Introduction

Establishing the characteristic membrane lipid composition of organelles is thought to be critical for their proper function. This is achieved by targeting lipid biosynthesis and modifying enzymes to specific compartments, intracellular transport of lipid molecules across these compartments, and homeostatic responses to perturbations in the environment. Surprisingly, a large fraction of the molecular components and pathways in this fundamental cellular process are still elusive. For instance, it is largely unknown which route(s) a given lipid utilizes to navigate the multiple membranes of eukaryotic cells. Lipid transport is carried out by vesicular traffic as well as by dedicated lipid transport proteins (LTPs). LTPs catalyze the transfer of lipid molecules between two membranes by shielding them from the aqueous environment. LTPs often function at regions of close apposition (10–30 nm) between organelles—termed membrane contact sites (MCSs)—found between various compartments, including membranes connected by vesicular trafficking pathways (Gatta & Levine, 2017; Reinisch & Prinz, 2021). Indeed, phospholipids and sterols are transported between the ER and the plasma membrane even after inhibition of the secretory pathway, indicating that non-vesicular trafficking pathways contribute to lipid transfer at these sites (Vance et al, 1991; Schnabl et al, 2005; Quon et al, 2018).

Moreover, there is redundancy in lipid exchange routes even within non-vesicular trafficking pathways, as exemplified with mitochondria. In the yeast Saccharomyces cerevisiae, two pathways have been implicated in lipid transport at mitochondria: the ER-Mitochondria Encounter Structure (ERMES) pathway and the Mcp1-Vps13 pathway (Kornmann et al, 2009; Lang et al, 2015; Park et al, 2016; John Peter et al, 2017; preprint: John Peter et al, 2021). Structural analysis revealed hydrophobic cavities, both in the SMP domains of multiple ERMES components and in Vps13, which facilitates lipid exchange in vitro (Jeong et al, 2016, 2017; AhYoung et al, 2017; Kumar et al, 2018; Li et al, 2020). The fact that mitochondria are also in contact with multiple organelles including the PM, peroxisomes, and lipid droplets (Elbaz-Alon et al, 2014; Hönscher et al, 2014; Schuldiner & Bohnert, 2017; Shai et al, 2018) makes it plausible that these organelles could also function as potential donor and/or acceptor organelles for lipid exchange. These redundancies in lipid exchange likely exist for other organelles as well.

An additional layer of redundancy is observed at the level of lipid synthesis; two pathways, namely the Kennedy and the cytidyl-diphosphate-diacylglycerol (CDP-DAG) pathways (Fig 1A, left panel), produce phosphatidylethanolamine (PE) and phosphatidylcholine (PC). The Kennedy pathway utilizes ethanolamine and choline—primarily obtained from the medium—to convert diacylglycerol (DAG) directly to PE and PC, respectively (Fig 1A, left panel) (Kennedy & Weiss, 1956). By contrast, the CDP-DAG pathway uses CDP-DAG generated from phosphatidic acid (PA) and a series of enzymatic reactions to produce phosphatidylserine (PS), PE and PC in sequential order (Carman & Henry, 1999). In yeast, the dually localized PS decarboxylase Psd1 produces PE in the mitochondrial inner membrane (MIM) and ER, while Psd2 synthesizes PE at endosomes (Gulshan et al, 2010; Friedman et al, 2018). The production of PC by trimethylation of PE is performed by the ER-localized enzymes, Cho2 and Opi3. Either pathway alone is necessary and sufficient for optimal growth, indicating that lipids are distributed throughout the cell from different locations.

Figure 1. Rewiring PE and PC synthesis to expose lipid trafficking routes

Scheme depicting the topology of the CDP-DAG and Kennedy (red) pathways producing PE and PC in wild-type (WT) yeast cells (left panel), and an example of a rewired condition in which PE synthesis is directed to the mitochondrial inner membrane (MIM) and PC synthesis to peroxisomes (pex) (right panel). Black arrows indicate PS (dashed) and PE (dotted) lipid transport that must occur to enable the sequential enzymatic reactions. Localization of the GFP-tagged PE-synthesizing enzyme (Psd) targeted to the indicated organelles in choppΔ cells, grown in SD medium supplemented with 10 mM choline. Images shown are either a single Z-slice (ER construct) or a maximum intensity projection of several Z-sections (other constructs); LD, lipid droplets; MM, mitochondrial matrix. A scheme depicting each chimeric Psd construct is shown under the corresponding microscopy image. Localization of the GFP-tagged PC-synthesizing enzyme (Pmt) targeted to the indicated organelles in choppΔ cells, grown in SD medium supplemented with 10 mM ethanolamine. Images shown are either a single Z-slice (ER construct) or a maximum intensity projection of several Z-sections (other constructs). A schematic depicting each chimeric Pmt construct is shown under the corresponding microscopy image.

