Decatransin, a new natural product inhibiting protein translocation at the Sec61/SecYEG translocon

Most secretory and membrane proteins are translocated across or inserted into the plasma membrane of bacteria or the endoplasmic reticulum of eukaryotes by the conserved SecY/Sec61 translocon complex (Park and Rapoport, 2012; Shao and Hegde, 2011). The core translocon consists of SecY or Sec61α (Sec61p in yeast) in bacteria and eukaryotes, respectively, forming a protein-conducting channel with its ten transmembrane domains (TM1–TM10), and of two peripherally attached single- to triple-spanning subunits SecG or Sec61β (Sbh1p), and SecE or Sec61γ (Sss1p). In its idle state, the potential pore is closed by a central constriction ring of six predominantly hydrophobic residues and a lumenal plug helix that blocks the periplasmic or luminal cavity. Substrate proteins are targeted either co- or post-translationally to the SecYEG/Sec61 complex. In the former case, a hydrophobic signal sequence is recognized by signal recognition particle (SRP) to bring the nascent-chain–ribosome complex to the SRP receptor in the appropriate membrane allowing the ribosome to bind to cytoplasmic loops of SecY/Sec61α. Alternatively, substrates are synthesized into the cytosol and reach the translocon pos-translationally, which, in eukaryotes involves the Sec62–Sec63–Sec66–Sec72 complex. Somehow the signal sequence engages with the translocon and intercalates as a helix in between TM2 and TM7 of SecY/Sec61α, which form a lateral gate into the lipid bilayer (Gogala et al., 2014; Park et al., 2014; Voorhees et al., 2014) In the process, the hydrophilic sequence is inserted into the pore, pushing away the plug and the constriction residues that act as a gasket around the polypeptide (Park and Rapoport, 2011). It is not clear how signal peptides with diverse primary sequences that only share a general hydrophobic character initiate translocation.

Chemical inhibitors have frequently been useful tools to elucidate the molecular mechanisms of complex processes. A general translocation inhibitor is eeyarestatin, a chemical most likely binding to the translocon and preventing the transfer of the nascent-chain–ribosome complex from the SRP–SRP receptor targeting complex to the translocon in mammalian cells (Cross et al., 2009). Apratoxin A, a cyanobacterial metabolite, has been shown to inhibit cotranslational translocation in vitro (Liu et al., 2009), but the blocked step is unknown. Furthermore a group of closely related cyclic heptadepsipeptide inhibitors including HUN-7293, CAM741 (Besemer et al., 2005) and a simplified version thereof called cotransin (Garrison et al., 2005) have been found to inhibit cotranslational translocation of VCAM1 and other specific substrates (Maifeld et al., 2011; Westendorf et al., 2011). Photoaffinity labeling has identified Sec61α as the target (MacKinnon et al., 2007), which was confirmed by isolation of resistance mutations in SEC61A1 (MacKinnon et al., 2014). Using in vitro translation–translocation assays, it has been shown that SRP-dependent targeting and binding of the ribosome, and interaction of the signal with the translocon in the cytosolic vestibule, are unaffected by these cyclic heptadepsipeptide inhibitors, but signal insertion is blocked (Besemer et al., 2005; Garrison et al., 2005; MacKinnon et al., 2014). The origin of signal specificity of inhibition is not clear (Harant et al., 2006).

In this study we identified a new bioactive cyclodepsipeptide that inhibits cell growth. To identify its target, we used yeast chemogenomic profiling and unbiased genome-wide mutagenesis, followed by selection and sequencing of resistant clones in yeast and mammalian cells. All assays identified the endoplasmic reticulum (ER) translocon component Sec61 as the conserved target in eukaryotic cells. Biochemical characterization of the inhibition mechanism in both yeast and mammalian cells indicated that the compound blocks all translocation through the Sec61 channel. We thus suggest ‘decatransin’ as the name for this new decadepsipeptide translocation inhibitor.

Screening new compounds of natural origin for growth inhibition of HCT116 human carcinoma cells identified compounds of the saprophyte fungus Chaetosphaeria tulasneorum with potent activity. Scaled up cultivation, isolation, purification and structure elucidation (see Materials and Methods as well as supplementary materials) led to the discovery of a cyclic decadepsipeptide (Compound 1, Fig. 1A).

