Integrating multiple single-cell phenotypes links stress acclimation to prior life history in yeast

Stress  defense  and  cell  growth  are  inversely  related  in  bulk  culture  analyses;;    however, these  studies  do  not  capture  cellular  heterogeneity,  thus  obscuring  true  phenotypic relationships.    Here,  we  devised  a  microfluidics  system  to  characterize  multiple phenotypes  in  single  yeast  cells  responding  dynamically  to  stress.    We  simultaneously followed  cell  and  colony  growth,  cell  size  and  volume,  and  cell-­cycle  phase  plus  nuclear trans-­localization  of  two  transcription  factors:  stress-­responsive  activator  Msn2  and repressor  Dot6  that  are  co-­regulated  during  stress.    Coordinated  activation  reflects  a


Introduction
All organisms respond to stressful environments. Stress can come from external conditions including drugs and environmental shifts, or internal perturbations including mutation and disease. Thus, at the cellular level, organisms must be able to sense both external and internal signals to mount a proper response. Yet in both single- and multicelled organisms, there can be large variation in how individual cells respond to environmental stress, even among genetically identical cells in the same environment.
For example, cell-to-cell variation in signaling and gene expression have been linked to differential survival of isogenic cancer cells responding to drugs ; Paek et al. 2016;; Shaffer et al. 2017;; Inde and Dixon 2018). Similarly, cellular heterogeneity in bacterial growth and gene expression can produce variation in survival upon antibiotic treatment (Balaban et al. 2004;; Keren et al. 2004). Understanding the nature of this variation could facilitate the modulation of stress survival, with therapeutic applications.
One marker of heterogeneity in stress responses is dynamic localization of stress-activated transcription factors. Several canonical factors, including p53 in mammalian cells (Purvis et al. 2012;; Kracikova et al. 2013;; Paek et al. 2016) and Msn2 and its paralog Msn4 in fungi (Görner et al. 1998), reside in the cytosol in the absence of stress but rapidly translocate to the nucleus upon activation. These and other stressactivated factors can vary substantially in their responsiveness, in ways that can impact cellular outputs including gene-expression. For example, Msn2 localization dynamics differ depending on the nature of the stress (Hao and O'Shea 2012;; Petrenko et al. 2013;; Granados et al. 2018), and these differences impart distinct effects on different target genes (Hao and  patterns of p53 localization dynamics can alter expression of different gene sets, and these differences correlate with differences in stress survival (Purvis et al. 2012).
Cell-to-cell variation in transcription factor localization dynamics could arise for several reasons. Changes in the state of a single transcription factor may alter its localization independent of the cellular system (defined as factor-specific variation). In contrast, activity-state changes in the upstream signaling networks or cellular system itself could produce coordinated activation of the stress response (referred to as systemic variation). Distinguishing between local versus systemic variation has been difficult, since most studies to date have followed only single transcription factors. We recently developed strains in which two differentially tagged transcription factors regulated by the same signaling network are expressed in the same yeast cell. Msn2 activator fused to mCherry is co-expressed with the transcriptional repressor Dot6 fused to GFP. Both factors help to coordinate the yeast environmental stress response (Gasch et al. 2000;; Causton et al. 2001): whereas Msn2 activates defense genes that are induced in the ESR (iESR genes), Dot6 represses growth-promoting genes involved in ribosome biogenesis that are correspondingly repressed in the ESR during stress (rESR genes) ; Bergenholm et al. 2018). Both factors are controlled by the Protein Kinase A (PKA) and mTOR pathways, which are generally associated with promoting growth ( Figure  1A): PKA/TOR-dependent phosphorylation of Msn2 and Dot6 maintains the factors in cytosol, whereas Msn2 and Dot6 dephosphorylation after PKA/TOR inhibition leads to their nuclear localization. (Görner et al. 1998;; Smith et al. 1998;). Thus, we expect the two factors to be coordinated in their localization when the stress response is activated systemically but discordant in response to factor-specific differences in regulation. The challenges in distinguishing factor-specific versus systemic variation have obscured how systemic activation of the stress response relates to other physiological responses. One important factor is growth rate. Growth rate and stress tolerance are competing interests in the cell and are often antagonistically regulated: fast growing cells tend to be the most susceptible to stress and toxins, whereas slow growing or quiescent cells generally survive extreme conditions (Balaban et al. 2004;; Lu et al. 2009;; Zakrzewska et al. 2011;; Levy et al. 2012). Part of this antagonistic correlation is thought to be controlled, at least under specific situations, by the RAS-PKA pathway, which promotes growth and suppresses the stress response (Smith et al. 1998;; Gasch et al. 2000;; Zaman et al. 2008;; Zaman et al. 2009). Li et al. (2018) used a microfluidics system to show that slower growing cells in an isogenic culture showed lower levels of the PKA allosteric activator cAMP and that artificial activation of PKA diminished the slow growing population (Li et al. 2018). They further showed a slight but statistically significant negative correlation between Msn2 nuclear localization and growth of microcolony size over the subsequent 10 hours in the absence of stress. This suggests that activation of Msn2 is coupled to reduced growth rate, a theory put forward and debated in other bulk-culture studies (Regenberg et al. 2006;; Castrillo et al. 2007;; Brauer et al. 2008;; Ho et al. 2018). The inability to distinguish between factor-specific variation and systemic activation of the stress response likely obscures the true relationship with growth.
Here we monitored dynamic localization changes of both Msn2 and Dot6 in the same yeast cells, along with a panel of other single-cell measurements, to dissect local and systemic variation and illuminate the relationship between ESR activation and growth rate. We optimized a microfluidics system that can monitor single-cell localization levels and dynamics of both Msn2-mCherry and Dot6-GFP along with single-cell and colony growth rates, size, shape, cell-cycle phase and volume changes before and after an acute dose of sodium chloride (NaCl) as a model stressor. Our results revealed several insights, including surprising levels of discordance in Msn2 and Dot6 activation that suggest previously unknown regulatory distinctions and uncover subpopulations of cells. We developed a multi-factorial model explaining cell growth rate after stress acclimation to demonstrate that stress acclimation is partly predictable based on prior life history states. Our work highlights that quantifying and integrating multiple phenotypes, including the behavior of multiple transcription factors in the same cells, increases predictive power and contributes a deeper understanding of the causes and consequences of heterogeneous stress responses.

