Basal P21 Controls Population Heterogeneity in Cycling and Quiescent Cell Cycle States

Basal p21 controls population heterogeneity in cycling and quiescent cell cycle states

K. Wesley Overtona, Sabrina L. Spencerb, William L. Noderera, Tobias Meyerb, and Clifford L. Wanga,1

Departments of aChemical Engineering and bChemical and Systems Biology, Stanford University, Stanford, CA 94305

Edited by Charles S. Peskin, New York University, Manhattan, NY, and approved August 27, 2014 (received for review May 27, 2014) Phenotypic heterogeneity within a population of genetically identical SCF/Skp2 then ubiquitinates p21, targeting it for proteasomal cells is emerging as a common theme in multiple biological systems, degradation (13). Thus, p21 both regulates and, through the ac- including human cell biology and cancer. Using live-cell imaging, flow tion of E3 ubiquitin ligase complexes that target p21, is regulated cytometry, and kinetic modeling, we showed that two states—quies- by active CDK2 bound to Cyclin E. cence and cell cycling—can coexist within an isogenic population of The p21–CDK2 control scheme is an example of a double- human cells and resulted from low basal expression levels of p21, a negative feedback loop. When stochastic gene expression leads Cyclin-dependent kinase (CDK) inhibitor (CKI). We attribute the p21- to fluctuations in factors involved in positive or double-negative dependent heterogeneity in cell cycle activity to double-negative feedback regulation, distinct cellular states within a population feedback regulation involving CDK2, p21, and E3 ubiquitin ligases. can arise (1, 14–18). Because of the role of p21 in the double- In support of this mechanism, analysis of cells at a point before cell negative feedback regulation of cell cycle activity, we hypothesized cycle entry (i.e., before the G1/S transition) revealed a p21–CDK2 axis that p21 controlled population heterogeneity in quiescent and that determines quiescent and cycling cell states. Our findings suggest cycling cell states. In mammalian systems, there are few examples a mechanistic role for p21 in generating heterogeneity in both normal where inactivation of a single gene leads to loss of heterogeneity in tissues and tumors. cell cycle activity. Here, in comparing wild type (WT), deficient, and ectopically restored p21 genetic backgrounds, we reveal a di- tumor heterogeneity | cell dormancy | synthetic uORF | rect role for low basal levels of p21 in controlling population nongenetic cell heterogeneity | positive feedback loop heterogeneity in cell cycle activity.

population of genetically identical cells can exhibit phe- Results A notypic heterogeneity (1, 2). That is, even when cells with Live-Cell Imaging Revealed p21-Dependent Population Heterogeneity. the same DNA sequence and epigenetic markings experience the To gauge cell cycle activity, we tracked the activity of CDK2 same environment, there can be cell to cell variability. Researchers in single cells by using a fluorescent reporter consisting of the have reported that cells within a clonal population can vary by size C-terminal CDK2 phosphorylation domain of DNA helicase and morphology (3), cell cycle activity (4, 5), lifespan (6), and B (DHB) fused to YFP (19). In G0 or G1 phase cells, the un- receptor sensitivity to cytokines (7, 8). Furthermore, population phosphorylated reporter is located primarily in the nucleus (Fig. heterogeneity can determine biological outcome: instead of a 1A and Fig. S1A). During the G1/S transition, S, and G2 phases, homogenous stem cell population where cells renew equally, a CDK2 phosphorylates the reporter, causing it to translocate from subpopulation can remain dormant until needed (5), a bacterial the nucleus to the cytoplasm (Movie S1). As a result, the cyto- colony can harbor a subpopulation resistant to antibiotics (3), plasmic-to-nuclear ratio of DHB-YFP can be used to monitor and a clonal tumor can harbor a subpopulation that survives CDK2 activity and cell cycle progression (12). radiation or chemotherapy (9). Although many instances of nongenetic population heterogeneity have been documented, Significance fewer underlying regulatory mechanisms governing this heterogeneity have been reported. When cells exhibit heterogeneity in cell cycle activity, one Population heterogeneity can make the treatment of tumors subpopulation actively cycles, whereas another subpopulation more challenging. Whereas a therapeutic agent may be effec- tive against one fraction of a population, it may be less effec- remains quiescent. The main activators of the cell cycle are tive against another fraction. Although heterogeneity can be Cyclin-dependent kinases (CDKs) bound to Cyclin proteins, and genetic and attributed to mutations, there can also be non- a major class of cell cycle inhibitors is CDK inhibitors (CKIs). genetic heterogeneity, where a clonal population can harbor One CKI, p21, was first discovered in the mid-1990s as an in- distinct subpopulations. Here, we identified a single gene, p21, hibitor of CDKs in G1 phase (10) and a factor that was tran- that was responsible for population heterogeneity in cell cycle scriptionally activated by p53 (11). Since its discovery, much of activity and explain that this heterogeneity can arise from the research involving p21 has focused on its role in arresting the regulatory relationships of p21 with Cyclin-dependent kinase 2 cell cycle in response to activation by p53 after DNA damage. (CDK2) and E3 ubiquitin ligases. We suggest that, instead of However, cells without exogenous DNA damage still express using CDK inhibitors (CKIs) in cancer therapy, CKIs themselves p21. For example, single-cell analysis of p21 and CDK2 activity should be targeted. Given concurrently with chemotherapy led Spencer et al. (12) to propose that p21 is involved in the agents, CKI inhibitors would reduce tumor heterogeneity and cellular decision to enter the cell cycle under normal conditions. thus increase chemotherapy efficacy. Because p21 is a potent inhibitor of Cyclin-CDK complexes,

the cell must actively regulate p21 levels under normal conditions Author contributions: K.W.O. and C.L.W. designed research; K.W.O. performed research; to promote the transition from G1 to S phase. At this stage of the S.L.S., W.L.N., and T.M. contributed new reagents/analytic tools; K.W.O. and C.L.W. ana- cell cycle, CDK2 activity begins to increase and causes inactivation lyzed data; and K.W.O. and C.L.W. wrote the paper. of the anaphase promoting complex (APC). The inactivation of The authors declare no conflict of interest. the APC allows Skp2 levels to increase, because Skp2 is a target of This article is a PNAS Direct Submission. the APC-Cdh1 complex. The E3 ubiquitin ligase complex of Skp1/ 1To whom correspondence should be addressed. Email: [email protected]. Cullin/F box (SCF) and Skp2 recognizes p21 that has bound This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. Cyclin E-CDK2 complexes and been phosphorylated by CDK2. 1073/pnas.1409797111/-/DCSupplemental.