The presence of the non-proteinogenic amino acids pipecolic acid and homoleucine at positions 2, 6 and 9, and 3, 4 and 7, respectively, indicated that the compound had been synthesized by a non-ribosomal peptide synthetase (NRPS). NRPSs are large multienzymes with a modular organization of catalytic domains, namely the adenylation (A), peptidyl carrier protein (PCP) and condensation (C) domains, responsible for the activation of, transfer of amino acids as thioesters and peptide bond formation, respectively. Using a software to predict microbial natural product pathways (Bachmann and Ravel, 2009), multiple NPRSs were identified in the genome of Chaetosphaeria tulasneorum. The NRPS responsible for the biosynthesis of the compound was expected to contain ten modules and four N-methyltransferases. Of the predicted NRPSs, three contained ten modules. The pattern of eight-residue signature sequences that define A-domain selectivity (Stachelhaus et al., 1999) uniquely identified a single NRPS with DPFMYLGI in the A domains at positions 2, 6 and 9, and DAWTYGVA at positions 3, 4 and 7, corresponding to the positions of pipecolic acids and homoleucines, respectively (Fig. 1A,B). In addition, N-methyltransferases were present at positions 4, 5, 7 and 10, consistent with the observed N-methylation patterns. Finally, the synthetase had a terminal condensation domain (C_T) catalyzing the cyclization of NRPS products in fungi (Gao et al., 2012). The presence of a 2-hydroxy-acid dehydrogenase gene in the neighborhood of the NRPS (Fig. 1C) supported incorporation of a 2-hydroxy-amino acid at position 1, and thus peptide closure through an intermolecular ester bond, which is the hallmark of the depsipeptide family. Genes encoding L-pipecolate oxidase and 2-isopropylmalate synthase were also found close to this megasynthetase gene (Fig. 1C; supplementary material Table S1). These genes are likely involved in pipecolic acid and homoleucine biosynthesis, respectively (Field et al., 2004; He, 2006). In summary, sequence analysis identified the NRPS and accessory genes involved in the synthesis of the decadepsipeptide.

Although the compound showed potent growth inhibition of mammalian cell lines, such as HCT116 human colon carcinoma cells and COS-1 monkey kidney cells at IC₅₀ 30–140 nM, it also inhibited growth of the yeast S. cerevisiae at IC₅₀ ∼2 µM (Fig. 1A), thus enabling us to apply chemogenomic profiling to identify target proteins or pathways (Giaever et al., 1999). Haploinsufficiency profiling (HIP) and homozygous profiling (HOP) are based on S. cerevisiae heterozygous and homozygous deletion collections (Hoon et al., 2008). HIP indicates proteins or pathways directly affected by the compound, whereas HOP reveals synthetic effects and identifies compensating factors or pathways. The results are visualized by plotting the relative growth reduction of individual strains by the compound (sensitivity) versus a measure of significance (z-score, see Materials and Methods). The cyclic decadepsipeptide produced profiles that revealed haploinsufficiency for all components of the Sec61–Sec63 core complex (Fig. 2A, HIP), as well as synthetic growth defects after homozygous deletion of the three non-essential genes SBH1, SEC66 and SEC72 of the Sec61–Sec63 complex (Fig. 2A, HOP). This strongly indicated that the Sec61–Sec63 translocon is the target of the inhibitor. Dose–response growth experiments using individual strains with compound 1 fully validated the HIP result (supplementary material Fig. S1B). The only other hypersensitive HIP strain of the original library, CWC21 (involved in RNA splicing), was found to contain a heterozygous frame-shift mutation in the SEC63 gene responsible for the phenotype (supplementary material Fig. S2).

Comparing the decadepsipeptide profile to archived datasets identified a striking similarity with the HIP and HOP profile of a cyclic heptadepsipeptide, denoted compound 2 (Fig. 2B,C). Structural similarity searches with compound 1 did not reveal any similarity to compound 2 [Tanimoto coefficient <0.25 for the entire molecule, and 0.26 when only comparing the scaffolds (Bender and Glen, 2004)] or any other relevant hits, indicating that they constitute distinct chemotypes. However, the structure of compound 2 is almost identical to HUN-7293 and its derivatives, the cotransins, which have been previously characterized as mammalian translocon inhibitors (Besemer et al., 2005; Garrison et al., 2005) (Fig. 2B). With an IC₃₀ of 200 µM, HUN-7293 was less effective on wild-type yeast than compound 2 and the decadepsipeptide by a factor of 30 and 100, respectively, but produced very similar HIP and HOP profiles at this higher concentration (Fig. 2D). Pairwise comparison of HIP z-scores (Fig. 2F) confirmed conserved hits and thus the conserved mechanism of action of the new decadepsipeptide and the heptadepsipeptide and cotransin chemotype. It is interesting to note that prominent hits in the HOP profiles of the decadepsipeptides were strains where IPT1 (inositolphosphotransferase) or SUR1 (mannosylinositol phosphorylceramide synthase) have been deleted, suggesting a new genetic link between the Sec61–Sec63 translocon and lipid metabolism.