Importantly, cells are likely to adapt to the absence of one of these pathways by activating homeostatic responses. Indeed, mechanisms are known for sensing lipid levels and lipid packing/saturation (Young et al, 2010; Covino et al, 2016), which are coupled to transcriptional, post-transcriptional, or allosteric responses. A spectacular example is the adaptation to the lack of PC synthesis which involves remodeling of fatty acids in phospholipids (Bao et al, 2021).

Genetically uncovering lipid transport routes and homeostatic adaptive mechanisms requires removal of the multiple layers of redundancy. One way to achieve this goal is to delete endogenous enzymes and redirect synthesis of specific phospholipids to defined organelles through the expression of chimeric enzymes, as was successfully used to demonstrate PE transport from peroxisomes (Raychaudhuri & Prinz, 2008) or PS transport from non-native organelles (Shiino et al, 2021). Importantly, in this strategy, data interpretation depends on the targeting accuracy of the chimeric enzymes to the intended organelles.

Here, we extended this rewiring approach (i.e., minimalizing and rerouting phospholipid synthesis) by confining PE and PC synthesis to distinct combinations of organelles. We then performed genetic screens to identify factors, the perturbation of which results in a fitness gain or loss in rewired conditions. We focus on CSF1, encoding a protein with a Chorein-N lipid transport motif (Levine, 2019; Lees & Reinisch, 2020), and uncover its role in homeostatic adaptation of the lipidome.

Results Rewired yeast phospholipid synthesis as a strategy to remove layers of redundancies

We postulated that by constructing yeast strains with simplified and rewired phospholipid synthesis pathways (Fig 1A, right panel), we could (i) remove layers of redundancy and (ii) expose homeostatic adaptations and trafficking routes between organelles of interest. To do this, we first generated a cho2Δ opi3Δ psd1Δ psd2Δ strain (hereafter referred to as choppΔ) that is incapable of producing PE and PC via the CDP-DAG pathway and therefore relies on supplementation with ethanolamine and choline for growth. To rewire lipid synthesis, we engineered lipid synthesizing enzymes to make PE and PC in distinct pairs of (or in the same) organelle, by fusing them to specific targeting sequences (Fig 1). Specifically, we utilized a PS decarboxylase (Psd) lacking the transmembrane domain, either from yeast (scPsd) or Plasmodium knowlesi (pkPsd) and a soluble PE methyltransferase from Acetobacter aceti (aaPmt, hereafter referred to as Pmt) for the production of PE and PC, respectively (Hanada et al, 2001; Choi et al, 2012; Kobayashi et al, 2014). We directed the Psd and Pmt enzymes to either the ER, the mitochondrial inner membrane (MIM), endosomes (endo), or lipid droplets (LD), with the Pmt enzyme targeted in addition to the peroxisomal lumen (pex) or the mitochondrial matrix (MM).

We expressed the chimeric enzymes on a plasmid under the control of the TEF promoter in the choppΔ strain, except for scPSD, which was expressed from its own promoter (Friedman et al, 2018). Targeting of the enzymes was verified by microscopy (Fig 1B and C, Appendix Figs S1–S13). Indeed, the ER- and mitochondria-targeted enzymes revealed fluorescence patterns characteristic of these organelles (Fig 1B and C, Appendix Figs S1A, S4, S5, S10 and S13). Likewise, Pmt targeted to the MIM stained the mitochondrial outline (Appendix Fig S1B and S9), but also accumulated as bright puncta, suggesting aggregation. This did not however preclude enzyme activity (see below). Both lipid droplet (LD)- and peroxisome lumen (pex)-targeted enzymes co-localized with a LD (Erg6-mCherry) and a peroxisome marker (mCherry-SKL), respectively (Appendix Fig S2). We targeted the endosome using a PI3P-binding FYVE domain, in order to avoid transit of the proteins through the early secretory pathway, and potential ectopic activity. Both endosome-targeted Psd and Pmt (Psd-FYVE and Pmt-FYVE) assembled in perivacuolar puncta that were positive for the endocytic tracker FM4-64, consistent with late endosome localization. In addition, the Pmt-FYVE construct localized to vacuoles themselves (Fig 1C, Appendix Fig S7). Thus, both engineered enzymes were successfully re-targeted to endocytic compartments.