E4386–E4393 | PNAS | Published online September 29, 2014 www.pnas.org/cgi/doi/10.1073/pnas.1409797111 Downloaded by guest on October 2, 2021 A PNAS PLUS G1 S G2 M G1 Time after Mitosis (hrs) 1.4 3.6 8.8 19 20.6 21 22.4 G1 G2 G1 hhrsrs hrsrsr hrshrss hrshhrrsrs hrs hrshrh s hrsrsr 2 S M H2B- mTurq 1 CDK2 Activity YFP DHB-

(DHB-YFP cyt/nuc) 0 0 12 24 DHB-YFP 0.62 0.78 1.20 1.32 0.64* 0.64* 0.67 Time after (cyt/nuc) Mitosis (hrs)

B C p21-/- 0.2 ng/mL EGF 5 ng/mL EGF 20 ng/mL EGF GFP-p21-ER 0.05% Serum 1.25% Serum 5% Serum 2 2 2

WT 1 1 1

0 0 0 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 /- YFP cyt/nuc)

2 2 2 - - CDK2 Activity CDK2 Activity ACC/ACC/ACC ACC/ACC ACC/ACC/ACC/ACC CGA GGG/ACC UAA WT UCU/ACC WT+IR p21 p21-/- 1 1 1 GCCACC GFP- (DHB p21-ER 0 0 0 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 Time (hrs) p21 Beta- actin D p21-/- RFP-p21-ER ACC ACC ACC FP-p21-ER 0.2 ng/mL EGF 5 ng/mL EGF 20 ng/mL EGF

ytivitc 0.05% Serum 1.25% Serum 5% Serum

) 2 2 2 -YFP nuc A 1 1 1 2KDC cyt/ (DHB 0 0 0 SYSTEMS BIOLOGY

3 10 103 103 -ER

1 1 1 p21 10 10 10

-1 -1 -1

RFP- 10 10 0 10 20 30 40 10 0 10 20 30 40 0 10 20 30 40 Time (hrs)

Fig. 1. p21 causes population heterogeneity in cell cycle activity. (A) Single-cell tracking of cell cycle progression using a DHB-YFP reporter of CDK2 activity. (Left) CDK2-dependent translocation of DHB-YFP from the nucleus to the cytoplasm occurs during G1/S, S, and G2 phases. DHB-YFP returns to the nucleus after mitosis (M) and remains during early G1. H2B-mTurquoise (H2B-mTurq) is the nuclear marker. *During mitosis, after nuclear membrane collapse, the cytoplasmic-to-nuclear ratio (cyt/nuc) is not well-defined. (Right) CDK2 activity (cyt/nuc DHB-YFP) vs. time for cell in Left; arrows indicate depicted time points. − − (B) Single-cell traces of CDK2 activity in MCF10A WT and p21-deficient ( / ) cycling (blue) and quiescent (red) cells. Traces for cycling cells were aligned to the second mitosis. B and D show results for populations grown under varying growth factor concentrations (EGF and serum). (C) Immunoblot of p21. The right three immunoblot lanes show WT and p21−/− cells without ionizing irradiation (IR) and WT with 10 Gy IR. The left eight immunoblot lanes show expression of GFP-p21-ER in p21−/− cells. Expression levels were tuned by varying translation initiation site bases preceding the GFP-p21-ER gene and/or synthetic uORFs. Slashes separate specified bases preceding each of one to three uORFs and bases preceding GFP-p21-ER. Bases without a slash precede GFP-p21-ER without any uORF. Expression of RFP-p21-ER in D was tuned with two uORFs (ACC/ACC/ACC and schematic) to approximate basal expression. (D) Traces for CDK2 − − activity and RFP-p21-ER expressed in p21 / cells.

Using live-cell imaging to monitor the reporter in MCF10A serum), all cells were quiescent (Fig. 1B, Upper Left) (when de- cells, we could study growth factor-dependent (here synonymous scribing quiescence, here we do not distinguish G0 phase cells with “mitogen-dependent”) heterogeneity in cell cycle activity. For from those in prolonged G1). However, at an intermediate level typical experiments, researchers often desire maximal pro- of stimulation [5 ng/mL EGF, 1.25% (vol/vol) serum], we ob- liferation and therefore provide 20 ng/mL EGF and 5% (vol/vol) served heterogeneity in cell cycle activity with 43% of cells in serum. However, this culture-optimized level of mitogenic a cycling state and 57% of cells in a quiescent state (Fig. 1B, stimulation is likely supraphysiological, and in comparison, hu- Upper Center and Movie S2). man serum has been reported to contain 1–4 ng/mL EGF (20). In contrast, when we tracked the cell cycle activity of MCF10A In our experiments, when we supplemented cells with 20 ng/mL cells deficient in p21, this heterogeneity was largely abrogated EGF and 5% (vol/vol) serum, almost all cells were actively cy- (Fig. 1B, Lower Center and Movie S3); 92% of cells were cycling. cling (Fig. 1B, Upper Right and Movie S1). In contrast, as p21 is commonly known as a cell cycle inhibitor that is induced by expected, at a low level of stimulation (0.2 ng/mL EGF, 0.05% the transcription factor p53 in response to DNA damage (Fig. 1C,

Overton et al. PNAS | Published online September 29, 2014 | E4387 Downloaded by guest on October 2, 2021 WT+IR). However, even under normal, undamaged conditions, A WT cells express a background or basal level of p21 (Fig. 1C, WT). We hypothesized that, in undamaged WT cells, the basal level of p21 was responsible for heterogeneity in cell cycle activity. This notion would be supported if expression of p21 independent of p53 restored population heterogeneity in p21-deficient cells. To test this hypothesis, we expressed p21 in cells with a p21-deficient genetic background using vectors where expression did not require p53 (i.e., constitutive expression). The ectopic p21 was expressed as a fusion with red fluorescent protein (RFP) to enable live single- cell fluorescence detection. The p21 was also fused to an estrogen receptor (ER) domain (RFP-p21-ER), and therefore, nuclear lo- calization could be induced by the addition of 4-hydroxytamoxifen (4-OHT). Because the activity of p21 depends on its interactions p21-/- with factors in the nucleus, we were thus able to induce the activity of p21 by addition of 4-OHT at the beginning of experiments. To achieve physiologically relevant levels of p21 expression, we used synthetic upstream open reading frames (uORFs) and different translation initiation site sequences (21, 22). We generated vectors that expressed the p21 fusion proteins over a broad range of levels (Fig. 1C and Fig. S1B). To approximate the relatively low basal level in undamaged WT cells, we chose a vector using two uORFs (Fig. 1C and Fig. S1 B–D, vector ACC/ACC/ACC). After ectopically expressing this low level of p21 in p21-deficient cells, we observed that, in the quiescent populations, p21 was sustained at high Factors Increasing Growth levels (Fig. 1D, Lower, red). In cycling populations, p21 was expressed at lower levels and exhibited oscillatory dynamics out of p21-/- phase with CDK2 activity (Fig. 1D, Lower,blueandFig. S1E). The GFP-p21-ER single-cell dynamics, which indicated that p21 decreases at the start of S phase and increases at G2, were in line with previous studies that used synchronized populations to show the cell cycle-dependent regulation of p21 (23–25). Finally, at intermediate growth factor stimulation [5 ng/mL EGF, 1.25% (vol/vol) serum], ectopic p21 restored the heterogeneity in cell cycle activity exhibited by WT cells (Fig. 1D, Upper Center and Movie S4). Under these conditions, 52% of cells were actively cycling, whereas 48% remained quiescent.

