All the Saccharomyces cerevisiae strains used in this study (Supplementary Table 2) derive from the Σ1278b wild type (Bechet et al., 1970). The plasmids used are listed in Supplementary Table 3. Unless otherwise stated, yeast cells were grown at 29°C in minimal citrate-buffered (pH 6.1) medium (code numbers 165 for liquid and 167 for solid media) (Jacobs et al., 1980), with 3% glucose as the carbon source and 0.1% L-glutamic acid as the nitrogen source. In experiments where gene expression was driven by the GAL1-10 promoter, cells were grown overnight in minimal medium containing 3% galactose and 0.3% glucose as carbon sources and then shifted to 3% glucose 1.5 h before analysis.

Yeast and Lactobacillus fermentum (Beijerinck 1901 AL) were co-cultivated on a minimal medium containing glucose (3%) as a carbon source and ammonium as a nitrogen source added as (NH₄)₂SO₄ (0.5%). This medium (code number 169) is equivalent to 165 (Jacobs et al., 1980) except that citrate is replaced with MES hydrate (19.5 g.L^–1) as a buffer (pH 6.1) and that additional compounds are provided: folic acid (0.056 mg.ml^–1), nicotinic acid (0.09 mg.ml^–1), 4-aminobenzoic acid (0.0056 mg.ml^–1), riboflavin (0.09 mg.ml^–1), reduced L-glutathione (1.5 mg.ml^–1), and cobalt(II) chloride (0.19 mg.ml^–1). Cultures of L. fermentum alone were also supplied with amino acids (yeast synthetic drop-out without histidine, Sigma-Aldrich). For co-cultures, yeast and L. fermentum were first cultivated separately in 169 medium, collected by centrifugation, and washed twice in 169 medium. Cell suspension samples of equivalent optical density (∼0.05 at 660 nm) were used to fill 6-well plates equipped with ThinCert cell culture inserts (GREINER-657640, as outlined in Figure 5B). Each well was first filled with 2.5 ml yeast cell suspension and the Thincert insert containing 2 ml L. fermentum cell suspension was placed on top of it. The co-cultures were incubated for 48 h at 29°C with shaking at 100 rpm. Samples collected just after and 48 h after inoculation were diluted for counting the colony-forming units (CFUs) on rich medium.

Culture samples of exponentially growing cells were laid on a thin layer of 1% agarose supplemented with glucose, vitamins and metals, and visualized at room temperature by wide-field or confocal microscopy as previously described (Gournas et al., 2017). For fluorescence quantification at the cell surface vs. internal membranes, a custom-made Fiji macro (available in the Supplementary Material) was used. In each figure, only a few cells representative of the whole population, observed in at least two independent biological replicate experiments, are shown.

Yeast cells were grown for 72 h on 165 glucose glutamate medium, collected, and extracts were prepared as previously reported (Saliba et al., 2018). The extracts were then subjected to derivatization of amino groups followed by amino acid analysis as previously described (Fayyad-Kazan et al., 2016).

Donor cells bearing plasmid pGK008 and receiver auxotrophic cells (MG734 or Σ-A3hu) were first grown on rich solid medium for 48 h. Cell samples were resuspended in water and auxotrophic cells (350 μl of 0.7 optical density at 660 nm) were spread with sterile glass beads on minimal solid medium (167) in a single large Petri dish (120 × 120 × 17mm). Then, donor cells (10 μl of 2.0 optical density at 660 nm) were dropped on top of the receiver layer to form large-size colonies. The Petri dish was incubated at 29°C for 3 days (Thr cross-feeding) or 4 days (Hom cross-feeding). To quantify growth of peripheral colonies of the auxotrophic strains, a high-resolution image of the Petri dish was first obtained by scanning. The Fiji image processing package (Schindelin et al., 2012) was then used to convert the scanned image to 8bit and adjust brightness and contrast to remove the background signal. These adjustments were identical for images corresponding to all strains analyzed in the same biological replicate. A defined square centered on each donor cell colony was then extracted from the original images. The size of the square was identical for all strains in the same biological replicate. The central colony was then covered with a black disk of the same dimensions for each donor strain. In the next step, the white colonies in the remaining area (same dimensions for each strain) were labeled using the same adjusted threshold. Finally, we used the “Analyze particles” tool with circularity values ranging from 0 to 1 (0 for no circle, 1 for perfect circle, so that all particles are measured independently to their shape) to measure the area covered by the satellite colonies around the donor cells. The values were normalized to those corresponding to colonies around the gap1Δ agp1Δ gnp1Δ strain displaying maximal excretion.