Results
We optimized a microfluidics system that could measure nuclear localization dynamics as well as one- and two-cell colony growth rates before and after exposure to 0.7 M NaCl ( Figure  1B and Methods). Using this system, we characterized the variation in cell responses for 72 min before and 144 min after exposure to NaCl, which induces ionic and osmotic stress, in biological triplicates done on separate days. This time frame captures phenotypic variation in cells growing in the absence of stress, during the acute stress-response phase (from 0 to 54 minutes after osmotic stress), and over later timepoints as cells acclimate to continuous NaCl. Microscopy imaging and analysis reports on Msn2-mCherry and Dot6-GFP nuclear localization dynamics in the same cells ( Figure  1C). We used MATLAB scripts to identify nuclear translocation events, which we refer to as "peaks" in the traces (see Methods). We also measured cell and colony growth phenotypes, including colony size, colony growth rates (defined by increase in pixel number of masked colony area) both before and after stress, and volume lost upon NaCl stress ( Figure  1D and Methods). To insure accurate measures of growth, we limited our analysis to colonies of only one or two cells at the beginning of the time series and to cells that passed several quality-control filters (see Methods). In total, we analyzed 221 cells passing these filters, collected from the three independent biological replicates.  ) for pre-stress growth, the acute-stress response, and the acclimation phase. D. Cell or twocell colony size was estimated by the number of pixels within the mask for each colony, and growth rates were calculated based on the regressions of those points during the pre- or post-stress phases. Cell volume change was calculated as the difference in pixel number before and after stress.
This system captured variation in all of the features measured. As expected based on previous studies (Levy et al. 2012;; Fehrmann et al. 2013;; Crane et al. 2014;; Li et al. 2018;; Jin et al. 2019), there was substantial variation in cellular growth rates before NaCl addition, confirming that cells vary considerably in their growth properties in the absence of stress ( Figure  2A). Most colonies reduced their growth rate in response to NaCl stress, but once again there was substantial variation: some cells showed dramatic growth reduction upon NaCl whereas others showed little to no change ( Figure   2B). There were even individual colonies that accelerated growth after stress: 11 of 14 of these cells showed a small bud immediately before, or at the time of salt exposure, suggesting a cell cycle connection ( Figure  2B). NaCl-induced osmotic pressure is expected to produce rapid water loss before cells acclimate, and indeed most cells lost volume immediately after stress despite substantial variation in volume change ( Figure   2C). Together, these results highlight the extensive cell-to-cell variation in behavior that is not identified in bulk measures of culture growth.  Cell-to-cell heterogeneity in the NaCl stress response. A-C. Shown are the distributions of the natural log of A. colony growth rates before stress, B. the change in growth rate after NaCl stress compared to before stress, and C. the maximum change in cell volume during the acute-stress response versus during the pre-stress phase.

