How yeast cells find their mates

Accurate detection of extracellular chemical gradients is essential for many cellular behaviors. Gradient sensing is challenging for small cells, which experience little difference in ligand concentrations on the up-gradient and down-gradient sides of the cell. Nevertheless, the tiny cells of the yeast Saccharomyces cerevisiae reliably decode gradients of extracellular pheromones to find their mates. By imaging the behavior of polarity factors and pheromone receptors during mating encounters, we found that gradient decoding involves two steps. First, cells bias orientation of initial polarity up-gradient, even though they have unevenly distributed receptors. To achieve this, they measure the local fraction of occupied receptors, rather than absolute number. However, this process is error-prone, and subsequent exploratory behavior of the polarity factors corrects initial errors via communication between mating partners. The mobile polarity sites convert the difficult problem of spatial gradient decoding into the easier one of sensing temporal changes in local pheromone levels.


Introduction
with respect to the pheromone gradient (Nern and Arkowitz, 1999;Butty, et al., 1998). Thus, 67 Far1 provides a direct spatial connection between upstream receptor-pheromone binding and 68 downstream Cdc42 activation, allowing the cells to exploit the pheromone gradient to find their 69 partners. 70 Like other eukaryotic cells, yeast are thought to compare the ligand concentrations 71 across the cell to determine the orientation of the gradient (Arkowitz, 2009). If the distribution 72 of pheromone-activated receptors reflects the pheromone gradient, then G-Far1-Cdc24 73 complexes will be enriched up-gradient, spatially biasing activation of Cdc42 to kick off positive 74 feedback at the right location for mating. However, the accuracy of such global spatial gradient 75 sensing is limited by the small yeast cell size (~4 m diameter) (Berg and Purcell, 1977), and 76 simulations constrained by experimental data on binding and diffusion parameters suggested 77 that the process would be inaccurate (Lakhani and Elston, 2017). Indeed, when yeast are 78 exposed to artificial, calibrated pheromone gradients, polarized growth often starts in the 79 wrong direction (Moore, et al., 2008;Segall, 1993). Such cells can nevertheless correct initial 80 errors by moving the polarity site (Dyer, et al., 2013). 81 Moving a Cdc42 patch that is constantly being reinforced by positive feedback seems 82 counterintuitive, but time-lapse imaging revealed that the patch "wandered" around the cortex Ayscough and Drubin, 1998). As the polarity site wanders around the cortex, this receptive 92 "nose" would sample pheromone levels at different locations. Studies of cells treated with 93 uniform pheromone concentrations showed that when pheromone levels are high, the patch 94 stops moving (McClure, et al., 2015;Dyer, et al., 2013). In principle, this "exploratory 95 polarization" mechanism can explain error-correction by positing that movement of the patch 96 continues until cells sense high pheromone levels indicating that the patch is directed towards a 97 mating partner (Hegemann and Peter, 2017). 98 The extent to which yeast cells rely on global spatial sensing to orient the formation of a 99 polarity patch, versus exploratory polarization after the patch has formed, remains unclear. A 100 recent study found that when cells were placed in an artificial pheromone gradient in a 101 microfluidics device, initial patch formation was random with respect to the gradient, and 102 orientation occurred by exploratory polarization and "local sensing" (Hegemann, et al., 2015). 103 However, it is unclear whether similar results might apply to different pheromone gradients, or 104 to more physiological conditions in which gradients are generated by mating partners. 105 To better understand how yeast actually locate their mating partners, we imaged 106 mating events in mixed populations of a and  cells. We found evidence for both global spatial 107 sensing and error correction by exploratory polarization. Encounters between partners were 108 characterized by (i) rapid and non-random initial clustering of polarity proteins biased towards 109 the partner; (ii) an "indecisive phase" in which dynamic polarity sites relocalized in an apparent 110 search process; and (iii) a "committed phase" in which cells polarized stably towards mating 111 partners, culminating in fusion. Transition from indecisive to committed behavior was 112 associated with a rise in MAPK activity. Initial polarization was surprisingly accurate given that it 113 occurred despite a highly non-uniform (and thus potentially misleading) distribution of 114 receptors. We found that the variation in receptor density was corrected for via "ratiometric" 115 sensing of the ratio of occupied vs unoccupied pheromone receptors across the cell (Bush, et  116 al., 2016). In aggregate, our findings reveal how yeast cells can overcome the challenges 117 imposed by small cell size and lack of cell mobility to locate mating partners. 118 119 Results 120 121 Indecisive and committed phases of mating cell polarization 122 123 To observe how cells find their mates, we mixed a and  cells expressing differently 124 colored polarity probes, Bem1-GFP () and Bem1-tdTomato (a), and imaged them at 2 min 125 resolution (Video 1). Fusion events were identified from movies and the cells were tracked back 126 to their time of "birth" (the cytokinesis that preceded the mating event). Fig. 1A (top) illustrates 127 selected frames from a representative mating cell. This cell formed a faint initial cluster of 128 Bem1 just 4 min after birth (blue panel), which then fluctuated in intensity and moved 129 erratically around the cell cortex for 34 min before stably polarizing adjacent to a mating 130 partner (orange panel). After another 16 min the two cells fused, as seen by the mixing of red 131 and green probes. We designate the time between initial cluster formation (Tic) and stable 132 polarization (Tp) as the "indecisive phase", reflecting the erratic behavior of the polarity probe. 