A BLAST search performed on the PDB sequences (Altschul et al., 1990) identified, as appropriate templates for building Aqr1 models, the E. coli glycerol-3-phosphate transporter GlpT (Lemieux et al., 2005) and the E. coli multidrug transporter multidrug facilitator A (MdfA) complexed with chloramphenicol (Heng et al., 2015), both in the inward-facing conformation. Sequence alignments of Aqr1 with MdfA or GlpT were performed with PROMALS3D (Pei et al., 2008) and HHPRED (Söding et al., 2005). Sequence identity/similarity was calculated with the LALIGN program (Pearson, 2016), using standard parameters. Transmembrane domain predictions were performed on the Aqr1 sequence with TOPCONS (Tsirigos et al., 2015). Ten models of Aqr1 in the inward-facing occluded state were built with MODELER (Webb and Sali, 2016), five on the basis of Aqr1-GlpT sequence alignments and five others on the basis of Aqr1-MdfA sequence alignments. Docking of homoserine and threonine was performed as described previously (Ghaddar et al., 2014).

A yeast strain expressing the dominant HOM3-R allele carried on a centromeric vector overproduces threonine (Thr) and its precursor homoserine (Hom) (Martin-Rendon et al., 1993). We have previously reported that when this strain additionally lacks the three main permeases catalyzing Thr and Hom uptake (Gap1, Agp1, and Gnp1), excretion of both amino acids becomes detectable. This can be visualized by spotting a dense suspension of HOM3-R gap1Δ agp1Δ gnp1Δ cells on a solid minimal medium onto which yeast Thr- or Hom-auxotrophic cells have been spread: after a few days of incubation, a halo of satellite colonies of auxotrophic cells develops around the large colony formed by the spotted cells (Velasco et al., 2004). In the present work, we repeated these experiments and noticed that cross-feeding of the auxotrophs was particularly pronounced on a medium containing glutamate as sole nitrogen source. This medium was therefore chosen for subsequent experiments. In experiments where intracellular Thr and Hom levels were measured by ultra performance liquid chromatography (UPLC), HOM3-R cells growing on this medium were found to accumulate levels 10 to 20 times as high as those found in w-t cells, and high levels were also observed in HOM3-R cells lacking the Gap1, Agp1, and Gnp1 permeases (Figure 1). In cross-feeding experiments, as expected, colonies of Thr and Hom auxotrophs formed around HOM3-R donor cells only when these lacked the Gap1, Agp1, and Gnp1 permeases (Figure 2A). We then used these HOM3-R gap1Δ agp1Δ gnp1Δ donor cells, hereafter called 3R-3PΔ cells, to assess the contribution of the Aqr1 transporter to Thr and Hom excretion. When the AQR1 gene was deleted in the 3R-3PΔ strain, colonies of the auxotrophic strains still developed around the donor cells (Figure 2A). We observed, however, upon quantifying the development of these colonies in several independent experiments, that cross-feeding of both auxotrophic strains was significantly reduced. This suggests that Aqr1 contributes to excretion of both amino acids (Figure 2B).

That both Thr and Hom are still efficiently excreted when Aqr1 is not functional suggests that additional exporters are involved. Given the close phylogenetic proximity of the Qdr1-3 proteins and Aqr1 within the DHA1 protein family (Dos Santos et al., 2014), we suspected that these transporters might also contribute to Hom and Thr excretion. We thus deleted QDR3 in the 3R-3PΔ aqr1Δ strain and repeated the cross-feeding experiment. The 3R-3PΔ aqr1Δ qdr3Δ strain largely failed to cross-feed the Hom auxotroph (Figures 2A,B), indicating that Aqr1 and Qdr3 are the main Hom-excretion proteins under the conditions of this experiment. This mutant still cross-fed the Thr auxotroph, though with reduced efficiency (Figures 2A,B), indicating that Aqr1 and Qdr3 mediate Thr excretion as well. Consistently, intracellular Hom and Thr measured by UPLC were significantly higher in 3R-3PΔ cells lacking Aqr1 and Qdr3 (Figures 1A,B), a probable consequence of reduced excretion.

The observation that the 3R-3PΔ aqr1Δ qdr3Δ strain still efficiently cross-feeds the Thr auxotroph suggests that one or several other proteins, besides Aqr1 and Qdr3, contribute to Thr excretion. Consistently, when the QDR2 gene was deleted in the 3R-3PΔ aqr1Δ qdr3Δ strain, cross-feeding of the Thr-auxotrophic cells was strongly impaired (Figure 2A,B and Supplementary Figure 1). This deletion also caused a further significant increase in intracellular Thr concentration (Figure 1B).

In conclusion, using yeast cells genetically engineered to overproduce and excrete Thr and its precursor Hom, we have shown that the Aqr1 and Qdr3 transporters mediate Hom excretion, and that these proteins together with Qdr2 promote Thr excretion. This illustrates the important role of DHA1-family transporters in amino acid excretion.

