Saccharomyces cerevisiae Requires CFF1 To Produce 4-Hydroxy-5-Methylfuran-3(2H)-One, a Mimic of the Bacterial Quorum-Sensing Autoinducer AI-2

Quorum sensing is a cell-to-cell communication process that bacteria use to monitor local population density. Quorum sensing relies on extracellular signal molecules called autoinducers (AIs).

6 as the carbon source. Following 48 h of incubation, cell-free culture fluids were prepared and 124 assessed for an activity capable of inducing light production in the V. harveyi AI-2 reporter strain 125 called TL-26 (29). V. harveyi TL-26 produces maximum light in response to supplementation with 126 125 nM pure AI-2 (S-THMF-borate) ( Figure S1A, dotted line). High level activity was present in 127 the S. cerevisiae cell-free culture fluids, suggesting that S. cerevisiae produces an AI-2 mimic 128 ( Figure S1A).

129
Based on our finding that human epithelial cells produce the mammalian AI-2 mimic when 130 starved in PBS, and, again with the goal of facilitating purification, we next assessed AI-2 mimic 131 production under starvation conditions. S. cerevisiae was grown to saturation in rich medium, 132 washed twice, resuspended in either water or PBS and incubated overnight at room temperature.

137
To discover whether AI-2 mimic production occurred broadly among wild yeasts or was 138 restricted to laboratory S. cerevisiae, a panel of wild S. cerevisiae isolates obtained from different 139 environments ranging from clinical settings to vineyards was tested for production of activity using 140 the V. harveyi TL-26 reporter ( Figure S1B) (30, 31). All the production profiles mirrored that of 141 laboratory S. cerevisiae, suggesting that the AI-2 mimic is broadly made by S. cerevisiae strains.

142
To garner sufficient yeast AI-2 mimic for structural analysis, we tested the limit to which 143 we could concentrate the activity. At the final step of the above preparation procedure, the washed 144 S. cerevisiae cells were resuspended in water at different cell densities, from OD600 = 1 to OD600 145 = 128. Following overnight incubation, cell-free culture fluids were analyzed for activation of light 146 production in V. harveyi TL-26. Yeast AI-2 mimic activity increased with increasing S. cerevisiae 147 cell density ( Figure S1C). Moreover, the activity was specific to the AI-2 quorum-sensing pathway, 148 as light production was not induced by the preparations when supplied to a V. harveyi reporter 7 strain (TL-25) that is incapable of detecting AI-2 but, rather, responds exclusively to the V. harveyi 150 quorum-sensing AI called AI-1 (3-hydroxy-C4-homoserine lactone) (29, 32) ( Figure S1D).

151
To identify the yeast AI-2 mimic, washed S. cerevisiae cells were resuspended in water at 152 OD600 = 100. Following overnight incubation, the cell-free fluids were collected and concentrated 153 by lyophilization. HPLC fractionation on a Luna C18 reverse phase column revealed one peak at 154 8.3 min that exhibited absorption at 254 nm (blue trace, Figure 1C, arrow and inset) and 280 nm 155 (red trace, Figure 1C, arrow and inset). The material did not absorb significantly at 214 nm ( Figure   156 1C inset, green trace). The peak contained high levels of yeast AI-2 mimic activity as judged by 157 the V. harveyi TL-26 reporter strain ( Figure 1D). We pooled this peak from multiple such column 158 runs and prepared the sample for NMR and mass spectral analyses as described in the Methods.   Figure S2B). The yeast AI-2 mimic structure A was 169 identified as 4-hydroxy-5-methylfuran-3(2H)-one (MHF, Figure 1E, S2A). Indeed, comparison of 170 mass spectral, NMR, and HPLC analytical data confirmed that MHF purified from S. cerevisiae 171 was identical to an authentic commercial sample of MHF.

173
The mammalian AI-2 mimic is not MHF 174 8 With the MHF structure in hand, we could investigate whether the S. cerevisiae and the 175 previously reported mammalian AI-2 mimic are identical or not. As noted earlier, the mammalian 176 AI-2 mimic has not been identified so we do not have purified compound (13). Rather, we made 177 a preparation from Caco-2 cells containing high level mammalian AI-2 mimic activity in PBS (13).

178
To determine the elution pattern for MHF in such Caco-2 cell preparations, we spiked commercial 179 MHF into the mammalian AI-2 mimic preparation prior to HPLC fractionation. MHF eluted at 14 180 min ( Figure S3A, arrow) in the context of Caco-2 culture fluids. Samples from Caco-2 cells that 181 had not been spiked did not have a peak at the expected elution time for MHF ( Figure S3B, arrow).