Msn2 and Dot6 nuclear localization shows only partial coordination
We next investigated co-variation in Msn2-mCherry and Dot6-GFP localization dynamics, before and as cells responded to NaCl. Both factors showed sporadic activation in unstressed cells, with brief and typically low levels of nuclear translocation ( Figure  3A). In roughly 21% of cells, pre-stress peaks of Dot6 and Msn2 were temporally coordinated with the other factor ( Figure  3B), which is significantly above chance (p << 0.0001, permutation analysis, see Methods) and suggests systemic activation of the stress response. This reveals both coordinated and independent fluctuations in Msn2 and Dot6 activation in the absence of stress, consistent with our prior results (Gasch et al. 2017). In the vast majority of cells, NaCl provoked a dramatic and coordinated increase in nuclear localization of both Msn2 and Dot6 (acute phase).
However, after stress Msn2 and Dot6 behavior deviated: whereas few cells showed post-stress Dot6 nuclear translocation, many cells showed asynchronous pulses of Msn2 ( Figure  3C), consistent with prior work (Petrenko et al. 2013). This was surprising, since we expected that Msn2 and Dot6 would be highly correlated during and immediately after NaCl treatment.
Plotting the aggregate and standard deviation of nuclear localization over all cells highlighted the different behaviors of the two factors ( Figure  3D-F). The population deviation for both transcription factors was similar during the pre-stress phase, consistent with the significant co-variation of the two factors. However, during the acute  response immediately after NaCl exposure, the standard deviation of Dot6 was much wider than that of Msn2, indicating more variable Dot6 responses across the cells. In contrast, the standard deviation of Dot6 was very low during the acclimation phase but remained high for Msn2 ( Figure  3F). These results suggested that, even during a coordinated stress response, there are substantial differences in heterogeneity for Msn2 versus Dot6.
In the course of this analysis, we realized another key difference between Msn2 and Dot6: the timing of Dot6 nuclear pulses was often correlated between unstressed cells in the same two-cell colony, indicated by co-occurring peaks in colonies (Table   S1). Permutation tests showed that this was significant compared to random chance (p = 9.3e-4, see Methods). In contrast, the number of co-occurring Msn2 peaks in cells in the same two-cell colony was not significantly different from random. Since these cells are in the same local environment and have a shared life history in that one cell is the daughter of the other, it suggests that some feature of Dot6 regulation is predictable but separable from Msn2 behavior (see Discussion).  (Table  S2). Thus, most of the cell groupings represent reproducible subpopulations with different stress-responsive behaviors.

cells passing quality control metrics were partitioned into sub clusters
based on their population-centered nuclear translocation dynamics shown on the right. Each row represents a cell and each column in a block represents a single timepoint;; time of NaCl addition is indicated with an arrow. Data on the left show the log 2 ratio of nuclear versus total Msn2 (left) or Dot6 (right) according to the orange-scale key, see Methods. Data on the right show the same data normalized to the population median at each timepoint: yellow values indicate higher-than-median nuclear localization and blue indicates lower-than-median nuclear localization. Cell clusters identified by the package mclust are labeled to the right. The subpopulations were differentiated by a combination of transcription-factor phenotypes. One distinguishing feature was the level of Dot6 activation during the acute-stress phase. Cluster 11 was characterized by lower than population-median magnitude of acute-stress Dot6 nuclear translocation, whereas cells in Clusters 6 and 7 showed higher-than-average Dot6 response. These results are consistent with the wider variance of Dot6 nuclear translocation levels during the acute phase ( Figure  2D-F). A second distinguishing feature was the level of nuclear Msn2 and Dot6 before stress.
Cluster 11 cells showed low levels of Dot6 before stress, whereas cells in Clusters 9 and 6 displayed higher-than-median nuclear Msn2 and Dot6 during this phase. Finally, the behavior of Msn2 during the post-stress acclimation phase was significantly different across subpopulations. Whereas Clusters 11 and to some extent 7 showed low levels of post-stress Msn2 nuclear localization, cells in multiple clusters showed high levels and/or pulsatile nuclear Msn2 as cells acclimated. We noticed that cells in Clusters 2 and 3 showed elevated levels of Msn2 that persisted over time compared to other cells.
Closer inspection of the microscopy images suggested that some of the signal may not reflect nuclear translocation but instead was likely vacuolar signal (see more below). As mentioned above, the variation in nuclear localization dynamics captured within these clusters occurred in all three biological replicates (See Table  S2), indicating reproducible distinctions in transcription factor behavior. Together, this analysis revealed important differences in cellular behavior across the phases of the NaCl response that are obscured by aggregate analysis of all cells in the population.

