Sodium acrylate (Sigma-Aldrich, Søborg, Denmark), sn-Glycero-3-phosphocholine, >98% (Santa Cruz Biotechnology, Inc., Heidelberg, Germany), cefadroxil 95–105%, (VWR, Søborg, Denmark), 2-methylpyrazine (VWR, Søborg, Denmark), and 2-benzoxazolinone, ≥98.0% (VWR, Søborg, Denmark), 2,5-furandicarboxylic acid (TCI Europe N. V., Zwijndrecht, Belgium) were used in the study. Xenopus oocytes were incubated in either HEPES-based Kulori buffer (90 mM NaCl, 1 mM KCl, 1 mM MgCl₂, 1 mM CaCl₂, 5 mM HEPES pH 7.4) supplemented with gentamycin (100 μg/mL), or MES-based Kulori buffer (90 mM NaCl, 1 mM KCl, 1 mM MgCl₂, 1 mM CaCl₂, 5 mM MES pH 5.0).

Yeast knockout mutant strains with BY4741 (MATa; his3Δ1; leu2Δ0; met15Δ0; ura3Δ0) background were purchased from Open Biosystems (catalog # YSC1053). For Growth Profiler 960 (Enzyscreen B.V., CR9001) measurements, yeast pre-cultures were inoculated in 2 mL synthetic dextrose medium (SD –Trp) (20 g/L glucose, 7 g/L yeast nitrogen base, 2 g/L histidine, 3 g/L leucine, 2 g/L methionine, 2 g/L uracil) in 15 mL growth tubes and grown for 24 h at 30°C and 250 rpm. Optical density at 600 nm (OD₆₀₀) was then measured and adjusted to 0.01, in a 24 deep-well plate (Enzyscreen B.V., CR1424d), containing 2 mL SD –Trp supplemented with the indicated concentrations of additives. The cultures were grown for 72 h at 30°C with 300 rpm agitation and measurements acquired every 20 min.

Pooled human blood serum, 0.1 μm filtered was purchased from BioIVT (West Sussex, United Kingdom). S. cerevisiae strain CEN.PK113-7D (MATa URA3 HIS3 LEU2 TRP1 MAL2-8c SUC2) was a gift from Peter Kötter (Goethe University, Frankfurt/Main, Germany).

An overnight preculture of CEN.PK113-7D was grown in YPD overnight, resuspended in YPD to an OD₆₀₀ = 2. At OD₆₀₀ = 8, 1 mL of culture was taken and the cells were centrifuged at 3,000 × g for 5 min. The cells were washed, and resuspended in MilliQ to a final OD₆₀₀ of 2. 1 mL of the yeast solution was centrifuged at 3,000 × g for 5 min, and resuspended in 200 μL pooled human blood serum, to a final OD₆₀₀ of 10. The experiment was performed in duplicate, with two independent biological replicates. After 30 min of incubation at 30°C in a shaking incubator at 250 rpm, the cells were collected by centrifugation at 3,000 x g for 5 min and 200 μL of the supernatant was mixed with 700 μL of ice-cold methanol. Filtered pooled human blood serum was used as control. The mixtures of supernatant or serum in methanol were mixed by vortexing, and cleared for protein precipitate by centrifugation. The supernatant was aliquoted and dried in a centrifugal evaporator with no temperature application. Prior to LC–MS analysis, samples were resuspended in LC–MS grade water.

All analyses were performed using an Ultimate 300 LC system (Thermo Fisher Scientific) using a HypersilGold a C18 2.1 mm x 100 mm, 1.9 μm column (Thermo Fisher Scientific), coupled to an Orbitrap Fusion Tribrid Mass Spectrometer (Thermo Fisher Scientific). Parameters for LC elution gradient and parameters for data acquisition on the MS/MS system are described in Wright Muelas et al., (2020).

The raw data files were imported into Progenesis QI (Non-linear Dynamics, Newcastle upon Tyne, United Kingdom) for processing. Peak alignment, picking and compound identification were performed using the software. Basic descriptive analysis, as well as statistical analysis to calculate fold changes in metabolites of yeast incubated pooled human blood serum versus normal pooled human blood serum was done in R (version 3.5.1). A threshold of Log₂ fold change of compound levels in yeast supernatant relative to serum >0.5 with a CV < 30% was used to identify consumed and secreted compounds.