182
To eliminate the possibility that MHF was present in the Caco-2 cell preparations but at a level 183 below the UV detection limit on the HPLC instrument, we tested all of the fractions for activity in 184 the V. harveyi TL-26 reporter assay. While the reporter assay showed that the mammalian AI-2 185 mimic was indeed present in the non-MHF-spiked Caco-2 preparations ( Figure S3C, black), there 186 was no activity in the 12-14 min HPLC fraction ( Figure S3C, red). Collectively, these data 187 demonstrate that the mammalian AI-2 mimic is not MHF. In future studies, we will focus on 188 identification of the mammalian AI-2 mimic. to generate an activity-based standard curve. The S. cerevisiae cell-free fluids were identically 202 assayed, and the MHF concentration in each preparation was estimated from the activity standard 203 curve (Figure 2A, white bars). The concentrations of MHF in the preparations calculated by the 204 two methods were in close agreement. Assuming MHF production has a linear relationship with 205 OD600, we can use our data to estimate that S. cerevisiae produced 1.2 ± 0.4 µM MHF per OD600 206 of cells. In the context of detection by the V. harveyi quorum-sensing apparatus, the EC50 for AI-207 2 is 3 nM and that for MHF is 300 nM ( Figure 2B). Thus, while both compounds exert activity in 208 this system within the range reported for bacterial AIs (33)(34)(35)(36), the LuxP receptor prefers AI-2 209 over MHF.

211
Identification of CFF1 as an S. cerevisiae gene essential for MHF production 212 To identify the component(s) responsible for MHF production in S. cerevisiae, we 213 screened the yeast deletion library for an S. cerevisiae mutant that was defective in MHF 214 production (37-39). As described in the Methods, cell-free fluid preparations were made from 215 >5,000 S. cerevisiae mutants and incubated with the V. harveyi TL-26 reporter strain.

216
Bioluminescence was measured to assess the ability of each S. cerevisiae mutant to make MHF 217 ( Figure 3A). Mutants were identified that elicited at least two standard deviations less light from 218 the reporter strain than the mean amount of light production elicited from all strains ( Figure 3A).

219
Eight putative mutants were retested for the ability to make activity ( Figure 3B). Two mutants,

224
To verify the phenotypes of the mutants, clean deletions of CFF1 and RPS1B were 225 constructed in S. cerevisiae. The cff1∆ mutant displayed no defect in growth rate ( Figure S4A, 10 circles; compare to WT growth shown by the squares). As previously reported (41), the rps1b∆ 227 mutant had a growth defect ( Figure S4A, triangles). Neither mutant exhibited sensitivity to 228 overnight incubation in water ( Figure S4B). Importantly, culture fluids prepared from the clean 229 rps1b∆ mutant produced nearly the wildtype level of AI-2 mimic activity, as determined by the 230 ability to induce light production in the V. harveyi TL-26 reporter ( Figure 3C, triangles). By contrast, 231 preparations made from the cff1∆ mutant had no activity ( Figure 3C, circles). PCR analysis 232 revealed that the mutant annotated as rps1b∆ in the yeast deletion library, in fact, possesses a 233 deletion in CFF1, explaining its inability to stimulate the reporter strain as well as the ability of our 234 newly constructed rps1b∆ mutant to produce activity. Thus, CFF1 is the only gene revealed by 235 our screen to be required for production of the activity we are monitoring.

236
To confirm that the yeast AI-2 mimic activity produced by the protein encoded by CFF1 is 237 MHF, we prepared and fractionated cell-free culture fluids from the cff1∆ strain using the identical 238 procedure we used for isolation of MHF from wildtype S. cerevisiae. No MHF peak could be 239 detected in the cff1∆ mutant preparation ( Figure 3D). Consistent with this finding, the relevant 240 HPLC column fraction had no activity in the V. harveyi TL-26 reporter assay ( Figure S4C). These 241 data suggest that Cff1p has a required role in MHF biosynthesis in yeast.