Since an altered lipid composition in the rewired strains could affect organelle integrity and interfere with protein targeting, we validated the localization of all enzyme constructs in the various combinations of PE and PC synthesis (Appendix Fig S4–S13) in the choppΔ background. As both Psd and Pmt enzymes were GFP-tagged, we mutated the GFP of one of the enzymes to a “dark” state to verify proper targeting of the other enzyme. We did not observe gross differences in enzyme targeting in any rewired condition.

It is important to note however that we cannot exclude that some enzymatic activity is borne by a small fraction of mislocalized protein (with the exception of mitochondria matrix-directed Pmt, as discussed below). Indeed, a minor fraction of LD- and peroxisome-targeted Pmt was detected in the ER in some cells (Appendix Fig S6 and S8). Similarly, because most constructs respect the native topology of the enzymes with the active site facing the cytosol, it is possible that some of the activity might happen in trans on other membranes at sites of interorganelle contact.

Rewired yeast strains are viable and produce all phospholipids

To examine whether the rewired CDP-DAG pathway supports growth when the Kennedy pathway is inactive, we performed growth assays either in the presence or absence of exogenous ethanolamine and choline (conditions hereafter termed KennedyON and KennedyOFF, respectively). As expected, the choppΔ strain could not grow in the KennedyOFF conditions (Fig 2A and B). Single supplementation with ethanolamine or choline alone allowed growth of the choppΔ strain at various rates. Indeed, it is known that psd1 psd2 double mutants can grow in the sole presence of choline. This is due to the fact that some PE can be synthesized from the breakdown product of long-chain bases (Birner et al, 2001). This pathway is sufficient to generate a pool of PE but unable to supply enough precursor to PC biosynthesis via the CDP-DAG pathway, explaining the auxotrophy of these strains for either ethanolamine or choline. It has also been observed that cho2 opi3 double mutants can grow to a certain degree in the absence of choline (Summers et al, 1988). The basis for this residual growth is unclear but might be related to the recent finding that PC is not absolutely essential for cell function (Bao et al, 2021) or to impurities in growth media (Renne et al, 2020). In line with this, expression of any PSD alone led to a growth restoration equivalent to ethanolamine supplementation. However, cells expressing either Psd or Pmt targeted to any organelle grew better when supplied with both ethanolamine and choline (Fig 2A), showing the importance of both PE and PC for optimal cell growth.

Figure 2. Rewired yeast cells are viable and produce all phospholipids

Cells of the indicated genotypes were pre-cultured in medium with 10mM ethanolamine and 10mM choline and subsequently in medium lacking them for 5 h, before being diluted in the indicated medium for growth curve analysis. Emp. vec. refers to cells bearing an empty vector. Data shown are the mean ±SEM of three independent replicates, except for Psd(MIM) (two experiments). Five-fold serial dilutions of strains of the indicated genotypes (wt cells and choppΔ cells harboring empty vectors or choppΔ cells expressing chimeric Psd and Pmt enzymes). Cells were grown on SD-HIS-LEU (-HL) medium (KennedyOFF) or SD-HL containing ethanolamine and choline (KennedyON). Thin-layer chromatography (TLC) analysis of steady-state phospholipid profiles of cells of the indicated genotypes. Where Psd and/or Pmt are missing (-), choppΔ cells were grown in the presence of ethanolamine and/or choline. Bands corresponding to the phospholipids PA, PS, PE, PI, and PC are indicated.

We engineered a total of 24 combinations in which Psd and Pmt were targeted to various organelles. Remarkably, growth was restored in all rewired strains in the absence of ethanolamine and choline, albeit to varying extents, indicating that the chimeric enzymes bypass the need for the Kennedy pathway (Fig 2B). Growth was most robust when Psd was localized to the mitochondrial inner membrane (MIM), which is normally the major site of PE synthesis (Rosenberger et al, 2009). Cell growth was especially limited when Pmt was targeted to the mitochondrial matrix (MM). This was not likely due to the inability of matrix-targeted Pmt to meet PC synthesis demand, since in some rewired conditions, cells grew even slower in the KennedyON conditions (Fig 2B). Instead, slow growth was possibly due to lipostatic or proteostatic challenges.