To express the transporter coding genes in oocytes, genes were cloned downstream of the T7 promoter in the USER compatible Xenopus expression vector pUSER016 (Nour-Eldin et al., 2006). The empty vector was digested by PacI and Nt.BbvCI (New England Biolabs). The transporter genes were amplified from genomic DNA of CEN.PK113-7D or BY4741, where indicated. The amplified DNA fragments were gel purified and together with the linearized plasmids incubated with USER enzyme (New England Biolabs) for 25 min at 37°C, followed by incubation at 25°C for 25 min. The reactions were transformed into chemically competent E. coli cells. All the cloned plasmids were verified by Sanger sequencing. See Supplementary Tables S1–S3 for the sequences of primers and a list of the plasmids constructed in this study (Supplementary Table S4).

The selected transporters were cloned into expression vectors using USER cloning. The expression vector contained the T7 transcription initiation site and a poly-A tail and terminator. The full construct was amplified by PCR, and in vitro transcribed using the mMessage mMachine T7 transcription kit from Thermo Fisher Scientific. The quality of the generated RNA was analyzed using an Agilent 2,100 Bioanalyzer. Defollicated stage V-IV oocytes from Xenopus laevis were purchased from Ecocyte Bioscience (Dortmund, Germany). Individual oocytes were distributed into a 96 well plate with V-bottom containing Kulori buffer pH 7.4, and injected with 50 nL of a mix of 100 ng/μL RNA of GFP and 400 ng/μL RNA of a transporter of interest using a Roboinject 96 well multichannel injection system (Multi Channel Systems MCS GmbH, Reutlingen, Germany). Control oocytes were injected with 100 ng/μL RNA of GFP only. The oocytes incubated for 3 days at 18 °C.

The uptake assay was carried out as described in Nour-Eldin et al. (2006) and Wulff et al. (2019). After preincubation the oocytes were transferred to the assay solution containing Kulori and the compounds at a concentration of 2 mM each, with the pH of the solution adjusted to pH 5. The oocytes were incubated in the assay solution for 3 h at 18 °C. After washing with Kulori pH 7.4, pools of 10 oocytes were transferred to an Eppendorf tube, and the oocytes were disrupted by pipetting together with 50% ice cold MeOH. The oocyte mix was stored for min 2 h at −20°C, and the supernatant collected by centrifugation at 20,000 g for 15 min. The supernatant was further diluted with MilliQ to a final concentration of 29% MeOH before the samples were analyzed by LC–MS. LC–MS data was collected on EVOQ EliteTriple Quadrupole Mass Spectrometer system coupled with an Advance UHPLC pump (Bruker, Fremont, CA, United States). See Supplementary Table S5 for further information on the LC–MS method.

LC–MS data was collected on EVOQ EliteTriple Quadrupole Mass Spectrometer system coupled with an Advance UHPLC pump (Bruker, Fremont, CA, United States). Samples were kept cold in the CTC HTS PAL autosampler at a temperature of 5.0°C during the analysis. Injection volume was 1 μL, injected onto a Waters ACQUITY HSS T3 C18 UHPLC column (1.8 μm, 2.1 mm x 100 mm long). The column was held at a temperature of 35.0°C. The solvent system used was Solvent A “MilliQ water with 0.1% formic acid” and Solvent B “Acetonitrile with 0.1% formic acid.” The flow rate was 0.400 mL/min with an initial solvent composition of % A = 100, % B = 0 held until 0.50 min, the solvent composition was then changed following a linear gradient until it reached % A = 5.0 and % B = 95.0 at 1.00 min. This was held until 2.79 min when the solvent was returned to the initial conditions and the column was re-equilibrated until 4.00 min. The column eluent flowed directly into the Heated ESI probe of the MS which was held at 200°C and a voltage of 4,500 V. MRM Data was collected in positive ion mode. Other MS settings were as follows, Sheath Gas Flow Rate of 50 units, Nebulizer Gas Flow Rate of 50 units, Cone Gas Flow Rate of 20 units Cone Temp was 350°C, and collision gas pressure 1 mTorr.

