Cellular quiescence is a nonproliferative state required for cell survival under stress and during development. In most quiescent cells, proliferation is stopped in a reversible state of low Cdk1 kinase activity; in many organisms, however, quiescent states with high-Cdk1 activity can also be established through still uncharacterized stress or developmental mechanisms. Here, we used a microfluidics approach coupled to phenotypic classification by machine learning to identify stress pathways associated with starvation-triggered high-Cdk1 quiescent states in Saccharomyces cerevisiae. We found that low- and high-Cdk1 quiescent states shared a core of stress-associated processes, such as autophagy, protein aggregation, and mitochondrial up-regulation, but differed in the nuclear accumulation of the stress transcription factors Xbp1, Gln3, and Sfp1. The decision between low- or high-Cdk1 quiescence was controlled by cell cycle–independent accumulation of Xbp1, which acted as a time-delayed integrator of the duration of stress stimuli. Our results show how cell cycle–independent stress-activated factors promote cellular quiescence outside G1/G0.
Introduction
Quiescence is a nonproliferative cellular state regulated by stimuli including starvation, DNA damage, and developmental signals (Cheung and Rando, 2013; van Velthoven and Rando, 2019). In unicellular organisms, quiescent states are essential for surviving adverse environmental conditions ranging from starvation to antibiotic exposure (Rittershaus et al., 2013). In multicellular organisms, quiescence is essential for tissue homeostasis and the maintenance of adult stem cells (Sun and Buttitta, 2017; Bainbridge, 2013; Apte et al., 2015; Nakamura-Ishizu et al., 2014; Li and Clevers, 2010).
Quiescence in eukaryotes commonly occurs after cells halt proliferation in a state where the activity of the proliferation-promoting kinase Cdk1 is low (low-Cdk1 quiescence). In many organisms, however, cells can enter quiescence in states of high-Cdk1 activity (high-Cdk1 quiescence), which are essential for development and single-cell survival (Wei et al., 1993; Hajeri et al., 2005; Velappan et al., 2017; Baisch, 1988). High-Cdk1 quiescence occurs in mammalian oocyte development (Nixon et al., 2002), embryonic development of invertebrates (Hajeri et al., 2005; Nystul et al., 2003; Otsuki and Brand, 2018), plant meristems (Velappan et al., 2017), microorganisms (Laporte et al., 2011; Klosinska et al., 2011), and metastasis-initiating cancer cells (Wang et al., 2015). Despite its ubiquitous nature, mechanisms for high-Cdk1 quiescence remain largely uncharacterized (Sun and Gresham, 2021).
Several conserved mechanisms maintain low-Cdk1 quiescence including (1) the accumulation of Cdk1 transcriptional repressors, such as Whi5/Rb; (2) the accumulation of Cdk1 inhibitors, such as Sic1/p27/p21; and (3) the destruction of Cdk1 activators by anaphase promoting complex/cyclosome (APC/C)-Cdh1–mediated proteolysis (Moser et al., 2018; Cappell et al., 2016; Hopkins et al., 2017). By contrast, only a few factors promoting high-Cdk1 quiescence have been identified, such as Caenorhabditis elegans spindle assembly components SAN-1 and MDF-2 (Nystul et al., 2003) or the pseudokinase Tribbles in Drosophila melanogaster neural stem cells (Otsuki and Brand, 2018).
Stress-activated pathways are good candidates to promote quiescence regardless of Cdk1 activity due to their capacity to decrease overall translation and transcription and modulate cell cycle protein networks (Miles et al., 2013; Marion et al., 2004; Escoté et al., 2004; Wysocki et al., 2006). However, quiescence-promoting stimuli, such as starvation, often trigger multiple stress signaling pathways that are experimentally challenging to simultaneously measure in single cells (De Virgilio, 2012), and high-Cdk1 quiescent states are particularly difficult to characterize because they are less common than low-Cdk1 quiescence or occur in complex tissues (Velappan et al., 2017; Otsuki and Brand, 2018). These challenges can be overcome by studying quiescence in a tractable unicellular model organism such as Saccharomyces cerevisiae; however, most protocols to induce quiescence require long-term batch cultures, where interactions between cells and changing chemical and physical medium parameters confound the assessment of whether stress factors are associated or causative of quiescent states (Klosinska et al., 2011). Furthermore, because single cells are not tracked in batch cultures, it remains controversial whether high-Cdk1 quiescent states in S. cerevisiae correspond to a failure of arresting cell cycle in response to starvation, a population of slowly dividing cells, or a proper cellular strategy to survive acute starvation stress after G1 exit.
In this work, we used a microfluidics assay with controlled environmental conditions to study S. cerevisiae’s transition from proliferation into quiescence after acute starvation. To characterize quiescent states with high-Cdk1 activity, we used machine learning algorithms to automatically distinguish proper high-Cdk1 quiescent states from slowly dividing or senescent cells. Quiescent states were classified based on established markers of the high-Cdk1 activity state, such as destruction of the Cdk1 inhibitor Sic1, inactivation of APC/C-Cdh1–mediated proteolysis, and assembly of the septin ring at the budding site after G1 exit (Zhang et al., 2011).
We found that high-Cdk1 quiescent states reproducibly occurred when a proliferating population was challenged by acute starvation. The frequency of high-Cdk1 quiescent cells (Q-cells) depended on the stress status of the population at starvation onset. Although all quiescent states shared a core of stress-associated processes such as the up-regulation of autophagy, protein aggregation, and mitochondrial biomass, high-Cdk1 quiescence was distinguished by increased nuclear accumulation of the general transcriptional repressor Xbp1, the nitrogen stress transcription factor Gln3, and the ribosome biogenesis factor Sfp1. The establishment of high-Cdk1 quiescence was controlled by cell cycle–independent nuclear accumulation of Xbp1, which acted as a time-delayed integrator of stress stimuli. Our results show that cell cycle–independent integration of stress stimuli by transcriptional repressors is a viable cellular response to establish quiescence outside a G1/G0 state.