242
Our finding that Cff1p is required for MHF production is surprising. The presence of MHF 243 in fermented food products made from S. cerevisiae has been reported; however, the suggested 244 route to MHF is either spontaneous starting from D-ribulose-5-phosphate (42-45) or under 245 extreme conditions, via the Maillard reaction (46, 47). Quite to the contrary, our data suggest that 246 MHF production in S. cerevisiae is enzyme-catalyzed and under physiological conditions. Cff1p 247 has not been characterized. However, there does exist a crystal structure (26). It shows a putative 248 ligand binding pocket containing amino acid residues identical to those required for catalysis by 249 epimerases and isomerases that share the cupin fold (48). Specifically, the conserved E44 250 residue is proposed to have a catalytic role. We made an E44A substitution in Cff1p and assayed 251 the mutant protein for MHF production. The substitution did not alter Cff1p stability as judged by 11 visualization of a fused HALO tag ( Figure S4D); however, culture fluids prepared from the S.

298
It is intriguing that bacterial species may possess cff genes (note: we designate the 299 bacterial genes and proteins as cff and Cff, respectively, and the fungal genes and proteins CFF1 300 and Cff1p, respectively). Thousands of bacterial species are known to synthesize the inter- 316 lucimarinus and S. kowalevskii) that may coexist with bacteria that use AI-2-LuxP-mediated 317 quorum sensing. Additionally, we tested proteins from organisms representing unique phyla and 318 proteins with varying levels of amino acid identity relative to S. cerevisiae Cff1p (Table 1). To 319 assay activity, we expressed the candidate genes in our cff1∆ S. cerevisiae strain under control 320 of the endogenous S. cerevisiae CFF1 promoter. All of the homologs except those from U.

332
We were especially intrigued that the Interpro database search revealed a virus with a 333 potential Cff homolog, Pandoravirus salinus. Using the above strategy, we tested the functionality 334 of the P. salinus Cff homolog and found that, indeed, the viral cff gene complements the loss of 335 CFF1 in S. cerevisiae ( Figure 5C). This finding provides initial validation for the output of the larger 336 Interpro dataset and presages the existence of many additional functional Cff1p homologs.

337
Moreover, this final result, coupled with the other findings here, shows that MHF production could 338 occur across the viral, archaeal, bacterial, and eukaryotic domains.

341
Quorum sensing is the process by which bacteria monitor their local cell population density 342 and determine when it is appropriate to engage in collective behaviors (1-3). Bacteria often 343 employ multiple AIs, encoding distinct information about species relatedness, which presumably 344 enables them to take a census of "self" and "other". The AI-2-LuxP quorum-sensing pathway is 345 proposed to be used for the latter, to monitor the total cell density of the vicinal community.

346
Previously, we showed that mammalian cells can make a mimic of AI-2 that activates quorum

377
MHF is a volatile compound that is used as a flavorant. Food scientists have previously 378 shown that low levels of MHF exist in fermented foods, such as soy sauce and malt (62-64). As 379 alluded to above, MHF is hypothesized to form spontaneously from pentose sugars that undergo 380 the Maillard reaction (i.e., during cooking) or as a byproduct (in fungi). Specifically, in the fungus 16 the pentose phosphate pathway, MHF forms spontaneously as a breakdown product (42).

383
Notably, MHF is also a breakdown product of DPD; however, S. cerevisiae lacks LuxS, so DPD 384 is an unlikely source of MHF in fungi (44, 65). Importantly, production of MHF by the pentose 385 phosphate pathway and by the Malliard reaction is non-enzymatic. Here, we show that Cff1p, 386 which is presumably an enzyme, is required for MHF production in S. cerevisiae. Therefore, if the

399
However, no sugar was present in the Cff1p crystal. The authors hypothesized that the pocket 400 may bind a sugar-nucleotide. Accordingly, the jelly-roll fold motif is shared with enzymes such as 401 phosphoglucose isomerase and dTDP-4-keto-6-deoxy-D-hexulose-3,5-epimerase (26). Given 402 this relatedness and the fact that DPD, the non-borated precursor to AI-2, is a sugar, we suspect 403 that Cff1p could be the synthase for MHF, and that MHF is made from a sugar substrate.

404
Our study suggests that the scope of organisms that can participate in quorum sensing 405 through AI-2-type pathways continues to increase, hinting that AI-2 quorum sensing mediates   Table S1 and Table S2, respectively.

507
The phylogenetic tree in Figure S7 was constructed in MEGA-X using the Maximum Likelihood 508 method and JTT matrix-based model with 500 bootstrap replications as described previously (79).

509
The alignment was visualized using ggmsa in R, pruning the ends of the alignment to the first and

517
The fraction was dried by roto-evaporation to remove methanol. To enable further concentration, 518 the aqueous solution was saturated with sodium chloride and extracted with dichloromethane.

519
The dichloromethane layer containing the product of interest was dried and the sample was