133 We designate the time between stable polarization (Tp) and fusion as the "committed phase" of 134 mating, reflecting the strong and stably located polarity site. 135 To quantify the degree of Bem1 polarization, we used a metric that uses the pixel 136 intensity distribution within the cell to assess the degree of signal clustering (hereafter, 137 "clustering": see methods). Fig. 1A (graph) illustrates that Bem1 clustering fluctuated during the 138 indecisive phase but remained high during the committed phase. This pattern was 139 characteristic of mating cells (Fig. 1B), supporting the idea that mating involves a two-stage 140 process. A similar two-stage process was observed for cells expressing fluorescent versions of 141 Spa2, a polarisome component that binds and helps to localize the formin Bni1 (Video 2) 142 (Pruyne, et al., 2004;Fujiwara, et al., 1998;Sheu, et al., 1998). Analysis of cells expressing both 143 Bem1 and Spa2 probes revealed that although (as described previously) Spa2 clusters were 144 more tightly focused than Bem1 clusters, the probes clustered, dispersed, and moved together 145 (Fig. 1C). As for Bem1, Spa2 clustering fluctuated during the indecisive phase and remained 146 stably high during the committed phase (Fig. 1D). We conclude that cells undergo a 147 reproducible pattern of polarization during mating, with sequential indecisive and committed 148 phases. 149 The earliest observable clustering of polarity factors occurred shortly after birth ( Fig. 1E:  150 median time 4 min after initiating cytokinesis). This initial clustering was usually weak and 151 frequently at a different location than that of the final stable polarization (see below). During 152 the ensuing indecisive phase, cells appeared to search for mating partners, often assembling 153 polarity clusters adjacent to different potential partners before settling at a final location (Fig.  154 4 1F). The duration of the indecisive phase (Fig. 1G: median 42 min) was very variable, ranging  155 from 10 to 120 min. This is consistent with a search process that would take a variable amount 156 of time depending on the availability and proximity of potential mating partners. In contrast, 157 the subsequent committed phase was consistently about 20 min (Fig. 1G), which we speculate 158 is the time required to remodel the local cell walls to allow for cell fusion. 159 160 Commitment is synchronous for both partners 161 162 In our protocol, cells of each mating type are proliferating asynchronously before they 163 are abruptly mixed. Thus, in a large majority of cases, one cell of each mating pair is born (i.e. 164 enters G1 phase) before the other. Nevertheless, fusion is a unitary event that occurs at the 165 same time for both. This means that the "first-born" partner must extend one or both phases of 166 polarization while the "second-born" partner completes the previous cell cycle and "catches 167 up" (Fig. 2A). Does the first-born locate and commit to its partner first, and then wait ( Fig. 2A,  168 top), or does the first-born remain indecisive until the second-born has caught up ( Fig. 2A,  169 bottom)? We found no difference in the average duration of the committed phase between 170 first and second-born cells (Fig. 2B), and partners in each individual mating pair generally 171 committed at nearly the same time (Fig. 2C). Conversely, the indecisive phase was significantly 172 longer in first-born cells (Fig. 2D), suggesting that first-born cells remain indecisive while 173 second-born cells complete the cell cycle, and that cells only polarize stably towards partners 174 that are in G1 ( Fig. 2A, bottom). 175 Synchronous commitment implies that there is some communication between partners 176 that occurs only when both are in G1 phase of the cell cycle. As the only known mode of 177 communication is via the secretion of pheromones, the simplest hypothesis to explain why 178 commitment must wait until both cells are in G1 would be that pheromone secretion changes 179 when cells enter G1. To assess the rate of pheromone synthesis, we introduced a fluorescent 180 reporter whose production was driven by the major α-factor gene (MFα1) promoter 181 (Achstetter, 1989;Singh, et al., 1983). Reporter signal fluctuated regularly through the cell 182 cycle, rising in G1 and falling (due to dilution) after bud emergence (Fig. 2E with the Spa2 probe. The MAPK sensor is a fluorescent probe that moves from the nucleus to 193 the cytoplasm when it is phosphorylated by active MAPK. In the absence of pheromone, the 194 sensor was predominantly nuclear, although the nuclear to cytoplasmic ratio fluctuated 195 somewhat through the cell cycle, peaking during anaphase ( Fig. 3A and video 3). In a mating 196 mix, the sensor distribution became uniform prior to fusion, reflecting an increase in MAPK 197 activity ( Fig. 3B and video 4). To quantify the degree of nuclear concentration of the MAPK 198 sensor, we measured the coefficient of variation (CV) in pixel intensity across the cell. When the 199 probe is nuclear, the bright nuclear and dim cytoplasmic pixels yield a high CV, but when the 200 probe distribution is uniform there is a low CV. We found considerable cell-to-cell variability in 201 this signal, which could be largely accounted for by differences in the level of expression of the 202 probe (Fig. S1A, B). This variability could be reduced by normalizing the CV to the maximum and 203 minimum CV for each cell, and we developed a MAPK activity metric based on the normalized 204 CV of the probe (Fig. S1C). 