Results
Acute starvation induces heterogeneous quiescent states
To study the establishment of quiescence in single cells, we modified a previously published microfluidics assay for quiescence induction in S. cerevisiae (Argüello-Miranda et al., 2018). Briefly, proliferating W303 diploid cells were cultured in a flow cell with rich medium (synthetic complete defined [SCD]) before being exposed to 20 h of acute starvation medium (0.6% potassium acetate, pH 7.2, 0.6 psi, which is a modified version of meiosis-inducing medium; Buonomo et al., 2000). Upon acute starvation, cells recapitulated quiescence properties previously observed in batch cultures (Miles et al., 2019). Proliferation was stopped after a slow and asymmetric final cell division, which required 5 ± 3 h under starvation conditions instead of 93 ± 7 min in SCD (Fig. 1 A). The daughter/mother cell size ratio at cytokinesis decreased from 0.58 ± 0.5 in rich medium to 0.25 ± 0.04 for cells born after starvation (Fig. 1, B–D). As reported in quiescent batch cultures (Shi et al., 2010), cells up-regulated trehalose and glycogen metabolism during quiescence entry as measured by tracking the level of the mNeonGreen-tagged trehalose synthase catalytic subunit, Tps1, and the mNeonGreen-tagged glycogen synthase, Gsy2 (Fig. 1 E). Q-cells maintained high viability as judged by return to proliferation (Fig. 1 F; 72 h is the limit of the microfluidics device) and acquired a stress-resistant state after 20 h of starvation as judged by 94 ± 3% survival (n = 7; 443 cells) after exposure to 4 M NaCl for 4 h (Video 1). Strikingly, although most cells arrested in G1, as judged by maintaining an unbudded state, a reproducible minority of 7 ± 3% cells remained arrested in a budded state (Fig. 1, G and H), which was highly viable upon return to rich medium (Fig. 1 I) and was reminiscent of budded quiescent states previously observed in stationary cultures in carbon-limited medium (Laporte et al., 2011). Single-cell tracking of budded and unbudded Q-cells showed no significant differences in cell volume (Fig. 1 J), Tps1 or Gsy2 up-regulation (Fig. 1, K and L), or cell morphology (Fig. S1 A), as assessed by the overlap of the 95% confidence intervals. We concluded that a fraction of proliferating S. cerevisiae cells entered viable quiescent states outside G1/G0 when challenged by acute starvation.
Acute starvation induces quiescence in states of low- or high-Cdk1 activity
To analyze the Cdk1 activity state of budded and unbudded Q-cells, we tracked the formation of the septin ring at the bud site—which is absent during the low-Cdk1 state of G1 but present during the high-Cdk1 state of S-M phase—by C-terminally tagging the septin Cdc10 with the fluorophore mCyOFP1 (McMurray and Thorner, 2009). The presence of the septin ring was measured by the SD of Cdc10 fluorescence at the cell periphery (Cdc10 signal; Fig. 2 A and Fig. S1 B). To analyze the Cdc10 signal of single cells during the proliferation–quiescence transition, we developed the algorithm “Time series Profiling by Machine Learning” (TPML), which groups single-cell time series according to similarity of cell cycle arrest into hierarchically sorted heatmaps (Fig. 2 B; Fig. S1, C and D; and Video 2). TPML analysis of Cdc10-mCyOFP1 time series established three major groups or “Cdc10 clusters” named according to whether cells reached a G1 cell cycle arrest during the proliferation–quiescence transition. Cells in cluster one—or “post-mitotic G1”—were in a high-Cdk1 state at starvation onset, as judged by the presence of the septin ring, and finished one last division before arresting in G1 (Fig. 2, C and D, blue). Cells in cluster two—or “Non-G1”—were also in a high-Cdk1 state at starvation onset but remained arrested in this state without reaching G1 conditions, as judged by a persistent septin ring (Fig. 2, C and D, red). Cells in cluster three—or “direct G1”—were in the low-Cdk1 state of G1 at starvation onset and remained arrested in this state (Fig. 2, C and D, black). Cdc10-cluster classification was independent of ploidy or strain genetic background (Fig. S1, E–H).
To assess the Cdk1 activity state of Cdc10 clusters, we imaged quiescence entry in cells bearing Cdc10-mCyOFP1 as a marker for the high-Cdk1 state of S-M phase and mNeonGreen C–terminally tagged Sic1 as a marker for the low-Cdk1 state of G1 (Hopkins et al., 2017). After starvation onset, nuclear Sic1 accumulated in post-mitotic G1 and direct G1 Q-cells according to their cell cycle progression (Fig. 2 E, black, blue). In contrast, nuclear Sic1 did not accumulate in Non-G1 Q-cells (Fig. 2 E, red), suggesting persistent Cdk1 activity.
The high-Cdk1 kinase state of S-M phase is characterized by the inactivation of APC/C-Cdh1–mediated proteolysis (Zachariae et al., 1998). To measure the activity of APC/C-Cdh1 in Q-cells, we tracked the nuclear levels of the APC/C-Cdh1 activity sensor mNeonGreen-Ase1(R632-I885), expressed from the MET3 promoter (Ondracka et al., 2016). During proliferation, the accumulation–degradation of the sensor followed the Cdc10 signal as expected for APC/C-Cdh1 substrates (Fig. S1 I). After starvation, the APC/C-Cdh1 sensor was degraded in post-mitotic G1 and direct G1 Q-cells (Fig. 2, F and G, black, blue) but persisted in Non-G1 Q-cells (Fig. 2, F and G, red), indicating APC/C-Cdh1 inactivation. Consistent with this observation, spindle pole body (SPB) separation, a process that requires APC/C-Cdh1 inactivation (Crasta et al., 2006), occurred, although with a delay, in Non-G1 Q-cells as judged by tracking of the SPB component Spc42 tagged with mTFP1 (Fig. 2 H; and Fig. S1, J and K).
To assess whether Non-G1 Q-cells entered S-phase, we tracked the DNA replication kinase Cdc7 C-terminally tagged with mScarlet-I. During starvation, nuclear Cdc7 persisted at higher levels in Non-G1 Q-cells (Fig. 2 I, red) in comparison to the other clusters (Fig. 2 I, blue, black). To measure DNA content in Q-cells, Cdc10-mCyOFP1 cells were microfluidically stained with propidium iodide after 20 h in starvation medium. On average, Non-G1 Q-cells had duplicated DNA content (Fig. 2, J and K).
Together, these results indicated that Non-G1 Q-cells persisted in a state of high-Cdk1 activity as judged by APC/C-Cdh1 inactivation, SPBs separation, DNA replication, and lack of Sic1 accumulation. High-Cdk1 quiescent states were equally viable to low-Cdk1 quiescence, as judged by comparing proliferation resumption among Cdc10 clusters (Fig. 2 L). We concluded that proliferating S. cerevisiae cells challenged by acute starvation after exiting G1 faced a decision-making process between remaining arrested in a high-Cdk1 state or completing a final cell cycle to reach a low-Cdk1 quiescent state in G1. In the rest of this study, we focused on cells experiencing starvation after G1 exit (post-mitotic and Non-G1 Cdc10 clusters) and used the TPML algorithm to classify them into “low-Cdk1 quiescence” if they finished one last division or “high-Cdk1 quiescence” if they remained arrested.