Cell subpopulations show different relationships with cell growth
Are subpopulations of cells identified above biologically meaningful? We turned to the other cellular measurements to look for co-variates in cellular behavior that reflect on higher-order relationships. We tested each of the cell subpopulations for statistically significant differences in pre-stress growth rate, post-stress growth rate, starting size, volume change, and cell-cycle phase at the time of NaCl exposure (inferred by visual inspection of bud size and nucleus location in the cell, see Methods). We found no significant correlations with cell volume changes or cell-cycle phase (although there was a minor signal for cell cycle, Figure  S1). This is consistent with the lack of strong connection between cell-cycle phase and stress response found in several other studies (Paek et al. 2016;; Gasch et al. 2017;; Bagamery et al. 2020). In contrast, several clusters showed significant differences in growth rates.
Overall, there was a positive correlation between pre-stress colony growth rate compared to post-stress growth rate ( Figure  5A);; however, the association was different for subpopulations of cells. Cells in Cluster 11, which were characterized by below-average Dot6 response before and during stress, showed slower growth rates before being exposed to stress and after NaCl treatment ( Figure  5B-C), and the slower growth was consistent across biological replicates (p < 0.02, Wilcoxon rank sum for each replicate analyzed separately). In contrast, cells in Cluster 7 showed higher than average recovery growth rates - these cells were characterized by larger-than-average Dot6 nuclear localization responses and somewhat below-average nuclear translocation of Msn2 during the acute-response phase. The relationships between post-stress growth rate and Dot6 response during the acute phase raised the possibility that this factor's activation is more closely tied to growth rate than Msn2, even when both factors are activated in a systemic response. Interestingly, cells in Cluster 2 that had unusually high (and potentially vacuolar) red-channel fluorescence before stress displayed very slow growth recovery after stress, demonstrating the biological validity of the subpopulation and raising the possibility of poor stress acclimation in these cells. (We note that cells with apparent vacuolar signal were excluded from subsequent analyses).  Figure  4. B-C. Distribution of mediancentered growth rates before (B) and after (C) NaCl addition, for cell clusters shown in Figure  4. Boxes are colored yellow or blue if the distribution was significantly higher or lower, respectively, from all other cells in the analysis (Wilcoxon Rank Sum test, FDR < 0.022). Dashed line indicates the median of all cells analyzed.

Combining multiple characteristics increases the predictive power to explain growth
The above results hinted that how well cells acclimate to NaCl stress, as indicated by post-stress growth rate, may be predicted by cellular responses both before and during the stress response. Based on the work of (Li et al. 2018), we expected a negative correlation between Msn2 nuclear localization and growth rate.
While there was no correlation with pre-stress growth rate (p = 0.65), we indeed observed a negative correlation between pre-stress Msn2 activation (taken as the area under the nuclear-localization curve (AUC) for pre-stress timepoints) and post-stress growth rate ( Figure  6A). However, the correlation explained only 3% of the variance (p = 0.016, linear regression), indicating that the pre-stress behavior of Msn2 has little power to predict growth rate.
We next investigated other features that could explain differences in post-stress growth rate ( Figure  6B). Pairwise correlations revealed that some individual features, such as the magnitude of Dot6 acute-stress response, correlated well with post-stress growth rate but others did not ( Figure  S2). However, the most impactful single factor, pre-stress growth rate, explained only 20% of the variance in post-stress growth rate (Table  S3). We next asked if combing factors into a single multiple linear model could explain more of the variance in growth. We considered multiple metrics for summarizing prestress nuclear localization, including AUC (which is a measure of the overall nuclear abundance) and the sum of called translocation peak heights (which is influenced by the magnitude and frequency of pre-stress pulses), along with acute-stress translocation  Table  S3 for all parameters used).
Factors that did not contribute significantly (adjusted p > 0.05) were progressively removed until the variance explain decreased (Table  S3). The final regression identified 4 factors that contributed significantly to explain post-stress growth rate ('Model 3' in Table  S3): pre-stress AUC of Dot6 nuclear localization, the sum of pre-stress Msn2 peak heights, the pre-stress growth rate of the cells, and the magnitude of Dot6 nuclear localization change immediately after NaCl. Together, these factors - all but one of which represent pre-stress cellular phenotypes - explained 35% of the variance in poststress growth rate ( Figure  6C), nearly doubling the explanatory power of any single feature alone ( Figure  6D,  S2). Surprisingly, the sum of pre-stress Msn2 peak heights and pre-stress AUC of Dot6 nuclear localization both contributed significantly to the multi-factor model but were not predictive when considered individually ( Figure  S2).
This suggests that these measures of transcription-factor nuclear localization can explain residual variance even though they explain a small portion of the data when considered alone.