To identify the direct target of the inhibitors using an orthogonal approach, we performed unbiased, genome-wide chemical mutagenesis and selection for resistance to the inhibitors in parallel for S. cerevisiae and for mammalian HCT116 cells. In drug-efflux-compromised yeast, we obtained 45 colonies resistant to 30 µM compound 1. These resistant cells were mated with wild-type cells, and the heterozygous clones maintained the resistance, indicating that the underlying mutations were dominant. Direct Sanger sequencing of the SEC61, SEC62 and SEC63 gene loci revealed 13 different single-amino acid mutations exclusively in SEC61 (Fig. 3A, top; Table 1). The mutant alleles were introduced into drug-efflux-compromised wild-type cells, replacing the wild-type copy of SEC61. The resulting cells were all viable and retained resistance equivalent to the original resistant colonies, demonstrating that the SEC61 gene encoded the critical target of compound 1. Several mutations increased the IC₅₀ value by more than 100-fold, whereas others only caused moderate increases in the inhibitory concentration (Table 1; Fig. 3B). Given that the Sec61 translocon had also been shown to be the target of the heptadepsipeptide chemotype (compound 2, compound 3; HUN-7293 and cotransin), we tested for cross-resistance to compound 2. Except for mutations G97D, A186T and G430D, which were completely sensitive to compound 2, the mutants were cross-resistant to compound 2 (Table 1; Fig. 3C). This indicates similar, yet distinct, modes of action of the deca- and hepta-depsipeptide inhibitors on Sec61. In addition, we also tested our existing library of Sec61p mutants that had been isolated in the context of membrane protein topogenesis (Junne et al., 2007) with respect to sensitivity to compounds 1 and 2 (supplementary material Fig. S3). Mutation of the six residues of the constriction ring, partial and full deletion of the plug domain, and deletion of TM2 also resulted in strong resistance to both inhibitors.

We also performed genome-wide mutagenesis of human HCT116 cells using ethyl methanesulfonate (EMS) and N-ethyl-N-nitrosourea (ENU). Owing to limited availability of the new decadepsipeptide, mutagenized cells were cultured in the presence of 1 µM compound 2. After 2 weeks, 12 strongly growing colonies were picked for further characterization. The majority of the clones showed more than 20-fold increased IC₅₀ for compound 2 (Fig. 4A). With the exception of one single clone, they also were cross-resistant to compound 1, however, most of them only had a 2–3-fold increased IC₅₀ values (Fig. 4B). Resistance was specific because no change was observed in sensitivity to the control compound taxol (Fig. 4C).

To identify the resistance-conferring mutations, the genomic DNA was isolated from ten resistant clones and two sensitive parental samples, and used to prepare exon-enriched libraries for Illumina sequencing. The resulting reads were aligned to the human genome, and variants [single nucleotide polymorphisms (SNPs) and insertions/deletions (indels)] were called for each of the samples and then filtered according to several quality metrics (e.g. sequencing depth). Variants in the parental sample were excluded. For further analysis, we considered only missense and nonsense mutations in coding regions. We further excluded missense SNPs that were not predicted to be deleterious to protein function (see Materials and Methods). Finally, we included only mutations in genes that were expressed according to RNAseq analysis of the parental samples. Applying these three filters, the gene with the highest number of mutations was SEC61A1, which encodes Sec61α1, the human ortholog of yeast Sec61, with mutations in eight out of the ten clones (Fig. 4D). Based on the SNP patterns, the ten clones originated from eight independent progenitors. The two M65R and two D60E clones showed highly similar SNP patterns, whereas the patterns of the two S71F clones indicated independent events (Table 3). We further sequenced the SEC61A1 cDNA of six additional resistant HCT116 clones obtained by N-ethyl-N-nitrosourea (ENU) mutagenesis, all six of which also harbored mutations (Table 3). The mutated residues D60, R66 and S71 of human Sec61α1 correspond to D61, R67 and S72 in yeast Sec61p, which also produced resistance when mutated (Fig. 3). Thus the identical mutations S71F in Sec61α1 and S72F in Sec61p have even been identified independently in different species, although yielding different levels of resistance. In summary, selection for resistant clones identifies the Sec61 translocon as the conserved target of deca- and hepta-depsipeptide inhibitors in both yeast and mammalian cells.