To confirm that the rewired strains produced PE and PC, we used lipid thin-layer chromatography (TLC). In agreement with their ability to grow, all rewired strains efficiently produced both lipids (Fig 2C). Differences in overall lipid profiles were nevertheless obvious. For example, phosphatidylinositol (PI) levels appear to be higher in most rewired strains. This increase could be a compensatory mechanism to deal with altered PE/PC ratios. Moreover, in the strain expressing Psd targeted to the MIM, PS levels were consistently low, suggesting that PS is most efficiently metabolized into PE in these conditions. Altogether, the growth and lipid profile of rewired yeast strains indicate that yeast cells can tolerate topological rewiring of their PE and PC biosynthesis pathways. This suggests that cells must have mechanisms to re-distribute lipids from these non-native locations to other membranes around the cell, or membrane tethers that might promote activity of the enzymes in trans, as well as mechanisms to cope with lipid imbalance.

A transposon mutagenesis screen to identify genetic components essential for survival of yeast with rewired lipid biosynthesis

To identify the genetic components necessary to handle lipid imbalances and reroute lipid trafficking in the rewired strains, we performed genetic screens using SAturated Transposon Analysis in Yeast (SATAY, Fig 3A) (Michel et al, 2017; preprint: Michel et al, 2019). Briefly, we generated libraries consisting of millions of independent transposon insertion mutants. The transposon libraries for a given genotype were then grown for several cycles in the presence or absence of ethanolamine and choline (KennedyON or KennedyOFF) to assess the growth of the mutants in the rewired conditions. Next-generation sequencing then allows mapping and quantifying the transposon insertion sites (TNs) across the genome. The number of transposons and associated sequencing reads in each ORF is then used to evaluate the fitness of the mutants in the given conditions.

Figure 3. Transposon mutagenesis screens to identify genes required for adaptation and lipid trafficking in the rewired strains

Outline of the SATAY screening procedure. Cells expressing plasmids containing the PE- and PC-synthesizing enzymes and a plasmid harboring the galactose inducible transposase (TPase) and a transposon (TN) disrupting the TRP1 gene were grown for ~ 56 h in galactose-containing medium to induce transposition (step 1). Cells were inoculated in synthetic medium lacking tryptophan in either KennedyON or OFF conditions, grown for several generations (step 2) and harvested for DNA extraction and sequencing of transposon insertion sites (TNs) (step 3). TNs are mapped to the genome to identify genes that become required under certain rewired conditions (step 4). An illustrative example of TN insertions of four libraries visualized in the UCSC genome browser highlights a genetic requirement for DRS2 (green bar) in KennedyOFF conditions. GO term enrichments in the top 200 (red) and top 500 (blue) most variable genes (defined as genes with the highest standard deviation of TN insertions across libraries) identified using the YeastMine web server. Significance was determined using Holm–Bonferroni corrected P-values with a cut-off of 0.05, the size of the circles is inversely proportional to the P-values (i.e., bigger circles indicate higher significance). VPS13 is required in KennedyOFF conditions when chimeric enzymes are targeted to mitochondria or endosomes. Volcano plots show the fold change of number of transposon insertions per gene of libraries grown in KennedyOFF versus ON conditions. This comparison included all six libraries where the enzymes are targeted to endosomes and/or mitochondria (top left panel) or six libraries with enzymes targeted to other organelles (bottom left panel). Data points for VPS13 are highlighted in red. Schematic illustrating Vps13-mediated lipid transport at endosomes and mitochondria contact sites (right panel). Schematic (top panel) illustrating the requirement for SAM transport across the mitochondrial inner membrane by Sam5 for PC production by Pmt in the mitochondrial matrix (MM). Lipids can be produced by the Kennedy pathway (red) independent of Sam5 activity. Transposon insertion maps generated in the UCSC genome browser for SAM5 for the library where PC is produced in the MM in either KennedyON or OFF conditions (bottom panel).

Source data are available online for this figure.