205 In mating cells, MAPK activity fluctuated but then climbed to a plateau about 20 min 206 prior to fusion (Fig. 3C). As this was similar to the clustering behavior of polarity probes, we 207 directly compared MAPK activity with Spa2 clustering in individual mating cells (Fig. 3D). These 208 measures aligned well with one another in most cells, with both Spa2 clustering and MAPK 209 activity fluctuating during the indecisive phase before rising to a stable plateau during the 210 committed phase (Fig. 3D, E). A cross-correlation analysis of Spa2 clustering and MAPK activity 211 during the indecisive phase revealed that they fluctuated in tandem (Fig. 3F). This correlation 212 suggested that MAPK activity might promote stable polarization, or that polarization might lead 213 to an increase in MAPK activity, or both. 214 To more directly ask whether an increase in MAPK activity promotes stable polarization, 215 we induced MAPK activity in the absence of pheromone by expressing a membrane-tethered 216 version of the MAPK scaffold Ste5 (Pryciak and Huntress, 1998 vicinity of receptors on the partner cell. That would produce a higher pheromone signal than 233 when polarity clusters point away from each other (Fig. 4A). To visualize the location of α-factor 234 secretion, we imaged Sec4-GFP, a Rab GTPase highly concentrated on the secretory vesicles 235 that deliver α-factor to the cell surface (Mulholland, et al., 1997;Walch-Solimena, et al., 1997). 236 Sec4 accumulated in regions enriched for Bem1, during the indecisive phase as well as the 237 committed phase of polarization (Fig. 4B). 238 Putting together our findings thus far, we propose that mating cells undergo the 239 following sequence of events. As cells undergo cytokinesis, newborn cells in G1 phase detect 240 enough pheromone in their surroundings to arrest the cell cycle and initiate a weak and mobile 241 level of polarization. Cells in G1 also increase their rate of pheromone production, signaling to 242 potential partners that they are ready to mate. As mobile polarity clusters explore the 243 surrounding pheromone landscape during the indecisive phase, they also locally secrete 244 pheromones to be sensed by their partners. When potential partners orient their polarity 245 clusters towards each other, both cells perceive higher pheromone concentrations, leading to a 246 simultaneous rise in MAPK activity in each partner. Higher MAPK activity leads to stronger 247 polarity, and when MAPK activity crosses some threshold, the polarity clusters stop moving, 248 now properly facing their partners. This leads to sustained high MAPK signaling during the 249 committed phase, maintaining polarity until fusion can occur. 250 251 Gradient sensing before initial polarity clustering 252 253 The view of the mating process outlined above proposes an important role for polarity 254 clusters in tracking pheromone gradients to locate partners, as recently suggested by other 255 studies which noted mobile polarity sites in cells exposed to uniform pheromone ( mutually exclusive, and it could be that significant gradient sensing takes place prior to the 261 initial clustering of polarity factors. Indeed, we found that in our mating mixtures, cells biased 262 the locations of their initial clusters towards their eventual mating partners (Fig. 5A), suggesting 263 that a form of global spatial gradient sensing occurs in the few minutes between cell birth and 264 initial clustering. In contrast, we found no bias towards the previous cytokinesis site (neck) 265 under our conditions (Fig. 5B). The directional bias towards partners was similar in first-born 266 and second-born cells (Fig. 5C). Among second-born cells, those that formed their initial 267 clusters within 60 of their partners took less time to commit than those whose initial clusters 268 were less well-oriented ( Fig. 5D: median indecisive phase duration 32 min vs 48 min). Thus, 269 gradient sensing before polarity cluster formation can shorten the search for a partner. 270 271 Non-uniform pheromone receptor distribution 272 273 How would spatial gradient sensing occur? Studies in other model systems like 274 Dictyostelium discoideum indicated that receptors and their coupled G proteins were 275 distributed uniformly around the cell surface, with active G proteins reflecting the external 276 ligand gradient (Janetopoulos, et al., 2001;Jin, et al., 2000). Receptor distribution has been 277 harder to assess in yeast cells, for technical reasons stemming from the rapid secretion and 278 recycling of receptors (Suchkov, et al., 2010). Transit of pheromones and pheromone receptors 279 through the secretory pathway is rapid (5-10 min) (Losev, et al., 2006;Govindan, et al., 1995). In 280 the presence of α-factor, Ste2 is then endocytosed on a 10-min timescale and delivered to the 281 vacuole for degradation Hicke and Riezman, 1996;Schandel and Jenness, 282 1994; Jenness and Spatrick, 1986). As GFP maturation occurs on a 30-min timescale (Iizuka, et 283 al., 2011;Gordon, et al., 2007), much of the GFP-tagged receptor at the cell surface is not yet 284 fluorescent. Moreover, the GFP moiety survives intact in the vacuole following receptor 285 degradation, yielding a high vacuolar fluorescence signal. To partially resolve these issues, we 286 tagged Ste2 with sfGFP, which matures on a 6-min timescale (Khmelinskii, et al., 2012). 287 Although bright vacuoles remained, the surface Ste2-sfGFP was clearly visible (Fig 6A), allowing 288 us to assess Ste2 distribution. In cells that were not exposed to α-factor, Ste2 distribution 289 varied throughout the cell cycle, accumulating in the bud (enriched at the tip) and depleted in 290 the mother during bud growth, and then accumulating at the neck during cytokinesis (Fig. 6B). 