Dot6 responses to stress are influenced by pre-stress life history states
We were surprised to find that the magnitude of the Dot6 nuclear translocation during the acute stress response was among the most predictive of post-stress growth rate, whereas the Msn2 behavior contributed little explanatory power. We therefore sought factors that could explain the variation in Dot6 peak height during the acute response. We measured co-variation of Dot6 acute-stress peak height and a panel of features and realized several relationships. Although the magnitude of Msn2 nuclear   Figure  6B, and cell points are colored according to their cell cluster from Figure  4 as shown in the key. A-B. Acute-stress peak height plotted against pre-stress AUC for Msn2 (A) or Dot6 (B). C-D. Acute-stress peak height for Msn2 (A) or Dot6 (B) plotted against the difference in pre-stress AUC. E. The difference in pre-stress Msn2 and Dot6 nuclear localization best explains the difference in acute-stress peak height for Msn2 and Dot6. P-values and R 2 of the fit are shown on each plot. All p-values were significant at a Benjamini-Hochberg corrected false discovery rate of 0.05.
localization upon NaCl stress was not strongly influenced by pre-stress Msn2 activation ( Figure  7A), we did observe a significant positive relationship between Dot6 acutestress peak height and Dot6 AUC in pre-stressed cells - cells with more pre-stress activation of Dot6 had a larger Dot6 response to NaCl ( Figure  7B). The trend was recapitulated in a strain in which the fluorescent tags were swapped (Msn2-GFP, Dot6-mCherry, see Figure  S3), indicating that it is not an artifact of the tags. It is important to note that the relationship was influenced by cells in cell Cluster 11, since removing those cells yielded a significant correlation but with reduced the variance explained (from 29% to 6%, see Figure  S4B). We noticed when considering cell clusters in Figure  4 that many subpopulations showed inverse trends in pre-stress Msn2 versus Dot6 activation. Several clusters that had higher nuclear levels of Dot6 before stress had lower levels of Msn2, and vice versa. This was especially surprising, since Msn2 and Dot6 are thought to be coordinately activated to mount the yeast ESR. We therefore wondered if the relative activation of Dot6 versus Msn2 before stress was any indication of different cellular states. In fact, the relative difference in Msn2 AUC versus Dot6 AUC before stress was more predictive of the acute-stress nuclear translocation peak of both Msn2 and Dot6 than considering either pre-stress factor alone ( Figure  7C-D). To ensure that this was not an artifact of the fluorescent tags, we repeated the analysis in a strain in which the tags were swapped (Msn2-GFP, Dot6-mCherry). The trend was repeated for Msn2 but not Dot6 (perhaps owing to the unusually high level of variance explained in this experiment by pre-stress Dot6 AUC alone, Figure  S3). For both strains, the variance explained for acute-stress response was in fact highest when assessing the relative peak heights of Msn2 versus Dot6 during the acute-NaCl response compared to the relative Msn2 and Dot6 activation during the pre-stress treatment ( Figure  7E and S3E).
Although we cannot exclude the possibility that the improved fit is a technical artifact, it is notable that the fits with other features, including growth rate modeling, was not improved by considering the relative pre-stress activation of Msn2 versus Dot6 compared to considering Dot6 pre-stress behavior alone. This raises the possibility that the differences in Msn2 and Dot6 behavior differentiate cells in a way that is predictive of distinct subsequent stress responses (see Discussion). is that activating Msn2 by wholesale inhibition of PKA likely loses much of the heterogeneity seen in natural responses. Our study thus provides an important complement to artificial system activation.