Cellular Incorporation of BrdU Revealed a Range of Growth Factor Quiescent Cycling Stimulation That Supported p21-Dependent Cell Cycle Heterogeneity. Having established p21-dependent heterogeneity at an inter- Relative BrdU-AF488 mediate growth factor concentration using live-cell imaging, we Fluorescence next investigated the growth factor concentration range where B 100 quiescent and cycling states could coexist. We administered a WT Vector Only 48-h pulse of BrdU to cells incubated in 12 different growth p21-/- Vector Only 75 factor conditions and used flow cytometry to evaluate the frac- p21-/- GFP-p21-ER tion of each population that remained quiescent vs. actively cycling. Cells that did not incorporate any BrdU over a 48-h period 50 were considered quiescent (Fig. S2A). At low growth factor stimulation (between 0 ng/mL EGF, 0% serum and 3 ng/mL EGF, 0.75% serum), both WT and 25 p21-deficient cells were largely quiescent (Fig. 2A, Top and Middle),

and at high growth factor stimulation [between 8 ng/mL EGF, 2% quiescent population of % (vol/vol) serum and 20 ng/mL EGF, 5% (vol/vol) serum], both WT 0 012345 and p21-deficient cells were mostly cycling (Fig. 2A, Top and Mid- % Serum dle). However, over an intermediate range of stimulation [between 4 ng/mL EGF, 1% serum and 7 ng/mL EGF, 1.75% (vol/vol) se- 0 5 10 15 20 rum], the WT population distributions were bimodal in BrdU in- EGF (ng/mL) corporation, indicating that both quiescent and cycling states coexisted (Fig. 2A, Top). When the percentage of each population Fig. 2. p21 enables coexistence of cycling and quiescent cells over a range of that remained quiescent was plotted against growth factor concen- growth factor stimulation. (A) Histograms for cell populations (y axes show tration, it could be seen that the response manifested as a graded numbers of cells); the cycling subpopulation is distinguished from the quiescent subpopulation by BrdU incorporation. (B) Percentage of quiescent cells from A. change in the heterogeneity of the population (Fig. 2B). In contrast, The shaded region indicates conditions at which 25–75% of WT cells remained p21-deficient populations exhibited significantly less heterogeneity quiescent. Data are represented as means ± SDs of triplicate samples. at all growth factor levels. They responded to changes in growth factor stimulation with more switch-like behavior (Fig. 2 A, Middle and B) (represented by the steeper drop in the quiescent subpop- EGF, 1% serum), the populations switched from being pre- ulation)—when stimulation surpassed a threshold level (4 ng/mL dominantly quiescent to predominantly cycling.