291 G1 cells displayed quite variable Ste2 distributions, ranging from nearly uniform to highly 292 polarized ( Fig. 6C: left). Quantification of surface Ste2 distribution revealed a 3-fold difference 293 (on average) in Ste2 concentration from one side of the cell to the other ( Fig. 6C: right). 294 The non-uniform receptor distribution poses a significant problem for accurate gradient 295 sensing, because cells would be preferentially sensitive to pheromone on the side where 296 receptors are enriched, which would not necessarily correspond to the side facing a mating 297 partner. To directly observe the relationship between Bem1 clustering and Ste2 distribution, we 298 imaged MATa cells carrying both Ste2-sfGFP and Bem1-tdTomato, mixed with MAT cells in 299 mating reactions. If receptor density impacts the location of initial clustering, we would expect 300 that Bem1 clustering would occur preferentially on the side with higher Ste2 signal. Individual 301 cells clustered Bem1 at various different locations relative to the Ste2 distribution, and in 302 several cells the initial Bem1 cluster formed adjacent to a mating partner even though Ste2 was 303 concentrated at the opposite end of the cell (Fig. 6D). Averaging revealed no clear spatial 304 correlation between the location of Bem1 initial clustering and the Ste2 distribution (Fig. 6E). 305 We conclude that cells are able to perform a surprisingly accurate degree of gradient sensing 306 prior to polarization, despite having non-uniform receptor distributions. 307 308 Effect of changing receptor distribution on gradient sensing 309 310 To probe the degree to which receptor distribution influences the accuracy of gradient 311 sensing, we sought to manipulate receptor distribution. Ste2 distribution reflects a dynamic 312 balance between polarized secretion of new Ste2, slow diffusion at the plasma membrane, and 313 retrieval by endocytosis (Suchkov, et al., 2010;Valdez-Taubas and Pelham, 2003). Endocytosis is 314 more rapid for ligand-bound Ste2 (which undergoes phosphorylation and ubiquitination) than 315 for unbound Ste2 (which is endocytosed at a slower basal rate) Terrell, et al., 316 1998; Hicke and Riezman, 1996). To manipulate Ste2 distribution, we used Ste2 mutants that 317 either lacked endocytosis signals (Ste2 7XR-GPAAD , allowing accumulation all over the membrane) 318 (Ballon, et al., 2006;Terrell, et al., 1998) or had a constitutively active strong endocytosis signal 319 (Ste2 NPF yielding a highly polarized distribution with a bias toward the mother-bud neck)( Tan, et  320 al., 1996) (Fig. 7A, B). As endocytosis is needed for Ste2 degradation, Ste2 7XR-GPAAD was more 321 abundant than Ste2 or Ste2 NPF (Fig. 7C), and in halo assays cells expressing Ste2 NPF were slightly 322 less sensitive to pheromone while cells expressing Ste2 7XR-GPAAD were more sensitive to 323 pheromone (Fig. 7D). Correspondingly, in mating mixes cells with Ste2 NPF sometimes re-entered 324 the cell cycle despite being adjacent to potential partners, while cells with Ste2 7XR-GPAAD were 325 more likely to remain arrested and mate (Fig. S2). This was reflected in the duration of the 326 indecisive phase, which we quantified among all cells that were born and remained 327 immediately adjacent to a G1 cell of opposite mating type until they either mated or budded 328 (Fig. 7E). Cells that budded instead of committing to a partner were recorded as never entering 329 the committed phase. Among the cells that successfully mated, indecisive phases had similar 330 durations, suggesting that indecisive phase dynamics were unaffected by the changes in 331 receptor distribution. 332 To quantify the accuracy of initial clustering, we recorded the location of Bem1 clusters 333 among all cells that were born immediately adjacent to a G1 cell of opposite mating type, 334 including those that mated, budded, or failed to do either by the end of the movie. We found 335 that cells despite the dramatic difference in receptor distribution (Fig. 7B), cells with Ste2 NPF or 336 Ste2 7XR-GPAAD were no less accurate than wild type cells at orienting their initial clusters towards 337 adjacent partners (Fig. 7F). Cells with Ste2 7XR-GPAAD were a little more accurate than wild-type 338 cells (Fig 7F), perhaps indicating that abundant and uniformly distributed Ste2 improves 339 gradient sensing, but the difference was not statistically significant for the number of cells 340 analyzed. The finding that even cells with a highly asymmetrical receptor distribution can 341 respond to a pheromone gradient suggests that yeast have a mechanism to correct for 342 variations in receptor density. 343 344 Ratiometric sensing of receptor occupancy 345 346 One way to correct for variations in receptor density would be to measure the local ratio 347 of ligand-bound to unbound receptors (i.e. ratiometric sensing). If cells were to respond to the 348 spatial distribution of the ratio of active/total receptors, rather than the spatial distribution of 349 active receptors, then differences in the local receptor density would not distort a cell's ability Pheromone-bound Ste2 loads GTP on G, whereas unbound Ste2-Sst2 promotes GTP hydrolysis 354 by G, so the level of activated G depends on the ratio between pheromone-bound and 355 unbound Ste2 (Fig. 8A). Although originally proposed as a global mechanism to integrate 356 signaling from all Ste2 (Bush, et al., 2016), in principle this mechanism could also apply locally to 357 extract the gradient of pheromone from the spatial distribution of bound/unbound receptor. 358 Sst2-based ratiometric sensing can be disrupted by replacing Sst2 with a human paralog, 359 hsRGS4, which has similar GAP activity towards G but does not associate with Ste2 (Bush, et  360 al., 2016). hsRGS4 is myristoylated and localized uniformly to the plasma membrane ( Fig. 8B). 