Low- and high-Cdk1 quiescent states share a core of stress responses but differ in the nuclear accumulation of Sfp1, Gln3, and Xbp1
To study how single cells challenged by starvation after G1 exit subsequently enter low- or high-Cdk1 quiescent states, we used a six-color fluorescent imaging system to assess to what extent differences in cell cycle progression or stress-activated factors determined high-Cdk1 quiescence entry (Argüello-Miranda et al., 2018). Imaging was optimized for using Cdc10-mCyOFP1 as cell cycle marker while simultaneously tracking five reporters for the major stress responses associated with the establishment of quiescence (Fig. 3 A; Fig. S2 A–H; and Video 3). Metabolic stress was measured by the nuclear translocation of the nitrogen stress transcriptional regulator Gln3 (Crespo et al., 2002) and the carbon stress/retrograde pathway transcriptional regulator Rtg1 (Komeili et al., 2000). The biosynthetic capacity of the cell was tracked using the nuclear translocation of the ribosome biogenesis factor Sfp1 (Marion et al., 2004). Transcriptional repression was tracked using the histone deacetylase regulators Stb3 and Xbp1, which are key transcriptional repressors during quiescence (Mai and Breeden, 1997; McKnight et al., 2015).
Sfp1-Stb3-Rtg1-Gln3-Xbp1-Cdc10 (6C1) cells modulated the nuclear levels of stress markers upon starvation (Fig. S2 H). Low- and high-Cdk1 quiescent states were identified by TPML, and the differences in the average time series of each group were assessed using the overlap in the 95% confidence interval. Strikingly, the most significant difference between low- and high-Cdk1 Q-cells was not cell cycle progression at starvation onset (Fig. 3 B), but the accumulation of nuclear Sfp1 (Fig. 3 C), Xbp1 (Fig. 3 D), and Gln3 (Fig. 3 E). In contrast, nuclear Rtg1 (Fig. 3 F) and Stb3 (Fig. 3 G) were accumulated to the same extent in low- and high-Cdk1 Q-cells (Fig. S2 I).
To assess whether low- and high-Cdk1 Q-cells presented differences in stress responses not directly related to starvation, we imaged a second six-color (6C2) strain bearing fluorescent reporters for calcium signaling (Crz1-mTFP1; Cyert, 2003), cell wall stress (Pkc1-mRuby3; Mishra et al., 2017), osmoregulation (Hog1-mNeptune2.5; Westfall et al., 2004), general stress (Msn2-mNeonGreen, Sfp1-mKOκ; Granados et al., 2018), and cell cycle (Cdc10-mCyOFP1). 6C2 cells readily reported stress stimuli if perturbed during proliferation (Fig. S2 J). Upon starvation, the nuclear levels of Msn2 (Fig. 4 A), Crz1 (Fig. 4 B), and Hog1 (Fig. 4 C) and the cell membrane-associated levels of Pkc1 (Fig. 4 D) were similar in low- and high-Cdk1 Q-cells.
To investigate the status of quiescence-associated processes (Sagot and Laporte, 2019; Laporte et al., 2018) in low- and high-Cdk1 Q-cells, we induced starvation in Cdc10-mCyOFP1 strains containing fluorescent sensors for autophagy, protein aggregation, mitochondrial biomass, and DNA damage. Autophagy was measured by the foci number of an autophagosome marker, mNeonGreen-Atg8, expressed from its own promoter (Fig. S2 L; Nair et al., 2011). Upon starvation, Atg8 foci increased with no difference between quiescent states (Fig. 4, E and F). Protein aggregation was measured using a glutamine-asparagine–rich peptide (Antonets et al., 2016) bound to an mRuby3 fluorophore (QN-mRuby3) under the control of a copper-inducible promoter (CUP1-1). CUP1 promoter (CUP1p)expression of QN-mRuby3 during proliferation produced few foci that remained within mother cells, as expected from protein aggregates in S. cerevisiae (Fig. S2 M; Saarikangas et al., 2017). By contrast, expression of QN-mRuby3 during quiescence led to several foci per cell independent of cell cycle progression (Fig. 4, G and H). Total mitochondrial biomass, as measured by the mean fluorescent intensity of the translocase Tom70 (Hughes et al., 2016), showed no difference between quiescent states (Fig. 4, I and J). Rad52 nuclear levels and foci, which increase upon DNA damage (Alvaro et al., 2007), were not significantly different between quiescent states (Fig. 4, K and L).
These results showed that although quiescent states shared a core of stress-associated processes (Fig. S2 K), high-Cdk1 quiescence entry was characterized by increased nuclear accumulation of Sfp1, Gln3, and Xbp1.
High-Cdk1 quiescence entry is determined by the stress status of the cell
The increased nuclear accumulation of Xbp1, Gln3, and Sfp1 in high-Cdk1 Q-cells suggested that the stress status of the cell, rather than differences in cell cycle progression, promoted high-Cdk1 quiescence entry upon starvation. To test this hypothesis, 6C1 cells were cell cycle synchronized with a nocodazole arrest-release protocol before exposure to starvation (Fig. 5 A) and then clustered using TPML to identify cells with the same cell cycle synchrony around starvation onset (Fig. S3 A). This revealed that cells with identical synchrony, as judged by the Cdc10 signal overlap, could diverge into low- or high-Cdk1 quiescent states (Fig. 5 B and Fig. S3 B). A G1 synchronization protocol confirmed this result (Fig. S3 C). As in unsynchronized cultures, high-Cdk1 Q-cells accumulated higher nuclear levels of Xbp1 (Fig. 5 C), Gln3 (Fig. 5 D), and Sfp1 (Fig. 5 E), but not Rtg1 or Stb3 (Fig. S3 B). This showed that high-Cdk1 quiescence was not defined by cell cycle stage at starvation onset.
To assess whether high-Cdk1 quiescence was predisposed by the stress status of the cell at starvation onset, 6C1 cells were exposed to medium lacking glucose or nitrogen for 2 h before starvation (Fig. 5 F). Glucose depletion generated a multidimensional stress response characterized by down-regulation of nuclear Sfp1 and up-regulation of Stb3 and Rtg1 (Fig. 5, G and H). Nitrogen depletion produced a different multidimensional stress response with nuclear translocations for Gln3, Sfp1, and Stb3 (Fig. 5, I and J). Triggering acute starvation from these two different stress states led to a significant increase of high-Cdk1 Q-cells (P < 0.05; Fig. 5 K) with high final levels of nuclear Xbp1, Gln3, and Sfp1 (Fig. 5, L and M; and Fig. S4, D and E).