The exometabolome data was visualized using Umap, standard settings. We decided to use UMAP as it is often favored over PCA due to its ability to capture non-linear relationships, preserve both local and global structure in the data, produce more distinct clusters, and better separate different groups within the data. Significance was determined with Student’s t-test. P-levels >0.005 determines significance. Outliers were identified by the interquartile method. After the outliers were identified, they were removed and the statistical analysis redone. Blast was carried out using NCBI protein blast with standard settings. For blast searches in TCDB, Blosum62 was used as scoring matrix, and cut-off values were the standard settings. Data were visualized using UMAP, and analyzed using R and JMP. Phylogenetic tree was calculated using clustal Omega (Sievers et al., 2011) and visualized using iTOL (Letunic and Bork, 2007, 2021).

For expression, we selected transporter proteins from secondary active transporter superfamilies known to transport drugs and natural compounds. Bacterial transporter proteins of the drug/metabolite transporter (DMT) superfamily transport natural products and toxins, like camphor, chloroquine, and ethidium bromide (Berninsone and Hirschberg, 2000; Caffaro and Hirschberg, 2006; Västermark et al., 2011). While yeast transporters of the DMT superfamily are largely uncharacterized, we hypothesized that they could include transporters of xenobiotic compounds. The multidrug/oligosaccharide/polysaccharide (MOP) superfamily includes the multi-antimicrobial extrusion (MATE) family, which consists of well-characterized xenobiotic efflux pumps across several organisms (Hvorup et al., 2003; Tanihara et al., 2007; Staud et al., 2013). At the same time, the two MATE transporters present in the yeast genome, YDR338C and Erc1 (Shiomi et al., 1995), have not previously been characterized for the uptake of non-toxic xenobiotic compounds. We also included transporters from the major facilitator superfamily (MFS), specifically the aromatic acid exporter (ArAE) family, and the yeast Equilibrative Nucleoside Transporters (ENT) family. We further reduced the list of transporters by omitting sugar transporters and other experimentally validated nutrient transporters. We also excluded the ABC transporters, as they tend to be exporters rather than importers, and the direction of transport is normally fixed via the internal hydrolysis of ATP.

The workflow of Xenopus oocyte uptake assay is described in Figure 1A. Briefly, the oocytes were injected with mRNA encoding the transporter of interest and mRNA encoding GFP and incubated for 3 days to allow protein expression. Fluorescence was used as a quality control to select for the oocytes that received mRNA and expressed the proteins, supposedly both GFP and transporter. The transporter expressing oocytes were incubated with the solution of several selected xenobiotic metabolites in pH 5 buffer. The low pH was chosen because the yeast S. cerevisiae is known to proliferate in an acidic environment. After incubation, the concentration of xenobiotic metabolites was determined in the buffer and in the oocytes by LC–MS.

An initial screen to establish the small molecules that yeast cells can transport was conducted (Figure 1B). Yeast cells were incubated in human blood serum for 30 min, removed by centrifugation, and the supernatant analyzed on LC–MS. There was a concentration change (log₂FC > 0.5) in 385 compounds (Supplementary Dataset S1 and Figure 2A). We observed a considerable variation in response across replicates in the yeast-treated samples compared with the untreated samples (Figure 2A), though not to the extent where the compound changed the direction of transport. The log₂FC values in concentration compared to the untreated samples were predominantly negative (total of 221 compounds, 169 metabolites and 52 drugs/xenobiotics), indicating the import of compounds from the serum or the metabolism and degradation of compounds. On the other hand, the log₂FC values were also positive for 164 compounds (97 metabolites and 67 drugs/xenobiotics), suggesting an increase in extracellular concentration due to the export or antiport of yeast metabolites or enzymatic activity on serum compounds (Figure 2B). The UMAP visualization showed a skewness, with compounds categorized as drugs present on the right-hand side of the U1 axis and compounds categorized as metabolites on the left (Figure 2C). This distinction in compound class has seemingly no influence on which compounds yeast interacts with, indicating that in our experiment, yeast had no substrate preference to drugs or metabolites.