E4388 | www.pnas.org/cgi/doi/10.1073/pnas.1409797111 Overton et al. Downloaded by guest on October 2, 2021 In line with results from the live-cell imaging experiments, in However, a heterogeneous population consisting of a fraction of PNAS PLUS BrdU incorporation experiments, ectopic expression of p21 fused cells that is cycling and a fraction that is quiescent would exhibit to GFP and the ER (GFP-p21-ER) also restored heterogeneity an even greater percentage of cells with stabilized p21 (i.e., the to p21-deficient populations (Fig. 2 A, Bottom and B). At the GFP-high subpopulation). The existence of a quiescent G1-arres- same intermediate growth factor stimulation that supported ted population would further increase the percentage of GFP- heterogeneity in WT cells [between 4 ng/mL EGF, 1% serum and high cells in the total population, because p21 would be stabi- 7 ng/mL EGF, 1.75% (vol/vol) serum], populations expressing lized in all of the quiescent cells as well as a fraction of the ectopic p21 exhibited bimodality in BrdU incorporation. Ectopic cycling cells. Thus, at an intermediate stimulation condition that p21 expression caused the sharp, switch-like response to growth supported heterogeneity in cell cycle activity (Fig. 1), the in- factors observed in p21-deficient cells to become more graded, creased fraction of cells expressing high levels of GFP-p21-ER closely resembling the response of the WT population (Fig. 2B). (4 ng/mL EGF, 1% serum; 41% GFP-high) over that of the This result further supported the notion that p21 bestows pop- maximal stimulation condition [20 ng/mL EGF, 5% (vol/vol) ulation heterogeneity in cycling and quiescent states. serum; 33% GFP-high] can be attributed to the quiescent subpopulation (Fig. S2B). We consider this increased fraction of Population Distributions in p21 Expression and Cell Cycle Phase Reflected GFP-high cells in the heterogeneous population to be telling and Reversibility and Growth Factor-Dependent Heterogeneity in Cell States. that, in general, flow cytometry analysis of p21 and cell cycle We next sought to determine whether the growth factor-dependent population distributions supported the results from live-cell cycling and quiescent states were reversible. Reversibility would tracking and BrdU incorporation experiments. As a control, we − − suggest that the heterogeneity in cell cycle activity was not caused by also expressed GFP-ER in p21 / cells. Under all conditions mutations or heritable epigenetic markings. We used flow cytometry tested, GFP-ER expression was unimodal (Fig. S2 D and E). Thus, to analyze population distributions in GFP-p21-ER and GFP-ER the growth factor-dependent regulation of GFP-p21-ER was likely expression and cell cycle phase (Fig. S2 B–E). When flow cytom- dependent on the p21 domain. etry analysis could not detect GFP-p21-ER expression in cells, we identified those populations as GFP-low. In the live-cell imaging Kinetic Model of Double-Negative Feedback Regulation Simulated and BrdU incorporation experiments, we expressed GFP-p21-ER Bistability in Cell Cycle States. We postulated that the bistability using the ACC/ACC/ACC uORF-containing RNA leader sequence in reversible cell states arose from double-negative feedback (Fig. 1C and Fig. S1B) to reproduce low basal expression levels. regulation involving CDK2 and p21 (Fig. 3 A and B). We created Although we could distinguish GFP-high and -low subpopulations, a kinetic model to show the potential bistability in levels of p21 we could not confidently quantify their fractions in the total pop- and CDK2 activity at a single point in the cell cycle—namely, ulation because the GFP-p21-ER expression level was low. Thus, a period in G1 before the G1/S transition, when it is typically for analysis of GFP-p21-ER expression by flow cytometry, we used believed that cells have not committed to the next cell cycle (i.e., a leader that generated a moderate expression level (Fig. 1C,UCU/ the restriction point). ACC and Fig. S1 B–D). We obtained similar results in experiments The level of active CDK2 complexed with Cyclin E or Cyclin A using the ACC/ACC/ACC and the UCU/ACC RNA leaders, but (represented by CDK2a inourequations)isaffectedby(i)growth here, we have chosen to plot the results for the UCU/ACC leader. factor stimulation, (ii) inhibition by p21, and (iii) positive feedback To evaluate reversibility, we changed growth factor stimula- [a positive feedback loop was previously modeled by Yao et al. (17) SYSTEMS BIOLOGY tion from maximal levels [20 ng/mL EGF, 5% (vol/vol) serum] to and involved pRb, E2F, and Cyclin E]. The concentration of p21 is submaximal levels (0 ng/mL EGF, 0% serum and 4 ng/mL EGF, affected by its interaction with the Cyclin E-CDK2 complex through 1% serum) before returning to maximal stimulation. In line with the action of SCF/Skp2 and other E3 ubiquitin ligase complexes our previous results, decreasing the growth factor stimulation (Fig. 3A and SI Text). The parameters used in the model are de- levels shifted population distributions toward subpopulations in fined in Table S1. G1 with high p21 (Fig. S2B). When maximal stimulation was We generated steady-state balance plots by solving for CDK2 restored, the population distributions returned to ones dominated activity over a range of p21 concentrations (Fig. 3B, purple lines) by cycling cells expressing low p21 (Fig. S2C), indicating re- and then solving for p21 concentration over a range of CDK2 versibility in cycling and quiescent states. activity values (Fig. 3B, green lines) at three different growth factor We also sought to determine whether GFP-p21-ER expres- concentrations. Each intersection of the CDK2 and p21 lines rep- sion levels were associated with distinct cell cycle-phase dis- resented a steady state of the system. The system was monostable at tributions. By staining DNA with propidium iodide (PI), we both high and low growth factor concentrations (Fig. 3B, Left and analyzed the cell cycle phase of each GFP-high and -low sub- Right). At intermediate growth factor concentration, three inter- population. At all stimulation levels, the cells that highly ex- section points existed (Fig. 3B, Center), indicating that the system pressed GFP-p21-ER were predominantly in G1 (2 N DNA was bistable. The points corresponding to high p21/low CDK2 ac- content, and here, we do not distinguish G1 from G0) and, to tivity (Fig. 3B, Center, red circle) and low p21/high CDK2 activity a lesser extent, G2/M (4 N DNA content) (Fig. S2 B and C). No (Fig. 3B, Center, blue circle) represented quiescent and cycling cells expressing high levels of GFP-p21-ER were found in stable steady states, respectively. S phase (0% 2 N < DNA < 4 N). In contrast, with intermedi- Because our live-cell imaging experiments tracked p21, CDK2 ate and maximal growth factor stimulation, cells that were activity, and cell cycle phase over multiple cell generations, we expressing low or undetectable levels of GFP-p21-ER were could examine levels of p21 and CDK2 activity within the G1 actively cycling, which was evidenced by the sizable population window described by our model. Under all stimulation con- in S phase (∼30% 2 N < DNA < 4N). ditions, we found that high p21/low CDK2 activity was associated However, did the population distributions in GFP-p21-ER and with G1-arrested quiescent cells (Fig. 3C), whereas low p21/high cell cycle phase reflect population heterogeneity in cycling and CDK2 activity was associated with cycling cells in early G1, 3 h quiescent states? Keep in mind that, even if all or nearly all cells after mitosis and before the G1/S transition. Most importantly, in an unsynchronized population were cycling [e.g., the pop- under the intermediate stimulation condition [5 ng/mL EGF, ulation under conditions of 20 ng/mL EGF and 5% (vol/vol) 1.25% (vol/vol) serum], we found that coexisting quiescent and serum] (Fig. S2B), the cells in G1 and, to a lesser extent, G2/M cycling states were also characterized by high p21/low CDK2 would still express a high level of GFP-p21-ER. This high GFP- activity and low p21/high CDK2 activity, respectively. Therefore, p21-ER expression occurs because, regardless of whether it is together, modeling and experiments indicated that cell states cycling, p21 is stabilized at these points during the cell cycle. were reflected by each cell’s position along a p21–CDK2 axis.

Overton et al. PNAS | Published online September 29, 2014 | E4389 Downloaded by guest on October 2, 2021 A SCF/Skp2 Ub Ub Ub SCF/Skp2 Cell-Cycle Growth p21 Activity/ p21 Factors CDK2 CDK2 Cyclin E

B GF=0.25 GF=1.75 GF=4 p21

CDK2 Activity

C 0.2 ng/mL EGF 5 ng/mL EGF 20 ng/mL EGF 0.05% Serum 1.25% Serum 5% Serum 103 103 103

ER 102 102 102

101 101 101 RFP-p21-

100 100 100 0.4 0.6 0.8 1 0.4 0.6 0.8 1 0.4 0.6 0.8 1 CDK2 Activity (DHB-YFP cyt/nuc)

Fig. 3. Double-negative feedback regulation explains p21-dependent heterogeneity in cell cycle activity. (A) Schematic of (Left) interacting factors and (Right) double-negative feedback regulation. Ub, ubiquitin. (B) Steady-state balance plots generated by the model: steady-state p21 (green) and CDK2 (purple) activity levels, where intersections indicate stable quiescent (red) and cycling (blue) states at three growth factor (GF) concentrations. Open circles indicate an unstable steady state. At intermediate stimulation (GF = 1.75 relative units), two states (high p21/low CDK2 activity and low p21/high CDK2 activity) can stably coexist in a single population. (C) Live-cell imaging: RFP-p21-ER vs. CDK2 activity for cycling (blue) and quiescent (red) cells. Values for cycling cells were determined at a point in G1 3 h after mitosis and before cell cycle entry.