To test translocon function directly in the presence or absence of these compounds, yeast cells were metabolically labeled with [³⁵S]methionine for 5 min, and translocation of dipeptidylaminopeptidase B (DPAPB) and carboxypeptidase Y (CPY) into the ER was assessed based on the glycosylation and processing pattern after immunoprecipitation and SDS-PAGE. DPAPB and CPY are established co- and post-translationally translocated substrates, respectively. Both the decadepsipeptide compound 1 and the heptadepsipeptide compound 2 effectively inhibited co- and post-translational translocation as is apparent in a reduction of the glycosylated and an increase of the unglycosylated forms (Fig. 5A). Inhibition of post-translational translocation demonstrates that the mechanism of action is independent of SRP and SRP receptors and thus of protein targeting to the translocon, as has already been concluded for the cotransins from in vitro translation–translocation experiments (Garrison et al., 2005), and is independent of the presence of a translating ribosome.

Inhibition was dependent on the concentration (Fig. 5B) as well as on pre-incubation time (Fig. 5C,D): compound 2 required ∼30 min to reach maximal levels, and compound 1 even longer. This time dependence most likely reflects the penetration kinetics of the compounds (the time required to cross the cell wall and plasma membrane). In general, CPY translocation appeared to be more sensitive to inhibition than that of DPAPB.

Both types of inhibitors also acted on non-natural, generic signal sequences made of 13 or 16 consecutive leucine residues (Fig. 5E). This underlines the independence of the action mechanism of these inhibitors from the specific signal sequence in yeast. In agreement with the target conservation in mammalian cells, both compounds similarly inhibited translocation in mammalian COS-1 cells, as shown in Fig. 5F for the asialoglycoprotein receptor H1, a type II membrane protein, and derivatives with generic signal-anchors made of Leu₁₃ or Leu₂₅ segments.

The conserved action of both chemotypes on the fungal and human Sec61p homologs and conserved mutations that confer resistance across species raised the question as to whether bacterial translocons also can be targeted by compounds 1 and 2. We performed in vitro translocation experiments on purified E. coli translocons reconstituted into lipid membranes as described previously (Bauer et al., 2014). Both compounds inhibited translocation of a pro-OmpA model substrate in a dose-dependent fashion (Fig. 5G) with estimated IC₅₀ of 10 µM for compound 1 and 90 µM for compound 2. Interestingly, deletion of the SecY plug domain conferred resistance to both chemotypes as shown by a considerable shift of the curves.

The identified resistance-conferring mutations are conspicuously concentrated in or close to the plug domain (Fig. 3A; Fig. 4D, and see MacKinnon et al., 2014). Plug mutations have previously been found to cause a prl phenotype, i.e. the suppression of signal sequence mutations (Emr et al., 1981). prl mutations specifically destabilize the closed state of the translocon, and thereby facilitate pore opening, reducing the stringency for signal acceptance by the translocon. We tested the new resistant Sec61p mutants for a prl phenotype. Five of the 13 newly isolated resistant mutants indeed suppressed the translocation defect of CPYΔ3 in which the hydrophobic core of the signal peptide was truncated by three residues (Fig. 5H). All of them carry mutations involving the plug, four in the plug domain itself and one (G47D) in TM1 pointing towards it. From our older collection (Table 2), all prl mutants except one (I86T) were significantly resistant to at least one of the inhibitors. The correlation between the prl phenotype and resistance suggests that the inhibitors bind to the closed state of the wild-type translocon. In prl mutants this state is destabilized, resulting in reduced binding affinity for the compounds.