361 We found that two copies of hsRGS4 expressed from the SST2 promoter were sufficient to 362 restore wildtype pheromone sensitivity (as judged by halo assays) to cells lacking endogenous 363 Sst2 (Fig. 8C). Compared with cells containing Sst2, cells with hsRGS4x2 were significantly worse 364 at orienting at their initial Bem1 clusters towards their partners (Fig. 8D). Instead, the initial 365 clusters in hsRGS4x2 cells were strongly biased towards the previous mother-bud neck (Fig. 8E), 366 a region of high receptor density (Fig. 6). Indeed, a direct comparison showed that unlike wild-367 type cells, hsRGS4x2 cells displayed a strong tendency to establish initial clusters of Bem1 at 368 sites enriched for Ste2 (Fig. 8F). Thus, gradient sensing depends on the endogenous RGS 369 protein Sst2, which may assist in this process by linking G GTP hydrolysis to the location of 370 unbound receptor. 371 If the inaccurate gradient sensing exhibited by hsRGS4x2 cells is indeed due to 372 disruption of ratiometric sensing, then the orientation defect of hsRGS4x2 should be corrected 373 if the cells were to have uniformly distributed receptors (i.e. ratiometric sensing should be 374 unnecessary if receptor density is uniform). We used the more uniformly distributed Ste2 7XR-375 GPAAD to test this hypothesis, and found that Ste2 7XR-GPAAD restored the accuracy of initial Bem1 376 clustering to wildtype levels in hsRGS4x2 cells (Fig. 8D,E,F). These findings suggest that yeast 377 cells use Sst2-dependent local ratiometric sensing of receptor occupancy to extract accurate 378 information from the pheromone gradient despite having non-uniform receptor density. 379 380 Discussion 381 Initial polarity cluster location is surprisingly accurate 382 The rapid diffusion of peptide pheromones and the small size of the yeast cell led to the 383 expectation that there would be only a small difference in pheromone concentration between 384 the up-and down-gradient sides of the cell. This poses a fundamental difficulty in extracting 385 accurate directional information in the face of molecular noise (Berg and Purcell, 1977). Indeed, 386 a recent study on cells responding to a 0.5 nM/m pheromone gradient found that initial 387 polarity cluster location was close to random with respect to the gradient (Hegemann, et al., 388 2015). Moreover, simulations of gradient sensing suggested that even with uniformly 389 distributed receptors and error-free interpretation of the ligand-bound receptor distribution, 390 the signal from such gradients would be obscured by molecular noise and diffusion (Lakhani 391 and Elston, 2017). In principle, time-averaging of the ligand-bound receptor distribution could 392 extract the signal from the noise, but we show that initial clustering of polarity factors occurs 393 5.1 +/-2.7 min from cell birth, which is too fast to allow for significant time averaging have uniform receptor distributions. The polarized secretion, slow diffusion, and subsequent 407 endocytosis of pheromone receptors resulted in significant receptor asymmetry, with (on 408 average) three-fold more concentrated receptors on one side of the cell than the other. This 409 creates a receptor gradient that is significantly steeper than the assumed pheromone gradient 410 detected by the cells. As the receptor gradient is randomly oriented with respect to the mating 411 partner, this poses a serious hurdle in accurate gradient detection. 412 Despite the difficulties enumerated above, we found that the location of initial polarity factor 413 clustering in mating mixtures was highly non-random and surprisingly accurate, with more than 414 40% of cells clustering within 30 of the correct direction and less than 5% of cells clustering in 415 the opposite segment (a random process would have 17% of cells polarizing in each of these 416 segments). This finding suggests that physiological pheromone gradients may be considerably 417 steeper than previously assumed, and/or that cells possess unappreciated mechanisms to 418 overcome the difficulties in accurate gradient detection discussed above. 419 Orientation accuracy is enhanced by ratiometric sensing 420 One way to avoid being misled by an asymmetric receptor distribution would be to compare 421 the local ratio of occupied and unoccupied receptors, rather than simply the density of 422 occupied receptors, across the cell surface. additionally that a pheromone-bound receptor diffuse slowly relative to its lifetime at the 429 surface (~10 min) (Jenness and Spatrick, 1986), so that information about where receptors were 430 when they bound to pheromone is not lost. Similarly, GTP-G and G must diffuse slowly 431 relative to the timeframe for G GTP hydrolysis and G protein re-association, so that 432 information about where they were when they became activated is not lost. 433 We found that when RGS function was delocalized by replacing Sst2 (which binds unoccupied 434 receptors) with an equivalent amount of hsRGS4 (which binds the plasma membrane), the 435 accuracy of initial polarity clustering was severely compromised. Instead of polarizing towards 436 potential partners, these cells assembled polarity clusters at regions where receptors were 437 concentrated (often at the site of the last cell division or neck). Thus, abrogating the Sst2-based 438 ratiometric sensing mechanism allowed cells to be misled by the asymmetric receptor 439 distribution. Accurate orientation could be restored to these cells by manipulations that made 440 receptor distribution more uniform. In sum, our findings suggest that local ratiometric sensing 441 compensates for uneven receptor distribution and allows more accurate polarization towards 442 mating partners. 443

Why is receptor distribution non-uniform? 444