Discussion
In fact, our analysis revealed highly varied responses across subpopulations of cells responding to the same stress in the same environmental system: some cells responded to the osmotic/ionic stress induced by NaCl with a large nuclear pulse of Msn2 followed by near complete acclimation, as previously reported for sorbitol-induced osmotic stress - but other cells showed extensive and prolonged Msn2 fluctuations during the acclimation phase, akin to what has been reported for glucose starvation.
These subpopulations are obscured by aggregate analysis but have important implications, since the different dynamics of Msn2 (and likely also Dot6) activation produce different transcriptomic outputs, even for cells responding to the same stressor in the same environment. This hypothesis is consistent with past work from our lab investigating single-cell transcriptomics, in which isogenic cells in the same culture displayed different transcriptomes upon NaCl stress, including for ESR genes, indicating that they experience the stress differently (Gasch et al. 2017).

An important unanswered question in biology is why some cells persist and later
resume growing after what should be a lethal drug treatment, whereas other cells succumb. This has major medical importance, from bacterial cells that evade antibiotic treatment to cancer cells surviving chemotherapy to later relapse to growth and metastasis. Here, we found that a multi-factorial model of cell phenotypes could explain 35% of the variance in post-stress growth rate. Although Msn2 response had been previously linked to growth rate in single cells (Li et al. 2018), we found little predictive association with Msn2 behavior ( Figure  6A).
Instead, the response of Dot6 explained a much larger fraction of the variance in post-stress growth rate, when considered alone or in the multi-factor linear model ( Figure  6 and Table  3). Cells with a larger Dot6 response during the acute-stress phase had a faster recovery during the acclimation phase ( Figure  S2F). Dot6 is a repressor of ribosome-biogenesis genes, and this connection is intriguing since ribosome abundance is associated with faster growth rates in many (although not all) circumstances (Dai et al. 2016;; Metzl-Raz et al. 2017;; Dai et al. 2018;; Ho et al. 2018). Whereas a cell's post-stress recovery rate is partly explained by its pre-stress growth rate (with 20% of the variance explained), its connection to the Dot6 acute-stress response is separable, since the two factors together explained a significantly higher fraction of the variance in recovery growth rate (26%) than pre-stress growth rate did alone (See Model 5 in Table   S3). We propose that the relationship between Dot6 response and post-stress growth rate lies in the production of ribosomes at the time of NaCl, which can be separated from the growth rate at that moment. Cells with higher active transcription of ribosome biogenesis genes at the time of NaCl exposure may require stronger slamming-on of the transcriptional brakes, reflected in a larger Dot6 acute-stress response. In turn, cells with new ribosomes coming off the production line may be prepared for faster recovery after NaCl stress, even if their pre-stress growth rate was not already accelerated. This is known for bacterial cells responding to antibiotic stress, where cells with higher ribosome content at the time of stress relief resume faster, including cells not growing when exposed to stress/relief (Kim et al. 2018;; Remigi et al. 2019).
Another major discovery of this work is that the regulation of Msn2 and Dot6, which together activate and repress gene sets coordinately induced and repressed in the yeast ESR, show less coordination than we expected at the outset. This suggests significant, unrealized differences in how the two factors are regulated during a stress response. Although the pre-stress activation patterns of Msn2 and Dot6 were correlated above random chance, many of the nuclear pulses were uncoupled in timing and magnitude. Furthermore, the acclimation profiles were quite different and often unrelated. These differences could emerge through several possibilities, including differences in attenuation/feedback and differences in upstream activation inputs. On behavior may reflect subtle distinctions in cellular states, and these distinctions influence the subsequent stress response. We show that these states are unlikely to relate strongly to cell-cycle phase or cell size, but could reflect different metabolic states known from single-cell microscopy and transcriptomic studies (Gasch et al. 2017;; Bagamery et al. 2020). While the exact nature is not presently known, together these results highlight that much of the heterogeneity in stress responses is predictable and determined by a cell's prior life history. An FCS2 chamber (Bioptechs Inc, Butler, Pennsylvania) microfluidic system was used for time-lapse microscopy. In short, a 40 mm round glass coverslip and FCS2 lower gasket were assembled, and Concanavalin A solution (2 mg/mL Concanavalin A, 5 mM MnCl 2 , 5 mM CaCl 2 ) was applied to the coverslip, incubated for two minutes, then aspirated. Next, 350 µL of an ~ 0.5 OD 600 culture was placed on the coverslip and incubated 5 minutes for cells to settle and adhere to the Concanavalin A. 150 µL of the media was then removed and the rest of the FCS2 chamber was assembled. Media was flown through the FCS2 chamber using gravity flow. Input tubing was attached to elevated bottles containing either LFM or LFM + 0.7 M NaCl (See diagram in Figure  1B) with a valve to switch between media with and without 0.7 M NaCl. The outflow tubing was connected to an additional ~1 meter of BD Intramedic PE Tubing Images of each field of view were captured at 6-minute intervals. The z-focal plane focus was set on the center of cells, and images were taken 1 um above, at, and 1 um below this center of focus, generating a three-image z-stack for each channel. The three-image z-stacks were collapsed into a single image by taking the maximum projection of the 3 images using a custom MATLAB script.