We generated a total of 24 libraries comprising 12 rewired strains grown in both KennedyON and OFF conditions. Transposition insertion sites mapping across the entire yeast genome (SData Fig 3, dataset 1) can be viewed in the genome browser (http://genome-euro.ucsc.edu/s/benjou/rewiring_paper, Suppl. Data 1). Additionally, we compared different sets of libraries against each other (e.g., individual libraries against all other libraries and libraries with a common condition, such as the expression of the same chimeric enzyme, or KennedyON vs. OFF, against other libraries). We generated multiple interactive volcano plots available for browsing (SData Fig 3, dataset 2 and https://kornmann.bioch.ox.ac.uk/satay/rewiring) in which we plotted the fold change of the average number of transposons per gene of a test set and a reference set against a P-value associated with this difference to identify genes that provide a fitness advantage or disadvantage in certain libraries.

We performed hierarchical clustering of the libraries according to their pattern of transposon insertion in each gene (Appendix Fig S14). Libraries with the same genotypes, in the KennedyON and OFF conditions, cluster as nearest neighbors. Since these libraries are replicates grown under different conditions, and the number of transposons in most genes is unaffected by the activity of the Kennedy pathway, this clustering may be dominated by technical variation. Nevertheless, independent libraries generated in strains expressing the same PE- or PC-producing enzymes also tended to cluster together. In some cases, the location of the PE-producing enzyme dominates in the clustering (e.g., PE-LD), and in other cases, it is that of the PC-producing enzyme (e.g., PC-pex). Yet in general, there is no pattern based on the location of one or the other enzyme. Overall, this clustering implies that the rewiring dominates in determining fitness and genetic requirements rather than the availability of choline and ethanolamine. To identify genes required to adapt to the rewired conditions, we first sought to filter for those with the most variable number of transposons across the whole dataset, using the standard deviation (SData Fig 3). This subset may omit genes that become essential in only one or very few conditions yet is an unbiased way of identifying genes that generally affect fitness in rewired conditions. We selected the top 200 and top 500 most variable genes and searched for GO term enrichment using the YeastMine web server (Balakrishnan et al, 2012). Remarkably, these genes were enriched for GO terms related to lipid homeostasis (lipid transport, localization, and metabolism) and other processes such as intracellular transport (Fig 3B). Particularly striking examples of variable genes are DRS2, DNF2, and LRO1, which are completely devoid of transposons in some conditions (Fig EV1). DRS2 and DNF2 both encode phospholipid translocases required for maintaining phospholipid asymmetry (Sebastian et al, 2012). They may be required to maintain proper lipid bilayer distribution in rewired strains. Lro1 is an acyltransferase that converts DAG to triacylglycerol (TAG) using phospholipids as acyl chain donors (Barbosa et al, 2019). Thus, Lro1 may be required to degrade excess membrane. Importantly, these genes had variable but non-similar patterns of transposon coverage across libraries, indicating that it was not a common condition (e.g., KennedyON or OFF) that made them more or less required.

Click here to expand this figure.

Figure EV1. Lipid metabolism and lipid flippases genes are required for adaptation in specific rewired conditions

Transposon insertion maps of the DRS2, LRO1, DNF2, NEM1, and MGA2 genomic loci generated in the UCSC genome browser for libraries of strains expressing the chimeric Psd and Pmt enzymes to produce PE and PC at the indicated cellular locations in the choppΔ background. “KennedyON” refers to libraries that were grown in media supplemented with 10 mM ethanolamine and choline.

Intriguingly, the lipid transporter Vps13, even though never completely essential, appears to be consistently less covered in KennedyOFF conditions (Appendix Fig S15) in libraries in which Psd and Pmt are targeted to mitochondria and/or endosomes, organelles to which this lipid transporter localizes (Lang et al, 2015; Park et al, 2016; John Peter et al, 2017) (Fig 3C, Appendix Fig S15). Together, these results suggest that Vps13 may have a more prominent role at these organelles in KennedyOFF conditions and underlines that lipid transporters could become limiting when rewiring involves their native site. Notably, we find that the inner mitochondrial S-adenosylmethionine (SAM) transporter SAM5 (Marobbio et al, 2003) is essential in KennedyOFF conditions when PC synthesis is targeted to the mitochondrial matrix (Fig 3D). SAM is the methyl group donor for the conversion of PE to PC, and thus, SAM is necessary in the matrix for PC production, making Sam5 indispensable. This finding excludes the possibility that growth of this strain is attributable to PC production by a small fraction of mistargeted enzymes. Thus, our rewiring approach can identify genes that maintain cell viability in rewired conditions and these genes are, as expected, enriched for lipid and membrane-related processes.