Together, these experiments showed that high-Cdk1 quiescence entry was determined by the stress status of the cell and not by cell cycle stage at starvation onset.
Nuclear levels of stress transcription factors predict high-Cdk1 quiescence
To assess the contribution of stress responses to the establishment of high-Cdk1 quiescence, we used linear discriminant classifiers (LDCs) to find the best combinations of stress markers that discriminated between low- and high-Cdk1 quiescent states. To quantify the discrimination between quiescent states, the single-cell Mahalanobis distance to the mean of the low-Cdk1 Q-cells cluster was used. If a stress marker discriminates well between clusters, this metric should be small for low-Cdk1 Q-cells and large for high-Cdk1 Q-cells (Fig. 6 A). Time series obtained during quiescence entry in 6C1 (Fig. 3) and 6C2 (Fig. 4) cells were used as input for LDCs, and the average separation between clusters using single or combined parameters at each time point was calculated. LDCs based on Xbp1 and Gln3 could significantly discriminate between low- and high-Cdk1 quiescent states after 10 h into starvation (Fig. 6 B), whereas all other stress markers or their combinations did not (Fig. 6 C and Fig. S3 F).
To evaluate the predictive power of Xbp1 and Gln3, support vector machine classifiers (SVMs) were used to calculate probabilities of high-Cdk1 quiescence using data from assays in which the 6C1 strain was exposed to rich medium (Fig. 3), glucose depletion (Fig. S3 D), or nitrogen depletion (Fig. S3 E) before starvation. SVMs were trained using the final nuclear values of Xbp1 or Gln3, and single-cell probabilities of high-Cdk1 quiescence upon starvation were computed for all time points (Fig. 6 D and Fig. S3 G). Factors were considered predictive when correctly assigning probabilities of high-Cdk1 quiescence over the a priori probability threshold given by the fraction of high-Cdk1 Q-cells in the experiment.
Nuclear Gln3 produced predictions for high-Cdk1 quiescence over the a priori threshold only when cells were transferred from rich medium into starvation (Fig. 6 E) but not when cells were pretreated with glucose (Fig. 6 F) or nitrogen depletion (Fig. 6 G) before starvation. In contrast, nuclear Xbp1 produced predictions for high-Cdk1 quiescence over the a priori threshold regardless of quiescence entry conditions (Fig. 6, H and I) and as early as 1 h into starvation for cells pretreated with nitrogen depletion (Fig. 6 J). The average predictive power of Xbp1 across experiments, however, was significant after 8–9 h into starvation, suggesting that Xbp1 was correlated to high-Cdk1 quiescence when accumulated to high levels.
These results indicated that although different stress responses could contribute to high-Cdk1 quiescence depending on starvation conditions, increased nuclear Xbp1 was a general and potentially causative factor of high-Cdk1 quiescence.
Nuclear accumulation of the transcriptional repressor Xbp1 promotes high-Cdk1 quiescence entry during starvation
To evaluate whether increased nuclear Xbp1 levels were causative of high-Cdk1 quiescence while simultaneously measuring the stress status of the cell, we tracked quiescence entry in a Cdc10-mCyOFP1, Gln3-mKOκ, Rtg1-mNeptune2.5, Sfp1–mScarlet-I strain endogenously expressing Xbp1-mNeonGreen and also expressing Xbp1-mTFP1 from a CUP1 promoter that is mildly induced by starvation (Fig. 7 A; Peng et al., 2015). As an internal control for copper induction experiments, cultures were spiked with isogenic cells lacking CUP1p-XBP1-mTFP1, and both populations were algorithmically sorted using the mTFP1 signal before TPML analysis (Fig. S4 A).
CUP1p-expression of Xbp1-mTFP1 during starvation (Video 4) significantly increased the fraction of high-Cdk1 Q-cells (P < 0.05; Fig. 7 B) without altering stress responses (Fig. S4 B). Interestingly, CUP1p-expressed Xbp1-mTFP1 reduced the levels of endogenously expressed Xbp1-mNeonGreen (Fig. 7, C and D), suggesting autoinhibition.
To test whether the effect of CUP1p-XBP1-mTFP1 required its transcriptional repressor activity, we generated a DNA-binding domain mutant of XBP1 by introducing the mutations R349E and Q356E, which prevent DNA binding in related transcription factors (Mai and Breeden, 1997; Liu et al., 2015; Nair et al., 2003). CUP1p-expression of xbp1-EE-mTFP1 during starvation had no impact on the fraction of high-Cdk1 Q-cells (Fig. 7 B), the endogenous Xbp1-mNeonGreen levels (Fig. 7, C and D), or stress responses (Fig. S4 C).
To determine the specificity of Xbp1’s effect in promoting high-Cdk1 quiescence, we used the CUP1 promoter to express Stb3, which is a transcriptional repressor functionally related to Xbp1 (Liko et al., 2010), and Whi5, a major transcriptional repressor of cell cycle (Schmoller et al., 2015). CUP1p-expression of Stb3-mTFP1 or Whi5-mTFP1 did not alter the fraction of high-Cdk1 Q-cells (Fig. 7 B) or endogenous Xbp1 levels (Fig. 7 D) and had negligible effects on stress responses (Fig. S4, D and E).
The nuclear accumulation of CUP1p-expressed Whi5-mTFP1 strictly depended on cell cycle progression, as indicated by the opposing patterns of Cdc10 (Fig. 7 E) and Whi5 (Fig. 7 F) in Cdc10 clusters, remaining excluded from the nucleus of high-Cdk1 Q-cells (Fig. 7 G and Video 5). In stark contrast, nuclear accumulation of CUP1p-expressed Xbp1-mTFP1 was entirely independent of Cdk1 activity state (Fig. 7 H) or cell cycle progression (Fig. S4 B).