To investigate the subcellular localization of Aqr1, Qdr2, and Qdr3, we first visualized cells where each protein fused to GFP was produced from a gene placed under the control of the galactose-inducible GAL promoter (Figure 3). Aqr1-GFP was detected at the plasma membrane, and a substantial fraction was also found in the vacuolar lumen, as judged by colocalization with the vacuolar dye CMAC. In contrast, Qdr2-GFP and Qdr3-GFP were found mostly at the plasma membrane. The endocytosis of many yeast transporters and their sorting to the vacuole depend on their ubiquitylation via the essential Rsp5 ubiquitin (Ub) ligase (Lauwers et al., 2010). In the npi1-1 strain, a viable hypomorphic mutant of RSP5 in which many yeast transporters fail to be properly ubiquitylated (Hein et al., 1995), Aqr1-GFP localized mainly to the cell surface, like Qdr2-GFP and Qdr3-GFP (Figure 3). We also investigated the localization of Aqr1-GFP, Qdr2-GFP, and Qdr3-GFP when each was produced under the control of its own gene’s promoter at the endogenous chromosomal locus (Supplementary Figure 2). Although fluorescence signal intensity was much weaker, each protein displayed a localization pattern similar to that observed when it was overexpressed. In particular, Aqr1 differed from Qdr2 and Qdr3 in that a large fraction of the protein was detected in internal membranes and in the vacuole. We wondered if this localization pattern might be due to altered trafficking of the protein caused by its fusion to GFP, but the Aqr1-GFP construct is functional, as judged by its ability to promote excretion of Hom and Thr in a HOM3-R strain lacking Qdr2 and Qdr3 (Supplementary Figure 3). Further analysis of Aqr1-GFP localization in mutants altered at specific intracellular trafficking steps (Supplementary Figure 4 and supportive text) revealed that at least part of the protein cycles constantly between the plasma and internal membranes, in keeping with previous observations (Velasco et al., 2004).

We next sought to determine whether localization of the Aqr1, Qdr2, and Qdr3 exporters might be altered in starved cells. We first examined cells after 90 min of nitrogen starvation (Figure 3). This treatment caused total relocation of Aqr1-GFP from the cell surface to the vacuole. A fraction of the Qdr3-GFP was also targeted to the vacuole, whereas Qdr2-GFP remained at the cell surface. In the npi1-1 strain, both Aqr1-GFP and Qdr3-GFP remained at the cell surface. We also analyzed the localization of each exporter after shifting cells to a glucose-free medium (Figure 3). This condition is known to cause rapid Ub-dependent down-regulation of many plasma-membrane transporters, a response contributing to cell survival under these conditions (Lang et al., 2014). In wild-type cells shifted to glucose starvation for 90 min, Aqr1-GFP and Qdr3-GFP were efficiently targeted to the vacuole. This down-regulation was reduced though still effective in the npi1-1 mutant. In contrast, Qdr2-GFP remained at the plasma membrane (Figure 3).

In conclusion, although Aqr1, Qdr2, and Qdr3 share high sequence similarity and are all three involved in amino acid export, they differ appreciably as regards their turnover at the plasma membrane under nutrient sufficiency and starvation.

Solving at the atomic level the structures of Aqr1, Qdr2, and Qdr3 would be most helpful toward understanding how these proteins function as amino acid exporters. As such experimental data are unavailable, we adopted a structural modeling approach focusing on Aqr1, chosen as a representative DHA1-family amino acid exporter.

After comparing the Aqr1 sequence with those of proteins available in the Protein Data Bank (PDB), we selected the crystal structures of two Escherichia coli proteins as templates for building Aqr1 structural models: the MdfA multidrug-resistance protein (Multidrug Facilitator A; PDB code: 4ZOW) (Heng et al., 2015) and the GlpT glycerol-3-phophate antiporter (PDB code: 1PW4) (Lemieux et al., 2005), both in the Inward Facing (IF) conformation. These bacterial transporters belong to the MFS superfamily and their transport activity is driven by the proton gradient at the plasma membrane (Edgar and Bibi, 1997; Lemieux et al., 2005). For each bacterial transporter template, five 3D models of Aqr1 were built and their stereochemical quality was verified (see details in Supplementary Figure 5). One of these Aqr1 structural models is shown in Figures 4A,B.