Blocking receptor endocytosis allowed receptors to accumulate all over the cell surface in a 445 much more uniform distribution than that seen in wild-type cells. In our mating conditions, this 446 promoted a slightly more accurate orientation of initial polarity factor clustering towards 447 mating partners, and a small improvement in mating efficiency. Why, then, would cells 448 internalize their receptors and create the need for error correction by ratiometric sensing? One 449 possible answer stems from the fact that wild yeast (unlike lab strains) are able to switch 450 mating type. Without receptor endocytosis, cells may be unable to clear pre-existing receptors 451 during mating type switching, generating a situation in which cells arrest in response to their 452 own newly secreted pheromones after a switch. We speculate that receptor endocytosis is 453 necessary to clear the membrane of old receptors when switching mating types, and that 454 receptor asymmetry is the price that cells pay for this advantage. 455 Error correction following initial clustering of polarity factors 456 11 Although initial polarity clusters were biased to occur near potential mating partners, the 457 process was error-prone and about 60% of cells failed to orient initial polarity within 30 of the 458 correct direction. Nevertheless, these cells did eventually polarize towards partners and mate 459 successfully, indicating the presence of a potent error correction mechanism. We found that 460 after initial clustering, polarity factor clusters relocated erratically during an "indecisive phase" 461 of variable duration (48 +/-24 min). Even cells that had correctly assembled initial polarity 462 clusters close to mating partners exhibited an indecisive phase, although of somewhat shorter 463 duration. During this phase, clusters fluctuated in intensity (concentration of polarity factors in 464 the cluster), extent (broader vs more focused clusters), location, and number (transiently 465 showing no cluster or 2-3 clusters instead of a single cluster). The dynamic polarity clusters 466 were able to polarize actin cables, as we detected frequent accumulation of secretory vesicles 467 at cluster locations. We suggest that this erratic behavior represents a search process in which 468 weak polarity clusters act both as sources of pheromone secretion and locations of pheromone 469 sensing, allowing communication between potential mating partners. At the end of the 470 indecisive phase, cells developed strong and stable polarity sites correctly oriented towards 471 their partners. 472 The strongest evidence that partners are engaged in communicating with each other during the 473 indecisive phase is that mating pairs ended the indecisive behavior nearly simultaneously 474 (within 5 min of each other). As one partner was born before the other, the durations of their 475 indecisive phases were often quite different, but they transitioned to stable polarization 476 together. Strengthening of the polarity cluster was correlated with an increase in mating MAPK 477 activity, and synthetic induction of MAPK without pheromone led to a similar strengthening of 478 polarity clusters. We speculate that during the indecisive phase, the cells are exposed to a 479 dynamic and constantly changing pheromone landscape. When a mobile polarity cluster is 480 distant from its partner's cluster, both cells detect relatively low levels of pheromone, leading 481 to intermediate levels of MAPK activity. But if clusters happen to point directly at each other, 482 each cell detects a higher pheromone concentration, leading to an increase in MAPK activity. 483 Increased MAPK then strengthens and stabilizes the polarity cluster, perhaps leading to 484 increased local pheromone secretion and hence increased signaling in a positive feedback loop. 485

Exploratory polarization as a mechanism for partner selection 486
Because the search strategy discussed above depends on polarized pheromone secretion and 487 detection, we call this process "exploratory polarization". This behavior is strikingly similar to 488 the "speed dating" behavior recently described for mating cells of the distantly related fission 489 yeast Schizosaccharomyces pombe (Merlini, et al., 2016;Bendezu and Martin, 2013). In that 490 system, potential mating partners exhibit a prolonged period in which they sequentially 491 assemble and disassemble a weak polarity cluster at multiple locations. Clusters that happen to 492 assemble in the vicinity of a cluster from a mating partner become strengthened and stabilized, 493 presumably due to detection of a higher pheromone level. Thus, distantly related yeasts that 494 mate under very different physiological circumstances (rich nutrients for budding yeast, 495 starvation conditions for fission yeast) appear to have converged on a common and highly 496 effective search strategy. 497 Exploratory polarization is flexible and responsive to dynamic external conditions. We found 498 that cells abruptly reduced their level of pheromone production when they transitioned from 499 G1 to S phase. Thus, if a potential partner were to enter the cell cycle, a cell would quickly 500 detect reduced pheromone signaling and resume the search for other potential partners. We 501 also noticed that cells with two potential partners nearby could transiently orient clusters 502 towards both partners. However, that situation was unstable and cells only strengthened one 503 polarity cluster and committed to one partner. The basis for restricting polarity to a single site 504 in mating cells is unknown, but may be due to a competition phenomenon as documented for 505 vegetative yeast cells that pick a single bud site (Chiou, et

Live-cell microscopy 573