Additionally, although there can be numerous other factors that S3D). As with our previous experimental (Figs. 1B and 3C) and distinguish G1 proliferating cells from G1 quiescent cells, the modeling results (Fig. 3B), when p21 was present in the system convergence between modeling and experimental results suggests (kgen,p21 = 1 or 2), the system was bistable at intermediate growth that p21-dependent population heterogeneity can be explained, at factor concentrations (Fig. S3 B, Center and Right, C, Center and least in part, by double-negative feedback regulation. Right,andD, Center and Right). In the p21-deficient case (kgen,p21 = Although modeling p21 as the sole CKI was sufficient to de- 0), the system remained monostable at all growth factor concen- – scribe the coexistence of two positions along a p21 CDK2 axis, trations (Fig. S3 B, Left, C, Left,andD, Left). there are two other kinase inhibitor proteins (KIPs), p27 and Without the positive feedback on CDK2 (k4 = 0), the change p57, regulating CDK2 that needed to be taken into account to in CDK2 activity in p21-deficient cells was more graded (Fig. appropriately model the CDK2 activity in our experiments (SI S3B, Left). However, when the model included the positive Text and Fig. S3). In particular, by including the contributions feedback on CDK2 (k = 0.25), there was a sharp increase in from p27 and p57, we were able to simulate the behavior of p21- 4 CDK2 activity at a particular intermediate growth factor con- deficient cells. Like p21, KIPs are regulated by CDK2 through C Left the SCF/Skp2 E3 ubiquitin ligase, and they also inhibit CDK2 by centration in the p21-deficient case (Fig. S3 , ). This in- binding to the CDK2-Cyclin E complex (13). In our model, the crease also reflected the switch-like transition between quiescent kinetic constants governing CDK2 and KIP regulation were the and cycling states that we previously observed in p21-deficient same as those involving p21. cells at conditions between 3 ng/mL EGF and 0.75% serum and Using this expanded model, we generated stimulus–response 4 ng/mL EGF and 1% serum (the sharp decrease in the per- curves to evaluate bistability over a range of growth factor con- centage of quiescent cells in Fig. 2B). Thus, our model indicated centrations for three p21 generation rates. For each growth that the switch-like behavior exhibited by p21-deficient cells in factor concentration, we plotted the corresponding steady-state response to growth factor stimulation also depended on the CDK2 activity (Fig. S3 B and C) or CKI concentration (Fig. positive feedback regulation of CDK2.

E4390 | www.pnas.org/cgi/doi/10.1073/pnas.1409797111 Overton et al. Downloaded by guest on October 2, 2021 Cell Cycle Heterogeneity Is Dependent on WT p21 and SCF/Skp2 Activity. the CDK2 activity on the steady-state balance plot to remain PNAS PLUS Having proposed this double-negative feedback mechanism, we constant, resulting in monostability at all evaluated growth factor next sought to establish whether the negative regulation of p21 in concentrations (Fig. S4B, Right). This result further supported our experiments depended on SCF/Skp2. When we inhibited p21 the notion that the double-negative feedback mechanism plays regulation using MLN-4924 (an inhibitor of Cullin-dependent a key role in generating bistability and population heterogeneity. E3 ligases) in p21-deficient cells expressing GFP-p21-ER, flow cytometry analysis of the p21 population distribution showed Discussion that a majority of the population had stabilized p21 and become In summary, we report that low basal levels of p21 control growth quiescent, even at maximal stimulation (Fig. 4 A and B). We also factor-dependent population heterogeneity in cell cycle activity evaluated expression of GFP-p21K6R-ER, a fusion protein con- within an isogenic population of cells (Fig. 5A). Although p21 is taining a p21K6R mutant that could not be ubiquitinated by SCF/ perhaps best known as a p53-responsive gene expressed highly in Skp2 at six known lysine targets (26). Across all levels of growth response to DNA damage, our study revealed a physiological role factor stimulation, populations predominantly showed high ex- for basal expression levels. Using both live-cell imaging and BrdU pression levels of p21K6R (Fig. 4C). Despite this stabilized ex- staining, we showed that cells expressing p21 exhibited hetero- pression of p21K6R, the cells still responded to varying growth geneity in cycling and quiescent states at intermediate growth factor stimulation. This result was consistent with previous reports factor stimulation—a fraction of the population remained qui- that the mutant is not as active as WT p21 (26, 27). Because the escent, whereas the other fraction was cycling. In contrast, cells mutant was stabilized at all levels of stimulation but diminished in deficient in p21 did not exhibit this population heterogeneity. In its ability to inhibit cell cycle activity, the negative regulation of WT further support of a direct role for p21, ectopic expression at a p21 (both endogenous p21 and GFP-p21-ER) was likely dependent level approximating the basal endogenous level of p21 expression on ubiquitination by SCF/Skp2 (or other Cullin-dependent com- restored heterogeneity in p21-deficient cells. plexes) and subsequent proteasomal degradation. Although we have proposed a role for p21 in governing pop- We then used our CDK2–p21 model to perturb the double- ulation heterogeneity in response to growth factors, we point out negative feedback regulation mechanism. Stabilizing p21 (setting k2 that cells deficient in p21 still responded to changes in growth to zero) (Fig. S4A, Left) caused the system to remain in a quiescent factor stimulation. The response of these p21-deficient cells, how- state (high p21/low CDK2 activity), even at high growth factor ever, was monostable. Compared with WT cells, cells deficient in stimulation (Fig. S4A, Right). This result was consistent with our p21 behaved in a more switch-like manner—predominantly quies- experimental observation that inhibiting SCF/Skp2 and other cent at low growth factor stimulation and predominantly cycling at Cullin-dependent E3 ubiquitin ligase complexes with MLN-4924 intermediate and high growth factor stimulations. This result sug- caused populations to remain predominantly quiescent at high gests that mitogenic signaling contributes to the decision to remain growth factor stimulation (Fig. 4 A and B). Because it was not quiescent or enter the cell cycle through mechanisms independent possible to experimentally eliminate the regulation of CDK2 by p21 of p21—possibly a feedback loop comprised of pRb, E2F, Cyclin E, without also eliminating the regulation of p21 by SCF/Skp2 (26), we and CDK2 (17)—but that the presence of p21 causes a decision used our model (setting k3 to zero) to simulate a p21 mutant unable point to be bistable. This bistability could arise from stochasticity in to inhibit CDK2 but still negatively regulated by CDK2 through expression of p21. It has been shown that stochasticity in expression SCF/Skp2 (Fig. S4B, Left). This inhibition of p21 regulation caused of genes involved in regulatory mechanisms, especially genes that SYSTEMS BIOLOGY

0 µM MLN-4924 1.38 µM MLN-4924 A 20 ng/mL EGF B 20 ng/mL EGF 5% Serum 5% Serum GFP GFP G1: 43% GFP GFP G1: 59% Low: High: S: 17% Low: High: S: 4% 57% 43% G2/M: 31% 24% 76% G2/M: 17% # of Cells of #

GFP-p21-ER 2N 4N GFP-p21-ER 2N 4N PI PI GFP-Low GFP-High GFP-Low GFP-High G1: 29% G1: 73% G1: 34% G1: 67% S: 25% S: 0% S: 11% S: 1% G2/M: 34% G2/M: 27% G2/M: 12% G2/M: 16% # of Cells# of