Here, we present an integrated study of a new bioactive decadepsipeptide from its isolation and the identification of the responsible megasynthetase of the producer organism by genome-sequencing and in silico analysis to the determination of the conserved target in yeast and mammalian cells, as well as in bacteria, using genomic assays and biochemical analysis of the inhibited processes. Given that this compound inhibits co- and post-translational translocation across the Sec61/SecYEG translocon, we propose to name it decatransin. The name also alludes to the cotransins, a group of heptadepsipeptide compounds without detectable molecular similarity that have previously been described as mammalian translocon inhibitors.

The activity of decatransin on the yeast S. cerevisiae allowed for chemogenomic profiling to rapidly home in on the Sec61 complex as the potential fungal target: our high-resolution HIP and HOP platform (Hoepfner et al., 2014) identified all core subunits of the heptameric translocon complex. Essentially the same profile was obtained for a new yeast active heptadepsipeptide (compound 2) of the cotransin chemotype and for the original inhibitor HUN-7293 (compound 3).

Mutagenesis followed by sequencing of resistant clones confirmed Sec61 as their primary binding protein. Although this approach has a long history in yeast, to our knowledge this is the first study where an unbiased, genome-wide mutagenesis approach, followed by whole-genome-sequencing has identified the drug target in mammalian cells. Although the cotransin target in mammalian cells had already been well documented previously, this analysis was also performed as proof of principle. Wacker et al. (Wacker et al., 2012) have pioneered this approach for two substances with known binding proteins, using spontaneous resistance and total RNA sequencing. However, in that study the approach failed to unambiguously identify the targets. In our case, the analysis yielded Sec61 as the best-scoring candidate. This might be due to our strategy of inducing mutations rather than selecting spontaneously resistant clones. Multiple mutagenesis experiments in fungi and mammalian cells have revealed that there is a bias towards chromosomal aberrations and SNPs in pleiotropic drug-resistance genes if spontaneous mutants are selected and sequenced (Nyfeler et al., 2012; Richie et al., 2013; Sadlish et al., 2013; Shimada et al., 2013). The underlying mechanism might be that amino acid mutations in the essential, primary target can often be deleterious, and these cells are rapidly outcompeted. Thus in the absence of a strong selective pressure there is a bias against spontaneous mutations in these genes, in contrast to drug efflux components or gene copy alterations that have no detrimental phenotype under laboratory conditions.

It has been shown previously that the cotransins allow SRP-dependent targeting and binding of the ribosome–nascent-chain complex to the mammalian translocon (Besemer et al., 2005; Garrison et al., 2005). Our experiments show that decatransin, as well as a member of the cotransin family, are not limited to the inhibition of human or mammalian translocation, but act similarly on fungal and bacterial translocation. Their action is not specific to co-translational (SRP-dependent) translocation, but also inhibits post-translational (SRP-independent) substrates. Thus the presence of a ribosome bound to the translocon complex is not essential.

By extensive in vitro crosslinking of arrested nascent chains, MacKinnon et al., (MacKinnon et al., 2014) have recently showed that cotransin prevents the signal sequence from inserting into the translocation pore. The signal-anchor of TNFα could be crosslinked – preferentially by its C-terminal end and in a pattern suggesting a helical conformation – with cysteine residues engineered into Sec61α at the cytosolic top of the lateral gate (MacKinnon et al., 2014). To understand how the inhibitor blocks signal insertion we identified the resistance mutations. The five mutations causing resistance to cotransin CT08 recently published by MacKinnon et al. (MacKinnon et al., 2014) localize to the plug and to the lumenal end of the lateral gate helices TM2 and TM3. The conspicuous concentration to plug and lower gate suggested that this part of the structure constitutes the cotransin-binding site. Given that the affected residues point to the interior of the translocon in the closed state, the inhibitor could only bind there directly in an open conformation. Inhibitor binding was thus proposed to stabilize the plug and the partially opened gate, thereby preventing the signal from entering. The five mutations identified independently in human Sec61α1 in our study also localize to the same region, all in the plug domain. R66 was mutated in both studies to a total of three different residues (I, G and K). Given that the various affected plug residues point in very different directions, they are not likely to all contact the inhibitor directly.

In yeast, we identified a higher number of 22 mutations in 21 different residues conferring resistance to decatransin and/or cotransins; 16 mutations in 15 different residues were resistant to both chemotypes, indicating similar mechanisms of action. Several mutations also localize to the plug and gate region. In some instances, the same homologous residues were mutated, in one case even with the same amino acid exchange (S72F in yeast and S71F in human). However, the yeast mutations are distributed over a larger area than could be covered by a compact cyclic deca- or hepta-depsipeptide, suggesting conformational or allosteric effects to cause resistance by many mutations.