Subsequently, we overlaid the identified compounds with a log₂FC > 0.5 from the exometabolome experiment data with a small natural compound library (O'Hagan and Kell, 2019) (Figure 2C). The overlay revealed that the small compounds categorized as drugs tend to have higher similarity to natural products, which was also previously described (O’Hagan and Kell, 2017), with most marketed drugs and natural products displaying a Tanimoto similarity of 0.7 or higher.

The 385 compounds were evaluated for their commercial availability, solubility in water, and general commercial interest (see Supplementary Dataset S2). We then tested the top-ranking compounds for toxic effects on the oocytes and excluded compounds that resulted in an above-average death rate. Additionally, compounds not precisely detected by our MS method were also excluded. The six compounds selected for testing were sn-glycero-3-phosphocholine (GPC), 2,5-furandicarboxylic acid (FDC), 2-methylpyrazine (MPZ), cefadroxil (CFX), acrylic acid (ACA), and 2-benzoxazolol (BXZ). The compounds represent metabolites (GPC), compounds with natural origin (MPZ, BXZ, and FDC), and synthetic compounds (CFX and ACA). In the UMAP visualization, the six compounds are evenly distributed (Figure 2C), with MPZ and FDC being the most closely related compounds.

Both FDC and ACA are industrially relevant. FDC is one of the US Department of Energy’s top 12 compounds for establishing a biology-based chemical industry (Bozell and Petersen, 2010), while ACA is an essential monomer in the chemical industry. ACA is approved as a food contact material and acidity regulator (Silano et al., 2018). Both ACA and FDC are formed by excessive heating of food and enter the body via the diet. ACA and its precursor 3-hydropropionic acid have been produced in cell factories (Dishisha et al., 2015). CFX is a broad-spectrum antibiotic with several known transporters in humans [SLC15A1 (Ganapathy et al., 1995; Wenzel et al., 1996), SLC15A2 (Ganapathy et al., 1995; Terada et al., 1997; Ocheltree et al., 2004; Luckner and Brandsch, 2005), SLC22A5 (Ganapathy et al., 2000), SLC22A6 (Jariyawat, 2000; Jung et al., 2002; Takeda et al., 2002), SLC22A8 (Jung et al., 2002; Takeda et al., 2002)]. MPZ is a naturally occurring flavoring agent found in many foods, while BXZ is a compound naturally occurring in cereal and cereal products and rye seedlings. Lastly, GPC is a native metabolite in humans, yeast, plants, and Xenopus laevis.

The six selected compounds were used to screen the transport activity of transporters expressed in Xenopus oocytes. For normalization, we used the intracellular concentration of compounds in control oocytes that expressed only GFP, but no yeast transporters.

A positive Log₂FC is a measure of uptake activity compared to the control. Conversely, a negative Log₂FC is a measure of export activity compared to the control (Figure 3A). The compounds were added in excess (2 mM), to ensure saturation for the transport protein was reached. Likewise, the incubation time (3 h) was chosen to ensure saturation. Export activity of transporters can be evaluated if the compound is imported into the GFP control by oocyte endogenous transporters or if the compound compromises the membrane of the oocytes and leaks in Figure 3B. Otherwise, export activity measurement would require injection of metabolites into oocytes, so called “export assay,” which was not performed in this study.

We identified transporter candidates with a significant response compared to the GFP control for all the tested compounds (Figure 4). Eight transporter candidates were identified for the three natural compounds, BXZ, MPZ, and FDC. For BXZ, the transporter proteins Rtc2, Thi74, and Ecm3 generated a significant response (Figure 4B). For MPZ, the transporter proteins Ypq2, YLR152C, and YNL095C generated a significant response (Figure 4F), and for FDC, the transporter proteins Thi74, YNL095C, Sly41, and Hut1 generated a significant response (Figure 4C). It is worth noting that Thi74 showed a negative Log₂FC for BXZ, and Hut1 for FDC, suggesting Thi74 and Hut1 are exporters. All other transporters showed a positive Log₂FC in intracellular concentration, suggesting the transporter proteins are importers. As prolonged incubation with BXZ and FDC did not result in increased mortality in oocytes, we do not have a reason to believe that BXZ or FDC compromised the oocyte membrane. The considerable variation between samples for Thi74 across the tested substrates could suggest difficulties with expressing Thi74 in oocytes.