2N 4N 2N 4N 2N 4N 2N 4N PI PI C 0 ng/mL EGF 4 ng/mL EGF 20 ng/mL EGF 0% Serum 1% Serum 5% Serum GFP GFP G1: 91% GFP GFP G1: 68% GFP GFP G1: 57% Low: High: S: 0% Low: High: S: 9% Low: High: S: 17% 12% 88% G2/M: 6% 17% 83% G2/M: 21% 12% 88% G2/M: 23% # of Cells# of

GFP-p21K6R-ER 2N 4N GFP-p21K6R-ER 2N 4N GFP-p21K6R-ER 2N 4N PI PI PI

Fig. 4. Negative regulation of p21 is dependent on SCF/Skp2 and p21 lysine residues targeted for ubiquitination. (A) GFP-p21-ER was expressed in p21−/− cells. Population distributions for GFP-p21-ER were determined by flow cytometry, and GFP-low and -high populations were quantified. Cell cycle phase distributions were determined by PI staining of DNA content (2 N–4 N). (B) Inhibition of SCF/Skp2 by the cullin inhibitor MLN-4924. (C) Expression of GFP- p21K6R-ER, where ubiquitination by SCF/Skp2 has been reduced by mutating six lysine targets.

Overton et al. PNAS | Published online September 29, 2014 | E4391 Downloaded by guest on October 2, 2021 PCR-amplified from pCru5-GFP-IRES-mCherry and substituted for GFP to create A B pCru5-RFP-p21-ER-IRES-Puro. CSII-EF-DHB-YFP and CSII-EF-H2B-mTurquoise were Quiescent constructed as previously described (12). GFP fusion constructs were used for flow cytometry experiments, because GFP detection is compatible with detection p21-/- of PI staining. RFP fusion constructs were used for live-cell imaging experiments, because RFP detection is compatible with detection of the DHB-YFP

p21 CDK2 sensor. WT Response Cell Culture. MCF10A cells were maintained in DMEM/F12 media (11330–032; Life Technologies) supplemented with horse serum (H1138; Sigma-Aldrich), (% of Cells Cycling) Cells of (% Cycling 2 mM glutamine, EGF (SRP3027; Sigma-Aldrich), 10 μg/mL insulin (I1882; Sigma-Aldrich), 500 ng/mL hydrocortisone (H0888; Sigma-Aldrich), 100 ng/mL Stimulus CDK2 Activity cholera toxin (C8052; Sigma-Aldrich), 100 U/mL penicillin, and 100 μg/mL (Growth Factors) streptomycin at 37 °C and 5% (vol/vol) CO2. Full growth medium contained 5% (vol/vol) serum and 20 ng/mL EGF. During routine cell culture and – Fig. 5. A p21 CDK2 axis governs p21-dependent population heterogeneity throughout all experiments, cell populations were cultured at subconfluent in quiescent and cycling cellular states. (A) Cells deficient in p21 respond to concentrations to minimize contact inhibition and increased activation of p27. growth factor stimulation in a more switch-like manner, whereas cells To make a stable MCF10A WT line expressing the pCru5-IRES-Puro empty expressing basal levels of p21 respond in a more graded manner. The shaded − − vector control and MCF10A p21 / lines expressing GFP-p21-ER, GFP-ER, RFP- region represents growth factor levels that support p21-dependent het- p21-ER, GFP-p21K6R-ER, and the pCru5-IRES-Puro empty vector control, erogeneity in cell cycle states. (B) Cell states mapped onto a p21–CDK2 axis retroviral particles were made by cotransfecting the retroviral expression (shaded region). At a stage before cell cycle entry, cells with high p21 and plasmids with pCL-Ampho (31) in 293T cells using calcium phosphate pre- low CDK2 activity are quiescent (red region), whereas cells with low p21 and cipitation (CalPhos Mammalian Transfection Kit; Clontech Laboratories, Inc.). high CDK2 activity are cycling (blue region). Supernatant containing viral particles was filtered and added to MCF10A cells along with 5 μg/mL polybrene (hexadimethrine bromide) for 18 h. The viral supernatant was tittered, such that each transduced cell received only are expressed at a low level, can lead to bistability in a system (14). one copy of the vector. Cells infected with pCru5-GFP-p21-ER, pCru5-GFP-ER, It is, therefore, possible that stochastic changes in the basal expression pCru5-RFP-p21-ER, and pCru5-IRES-Puro empty vector were selected with level of p21 could lead to the heterogeneity that we observed. 1 μg/mL puromycin for at least 3 d and supplemented with 1 μg/mL puro- Double-negative feedback regulation between p21 and CDK2 mycin during routine culture and 1 μg/mL puromycin and 500 nM 4-OHT can explain how both cycling and quiescent states can coexist throughout all experiments. Cultures infected with pCru5-GFP-p21K6R-ER within a single isogenic population. Although we have suggested were supplemented with 10 μg/mL blasticidin for at least 3 d to select for a role for the SCF/Skp2 E3 ubiquitin ligase in this feedback regu- stable integration, and cells were maintained with 10 μg/mL blasticidin during routine cell culture and 10 μg/mL blasticidin and 500 nM 4-OHT throughout lation, we do not discount that other mechanisms that regulate p21 − − all experiments. To make MCF10A WT and p21 / lines that express DHB- (e.g., CRL4-Cdt2, APC-Cdc20, and possibly other unknown com- YFP and H2B-mTurquoise, lentiviral particles were made by cotransfecting plexes) also contribute to the behavior that we have described. In the lentiviral expression plasmids pCSII-EF-DHB-mVenus or pCSII-EF-H2B- addition, our experimental results combined with kinetic modeling mTurquoise (12) with pRSV-Rev, pHCMV-VSVG, and pMdlg in 293T cells using revealed a p21–CDK2 axis that was predictive of cycling vs. quies- calcium phosphate precipitation. Supernatant containing viral particles was cent cellular states (Fig. 5B). We project that this heterogeneity can filtered and added to MCF10A cells along with 5 μg/mL polybrene (hex- be mediated by not only p21 but also the CKI proteins p27 and p57, adimethrine bromide) for 18 h. Cells transduced with CSII-EF-H2B-mTurquoise because both are regulated by SCF/Skp2. Because the control and CSII-EF-DHB-mVenus lentivirus were sorted using a FACSAria (Beckton scheme has been adopted not only by mammals but also yeast (13, Dickinson) cell sorter to isolate a population expressing both constructs. To determine the reversibility of quiescence and p21 stabilization at low 28), we propose double-negative feedback regulation involving cell −/− cycle activators and inhibitors to be a general strategy for main- growth factor stimulation, MCF10A p21 cells expressing GFP-p21-ER or GFP-ER with a vector using the UCU/ACC uORF and RNA leader sequence (pCru5-UCU/ taining heterogeneity in cell cycle activity. This mechanism may ACC-GFP-p21-ER) were induced with 4-OHT for 24 h in media containing 20 ng/mL explain how stem cell populations use p21 to maintain sub- EGF and 5% (vol/vol) serum. Cells were then cultured with reduced growth populations of dormant and self-renewing cells (29, 30). factor stimulation (0 ng/mL EGF, 0% serum or 4 ng/mL EGF, 1% serum) for 72 h. Finally, heterogeneity in cell cycle activity may impact the Each culture was restored to full growth stimulation [20 ng/mL EGF, 5% (vol/vol) treatment of tumors, because quiescent cells may evade therapies serum] for 72 h. For experiments examining the SCF/Skp2 dependence of p21 that target proliferating cells. Thus, therapeutics that inhibit cyclin- degradation, MCF10A p21−/− cells expressing GFP-p21-ER were induced with dependent kinases (e.g., a small molecule that targets CDK2) 4-OHT for 24 h in media containing 20 ng/mL EGF and 5% (vol/vol) serum. Cells could potentially increase heterogeneity, promote quiescent sub- were then cultured with no growth factor stimulation (0 ng/mL EGF, 0% serum) populations, and ultimately show diminished efficacy. Instead, our for 72 h. Each culture was then restored to full growth stimulation [20 ng/mL EGF, 5% (vol/vol) serum] and concomitantly supplemented with either 0 or work suggests that CKIs, such as p21, could be targeted. Such 1.38 μM MLN-4924 (A-1139; Active Biochem) cullin inhibitor for 48 h. therapeutic agents would be given concurrently with chemotherapy agents, reduce tumor cell heterogeneity, and subsequently increase Flow Cytometry. For DNA content analysis by PI staining, cells were trypsi- chemotherapy efficacy. nized, rinsed with PBS, added to ice-cold 70% (vol/vol) ethanol, and stored at −20 °C overnight. Each sample was then rinsed with PBS and incubated in Materials and Methods a buffer containing PI [PBS with 0.1% (vol/vol) Triton X-100, 0.2 mg/mL Expression Vectors. The p21 gene was amplified from human cDNA by PCR. RNase A, and 20 μg/mL PI] for 30 min at room temperature. Monomeric GFP (EGFP A207K) was PCR-amplified from pEGFP-N1 (Clontech To measure cell cycling and quiescent states by BrdU incorporation, cells Laboratories), and ER domain was PCR-amplified from pBabe-Puro-OmoMyc- were seeded in six-well plates (353046; Corning) and incubated in complete ER (a gift from Gerard Evan, University of California, San Francisco). Using growth media containing 1 μg/mL puromycin and 500 nM 4-OHT for 24 h. these DNA fragments, we constructed the gene fusions encoding GFP-p21- Each well was then washed with PBS and incubated with media containing ER and GFP-ER; using standard molecular biology techniques, the gene fusions various concentrations of serum and EGF, 500 ng/mL hydrocortisone, 100 ng/mL were inserted into plasmids to generate pCru5-GFP-p21-ER and pCru5-GFP-ER, cholera toxin, 1 μg/mL puromycin, and 500 nM 4-OHT for 72 h; 10 μM