A variety of different mutations have previously been found to mediate the suppression of signal peptide defects, the prl (‘protein localization’) phenotype described in bacteria (Emr et al., 1981; Junne et al., 2007; Smith et al., 2005) and yeast (Junne et al., 2007). prl mutations specifically destabilize the closed state of the translocon, facilitating signal entry and initiation of peptide insertion and translocation. Mutations of the plug domain, constriction ring residues or of the lateral gate showed this phenotype, as well as mutations at other positions. The fact that almost all prl mutants showed some level of resistance to inhibition might be explained by ‘flexibilization’ of the interaction surfaces to reduce the binding affinity of the inhibitors. This might suggest that the inhibitors bind to the closed translocon or an early state of channel opening and stabilize it, thereby preventing signal insertion.

It is striking that both decatransin and cotransins inhibit translocons as distant in evolution as E. coli, yeast and man. Although the Sec61α subunit shares almost 60% identity between yeast and man (including 11 of 16 residues of the plug domain), there is less than 17% identity between the two eukaryotic sequences and bacterial SecY, with no sequence conservation in the plug. It thus appears unlikely that there are conserved specific inhibitor–protein interactions. Like signal sequences, the inhibitors are oligopeptides of hydrophobic (although in part unusual) amino acids. It is conceivable that they engage with the translocon in a similar manner as natural signals do – up to the point where their circular structure prevents the next step, such as the formation of an extended helix to intercalate into the lateral gate and contact the lipid phase. This state might block the translocon for an incoming signal. Destabilizing mutations (including the prl mutants) might allow for sufficient flexibility in the translocon to rapidly release the depsipeptides. According to this hypothesis, mutations that directly block inhibitor binding would also interfere with signal entry and might not be viable.

A fungal strain closely related to Chaetosphaeria tulasneorum, as determined by Internal Transcribed Spacer sequencing (White et al., 1990), was isolated from maple leaf debris in Germany. The genome was determined using Roche/454 sequencing and the Roche/454 Newbler assembler v.2.6. Two sequencing libraries were prepared, one shotgun library which generated 1,277,077 reads with an average read length of 676 bp, and a paired-end library with pair distance average of 2768 bp and a pair distance standard deviation of 692 bp. The paired-end library was sequenced twice, yielding a total of 953,342 paired reads with a peak depth of 29. Sequences were assembled using the Newbler 2.6 assembler with default options, except for specifying the scaffolding and four processors. A total of 50 scaffolds were generated, with an average size of 843,301 bp and an N50 of 2,374,876 bp (the average contig size within these scaffolds was 124,629 bp; N50 scaffold contig size 594,413 bp). Gene modeling and prediction of the genome was undertaken using Augustus (Keller et al., 2011). Augustus must be trained on each new fungus, and we used the CEGMA software (Parra et al., 2007) to provide an initial training set of spliced genes. In parallel, total RNA was isolated from the fungus using the RNeasy plant mini kit (Qiagen). RNA-seq libraries were prepared using an Illumina RNA prep kit, and sequenced using the Illumina HiSeq2000 platform. A total of 80 million 76-bp paired-end reads were generated. We supplemented the genome data with RNAseq following the Augustus RNAseq tutorial.

Using the gene predictions from Augustus, we predicted protein sequences. In addition, we identified all open reading frames >500 amino acids within the genome. Finally, we used Tophat 2.0.4 (http://ccb.jhu.edu/software/tophat/index.shtml) and Cufflinks 2.0.2 (http://cufflinks.cbcb.umd.edu) to identify the putative transcripts in the mRNA sample, and protein sequences were predicted from these. Using NCBI Blast with both Swissprot and the NCBI Non-Redundant protein data files, the predicted sequences were annotated. AntiSMASH was used to find secondary metabolite gene clusters.

The strain was cultivated in 200-ml shake flasks with 60 ml main culture medium (yeast extract 2 g/l, malt extract 1.6 g/l, soy protein 2 g/l, glucose 20 g/l, MgSO₄ 2 g/l, KH₂PO₄ 2 g/l) at 28°C and 200 rpm for 8 days after inoculation with 1.5 ml of a 6-day preculture (agar 1 g/l, yeast extract 4 g/l, malt extract 15.6 g/l). Compounds were isolated by normal and reversed phase chromatography. The structure was determined by mass spectroscopy and 1D- and 2D-NMR experiments. The spectroscopic methods structural and spectral data for compound 1 can be found in the supplementary materials (supplementary material Table S2).