TPML analysis of xbp1Δ cells bearing markers for stress (Msn2-mNeonGreen, Gln3-mKOκ, Rtg1-mNeptune2.5, Sfp1–mScarlet-I, Stb2-mTFP1) and cell cycle progression (Cdc10-mCyOFP1) showed that deletion of XBP1 led to altered patterns of cell cycle arrest during starvation with 16 ± 4% of cells entering an ectopic, less viable cell cycle under stress (Fig. S4, G and H; and Video 6). Comparison of cells that arrested for >6 h under starvation in WT and xbp1Δ showed that 40 ± 10% of WT cells remained arrested, leading to high-Cdk1 quiescence, whereas most xbp1Δ cells bypassed entry into high-Cdk1 quiescence by completing a last division, with only 7 ± 5% remaining arrested. XBP1 deletion reduced the proportion (Fig. 7 I) and viability (Fig. 7 J) of high-Cdk1 Q-cells, whereas the onset of stress responses (Fig. S4 F) or the viability of cells with stable cell cycle arrest (Fig. S4 I) remained unaffected.
We concluded that Xbp1 was a causative factor of high-Cdk1 quiescence entry, which was sufficient to increase the frequency of high-Cdk1 Q-cells when accumulated at high levels.
Nuclear Xbp1 acts as a time-delayed integrator to record single-cell history of stress stimuli
The capacity of Xbp1 to promote quiescent states suggested that it must be rapidly down-regulated upon return to proliferation. To test this hypothesis, we analyzed the nuclear levels of stress transcription factors in 6C1 cells that were returned to rich medium after 20 h of starvation. Surprisingly, we found that whereas stress factors such as Sfp1 or Stb3 readily translocated upon exposure to rich medium, the nuclear levels of Xbp1 remained invariable for 2 h ± 20 min after exposure to rich medium (Fig. 8 A and Fig. S5 A) and after that decreased by dilution through cell division (Fig. 8 B and Fig. S5 B). This indicated that Xbp1 was slowly down-regulated after return to rich medium, suggesting that information from past quiescent states persisted in the nuclear levels of Xbp1.
To test whether information from past stress stimuli (Morimoto and Koshland, 1991) was recorded in the nuclear levels of Xbp1, we exposed 6C1 cells to hourly changes between rich medium (SCD) and either glucose-depleted medium or nitrogen-depleted medium (Fig. 8 C). Under these conditions, stress responses oscillated (Fig. 8 D; Fig. S5, C and D; and Video 7), but only the nuclear levels of Xbp1 increased in proportion to the number of previous stress episodes (Fig. 8, E and F; and Fig. S5, C and D).
To test whether nuclear Xbp1 levels retained information from past starvation episodes, we tracked Xbp1 in cells exposed to starvation for 3, 6, 9, or 12 h before return to rich medium (Fig. 8 G). Starvation periods of increasing duration produced populations that maintained significantly different Xbp1 levels after transfer to rich medium, with the 3-h and 12-h starvation-exposed populations maintaining different Xbp1 levels beyond 4 h after return to rich medium (P < 0.05; Fig. 8 H; cells aligned to the time of return to rich medium). This behavior was analogous to a time-delayed integrator in control theory (the output signal—nuclear Xbp1 levels—corresponds to the time integral, or duration in this case, of the input signal—starvation period; Hao et al., 2013). To test whether other histone deacetylase regulators could act as temporal integrators of starvation signals, we tracked nuclear Whi5 and Stb3 levels in cells exposed to different starvation periods before return to rich medium; however, Whi5 (Fig. 8 I) and Stb3 (Fig. S5 E) nuclear levels did not reflect starvation duration.
A time-delayed integrator role for Xbp1 predicted that previous quiescence entries could increase the frequency of high-Cdk1 Q-cells in future quiescence entries by producing populations with higher Xbp1 levels. To test this hypothesis, we imaged 6C1 cells during two quiescence entries separated by a period in rich medium (Video 8). Although Stb3 accumulated to higher levels during the second quiescence entry (Fig. 8 J), only Xbp1 persisted in rich medium after the first quiescence exit (Fig. 8 K) and accumulated to maximum levels 2 h ± 29 min faster during the second quiescence entry (Fig. 8 L). TPML Cdc10 clustering according to the first or second quiescence entry (Fig. S5 F) showed that high-Cdk1 Q-cells were indeed enriched during the second quiescence entry in WT (Fig. 8 M) and stb3Δ cells, but not in xbp1Δ cells (Fig. 8 N).
These results showed that Xbp1 acted as an integrator of previous stress stimuli, which rendered the propensity of high-Cdk1 quiescence dependent on single-cell history.
Discussion
In this study, we showed how proliferating S. cerevisiae populations challenged by acute starvation enter quiescent states with low- or high-Cdk1 kinase activity. Our results obtained under controlled microfluidic conditions confirm previous batch studies that described quiescence states in cell cycle stages other than G1 after depletion of nutrients in a carbon-limited batch culture (Laporte et al., 2011) or by inducing cell cycle arrest before starvation (Wei et al., 1993). The halt of proliferation upon acute starvation was accompanied by increased storage carbohydrate metabolism, mitochondrial biomass, protein aggregation, autophagy, stress resistance, and size asymmetry during cell division. This indicates that the general processes of quiescence entry observed in batch cultures can also be triggered by starvation under constant environmental conditions.
The use of machine learning classifiers allowed us to in silico isolate and characterize cells entering quiescence in a high-Cdk1 state. By turning the values of single or combined stress markers into probabilities of cell fate, this approach distinguished correlated (Sfp1 and Gln3) from causative (Xbp1) factors of high-Cdk1 quiescence. Our results show that stress responses are remarkably homogeneous during the initial stages of quiescence entry and that cells can diverge into different quiescent states after starvation stimuli are integrated by the histone deacetylase regulator Xbp1. Although Xbp1 represses crucial cell cycle genes involved in proliferation-promoting positive feedback loops, such as cyclins (CLN3, CLB2), mitotic regulators (CDC5, CDC20), and mitotic transcriptional activators (NDD1; Miles et al., 2013), it has been considered dispensable for quiescence because xbp1Δ cells still access G1/G0 quiescence in batch cultures (Mai and Breeden, 1997); however, we find that Xbp1 is essential for viable high-Cdk1 quiescence.
The role of Xbp1 as a cell cycle–independent integrator of past stress signals shows that quiescence entry in S. cerevisiae depends on single-cell history. Remarkably, this implies that each cell can have a different point of commitment to finish cell cycle under stress or restriction point (Johnson and Skotheim, 2013; Zetterberg et al., 1995), determined not only by Cdk1 activity (given by cell cycle stage at starvation onset) but also by nuclear Xbp1 levels (given by single-cell history; Fig. 8 O).