respectively. The fusions were then moved into pCru5-(UUU)GGFP-IRES-Puro BrdU (B23151; Life Technologies) was added to each culture, and cells were (22) using the EcoRI and NsiI restriction sites. The EcoRI-NotI fragment from this incubated for an additional 48 h. Cells were then dissociated with 0.05% vector was then moved into the EcoRI-NotI sites of pCru5-GFP-IRES-mCherry trypsin, and DMEM/F12 medium with 1% BSA was used to neutralize the vectors containing various RNA leader sequences and regulatory uORFs (22) to trypsin. Cells were washed with PBS and incubated in 70% (vol/vol) ethanol create pCru5-GFP-p21-ER-IRES-Puro and pCru5-GFP-ER-IRES-Puro. For live-cell overnight. After washing with PBS, DNA was denatured by treatment with imaging experiments, RFP (specifically, the fluorescent protein mCherry) was PBS containing 0.1% Triton-X 100 (PBS-T) supplemented with 4 M HCl for

E4392 | www.pnas.org/cgi/doi/10.1073/pnas.1409797111 Overton et al. Downloaded by guest on October 2, 2021 30 min. The acid was neutralized by washing two times with 0.1 M sodium Immunoblotting. Immunoblotting was performed according to standard PNAS PLUS tetraborate, and samples were incubated with PBS-T containing 1% BSA for procedures. Lysates were loaded onto a 10% (vol/vol) Mini-Protean TGX Gel 1 h for blocking; 1 μL mouse anti-BrdU antibody conjugated to AlexaFluor488 (Bio-Rad) and transferred to a PVDF membrane; p21 was detected using (clone MoBU1, B35139; Life Technologies) was added to each sample and a primary anti-p21 mouse monoclonal antibody (2946; Cell Signaling Tech- incubated at 4 °C overnight before washing with PBS-T and incubating with PI nology) and a secondary anti-mouse IgG conjugated to HRP (G21040; Mo- staining solution for 30 min. lecular Probes). β-Actin was detected using a mouse monoclonal anti–β-actin Cell populations were analyzed on an LSRII flow cytometer (Beckton antibody conjugated to HRP (A00730; Genscript). WesternBright ECL HRP sub- Dickinson). GFP fluorescence intensity was determined using 488-nm excitation strate (K-12045-D50; Advansta) was used for HRP detection. and a 525/50-band pass emission filter (the emission collection was centered at – 525 nm with a total range of 50 nm) as well as 405-nm excitation and a 525/50- Modeling. Computational analysis of the CDK2 p21 double-negative regu- band pass filter. Flow cytometry data were analyzed using FlowJo (Tree Star). latory feedback system was performed using MATLAB (MathWorks).