The growth-inhibitory potency of test substances was determined using wild-type S. cerevisiae BY4743. The optical density at 600 nm (OD₆₀₀) of exponentially growing cultures in rich medium was recorded with a robotic system. Twelve-point serial dilutions were assayed in 96-well plates with a reaction volume of 150 µl; the start OD₆₀₀ was 0.05. Solutions containing DMSO were normalized to 2%. IC₃₀ values were calculated using logistic regression curve fits generated by TIBCO Spotfire v3.2.1 (TIBCO Software Inc.).

HIP, HOP and microarray analysis were performed as described previously (Hoepfner et al., 2014). Sensitivity was computed as the median absolute deviation logarithmic (MADL) score for each compound and concentration combination. z-scores are based on a robust parametric estimation of gene variability from >3000 different profilings and were computed as described in detail in previously (Hoepfner et al., 2014).

HIP and HOP profiles were validated by picking the individual strains from the HIP and HOP collections (OpenBiosystems, cat. number YSC1056 and YSC1055) and testing log-phase cultures in 96-well microtiter plates in YPD medium with serial dilutions of the compound. The assay volume was 150 µl/well, start OD₆₀₀ was 0.01, DMSO was normalized to 2%. Curves were calculated by taking the 11 h OD₆₀₀ measurements and applying a logistic regression curve fit in TIBCO Spotfire v3.2.1. Strain HO/YDL228C was used as the wild-type reference.

Strain BY4743Δ8 (Hoepfner et al., 2012) was incubated with 2.5% EMS until only 50% of the cells formed colonies. A total of 2×10⁷ mutagenized cells were plated on two 14-cm dishes with synthetic complete medium (0.7 g/l Difco Yeast Nitrogen Base without amino acids, 0.79 g/l MPbio CSM amino acid mixture, 2% glucose) containing 30 µM compound 1. After 4 days, 45 resistant colonies were isolated. Resistance due to mutated SEC61 was confirmed by cloning the corresponding mutations into fresh BY4743 cells and recording dose–response curves in YPD medium with serial dilutions of compounds 1 or 2 at 200 µM maximum concentration and 11 serial dilutions. DMSO was normalized to 2%. Curves were fitted by logistic regression curve fitting in TIBCO Spotfire v3.2.1 (TIBCO Software Inc.).

HCT116 cells were mutagenized by incubating with 2% EMS for 60 min aiming to kill 30% of the cells. A total of 1×10⁷ mutagenized cells were plated and allowed to recover for 1.5 doubling times under standard conditions (10% FBS supplemented medium, 5% CO₂, 37°C). Compound 2 was added at the minimal inhibitory concentration (MIC; 1 µM) when they were 50–80% confluent, and medium was changed every 3–4 days. Resistant colonies appeared within 2–4 weeks. Resistance was confirmed by retesting the cells for growth in a dilution series from one tenth to one hundred times the IC₅₀ (0.3 µM). For growth assays, CellTiter-Glo (Promega) cell viability assay reagent was used according to the manufacturer's instructions. The envision system was used for the readout, and IC₅₀ values were determined by using the logistic regression curve fit function in TIBCO Spotfire v3.2.1.

Stable resistant colonies were tested in growth curves with compound 1, compound 2 and taxol. Ten colonies were expanded to extract genomic DNA and total RNA for sequencing using the Qiagen ALL Prep DNA/RNA kit (Qiagen) and were quantified by Qubit Fluorometric quantification (Life Technologies). 100 ng DNA was fragmented using a Covaris E210 ultrasonicator to an average length of 300 bp. The DNA was end-repaired, and Illumina-compatible sequencing libraries were prepared using the NuGen DR ultralow library kit. Libraries were then multiplexed and captured using a combination of NuGen blockers and Agilent SureSelect XT capture oligonucleotides following the NuGen recommendations. Libraries for transcriptome sequencing were prepared using the TruSeq v2 mRNA sequencing protocol (Ilumina Corp.). Sequencing was performed on an Illumina HiSeq2500 using TruSeq v3 sequencing chemistry on a HiSeq v3 paired end flowcell. The read length for all sequencing runs was 2× 76 bp, according to manufacturer's instructions (Illumina). Sample demultiplexing was performed using CASAVA v1.8.2 with FASTQC v0.10.0.