For the synthetic compounds ACA and CFX, we identified nine transporters. The transporter proteins YDR338C, Ypq2, YMR253C, Ypq1, and Fsf1, showed a significant response for ACA (Figure 4E). For CFX, we identified both proteins in the paralog pair Thi74 and YML018C as candidates, together with the transporter proteins Mam3 and YLR152C (Figure 4A).

While all the compounds mentioned were identified as substrates for between 2 and 4 transporters, the metabolite GPC was the substrate of 12 transporters (Figure 4D). We identified both paralogs Thi74 and YML018C as GPC transporter candidates, but only Gup1 from the paralog pair Gup1/Gup2. Other identified GPC candidates were Ypq2, Yea4, Hvg1, Erc1, YLR152C, Ymd8, Rch1, YMR253C, and Fsf1. Ypq2, Yea4, YLR152C, and Fsf1 all showed a negative Log₂FC in intracellular concentration, suggesting the transporter proteins are exporters.

To describe the thermodynamic nature of the transport mechanism of the identified transporter proteins acting as importers [i.e., equilibrative vs. concentrative (Kell and Oliver, 2014)], we compared the normalized intracellular concentration of the compounds found in the oocyte with the extracellular concentration found in the medium (Figure 5A). As the compounds were added in excess, and the volume of the medium was significantly higher than the volume of the oocytes, we assumed that the concentration found in the medium was constant for all experiments. A Log₂FC > 0 suggests a concentrative mode of transport, indicating that the internal concentration within the oocytes surpasses that found in the medium. Conversely, a Log₂FC < 0 indicates an equilibrative mode of transport, where the internal concentration within the oocyte does not surpass that found in the medium. The comparison revealed that only the transporters of BXZ (Ecm3, Rtc2) and ACA (Fsf1, YDR338C, YMR253C, Ypq1, and Ypq2) import their substrates in a concentrative manner (Log₂FC > 0) (Figure 5B), while CFX, FDC, GPC, and MPZ as substrates are effectively imported in an equilibrative manner (Log₂FC < 0) (Supplementary Table S5). However, when the GFP expressing control is taken into consideration, it becomes apparent that the transporters of BXZ and ACA have a systemic bias. For BXZ and ACA, the GFP control has a Log₂FC > 0. This indicates uptake of the compounds by an endogenous transporter. While subtracting the effect from endogenous transporters suggests YDR338C and YMR253C transport ACA concentratively, further experiments are needed to verify the results.

In the oocyte uptake assay, we identified GPC as the substrate for the highest number of transporters (12). The comparison with the concentration found in the medium revealed that all transporters transported GPC equilibratively. We previously observed that GPC could enter the oocyte through an unknown mechanism, possibly through leakage or through an endogenous transporter. We rule out the possibility that the oocytes are compromised in general by GPC, as we did not detect all six compounds intracellularly in the GFP-expressing control oocytes.

If GPC is excluded, all concentrative transporters but Ypq2 transported only one compound (ACA), strictly in a concentrative manner. The remaining equilibrative transporters transported from one substrate (Sly41/FDC) up to three substrates for Thi74 (BXZ, CFX, FDC). The concentrative nature of the ACA transporters would suggest they are either primary or secondary active transporters, able kinetically and thermodynamically to transport their substrate in a uni-directional way under the conditions tested.

Examining the direction of transport found in the data from the exometabolite assay, ACA was found to be effectively imported rather than exported. Our results suggest that ACA, in general, is imported into the cell by the transporters YMR253C, YDR338C, Fsf1, Ypq2, and Ypq1, where YMR253C and YDR338C potentially transport ACA in a concentrative manner. When further assessing the oocyte uptake data, we identified Thi74 as an exporter of BXZ rather than an importer. In the exometabolite assay, we had identified BXZ to be effectively exported, meaning either Ecm3 and/or Rtc2 transports BXZ in a bi-directional manner, or other transporters exist, including Thi74. For CFX, FDC, and MPZ, all three compounds were found to be effectively exported in the exometabolome assay. In the oocyte uptake assay, we had a higher extracellular concentration, identifying solely equilibrative transport for CFX, FDC and MPZ.