Cells were grown to mid-log phase (OD600  0.4) overnight at 30C in complete synthetic 574 medium (CSM, MP Biomedicals, LLC.) with 2% dextrose (Macron). Cultures were diluted to 575 OD600 = 0.1. For mating mixtures, the relevant strains were mixed 1:1 immediately before 576 mounting on slabs. Cells were mounted on CSM slabs with 2% dextrose solidified with 2% 577 agarose (Hoefer), which were then sealed with petroleum jelly. For Ste5-CTM MAPK induction 578 (Fig. 4A), slabs also contained 20 nM -estradiol (Sigma). Cells were imaged in a temperature 579 controlled chamber set to 30C. 580 Images were acquired with an Andor Revolution XD spinning disk confocal microscope (Andor 581 Technology, Olympus) with a CSU-X1 5000 rpm confocal scanner unit (Yokogawa), and a 582 UPLSAPO 100x/1.4 oil-immersion objective (Olympus), controlled by MetaMorph software 583 (Molecular Devices). Images were captured by an iXon3 897 EM-CCD camera with 1.2x auxiliary 584 magnification (Andor Technology). 585 For high resolution images of Ste2-sfGFP, Ste2 NPF -sfGFP, and Ste2 7XR-GPAAD -sfGFP (Fig. 6A, C, Fig.  586 7A, B), z-stacks with 47 planes were acquired at 0.14 m intervals. The laser power was set to 587 30% maximal output, EM gain was set to 200, and the exposure for the 488 nm laser was set to 588 250 ms. For all other microscopy, z-stacks with 15 images were acquired at 0.5 m z-steps every 589 2 min, laser power was set to 10% maximal output for the relevant 488 nm, 561 nm, or 445 nm 590 lasers, EM gain was set to 200, and the exposure time was 200 ms. 591 All fluorescent images were denoised using the Hybrid 3D Median Filter plugin for ImageJ, 592 developed by Christopher Philip Mauer and Vytas Bindokas. 593

Analysis of the timing of cell cycle and mating events 594
Bud emergence was scored using the membrane-targeted Psr1-GFP reporter (Lai, et al., 2018;595 Kuo, et al., 2014). Cytokinesis was recorded as the first time point when a strong Bem1 signal 596 was visible at the neck. Initial clustering was recorded as the first time point after cytokinesis 597 when a Bem1 cluster was clearly visible and distinguishable from background noise. 598 Polarization was recorded as the time point when the Bem1 patch reached its final stable 599 location and increased in intensity. If the patch appeared at the correct location, but then 600 transiently moved to a new location before returning, polarization was recorded as the time 601 point when the patch returned. Fusion was recorded as the time when cytoplasmic mixing of 602 different color probes became detectable. 603

Analysis of polarity factor clustering 604
To quantify the degree of clustering of the polarity probes Spa2-mCherry, Bem1-tdTomato, and 605 Bem1-GFP, we calculated a "deviation from uniformity" metric from maximum projections of 606 fluorescent z-stack images. Deviation from uniformity, referred to here as clustering (CL), 607 compares the cumulative distribution of pixel intensities in an actual cell, with that in a 608 hypothetical cell with the same range of pixel intensities that are uniformly distributed. That is, 609 CL measures how different the pixel intensity distribution is from a uniform distribution, which 610 reflects the degree to which the signals are clustered. 611 An elliptical region of interest (ROI) was drawn around each cell at each time point. Raw pixel 612 intensities (p) within each ROI were normalized to a minimum of 0 and maximum of 1: 613 For a cell with uniformly distributed pixel intensities, the cumulative distribution (U) is: 617 ≈ 618 500 uniformly-spaced i-values from 0 to 1 were indexed in ascending order as n = 1, 2, 3, …, 619 500. The deviation from uniformity metric (CL) was calculated as: 620 CL approaches a maximum of 1 when a small fraction of pixels exhibit near the maximum 622 intensity, while most pixels are clustered near the minimum intensity -as seen in a highly 623 polarized cell. CL is sensitive to the size of the patch, and the distribution of intensities within 624 the patch -a small patch with sharp edges yields a high CL, while a broad patch with graded 625 edges yields a low CL. As a result, CL is a sensitive indicator of the transition between the 626 indecisive and committed phases. 627 CL was measured using a MATLAB-based graphical user interface called ROI_TOI_QUANT_V8, 628 developed by Denis Tsygankov. 629

Analysis of initial polarity cluster orientation 630
Initial orientation was measured at the time of initial clustering. For orientation relative to the 631 partner (Fig. 5A,C; Fig. 7F; Fig 8D), we measured the angle between the line from the center of 632 the cell being scored to the centroid of the initial cluster, and a line from the cell center to the 633 closest surface of the nearest G1 cell of the opposite mating type. For orientation relative to 634 the neck ( Fig. 5B; Fig. 8E), we measured the angle between the line from the center of the cell 635 being scored to the centroid of the initial cluster, and a line from the cell center to the center of 636 the previous division site. Angles were then grouped into segments of 30 increments. 637

Analysis of -factor synthesis through the cell cycle 638
The PMF1-sfGFP reporter drives synthesis of sfGFP from the MF1 promoter. MF1 is the major 639 -factor encoding gene. Average fluorescence intensity of the probe was measured from 640 maximum projection images within an elliptical region of interest drawn around each cell. 641 Intensity values were normalized to the value at the end of G1 by dividing by the intensity at 642 the time of bud emergence (for cells with >1 cell cycle, the first bud emergence was used). To 643 express intensity as a function of cell cycle, we set the time of the emergence of the first bud to 644 0, and the time of the emergence of the second bud to 100. 645

Analysis of MAPK activity 646