Methods
Cells were identifying using a MATLAB circle-finding function on the brightfield images. Individual cells were then tracked through all images using the MATLAB simpletracker function (Tinevez 2019). Cell colonies were defined by segmenting images into a binary black-and-white image, and single colonies were defined as We applied several quantity control filters to insure accuracy of growth rates.
First, to insure that colony growth rates were representative of nuclear localization dynamics within individual cells, we limited our analysis to colonies consisting of no more than two cells at the time points leading up to NaCl exposure. Most of these two-

* Area Under the Curve (AUC) of nuclear localization: For pre-stress time points,
AUC was calculated by summing the first 9 measurements of nuclear localization scores (top brightest 5% of pixels over the median cellular signal). This summation represents the total relative levels of nuclear localization between all cells. The same AUC calculation was done for the acclimation phase using time points 24 -37. The difference in AUC between the two signals (Msn2 - Dot6 AUC in Figures  6  and  7) is simply the difference of the two individual AUC measurements.

* Acute Stress Peak Height: The acute stress peak height was calculated by taking
the maximum of nuclear localization score during the acute stress response (time points 13 - 20) and then subtracting the minimum of the nuclear localization scores just before stress (time points 11 - 13).

Cell clustering to identify subpopulations
Nuclear localization scores were log 2 transformed, and for each cell and each factor, the value at each timepoint was normalized to the median of all cells for that factor and time point (Figure  4, blue/yellow scale data). The population-median-normalized vector for Msn2 and Dot6 were concatenated and clustered by mclust (Scrucca et al. 2016) using model EII and k=30 (which was collapsed to k=11 by mclust). The log 2 of unnormalized nuclear traces for each cell was added for display in Figure  4 (orange/white scale data).
Relationships with logged growth rate data before and after stress, calculated as described above, were scored for each cluster of cells compared to all other cells in the data ( Figure  5, Wilcoxon Rank Sum test). Visual inspection of cells within mclust clusters 2 and 3 indicated that some Msn2 signal was focused but outside the nucleus (evidence after NaCl treatment), likely in the vacuole. There were 18 cells were this was observed visually. Since the impact of this signal was uncertain, these cells were excluded from subsequent regression modeling (i.e. Figures  5A,  6  and  7).

Probabilities of the number cells from each of the three biological replicates.
Binomial probabilities were used to determine if each cluster contained more cells from one of the three biological replicates than would be expected by chance.
Specifically, if is the number of cells from a given biological replicate present in a cluster, then probability of having cells or more in the cluster is where is the number of cells in the cluster and is the expected probability of having a cell from a given replicates (that is, the total number of cells in the replicate divided by the total number of cells in all three replicates). Since clusters 8 and 10 had a total of 5 and 6 cells, respectively, they lacked statistical power and were excluded from the analysis.
The Holm-Bonferroni method was used for multiple hypotheses correction, where there were = 27 (9 clusters and testing the number of cells from 3 biological replicates in each cluster) and a = 0.05. Using this threshold, only Cluster 9 showed strong enrichment for cells from one replicate compared to expected after multiple-test correction. Of note, this was the only cluster (besides Clusters 8 and 10), that had zero cells from a given replicate (Table  S1).