Similar transposon insertion patterns across rewired conditions reveal clusters of functionally related genes

Previous genome-wide analyses have shown that genes that are required under the same set of conditions are usually part of the same biological process, pathway or protein complex (Costanzo et al, 2016). Therefore, we asked if we could identify genes that have similar patterns of transposon insertion across different rewired libraries. We selected the most variable genes and computed the Pearson correlation coefficient for pairs of genes across our libraries and performed hierarchical clustering on the correlation matrix (Fig 4A, SData Fig 3, dataset 3).

Figure 4. Clustering of gene correlations across libraries reveal gene groups that work together to handle rewired conditions

Hierarchical clustering of the Pearson correlation coefficients of transposon insertion profiles for gene pairs. Insets show clustering of the IRA1/IRA2 and SWC3/SWC4 gene pairs (bottom inset, black arrows) and high correlation of the SYS1/GEF1/SPF1 cluster with DRS2 and MVP1 (top inset, blue arrows). Volcano plot comparing the number of transposon insertions per gene of all 12 libraries in KennedyON and OFF conditions. Genes significantly required in KennedyOFF conditions are highlighted in red (P-value < 0.05, log2[fold change] < −0.5). Genes highlighted in the clusterogram in (A) (top inset) are indicated with blue arrows.

We find clustering of highly correlated gene pairs consistent with their known function and/or profile similarities computed in the whole-genome genetic interaction map (Costanzo et al, 2016) (Fig 4A). Examples include IRA1 and IRA2 (Tanaka et al, 1990), paralogs that negatively regulate the RAS-cAMP pathway and SWC3 and SWC4, both components of the chromatin remodeling SWR Complex (Mizuguchi et al, 2004) (Fig 4A, black arrows). We also find correlation between BRE2 and SWD1 (Fig 4A, grey arrows), both members of the same histone methylation complex (COMPASS) (Krogan et al, 2002).

A striking example, also consistent with the Costanzo dataset (Costanzo et al, 2016), is the cluster containing SYS1, SPF1, GEF1, and MVP1 (Fig 4A and B, blue arrows). We found that this cluster became essential in most KennedyOFF conditions, as shown by a volcano plot in which libraries grown with and without ethanolamine and choline are compared against each other (Fig 4B). This implies that these genes function together and act redundantly with the Kennedy pathway to maintain cell viability in the rewired strains. Given their known roles in vesicular trafficking and protein sorting, it is possible that the Kennedy pathway is required for efficient vesicular trafficking in rewired conditions, or that vesicular trafficking is needed to redistribute lipids in the cell. Alternatively, vesicular trafficking may affect proper localization of other factors required for lipid distribution (e.g., Drs2). These examples demonstrate that the transposon insertion patterns across libraries can reveal genes that function together to maintain cell viability upon lipostatic challenges, in an unbiased and genome-wide fashion.

Rewiring highlights the importance of transcriptional and post-transcriptional responses

Besides unique adaptations to specific conditions, we postulated that rewired strains have common strategies to deal with general phospholipid and membrane stress. To identify such genes, we compared the rewired libraries generated in this study, with two previously generated wild-type libraries (Fig 5A) (preprint: Michel et al, 2019). Expected changes in transposon insertion profiles were associated with technical differences in library generation, in the present and previous studies (e.g., genes for histidine, leucine, tryptophan, and adenine prototrophy, that were used for selection) (Fig 5A, grey dots). Besides these, striking differences were found in several genes involved in transcriptional adaptation (Fig 5A, SData Fig 5). One of the best hits is OPI1, a transcriptional repressor for a variety of lipid biosynthesis genes under the control of the Ino2-Ino4 transcription activation complex (e.g., INO1, CHO1, ITR1, genes of the Kennedy pathway) (Henry et al, 2012). OPI1 is required in all rewired libraries, indicating an increased need for transcriptional regulation of phospholipid biosynthesis. Another example is the transcription factor CBF1, which is implicated in enhancing Ino2-Ino4 transcriptional activation (Shetty & Lopes, 2010), mitochondrial respiration, and repressing ceramide biosynthesis (DeMille et al, 2019). Examples of genes required for growth only in specific libraries, include NEM1, a phosphatase controlling phospholipid biosynthesis (Siniossoglou et al, 1998), and MGA2, a transmembrane transcription factor involved in sensing membrane saturation (Covino et al, 2016) (Figs EV1 and EV2). These results suggest that distinct rewired conditions cause both common and unique membrane challenges that require a combination of adaptive responses.

