed in Table S6. Primers for genetic manipulations and their validation were designed using Primers-4-Yeast (Yofe and Schuldiner, 2014), as listed in Table S7.

A query strain was constructed by introduction of an ER marker, Spf1–GFP, and a peroxisome marker, Pex3–mCherry (yMS1233). The query strain was then crossed into the yeast deletion and hypomorphic allele collections (Breslow et al., 2008; Giaever et al., 2002) by the synthetic genetic array method (Cohen and Schuldiner, 2011; Tong and Boone, 2006). Representative strains of the resulting screening library were validated by inspection for the presence of peroxisomal mCherry and ER-localized GFP expression. This procedure resulted in a collection of haploid strains used for screening mutants that display peroxisomal abnormalities.

The screening collection, containing a total of 5878 strains, was visualized in SD medium during mid-logarithmic growth using an automated microscopy setup as described previously (Breker et al., 2013).

After acquisition, images were analyzed using the ScanR analysis software (Olympus), by which single cells were recognized based on the GFP signal, and peroxisomes were recognized based on the mCherry signal (Fig. S1). Measures of cell and peroxisome size, shape and fluorescent signals were extracted, and outliers were removed. A total of 5036 mutant strains were successfully analyzed, with a mean of 126±56 (s.e.m.) cells per strain (minimum of 30) that passed filtering (Table S1). For each mutant strain, the mean number of peroxisomes recognized by the software per cell was extracted, as well as the mean area (in pixels) of peroxisomes. Note that these are arbitrary units that may only be used to compare between strains, as resulting values are dependent on the parameters of image analysis and of the fluorescent markers used.

For follow up microscopy analysis, yeast growth conditions were as described above for the high-content screening. Imaging was performed using the VisiScope Confocal Cell Explorer system, composed of a Zeiss Yokogawa spinning disk scanning unit (CSU-W1) coupled with an inverted Olympus microscope (IX83; ×60 oil objective; excitation wavelength of 488 nm for GFP, and 561 nm for mCherry). Images were taken by a connected PCO-Edge sCMOS camera, controlled by VisView software. In the case of galactose induction of GALp-GFP-PEX35 (Fig. 3F), yeast were transferred to the microscopy plate, and after adhering to the glass bottom, medium containing 2% galactose was used to wash the cells and during the time-lapse imaging.

Strains were transformed with EYFP–SKL (plasmid PST1219) and analyzed as described previously (Kaplan and Ewers, 2015) using GFP nanobodies coupled to Atto647N (gba647n-100, Chromotek, Planegg-Martinsried). Slides were analyzed on a custom-made STED setup described previously (Göttfert et al., 2013). Images show unprocessed raw data. STED images shown in Fig. 3 were rescaled by a factor of two using bicubic interpolation (Adobe Photoshop). The size of yeast peroxisomes was calculated as follows. Random peroxisomes were marked by manually drawing a ROI and the center of masses was determined. Peroxisome diameter was measured after drawing a line scan through the minor axis of the peroxisome. The full-width at half-maximum (FWHM) of the Gaussian fits for each peroxisomal structure was determined using an ImageJ macro by John Lim. Average size was quantified for more than 50 peroxisomes and statistical significance was calculated using a two-sample t-test (two-tailed, unequal variance).

Growth on oleate, oleate consumption and halo formation was analyzed as described previously (Nötzel et al., 2016; Rucktäschel et al., 2009; Thoms et al., 2008).

Preparation of peroxisomes and mitochondria followed published protocols (Flis et al., 2015; Thoms et al., 2011b). Briefly, yeast cells were grown in YPD (1% yeast extract, 2% peptone, 2% glucose) to late exponential phase, transferred to YPO medium and grown for 48 h. Cells were harvested by centrifugation (3000 g for 5 min), washed with water and resuspended in buffer containing 0.1 M Tris-HCl, pH 9.4 and 1.5 mg/ml dithiothreitol (DTT). Spheroblasts were prepared by using 2 mg Zymolyase 20 T per 1 g wet weight. Spheroblasts were recovered by centrifugation (3000 g for 10 min), washed twice with 1.2 M sorbitol, homogenized and lysed on ice using a Dounce homogenizer in breaking buffer [5 mM 2-(N-morpholino)ethanesulfonic acid (MES), 1 mM potassium chloride, 0.6 M sorbitol, 0.5 mM EDTA, 1 mM phenylmethanesulfonylfluoride (PMSF), pH 6.0 (with KOH)]. Cell debris and nuclei were removed by centrifugation at 3000 g for 5 min. The resulting pellet was collected and subjected to two further rounds of resuspension in breaking buffer, homogenizing and centrifugation. The combined (post nuclear) supernatants (PNS) were centrifuged at 20,000 g in an SS34 rotor (Sorvall) for 30 min. The organelle pellets containing peroxisomes and mitochondria were collected and gently resuspended in a small Dounce homogenizer in breaking buffer plus 1 mM PMSF and centrifuged at low speed (3000 g) for 5 min to remove residual cellular debris. The supernatant containing the microsomal fraction was centrifuged at 27,000 g for 10 min. The pellet was resuspended in breaking buffer and loaded onto a Nycodenz step gradient [steps of 17%, 24%, 35% (w/v) in 5 mM MES, 1 mM potassium chloride, and 0.24 M sucrose, pH 6.0 (with KOH)]. Centrifugation was carried out in a swing out rotor (Sorvall AH-629) at 26,000 rpm for 90 min. Using a syringe, the mitochondrial and the peroxisomal fraction were collected from the top of the 17% and the 24%–35% interface, respectively. Purified organelles were diluted in four volumes of breaking buffer and centrifuged for 15 min at 27,000 g in an SS34 rotor. Pellets were stored at −20°C before determination of protein concentration and western blot analysis. Anti-Por1 rabbit antiserum (courtesy of Günther Daum, Institut für Biochemie, Technische Universität Graz, Austria) was used at a dilution of 1:2000 to detect mitochondria.