The raw sequence reads were aligned to the human genome (hg19) using BWA version 0.5.9 (Li and Durbin, 2009). SNPs were called in two different ways. GATK version 1.6-11 was used to call SNPs for each of the resistant samples as well as for two samples of the unmutagenized reference strain HCT116 (McKenna et al., 2010). The SNPs of the two reference strain samples were then subtracted from the SNPs of the resistant samples. Secondly, we used a slightly modified version of the SNP calling method described previously (Wacker et al., 2012) to obtain SNPs at positions where the resistant mutant differs from the parental strain. SNPs were only kept if called against both of the reference strain samples. The combined set of SNPs from both methods was annotated using VEP for Ensembl v71 (Li and Durbin, 2009; McKenna et al., 2010; McLaren et al., 2010).

Yeast strain RSY1293 {matα, ura3-1, leu2-3,-112, his3-11,-15, trp1-1, ade2-1, can1-100, sec61::HIS3, [YCplac111 (LEU2 CEN) containing SEC61]} was used (Pilon et al., 1997). The substrate proteins dipeptidyl aminopeptidase B (DPAPB), carboxypeptidase Y (CPY), and CPYΔ3 with C-terminal triple HA epitope tags were described by Junne et al. (Junne et al., 2007), and H1* and derivatives by Goder et al. (Goder et al., 2004). All were cloned into pRS426 with a GPD promoter. Yeast cells expressing substrate proteins were in vivo pulse-labeled for 5 min with 150 µCi/ml [³⁵S]methionine and [³⁵S]cysteine (PerkinElmer Life and Analytical Sciences, Boston, MA), lysed with glass beads, heated to 95°C for 5 min with 1% SDS, cleared by centrifugation, subjected to immunoprecipitation, and analyzed by SDS-gel electrophoresis and autoradiography as described previously (Junne et al., 2006). Compound 1 and 2 were added in DMSO (≤1% of the medium) 0–30 min before and during the labeling period. Signals were quantified using a phosphoimager. Mutant sec61 sequences were cloned into YCplac111 (LEU2 CEN) and introduced into VGY61 {matα, ura3-1, leu2-3,-112, his3-11,-15, trp1-1, ade2-1, can1-100, sec61::HIS3, [YCPlac33 (URA3 CEN) containing SEC61]} (Goder et al., 2004), and the wild-type SEC61 plasmid was eliminated using 5-fluororotic acid for metabolic labeling experiments. COS-1 cells were grown, transfected, and labeled as previously described (Goder and Spiess, 2003). Inhibitors were added in DMSO (≤1% of the medium) during the labeling period. H1, H1Leu13, and H1Leu25 were as described previously (Wahlberg and Spiess, 1997).

In vitro translocation assays with purified components from Escherichia coli were performed essentially as described previously (Bauer et al., 2014). SecA (residues 1–831 with all cysteine residues replaced by serine residues and a C-terminal His₆ tag) and SecYEG (with all cysteine residues replaced by serine residues and with a N-terminal His₆ tag in SecE) as well as a derivative in which the plug domain of SecY (residues 60–74) was deleted were purified. The SecYEG complexes were reconstituted into proteolipsomes containing E. coli polar lipid extract. pOA-DHFR, a fusion of the first 175 amino acids of proOmpA and E. coli dihydrofolate reductase, was synthesized and radiolabeled with [³⁵S]methionine by in vitro translation with rabbit reticulocyte lysate (Promega) as described previously (Bauer and Rapoport, 2009).

Proteoliposomes containing 0.1 µM SecYEG were mixed with 0.4 µM SecA, 5 mM ATP and pOA-DHFR diluted 1∶50 from the reticulocyte lysate in buffer containing 50 mM Hepes-NaOH pH 7.5, 50 mM KCl and 5 mM MgCl₂. The mixture was split into equal volumes and incubated with different concentrations of compounds 1 or 2 or DMSO for 30 s on ice. Translocation was then initiated by incubation for 5 min at 37°C. Translocation of pOA-DHFR was tested by proteinase K treatment followed by SDS-PAGE and autoradiography.

We would like to acknowledge the Novartis Extended Natural Products Unit team that supported fermentation, isolation, and characterization of the compounds used in this study.