Figure 5. Differential genetic requirements common to rewired libraries

Volcano plot comparing the number of transposon insertions per gene of all 24 rewired libraries and two wild-type (wt) libraries. Genes that are significantly (P-value < 0.05) required (red, log2[fold change] < −1.5; blue, log2[fold change] < −1) or dispensable (green, log2[fold change > 2]) in the rewired libraries are highlighted. Genes depicted in grey are essential due to differential growth requirements in the preparation of the two wt libraries compared to the rewired strains, or correspond to genes deleted in the rewired strains. Requirement for mitochondrial lipid transport and biosynthesis genes in rewired libraries. Left panel, schematic depicting PE and cardiolipin (CL) synthesis in mitochondria. The inner mitochondrial putative lipid transporter Mdm31 is depicted in yellow. Gray arrows indicate the need for lipid transport from the ER or other organelles to mitochondria and transport between the MOM and MIM. Right panel, transposon numbers (normalized to wt libraries) for the indicated (set of) libraries for MDM31, PGS1, and CRD1.

Source data are available online for this figure.

Click here to expand this figure.

Figure EV2. Requirement for lipid metabolic genes in rewired yeast

Schematic illustration of lipid synthesis pathways in yeast. Genes that are essential or absent in all libraries are depicted in red or gray, respectively. Genes variably required in one or more rewired libraries, as assessed by manual inspection or volcano plots, are depicted in orange. Genes that are essential in wild-type conditions are boxed in red.

In addition to transcriptional adaptation to the altered lipidome, we observed many additional changes in the genetic requirement of rewired strains (Fig 5A, SData Fig 5). These include an increased reliance on proteostasis regulators (RPN4, MDJ1), and on otherwise non-essential lipid biosynthesis pathways, such as cardiolipin (CL) synthesis (PGS1, CRD1) and the final step of ergosterol synthesis (ERG4; Fig 5A). The mitochondrial inner membrane-targeted Psd rescues the requirement for CL synthesis in the transposon screen, consistent with previously published work highlighting redundant functions for PE and CL in mitochondria (Fig 5B) (Gohil et al, 2005; Joshi et al, 2012). In contrast, synthesis of the cardiolipin precursor phosphatidylglycerol (PG) is required in all libraries (Fig 5B). Moreover, analysis of genes involved in lipid biosynthetic pathways suggest a variable requirement for the enzymes involved in the synthesis of phospholipids such as PA, PI, PE, CL, and the neutral lipid TAG (Fig EV2).

Interestingly, transposons in the gene encoding the inner mitochondrial membrane protein Mdm31 are tolerated in wild-type libraries, but not in the rewired strains. MDM31 genetically interacts with PSD1 and is crucial for mitochondrial lipid homeostasis (Miyata et al, 2017). Notably, PE synthesis at the mitochondrial inner membrane by our chimeric construct (resembling mitochondrial localization of Psd1) only marginally rescues the number of transposons in the choppΔ background (Fig 5B), implying that mdm31 mutants are particularly sensitive to altered PE/PC levels and distribution in the absence of the endogenous CDP-DAG pathway. Mdm31 harbors a Chorein-N motif, a signature domain found in lipid transport proteins including Vps13 (Levine, 2019). Thus, it is tempting to speculate that the requirement for MDM31 in the rewired conditions is due to a requirement for lipid transport across the mitochondrial intermembrane space.