Yeast strains used for protein extraction (BY4741 as wild-type, GPD1_prom-PEX35::natNT2 and Δpex35::natNT2) were grown in YPD (with clonat for Pex35 strains) to the stationary phase and the diluted in YPD to an optical density at 600 nm (OD₆₀₀) of 0.2 and further grown until they reached an OD₆₀₀ of 0.6–0.8. 2.5 OD₆₀₀ units of yeast cells were harvested by centrifugation (3000 g for 3 min). Cells were resuspended in 100 µl distilled water, 100 µl 0.2 M NaOH was added, and cells were incubated for 5 min at room temperature, pelleted (3000 g for 3 min), resuspended in 75 µl SDS sample buffer (0.06 M Tris-HCl pH 6.8, 10% glycerol, 2% SDS, 50 mM DTT, 2 mM PMSF and 0.0025% Bromophenol Blue), boiled for 5 min at 95°C and pelleted again (20,000 g for 2 min). 25 µl supernatant was loaded per lane for SDS-PAGE (12% acrylamide) and then transferred to a nitrocellulose membrane for western blotting. Primary antibody was rabbit anti-Arf1 (1:2000 dilution, a gift from Chris Fromme, The Weill Institute for Cell and Molecular Biology, Cornell University, Ithaca, USA) and secondary antibody was horseradish peroxidase (HRP)-conjugated anti-rabbit-IgG. Membranes were exposed in LAS4000 machine using chemiluminescence reagents. Quantification was performed by ImageJ relative to a histone H3 loading control.

We analyzed protein secretion as previously described (Copic et al., 2009). Specifically, yeast strains grown in a 96-colony array format were pinned by the RoToR colony arrayer onto YEPD agar plates, and a nitrocellulose membrane was applied onto the plate immediately thereafter. Overnight incubation at 30°C allowed cell growth and ensured that secreted proteins were transferred to the nitrocellulose membrane. The membrane was removed from the plate and then washed (10 mM Tris-HCl, 0.5 M NaCl, pH 7.5) to release remaining cells. Secreted Kar2 was used as a reporter for secretion level, and was probed with a primary rabbit antibody (rabbit polyclonal antiserum, 1:5000 dilution, kindly shared by Peter Walter, Department of Biochemistry and Biophysics, UCSF, San Francisco, USA). Next, we used secondary goat anti-rabbit-IgG conjugated to IRDye800 (926-32211, LI-COR Biosciences) and scanned the membrane for infrared signal using the Odyssey Imaging System (LI-COR Biosciences).

Quantitative lipid analysis was performed using a Triversa NanoMate ion source (Advion Biosciences, Inc.) coupled to a LTQ Orbitrap XL mass spectrometer (Thermo Fisher Scientific), as previously described (Ejsing et al., 2009).

The thiol–disulfide pair of roGFP2 equilibrates with the glutathione redox couple. roGFP2 exhibits two fluorescence excitation maxima at ∼400 nm and ∼480 nm for fluorescence emission is followed at 510 nm. The relative intensity of fluorescence emission at each excitation maxima changes in the opposing direction dependent upon whether the roGFP2 thiols are reduced or forming a disulfide bond, thereby allowing ratiometric fluorescence imaging of the roGFP2 redox state. Dynamic E_GSH measurements were conducted as previously described in detail (Morgan et al., 2011).

Amino acid analysis was conducted as described previously (Nötzel et al., 2016).

The protein–protein interaction screen was performed using the yeast DHFR PCA library as previously described (Tarassov et al., 2008). In brief, strains with Pex35 tagged at its C-terminus with DHFR[1,2] (in MATa), or DHFR[3] (in MATalpha), and regulated by either the natural PEX35 promoter (yMS2585/6) or the strong TEF promoter (yMS2589/90), were mated with the complementary DHFR libraries in 1536 colony array agar plates. The resulting diploids were subsequently selected for growth in the presence of methotrexate for positive DHFR PCA reconstitution for 5 days at 30°C. The colony array plates were scanned in two repeats and colony sizes were measured using the freely available Balony software (Young and Loewen, 2013), with each colony size normalized by row and column means (Table S4).

We thank Lihi Gal and Corinna Dickel for technical assistance, Monika Schneider and Ralph Krätzner for help with the amino acid analysis, Ariane Wolf and Anne Clancy for yeast library handling, and Hans Kristian Hannibal-Bach for assistance on lipidomics analysis. We would like to thank Ariane Wagner and Günther Daum for advice on organelle purification and for anti-Por1 antibodies, Peter Walter for anti-Kar2 antibodies and Chris Fromme for antibodies, useful discussions and ideas. We thank Nitai Steinberg for help with graphical design of figures and Jutta Gärtner for support.