The mechanism to control Xbp1 activity is unknown but likely involves auto-inhibition when accumulated to high levels (Fig. 7 C). Interestingly, quiescence in hair follicles relies on an inhibitory network motif whereby a Cdk1 inhibitor, p21, transcriptionally represses another Cdk1 inhibitor, p15 (Lee et al., 2013). Auto-inhibition or mutual inhibition of quiescence-promoting factors might be a universal mechanism to ensure the reversibility of the quiescent state.
How can cells persist in high-Cdk1 quiescence for a long time? Due to the bistable nature of cell cycle networks, the low-Cdk1 state of G1 and the high-Cdk1 state of S-M phase are stable steady states, which can be, by definition, sustained for prolonged periods (Novak et al., 2007). We envision that under persistent starvation, high levels of Xbp1 force cells to settle in the available stable steady state by inhibiting genes required for proliferation-promoting positive feedback loops through transcriptional repression. Consistent with this, high-Cdk1 quiescence is promoted by high levels of Xbp1 and abolished by XBP1 deletion. Increased nuclear Xbp1, however, did not stop all cells from completing cell cycle under acute starvation (Fig. 7 B), indicating that other events, such as completion of DNA replication under stress or accumulation of mitotic cyclins, could also regulate high-Cdk1 quiescence entry.
The control of quiescence entry by cell cycle–independent integrators of stress stimuli could provide an adaptive advantage in environments with strong fluctuations of nutrients, as is the case for parasites with complex life cycles and cancer cells attempting metastasis (Wang et al., 2015). In mammalian cells, transcriptional repressors that could play an analogous role to Xbp1 are HES1 (Sueda et al., 2019; Coller, 2011), the Dimerization partner, RB-like, E2F and Multi-vulval class B complex (Bainor et al., 2018; Miles and Breeden, 2017; Schade et al., 2019), and developmentally regulated histone deacetylase regulators (Wang et al., 2019); however, it is unknown whether they integrate stimuli during quiescence entry.
In this study, we have characterized, using microfluidics and machine learning algorithms, the emergence of low- and high-Cdk1 quiescent states in proliferating S. cerevisiae cells challenged by acute starvation. Our results outline the temporal integration of stress stimuli through cell cycle–independent transcriptional repressors as a mechanism to establish cellular quiescence outside of G1/G0.
Materials and methods
Media
Rich medium was SCD (1% succinic acid, 0.6% sodium hydroxide, 0.5% ammonium sulfate, 0.17% yeast nitrogen base without amino acids or ammonium sulfate, 0.13% amino acid dropout powder [complete], and 2% glucose). Glucose-depleted medium was synthetic complete (SC) glycerol medium (SC with 2% glycerol instead of glucose). Nitrogen-depleted medium was SCD minus ammonium sulfate and amino acid dropout powder. Starvation medium was 0.6% potassium acetate, delivered at 0.6-psi flow rate, pH 7.2, adjusted with 0.125 M Na2CO3. For copper expression of proteins, cells were maintained in SCD plus 50 µM bathocuproine disulfonate (Gross et al., 2000) until loading in the microfluidic device. CUP1p-expression was triggered with SCD containing 25 µM CuSO4, added as a 1:1,000 dilution from a 25-mM stock solution (Labbé and Thiele, 1999). MET3p-expression was triggered with SCD lacking methionine. 200 mM CaCl2-containing SCD had NH4SO4 replaced by NHCl3.
Plasmid construction
All plasmids were confirmed by sequencing and restriction analysis. Table S1 lists the used primers and Table S2 lists the used plasmids. The MET3p-mNeonGreen-Ase1(R632-I885) plasmid resulted from a quadruple ligation of the following fragments: (1) a SacI-PacI–flanked MET3 promoter sequence (493–1 to −1); (2) a PacI-AscI–flanked mNeonGreen (yeast optimized) fluorophore released from pYLB10 (Argüello-Miranda et al., 2018); (3) an AscI-NotI–flanked C-terminus of ASE1 (R632-I885); and (4) a pRS305 backbone cut with NotI and SacI. The MET3 promoter and ASE1 fragment were amplified from WT W303 genomic DNA. Linearization with AgeI allowed integration at LEU2. pRS304 carrying ATG8p-mNeonGreen-ATG8-ATG8ter was created by replacing the PacI-BamHI–flanked yeGFP sequence in pRS304-ATG8p-2xyeGFP-ATG8-ATG8ter (a gift from the Henne Lab, University of Texas Southwestern Medical Center) by a PacI-BamHI–flanked mNeonGreen fluorophore amplified from pYLB10 (Argüello-Miranda et al., 2018). The pRS306 plasmid carrying CUP1p–QN-mRUBY3 was created by triple ligation of the following fragments: (1) a SalI-HindIII–flanked CUP1-1 promoter sequence (434–16 bp upstream of start codon); (2) a HindIII-PacI–flanked QN-rich sequence of GLN3 (aa 166–242); and (3) a PacI-SalI pLondon266 backbone containing the mRuby3 fluorophore. CUP1p and GLN3 (aa 166–242) fragments were amplified from W303 genomic DNA. Linearization with StuI allowed integration at URA3. CUP1 expression plasmids were created by replacing the PacI-AscI–flanked mCherry fluorophore in pLondon266 (Argüello-Miranda et al., 2018) with a PacI-AscI–flanked mTFP1 fluorophore. The resulting pLondon266-mTFP1 was cut with SalI-PacI and triple-ligated to a SalI-HindIII–flanked CUP1 promoter and the HindIII-PacI–flanked open reading frame of XBP1, STB3, or WHI5, which were amplified from W303 genomic DNA. Linearization with StuI allowed integration at the URA3. CUP1p-xbp1-EE-mTFP1 was created by swapping the DNA-binding domain of XBP1, which is naturally flanked by ClaI-BsaBI sites in S. cerevisiae W303, by a ClaI-BsaBI fragment with arginine 349 and glutamine 356 changed to glutamic acid by PCR mutagenesis with the appropriate primers.
Strain construction
Strains were diploid S. cerevisiae W303 (leu2-3,112 his3-11,15 ura3-1 trp1-1 can1-100 ade2) unless otherwise noted (Table S3). Diploids were created by mating appropriate haploids obtained by tetrad dissection or by transformation with PCR tagging cassettes, deletion cassettes, or plasmids using the polyethylene glycol/lithium acetate protocol (Longtine et al., 1998). xbp1Δ::KANMX4 and stb3Δ::KANMX4 deletion cassette was amplified from BY4741 genomic DNA. C-terminal PCR tagging cassettes were amplified from Longtine (pFA6a) plasmids (Table S2). Tagged strains were confirmed by DNA sequencing, PCR, or functional assays. Tagged fluorescent proteins were heterozygous unless otherwise stated.