MdfA has been crystallized in a complex with chloramphenicol, one of its substrates (Heng et al., 2015). Using the 3D models of Aqr1 superimposed onto the MdfA structure template, we were able to predict that the substrate-binding pocket of Aqr1 is mainly shaped by residues in TM1, -2, -4, -5, -6, -7, and -11. To identify, among these residues, those potentially involved in forming substrate contacts, we performed docking of Thr and Hom in the predicted binding pocket of Aqr1. In a significant number of poses predicted by the models, a small number of Aqr1 residues were repeatedly found to form hydrogen bonds or salt bridges with both Thr and Hom (Gln 149, Gln 238) or with Hom only (Arg 504) (Supplementary Figures 5A,B and Supplementary Table 1). These residues of Aqr1 are thus predicted to be directly involved in recognition of the excreted amino acids.

To assess this prediction, we used CRISPR-Cas9 to introduce nucleotide substitutions into the chromosomal AQR1 gene of the 3R-3PΔ qdr2Δ qdr3Δ strain (Mans et al., 2015), where Aqr1 is the sole functional Thr/Hom exporter. The resulting strains expressing the Aqr1(Q149A), Aqr1(Q238A), or Aqr1(R504A) variant were then used as donors in Thr and Hom cross-feeding experiments (Figure 4C and Supplementary Figure 6). Excretion of both amino acids was found to be markedly reduced in cells expressing Aqr1(Q149A) or Aqr1(R504A), whereas cells expressing Aqr1(Q238A) displayed significant reduction of Hom excretion only. We also measured intracellular concentrations of Thr and Hom in the mutant strains (Supplementary Figure 7). In agreement with the results of the cross-feeding experiments, cells expressing the mutant Aqr1 variants displayed higher accumulation of both amino acids, most likely due to their impaired excretion. To assess the possibility that reduced Thr and Hom excretion in the mutant strains might be due to mislocalization of the Aqr1 variants, we analyzed cells in which the wild-type and mutant chromosomal AQR1 genes were fused in frame to GFP. Initial analysis showed the mutant forms of Aqr1-GFP to localize roughly like the wild-type protein, to the cell surface, internal puncta, and the vacuole (Figure 4D). Quantification of the fluorescence signals further revealed more abundant Aqr1(R504A)-GFP and Aqr1(Q149A)-GFP at the cell surface than in internal membranes (Figure 4E). This suggests that these Aqr1 mutants, which are the less active in Thr and Hom excretion (Figure 4C), are also less prone to endocytosis and turnover.

In conclusion, these results support the predictions based on Aqr1 structural modeling and show that residues R504, Q149, and Q238 in the substrate-binding cavity of Aqr1 play an important role in amino acid excretion. The results are also consistent with the view that endocytic downregulation of Aqr1 is promoted by its transport activity, as observed for several amino acid transporters (Ghaddar et al., 2014; Gournas et al., 2017).

Yeast and lactic acid bacteria (LAB) interact metabolically during natural fermentation of several foods and beverages (D’Souza et al., 2018). Previous works have shown that yeast can supply LAB with several amino acids which the latter cannot synthesize naturally (Challinor and Rose, 1954; Ponomarova et al., 2017). In addition to mediating excretion of Thr and Hom, Aqr1, Qdr2, and Qdr3 might export additional amino acids and thus contribute to cross-feeding LAB. To assess this possibility, we first adjusted the composition of the liquid minimal buffered medium used in the above experiments to make it compatible with growth of Lactobacillus fermentum. Specifically, citrate used as a buffer and found to impede growth of this bacterium was replaced with MES hydrate, and additional vitamins were provided (see section “Materials and Methods”). The results presented in Figure 5A show that L. fermentum grew well on this medium only when supplied with amino acids, confirming that this LAB species is auxotrophic for several amino acids (Deguchi and Morishita, 1992; Oliva Neto and Yokoya, 1997). We then cultivated cells in wells consisting of two compartments separated by a solute-permeable membrane (Figure 5B). One compartment was inoculated with yeast cells of the wild-type or the aqr1Δ qdr2Δ qdr3Δ mutant strain, and the other with L. fermentum. Just after inoculation and after 2 days of growth, the cell density was quantified in culture samples by counting the number of colony-forming units (CFU/ml). The results first revealed that L. fermentum, when co-cultivated with wild-type yeast, propagated rather well (Figure 5C), though less than the yeast cells. This shows that the bacterium was cross-fed with amino acids supplied by the yeast. Importantly, the relative propagation of L. fermentum vs. yeast was reduced _∼2.5-fold when the bacterium was co-cultivated with the aqr1Δ qdr2Δ qdr3Δ yeast mutant strain (Figures 5C,D). This reduced cross-feeding was not observed when L. fermentum was co-cultivated with an aqr1Δ, qdr2Δ, or qdr3Δ single mutant (Figure 5E). These results indicate that Aqr1, Qdr2, and Qdr3 contribute redundantly to the excretion of amino acids promoting growth of L. fermentum.