MAPK activity was measured using maximum projection fluorescent images of the sensor Ste7-647 NLS-NLS-mCherry. As demonstrated in (Durandau, et al., 2015), the sensor relocates from the 648 nucleus to the cytoplasm upon phosphorylation by Fus3 or Kss1, and the cytoplasmic to nuclear 649 ratio of the sensor reflects the MAPK activity. We used the coefficient of variation (CV) of pixel 650 intensities measured from maximum projection images to approximate the nuclear to 651 cytoplasmic ratio of the probe. The CV was quite variable from cell to cell, but that variability 652 could be limited by normalization. To approximate MAPK activity (m), an elliptical ROI was 653 drawn around each cell at each time point using ROI_TOI_QUANT_V8. CV was measured for 654 each cell for the 60 minutes prior to fusion, and normalized to a minimum of zero and 655 maximum of 1. Because CV falls as MAPK activity rises, activity was scored as: 656

Analysis of receptor distribution 658
Membrane distribution of Ste2-sfGFP and Bem1-tdTomato were measured from medial plane 659 fluorescent images. Using FIJI software, fluorescence intensity was averaged across the width of splines were fit to each Ste2 linescan using the smooth.spline function in R, with a 0.75 667 smoothing factor. The normalized curves for Ste2 or Bem1 from the previous step were then 668 centered on the maximum from the Ste2 spline fit and averaged. 669

Halo assays of pheromone sensitivity 670
Cells were grown to mid-log-phase (OD600  0.4) at 30C overnight in YEPD (1% yeast extract, 671 2% peptone, 2% dextrose). Cultures were diluted to 2.5 x 10 5 cells/mL, and 5 x 10 4 cells were 672 spread on YEPD plates in triplicate using sterile glass beads. Plates were allowed to dry for 673 several minutes, and then 2 L of 1 mM, 500 M, and 100 M -factor was spotted in three 674 separate spots on each plate. Plates were incubated for 48 h at 30C, and then images were 675 taken using a Bio-Rad Gel Doc XR+ system. Using FIJI software, circles were fit to the zone of 676 arrest surrounding each -factor spot, and the diameter of the circles was measured in pixels. 677 Immunoblotting 678 Cell cultures were grown in triplicate overnight to mid-log phase in YEPD. 10 7 cells were 679 collected by centrifugation, and protein was extracted by TCA precipitation as described 680 ( Variances" function ( Fig. 7 C, D, Fig. 8C). Two-sample Kolmogorov-Smirnov tests were 690 performed using the Real Statistics Resource Pack software (Release 5.4, developed by Charles 691 Zaiontz) Add-in for Microsoft Excel (Fig. 2D, Fig. 5A-D, Fig. 7F, Fig. 8D Fig. 6C, in G1 cells with Ste2-sfGFP (blue), Ste2 NPF -sfGFP (orange), and Ste2 7XR-GPAAD -sfGFP (green). (C) Ste2-sfGFP abundance. Left: representative Western blot. -GFP antibodies label two bands -full-length Ste2-sfGFP and vacuolar sfGFP (note absence of vacuole signal for Ste2 7XR-GPAAD ). Right: quantification of full-length Ste2 abundance (n=3 biological replicates, normalized to the average abundance of wild-type Ste2). (D) Halo assay for pheromone sensitivity of cells with wild-type Ste2 (blue), Ste2 NPF (orange), and Ste2 7XR-GPAAD (green). Top: images of representative halos. Bottom: quantification of halo diameter (n=9, 3 technical replicates at 3 pheromone concentrations, normalized to the average wild-type halo diameter; * t test, p < 0.05). (E) Cumulative distribution of the duration of the indecisive phase for MATa cells that were born immediately adjacent to a MATα partner in G1, and either budded or mated by the end of the movie. Cells harboring Ste2 (blue, n=71), Ste2 NPF (orange, n=53), or Ste2 7XR-GPAAD (green, n=47). (F) Left: Cumulative distribution of initial Bem1 cluster orientation relative to the nearest potential mating partner for MATa cells born immediately adjacent to a MATα G1 cell. Cells with wild-type Ste2 (blue, n=117), Ste2 NPF (orange, n=78, not significant), or Ste2 7XR-GPAAD (green, n=79, not significant). Right: polar histograms of the same data.  Left: Gα is activated by pheromone-bound receptor (Ste2 + α-factor), and inactivated by the RGS protein Sst2. Sst2 associates with inactive Ste2. When Ste2 is activated by α-factor, Sst2 dissociates from Ste2. The instantaneous activation state of Gα is determined by the state of the receptor with which it last interacted. Right: Gα switches between active (green arrows) and inactive (red arrows) states when it interacts with active (green circles) and inactive (red circles) receptors. The fraction of the local Gα that is active reflects the ratio of active to inactive receptors, regardless of receptor density. This means differences in pheromone level at different points on the cell surface can be compared even if there are differences in receptor density. (B) hsRGS4 is distributed uniformly on the membrane. Single-plane inverted image of hsRGS4-CFP. (C) Pheromone sensitivity measured via halo assay in wild-type cells (blue), and cells in which Sst2 has been replaced by one copy (gray, hsRGS4, * t test, p < 0.05) or two copies (red, hsRGS4x2, not significant) of hsRGS4 (n=9, 3 technical replicates at 3 pheromone concentrations, normalized to the average wild type halo diameter). (D) Left: Cumulative distribution of the location of initial Bem1 cluster orientation relative to the nearest potential mating partner in wild type cells (blue, n=117), hsRGS4x2 cells (red, n=85, * two sample KS test, p < 0.05), and hsRGS4x2 cells harboring Ste2 7XR-GPAAD (uniform receptor, green, n=65, not significant). Right: polar histogram of the same data for hsRGS4 strains. (E) Left: Cumulative distribution of the location of initial Bem1 cluster formation relative to the site of cytokinesis in wild type cells (blue, n=117), hsRGS4x2 cells (red, n=85, * two sample KS test, p < 0.05), and hsRGS4x2 cells harboring Ste2 7XR-GPAAD (uniform receptor, green, n=65, not significant). Right: The same data represented as a polar histogram (WT not plotted).(F) Bem initial cluster location is biased by Ste2 distribution in hsRGS4x2 cells. Average Bem1 distribution at the time of initial clustering relative to Ste2 maximum, plotted as in Fig. 6E (n=33). Scale bar, 3 μm. Strains: DLY22318 (B, C), DLY22321, DLY22520 (C-F), DLY12943, DLY22606 (D-F).