Microfluidic cell culture
A Y04C CellASIC microfluidic device (http://www.cellasic.com/) operated at 25°C, 0.6-psi flow rate, was used for culturing cells in all experiments. 50 µl of sonicated (4–6 s at 3 W) liquid SCD culture, OD600 = 0.3, were loaded with 1–2 pulses of 8 psi for 5 s. All medium changes were isobaric. To trigger quiescence, cells grew for 2 or 3 h in SCD before exposure to starvation medium for 20 h. For MET3p protein induction, cells grew 1 h in SCD before exposure to SCD minus methionine for 2 h. For CUP1p induction, cells were maintained in 50 µM of bathocuproine disulfonate–containing SCD before exposure to 25 µM CuSO4-containing SCD for 2 h. For osmoregulatory tests, cells grew 2 h in SCD before exposure to 200 mM CaCl2-containing SCD for 30 min.
Microfluidic staining of DNA
After 20 h in starvation, cells in the Y04C microfluidic device were treated at 25°C, 0.6-psi flow rate with (1) a solution of 4 M LiCl containing 50 mM Nystatin (from a stock solution of 3 mg/ml Nystatin in DMSO) for 4 h; (2) a solution of 50 mM Tris/HCl, pH 7.8, containing 0.3 mg/ml RNase A (from a 10-mg/ml RNase stock solution in 10 mM sodium-acetate, pH 5.2) for 4 h; and (3) FACS buffer (180 mM Tris/HCl, 190 mM NaCl, 70 mM MgCl2) containing 1.85 µM propidium iodide (added before use from a 0.5-mg/ml solution) for 2 h. Nuclear fluorescent intensity was measured in segmented cells as describe below.
Statistical analysis
All statistical analyses were done in MATLAB and considered each lane of the microfluidic device as a biological replicate. Statistical significance was assessed using the Kolmogorov-Smirnov test (K-S test) using a 0.05 P value. Bar plots display mean and SD. Box plots display the median as a central mark, the 25th and 75th percentiles are the bottom and top limits of the box, and the whiskers are the most extreme nonoutlier values. Outliers were excluded using the function isoutlier(). Average times series were plotted as a solid line surrounded by a shaded area corresponding to 95% confidence intervals defined by biological replicates.
Time-lapse microscopy
A motorized, temperature-controlled, ZEN software–operated Zeiss Observer Z1 microscope with Definite Focus 2.0 was used. A 40X Zeiss EC Plan-Neofluar 40X 1.3 NA oil immersion objective and an AxioCam HR Rev 3 camera were used for image acquisition. At least five fields of view were imaged per biological replicate using a 6- or 12-min acquisition rate. A set of custom dichroic mirrors and bandpass filters was used to simultaneously image six different fluorophores (Table S4; Argüello-Miranda et al., 2018). Fluorophores and exposure times were cyan-excitable orange fluorescent protein (mCyOFP1) 35 ms (Chu et al., 2016), teal-cyan fluorescent protein (mTFP1) 400 ms (Ai et al., 2006), green-yellow fluorescent protein (mNeonGreen) 100 ms (Shaner et al., 2013), mKusabira-Orange-κ fluorescent protein (mKOκ) 400 ms (Tsutsui et al., 2008), red fluorescent protein variant (mScarlet-I) 120 ms (Bindels et al., 2017), eqFP611 red fluorescent protein variant (mRuby3) 120 ms (Bajar et al., 2016), mKate far-red fluorescent protein variant (mNeptune2.5) 400 ms (Chu et al., 2014), and phase-contrast 20 ms. CUP1p-expressed fluorescent proteins were imaged using 20 ms. Light-emitting diode (LED) light sources were kept at 12.5% intensity (25% for mNeptune2.5). Correction for crosstalk between fluorescent channels was assessed using single-fluorophore–tagged strains by calculating the cross-correlation between pixels in the signal-containing channel and the empty channels; the code for crosstalk detection is available at GitHub: https://github.com/alejandrolvido/Spectral-Imaging. Crosstalk corrections were for strains OAM421 and OAM425, the corrected mKOκ image equals original mKOκ − 0.23 mCyOFP1, and for strains OAM455/458/466/497, the corrected mCyOFP1 image equals original mCyOFP1 − 0.33 mTFP1.
Image processing and quantification of cellular features
The code for image analysis, single-cell quantifications, and time series visualization is available at https://github.com/alejandrolvido/Quiescence-Entry. The code is organized in subfolders corresponding to each main figure of this work. Images were collected using ZEN pro software (Zeiss) with 2 × 2 binning, exported in noncompressed TIFF format, and converted to double format before quantification. Shot noise was reduced using a medfilt2() filter with a 3 × 3 structuring element. After segmenting cells with published algorithms (Doncic et al., 2013; Wood and Doncic, 2019), image background was calculated as the median intensity of the space not occupied by cells. A 2D Gaussian fit to the brightest pixel of nuclear proteins was used to calculate nuclear parameters (Doncic et al., 2013), where nuclear intensity is defined as the mean intensity of nuclear pixels minus the mean intensity of cytoplasmic pixels. The septin ring was detected by measuring the SD Cdc10-mCyOFP1 at cell periphery “Cdc10 Signal” (Fig. S1 B).
TPML algorithm
This algorithm aims to use a small number of single-cell time series classified by experimentalists to construct a clustering strategy to classify cells from future experiments automatically. Single-cell time series for the Cdc10-mCyOFP1 signal were normalized per biological replicate before being pulled together into matrices where each row represented a cell and each column corresponded to a time point. The order of the rows was then randomized, and the data were divided into 50% training data, 30% validation data, and 20% test data (Chicco, 2017). The silhouette and the Calinski-Harabasz methods defined the minimum number of clusters in the data, which were confirmed by assessing higher cluster numbers using the function kmeans(). Unsupervised clustering algorithms were used to classify cells in the training dataset into three clusters according to their Cdc10-mCyOFP1 signal. The algorithms differed in the clustering method (k-means, k-medoids, hierarchical clustering), the number of time points, the interval of time points, and the cluster centroid distance metric (sqeuclidean, correlation, cityblock, cosine). Clustering algorithms’ performance was evaluated using the validation data, which were manually labeled by three cell biologists tasked with classifying 300 cells from movies as “direct G1 arrest after starvation onset,” “completion of one final cell division after starvation onset,” and “persistent arrest as a budded cell after starvation onset.” Clustering algorithms were then asked to classify the human-labeled validation data using the function pdist2(), and the coherence of the cluster assignments was scored by a multiclass Matthews correlation coefficient (MCC) (Gorodkin, 2004). The best-performing classifier algorithms were optimized by adjusting clustering parameters until their clustering closely recapitulated the human cluster assignments as judged by MCC. The centroids of the best clustering solutions were used to write the function fcentroids(), which uses the function pdist2() and the saved cluster centroids to classify new time series from new experiments into Cdc10 clusters automatically. In strains carrying multiple fluorescent markers, clustering according to Cdc10 was used to arrange all fluorescent channels.