Overall, we found the equilibrative transport of compounds to be a widespread phenomenon among the tested transporters and more common than concentrative transport. Moreover, we found that the ability to transport a compound concentratively was not restricted to a specific compound class, though assessing a larger chemical library is needed to investigate the phenomenon further.

We further tested the effect of the identified transporters from the oocyte screen in vivo by growing S. cerevisiae strains harboring deletions of the single genes in medium with the supplementation of one of the six selected compounds. The strains were grown at the LC50 concentration of the compounds, to allow both improvements in growth and growth inhibitions to be detectible (the data forming the basis for the LC50 calculations is available in Supplementary Table S6). We were not able to obtain a growth defect approximating to 50% for the reference BY4741 strain grown in SD-Trp supplemented with GPC or CFX. Therefore, we omitted the two compounds from the final growth experiment. We also omitted FDC from the final growth experiment, as the compounds precipitated at the calculated LC50, which biased the results. Using the calculated LC50 concentrations for ACA, BXZ and MPZ, strains harboring deletions of the identified transporters were cultivated in SD-Trp with and without the compounds in a growth profiler, the growth curves can be found in Supplementary Figures S1–S4. Figure 6 shows the calculated maximum OD (Max OD) and the maximum specific growth rate (μ_max) using the growth curves For acrylic acid, the strain harboring a deletion of FSF1 (fsf1∆), showed a decrease in μ_max rate when grown in SD-Trp (Figure 6A). The decrease in growth rate was rescued in the presence of acrylic acid. In addition, there was no difference in maximum OD for the ACA transporters deletion strains in SD-Trp, while when grown in the presence of ACA, all the ACA transporters deletion strains had a significantly lower maximum OD compared to the wildtype strain (Figure 6A).

For BXZ transporters, rtc2∆ mutant had lower μ_max and thi74∆ had higher μ_max in SD-Trp than the reference strain (Figure 6B). In the presence of BXZ, neither of the three strains tested had a significant difference in μ_max, while all tested strains had a significant decrease in maximum OD. For MPZ transporters, only YLR152C∆ exhibited a growth phenotype (Figure 6C). YLR152C∆ showed a significant increase in μ_max compared to the reference when grown on SD-Trp, while this μ_max was not significantly different from the reference when grown in the presence of MPZ (Figure 6C).

For transporters, divergent evolution is overwhelmingly the most common means of evolution, i.e., similar sequences evolve to transport different substrates (Saier, 1994; Wang et al., 2016; Jørgensen et al., 2017). To look into the conservation of substrate specificity among the identified transporters, we generated a phylogenetic tree as well as assessed their transporter classification for indications of a potential co-substrate.

Among the chosen transporters were a few pairs of paralogs with a considerable sequence similarity: Thi74/YML018C, Rtc2/Ypq1, Gup1/Gup2, Ecm3/YNL095C (Figure 7). For Thi74 and YML018C, we identified CFX and GPC as shared substrates, while FDC and BXZ were unique to Thi74. Both Rtc2/Ypq1 and Gup1/Gup2 failed to show an overlap in the substrate. The lack of functional overlap for the paralog pairs could suggest evolutional diversification through relatively few mutations. Indeed, paralog transporters have been previously shown to have different substrate specificities (Ramaswamy et al., 2017; Jiang et al., 2023). Consequently, the substrate of a transporter protein is hard to predict based on the amino acid sequence alone. There was no clear correlation between sequence similarity and substrate specificity in our study.

As a means of further validating our results, we sought to verify the presence of co-substrates for the transporter families. The presence of a co-substrate would indicate whether there is a known source of electrochemical power and hence would suggest the transporter can transport substrates concentratively. Assessing the identified transporters’ TCDB classification (Saier, 2006; Saier et al., 2016) did not give any further insight. All transporters, except Mam3, belong to the pores transporter class, including both symporters, uniporters, and antiporters (Supplementary Table S7), giving no additional information concerning potential co-substrates.