Permutations of nuclear localization peak matches
To identify if matched peaks of Msn2 and Dot6 were more coordinated than expected by chance, permutations were performed where a random Msn2 and Dot6 trace, including the time points of the called peaks, were randomly paired from the entire dataset. Coordinated peaks were then calculated from these random Msn2/Dot6 trace pairs. These permutations indicated that the number of matched Msn2/Dot6 peaks per cell was much higher than expected by random combinations (zero permuted datasets out of 100,000 total had 0.21 matched pre-stress peaks per cell or more). The same test was done for the matched peaks during the acclimation phase, and although the number of matched peaks per cell was significant for the acclimation time points (a fraction of 4 x 10 -4 of permuted datasets had 0.07 matched peaks per cell or more), this was significantly less than that for the pre-stress time points. This again demonstrated that there was more coordination in nuclear localization between Msn2 and Dot6 during the before stress compared to after stress.

Permutations of nuclear localization peak correlations between cells in two-cell colonies
There were 56 two-cell colonies in the dataset. Of these, 15 colonies showed coordinated Dot6 peaks between the two cells, defined as peaks occurring within one time point of each other. Permutations were performed where the 112 cells from these 56 colonies were randomly assigned in pairs and the same coordinated peak measurements were performed. Similarly, permutations were performed on matched peaks of Msn2. Results are shown in Table  S3.

Linear models
Multiple linear regressions shown in Figure  6 and Table  S3 were performed using fitlm in MATLAB. For each model was represented by = ; + = = + > > + ⋯ + 0 0 + e where the dependent variable is the post-stress growth rate, ; is the intercept, each subsequent is the estimate of the slope for each independent variable , and e is the error term. A list of independent variables is shown in Table  S3 for each of the multiple linear regression performed. The p-values shown in Table  S3 were determined from the t-statistic of each coefficient was not equal to zero. In Table  S3 Figure  S1. The strength of transcription factor nuclear localization is weakly related to cell cycle phase and budding. Cells grow the fastest in G1 and M-phase (Goranov et al., 2009), thus cells were binned if they were in G1 or M phase or S/G2 phases at the time of NaCl exposure. A. The peak height of nuclear localization for Msn2 and Dot6 during the acute-stress phase are plotted, with cells in S/G2 or M/G1 indicated in red or blue, respectively. B. and C. The distribution of acute-stress peak heights for Dot6 (B) or Msn2 (C) was plotted for cells in M/G1 or S/G2. Wilcoxon rank sum tests show that cells in S/G2 had slightly higher nuclear accumulation of both factors (p<0.001).
suggesting that much of cell-size contribution is correlated with and thus absorbed by other factors in the model. K. Fit from a multiple linear model similar to that shown in Figure  6 except in which pre-stress growth rate was not included (coefficient set to 0). The figure demonstrates that a substantial amount of the variation in post-stress growth rate is explained by features other than pre-stress growth rate of the cells. Figure  S3. Comparison of pre-stress Area Under the Curve with acute stress peak height. Same as Figure  7 in main text, but for strain AGY1666 (Msn2-GFP, Dot6-mCherry). Acute stress peak heights and pre-stress area under the curve (AUC) are as defined in Figure  6B in main text. Here, two experimental batches done with strain AGY1666 were combined. Here, (D) shows that taking the difference between Msn2 and Dot6 pre-stress AUC does not improve the correlation as it had done for strain AGY ( Figure  7D in main text). However, there is a slight improvement when the pre-stress AUC difference is correlated with the difference in acute stress peak heights (E). Figure  S4. Cells in Cluster 11 do not explain correlations in transcription factor behavior. As shown in Figure  7 except with cells from cell Cluster 11 removed from the analysis. Table  S1. Permutations of coordinately timed peaks in cells in two-cell colonies number of colonies with coordinated peaks (out of 56 total) Fraction of permutations with coordinated peak number equal or greater to actual number of colonies with coordinated peaks* Dot6 coordinated peak 15 9.5 x 10 -4 Msn2 coordinated peak 6 0.19 * There were 50,000 iterations of permutations for each comparison.  Figure  4 is shown along with the number of those cells from each of three biological replicates. P-values from binomial probability tests (see Methods) are shown and those significant after Holm-Bonferroni correction (namely Cluster 9 which was enriched for cells from replicate 3) are indicated with an asterisk. *Cell-cycle phase included four binary parameters (i.e. dummy variables) corresponding to the four cell-cycle phases (G1, S, G2 and M phase). See the Methods section for how these were scored. ** Model 2 only included all the significant dependent variables from Model 1. *** Model 3 excluded Msn2 acclimation AUC and Cell/colony size from Model 2 as the p-values did not pass Holm-Bonferroni correction.