Clustering analysis of synchronized cultures
For M-phase synchronization, exponentially growing cells, OD600 = 0.3, were spun down and resuspended in 20 µg/ml nocodazole-containing SCD and cultured for 2 h (∼90% of dumbbell-shaped cells) before loading in the microfluidics device and being released by exposure to nocodazole-free SCD. For G1 synchronization, an early stationary culture OD600 = 2 was centrifuged at 500 rpm for 2 min on a discontinuous PEG 5000: SC gradient (1:3, 1:27, 1:81), and cells in the top 200 μl, which were mostly unbudded, were washed and resuspended in SCD before loading in the flow cell. As soon as 15% of the synchronized population started to bud, cells were transferred to starvation medium.
Linear discriminant classifier (LDC) analysis
Affiliation to “low-Cdk1” or “high-Cdk1” clusters was used as a categorical variable to establish LDCs that measured how well stress and cell cycle markers discriminated between the two clusters at each time point. LDC hyperparameters were optimized using the function fitcdiscr(), holding out 30% of the cross-validation data. Discrimination between clusters at a given time point was measured by the Mahalanobis distance calculated with the function mahal().
Support vector machine (SVM)–based calculation of single-cell probabilities
Single-cell probabilities of high-Cdk1 quiescence were calculated using an SVM given by the function fitcsvm() with radial basis kernel. SVMs were trained using the last time point of the starvation period, assuming that stress and cell cycle markers relevant for cluster determination displayed their maximum divergence at this stage. A priori probabilities equal to the average ratio of high-Cdk1 cells to the total number of cells were assumed. The trained SVMs were used to classify every cell at every time point according to each stress or cell cycle marker’s values. The function fitPosterior() was used to obtain the posterior probability that a cell was correctly assigned to its cluster at a given time point.
In silico cell sorting according to cellular fluorescence
In copper induction experiments, mTFP1(+) and control mTFP1(−) strains were mixed in the same microfluidic space and sorted in silico using the following algorithm (Fig. S4 A): (1) single-cell time series for mTFP1 were transformed into log10 values; (2) the mean of the log10 time series was calculated; and (3) the mean log10 values derived from each single-cell time series were clustered using kmeans(), distance metric cityblock, which separated mTFP1(−) from mTFP1(+) cells before further analysis.
Semi-automatic quantifications of single cells and intracellular features
The quantification of the time of SPB separation, time point of cell division, time of maximum nuclear accumulation, and foci detection was made on a previously published pipeline for single-cell analysis with a semi-automated user interface for scoring of single-cell data (Argüello-Miranda et al., 2018), available at https://github.com/alejandrolvido/Quiescence-Entry.
Online supplemental material
Fig. S1 shows morphological quantifications of Q-cells under starvation, the algorithm for Cdc10 signal quantification, the workflow of the TPML algorithm, cluster number analysis, and the mitotic induction of the APC/C-Cdh1 activity sensor MET3p-mNeonGreen-Ase1(R632-I885). Fig. S2 shows the assessment of signal-to-noise ratios and crosstalk between fluorescent channels in the six-color imaging setup, the osmoregulatory response in the 6C2 strain, and representative micrographs of cells expressing mNeonGreen-Atg8, QN-mRuby3, or Rad52-mNeonGreen. Fig. S3 shows the clustering analysis of cells treated with nocodazole and nutrient depletion before starvation, as well as the linear discriminant analysis for low- and high-Cdk1 Q-cells. Fig. S4 shows the algorithm to sort cells according to the presence/absence of a fluorescent signal, the extended analysis of the CUP1p-expression of Xbp1, xbp1-EE, Stb3, and Whi5 during quiescence entry, and the TPML analysis of xbp1Δ cells. Fig. S5 shows the analysis of nuclear translocation of stress transcription factors during return to rich medium, cyclic episodes of nutrient depletion, or two consecutive quiescence entries. Video 1 shows the assay to study the proliferation–quiescence transition, including a stress-resistance test by exposure to hyperosmotic medium. Video 2 shows the TPML algorithm being optimized to classify cells in three main clusters. Video 3 shows six-color imaging of the proliferation–quiescence transition in 6C1 cells with single-cell tracking of a representative high-Cdk1 Q-cell. Video 4 shows six-color imaging of quiescence entry in 6C1 cells CUP1p-expressing Xbp1-mTFP1. Video 5 shows six-color imaging of quiescence entry in 6C1 cells CUP1p-expressing Whi5-mTFP1. Video 6 shows six-color imaging of quiescence entry in xbp1Δ cells bearing cell cycle and stress markers. Video 7 shows six-color imaging of 6C1 cells during cyclic exposure to stress. Video 8 shows six-color imaging of 6C1 cells during two consecutive quiescence entries. Table S1 lists primers used. Table S2 lists plasmids used. Table S3 lists strains used. Table S4 lists the optics set up for six-color imaging.
Acknowledgments
We thank Gaudenz Danuser, Mike Henne, Sandra Schmid, Jan Skotheim, Jennifer Ewald, and Peter Michaely. We thank Andreas Doncic posthumously.
This work was supported by the Cancer Prevention and Research Institute of Texas (RP150596 and RR150058), the Welch Foundation (I-1919-20170325), and the National Institute of General Medical Sciences of the National Institutes of Health (K99GM135487).
The authors declare no competing financial interests.
Author contributions: O. Argüello-Miranda: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Software, Validation, Writing - original draft, and Writing - review and editing. J. Noh: Conceptualization, Data curation, Formal analysis, Software, Supervision, Validation, Writing - original draft, and Writing - review and editing. A.J. Marchand, T. Kennedy, and M.A.X. Russo: Investigation, Methodology, Resources, and Validation.
References
A preprint of this paper was posted in bioRxiv on March 24, 2021.
Author notes
O. Argüello-Miranda’s present address is Department of Plant and Microbial Biology, North Carolina State University, Raleigh, NC.