Csf1 is required for lipid adaptation in rewired conditions

As we relocalized lipid synthesis to distinct compartments, we expected that LTPs might become particularly necessary to distribute lipids in specific rewired KennedyOFF conditions. Genes satisfying this condition were scrutinized based on previous literature and structural predictions to identify LTP candidates. Csf1 emerged as a promising hit, as a HHpred analysis (Zimmermann et al, 2018) predicts an N-terminal Chorein-N motif, homologous to that found in Vps13, Atg2, and other potential lipid transporters including Mdm31 (Fig 6A) (Levine, 2019; Li et al, 2020). Csf1 is a large protein like Vps13 and Atg2. They are respectively the 10th, 5th, and 92nd largest S cerevisiae polypeptides. While Vps13 and Atg2 are soluble proteins, Csf1 bears an N-terminal transmembrane domain. Transposons targeting the CSF1 ORF were strongly depleted in KennedyOFF conditions (Figs 6A and EV3) in the strain where Psd is targeted to the MIM and Pmt production to peroxisomes (hereafter referred to as Psd-MIM/Pmt-pex). However, contrary to VPS13, which becomes required for growth in most libraries involving PE or PC synthesis at endosomes or mitochondria (Figs 3C and 6A), the requirement for CSF1 was specific to the Psd-MIM/Pmt-pex conditions.

Figure 6. Csf1 is required specifically in the Psd-MIM/Pmt-pex rewired condition

Left: comparison of the number of transposon insertions per gene (dots) for libraries with PE synthesis targeted to the MIM and PC synthesis targeted to peroxisomes, in KennedyON and OFF conditions. CSF1 and VPS13 (highlighted in red) are more required in KennedyOFF conditions. Right: Transposon insertion profile displayed in the UCSC genome browser for the indicated libraries for the genomic region of CSF1. Bars correspond to transposons, rows to libraries. The boundaries of the full-length Csf1 protein, the truncated Csf1ΔC protein, the Chorein-N motif and the transmembrane (TM) domain are indicated. Five-fold serial dilutions of strains of the indicated genotypes grown at 30°C on SD medium supplemented with 10 mM ethanolamine and/or 10 mM choline. Growth assays are representative images of biological replicates. PS and PE transfer assay. Rewired strains were pulse labeled with 15N-serine. Lipids were extracted at the indicated time points and measured by mass spectrometry. Graphs depict quantification of the appearance of labeled species of the indicated phospholipids normalized by the amount of its precursor lipid. Click here to expand this figure.

Figure EV3. The requirement for Csf1 is specific to the Psd-MIM/Pmt-pex library grown in KennedyOFF conditions

Volcano plot comparing the number of transposon insertions per gene in the Psd-MIM / Pmt-pex KennedyOFF condition versus all other libraries. CSF1 is highlighted in red.

To validate the screen results, we generated a strain in which CSF1 was deleted in the choppΔ Psd-MIM/Pmt-pex background. As expected, CSF1 was required for growth in KennedyOFF conditions (Fig 6B), but dispensable when PE production was redirected to endosomes and PC production to lipid droplets (Appendix Fig S16). We observed variable phenotypes when CSF1 was deleted with different gene replacement cassettes. This phenomenon is likely attributable to interference with CSF1's direct neighbor, GAA1. GAA1 encodes an essential component of the glycosylphosphatidylinositol (GPI) anchoring machinery. Previous screens utilizing full csf1 deletion alleles found that these strains had a phenotype related to that of mutants with defective GPI protein anchoring (Čopič et al, 2009; Jonikas et al, 2009; Costanzo et al, 2010). It is therefore most likely that the slow growth and the phenotype similarities are due to inhibition of GAA1 by CSF1 deletion. To avoid interference with neighboring genes, we truncated (rather than deleting) CSF1 after amino acid 797. This allele (Csf1∆C) recapitulated the growth phenotype of the CSF1 deletion in the rewired KennedyOFF condition and was therefore used for further loss-of-function experiments.

We found that ethanolamine supplementation promotes growth of rewired csf1 mutants better than choline (Fig 6B), indicating that homeostasis of PE, and not PC, likely underlies the growth defect of the csf1 mutant in the rewired conditions. Yet, no detectable change in phospholipid abundance could be uncovered by thin-layer chromatography of whole cell lipid extracts of csf1∆C cells (Fig EV4). The defect of csf1 mutants in one specific rewired condition suggested that these cells had a defect in transporting PE made in the mitochondria to the peroxisome for conversion to PC. This transport could be direct or involve any number of steps. To assess the flux of PE made by the MIM-targeted Psd, and converted to PC by the peroxisome-targeted Pmt, we pulse-labeled the rewired strain with 15N-serine an