− − ’ Live-Cell Imaging. MCF10A WT and p21 / cells expressing H2B-mTurquoise, ACKNOWLEDGMENTS. This work marks the end of C.L.W. s time at Stanford University. C.L.W. would like to thank Jay Keasling and Matthias Wabl for DHB-YFP, and RFP-p21-ER (pCru5-ACC/ACC/ACC-RFP-p21-ER) were seeded in their tutelage and support along the way, and his students and laboratory μ a 96-well plate (3603; Costar) in complete growth media containing 1 g/mL members for believing in his laboratory’s vision and for doing science the puromycin and 500 nM 4-OHT and incubated for 24 h. Each well was washed right way. He’d still take reincarnation as a scientist. The authors thank Steve two times with PBS, and 300 μL phenol red-free media containing various Cappell for providing MLN-4924 and MATLAB code and Mingyu Chung for concentrations of serum and EGF, 500 ng/mL hydrocortisone, 100 ng/mL MATLAB code. Flow cytometry measurements were performed at the Stan- cholera toxin, 1 μg/mL puromycin, and 500 nM 4-OHT were added. Cells ford Shared FACS Facility, and sorting was performed on an instrument at the Stanford Shared FACS Facility obtained using National Institutes of Health were incubated for 72 h before imaging for 48 h on an IXMicro microscope (NIH) S10 Shared Instrument Grant S10RR025518-01. This study was supported × (Molecular Dynamics) with a heated stage and 5% (vol/vol) CO2.A10 ob- by NIH Grant 5R21AG040360-02 and Ellison Medical Foundation Grant AG-NS- jective was used, and total exposure time was kept below 900 ms. Images 0550-09. K.W.O. was supported by a National Science Foundation Graduate were analyzed using MATLAB (MathWorks). Research Fellowship and a Stanford Graduate Fellowship.

1. Maheshri N, O’Shea EK (2007) Living with noisy genes: How cells function reliably with 17. Yao G, Lee TJ, Mori S, Nevins JR, You L (2008) A bistable Rb-E2F switch underlies the inherent variability in gene expression. Annu Rev Biophys Biomol Struct 36:413–434. restriction point. Nat Cell Biol 10(4):476–482. 2. Niepel M, Spencer SL, Sorger PK (2009) Non-genetic cell-to-cell variability and the 18. Yao G, Tan C, West M, Nevins JR, You L (2011) Origin of bistability underlying mam- consequences for pharmacology. Curr Opin Chem Biol 13(5-6):556–561. malian cell cycle entry. Mol Syst Biol 7:485. 3. Aldridge BB, et al. (2012) Asymmetry and aging of mycobacterial cells lead to variable 19. Hahn AT, Jones JT, Meyer T (2009) Quantitative analysis of cell cycle phase durations growth and antibiotic susceptibility. Science 335(6064):100–104. and PC12 differentiation using fluorescent biosensors. Cell Cycle 8(7):1044–1052. 4. Lee TJ, Yao G, Bennett DC, Nevins JR, You L (2010) Stochastic E2F activation and 20. Westergaard LG, Andersen CY (1989) Epidermal growth factor (EGF) in human pre- – reconciliation of phenomenological cell-cycle models. PLoS Biol 8(9):e1000488. ovulatory follicles. Hum Reprod 4(3):257 260. 5. Wilson A, et al. (2008) Hematopoietic stem cells reversibly switch from dormancy to 21. Ferreira JP, Noderer WL, Diaz de Arce AJ, Wang CL (2014) Engineering ribosomal self-renewal during homeostasis and repair. Cell 135(6):1118–1129. leaky scanning and upstream open reading frames for precise control of protein – 6. Passos JF, et al. (2007) Mitochondrial dysfunction accounts for the stochastic hetero- translation. Bioengineered 5(3):186 192. geneity in telomere-dependent senescence. PLoS Biol 5(5):e110. 22. Ferreira JP, Overton KW, Wang CL (2013) Tuning gene expression with synthetic – 7. Gaudet S, Spencer SL, Chen WW, Sorger PK (2012) Exploring the contextual sensitivity upstream open reading frames. Proc Natl Acad Sci USA 110(28):11284 11289. 23. Dulic V, Stein GH, Far DF, Reed SI (1998) Nuclear accumulation of p21Cip1 at the onset of factors that determine cell-to-cell variability in receptor-mediated apoptosis. PLoS of mitosis: A role at the G2/M-phase transition. Mol Cell Biol 18(1):546–557. Comput Biol 8(4):e1002482. 24. Tinkum KL, et al. (2011) Bioluminescence imaging captures the expression and dy-

8. Spencer SL, Gaudet S, Albeck JG, Burke JM, Sorger PK (2009) Non-genetic origins of SYSTEMS BIOLOGY namics of endogenous p21 promoter activity in living mice and intact cells. Mol Cell cell-to-cell variability in TRAIL-induced apoptosis. Nature 459(7245):428–432. Biol 31(18):3759–3772. 9. Cohen AA, et al. (2008) Dynamic proteomics of individual cancer cells in response to 25. Yamamoto H, et al. (1999) Comparative effects of overexpression of p27Kip1 and a drug. Science 322(5907):1511–1516. p21Cip1/Waf1 on growth and differentiation in human colon carcinoma cells. On- 10. Harper JW, Adami GR, Wei N, Keyomarsi K, Elledge SJ (1993) The p21 Cdk-interacting cogene 18(1):103–115. – protein Cip1 is a potent inhibitor of G1 cyclin-dependent kinases. Cell 75(4):805 816. 26. Wang W, Nacusi L, Sheaff RJ, Liu X (2005) Ubiquitination of p21Cip1/WAF1 by SCFSkp2: 11. el-Deiry WS, et al. (1993) WAF1, a potential mediator of p53 tumor suppression. Cell Substrate requirement and ubiquitination site selection. Biochemistry 44(44):14553–14564. – 75(4):817 825. 27. Sheaff RJ, et al. (2000) Proteasomal turnover of p21Cip1 does not require p21Cip1 12. Spencer SL, et al. (2013) The proliferation-quiescence decision is controlled by a bi- ubiquitination. Mol Cell 5(2):403–410. – furcation in CDK2 activity at mitotic exit. Cell 155(2):369 383. 28. Venta R, Valk E, Kõivomägi M, Loog M (2012) Double-negative feedback between S-phase 13. Lu Z, Hunter T (2010) Ubiquitylation and proteasomal degradation of the p21(Cip1), cyclin-CDK and CKI generates abruptness in the G1/S switch. Front Physiol 3:459. p27(Kip1) and p57(Kip2) CDK inhibitors. Cell Cycle 9(12):2342–2352. 29. Cheng T, et al. (2000) Hematopoietic stem cell quiescence maintained by p21cip1/ 14. Elowitz MB, Levine AJ, Siggia ED, Swain PS (2002) Stochastic gene expression in waf1. Science 287(5459):1804–1808. a single cell. Science 297(5584):1183–1186. 30. Kippin TE, Martens DJ, van der Kooy D (2005) p21 loss compromises the relative 15. To TL, Maheshri N (2010) Noise can induce bimodality in positive transcriptional quiescence of forebrain stem cell proliferation leading to exhaustion of their pro- feedback loops without bistability. Science 327(5969):1142–1145. liferation capacity. Genes Dev 19(6):756–767. 16. Wong JV, Yao G, Nevins JR, You L (2011) Viral-mediated noisy gene expression reveals 31. Peacock RW, Wang CL (2011) A genetic reporter system to gauge cell proliferation biphasic E2f1 response to MYC. Mol Cell 41(3):275–285. rate. Biotechnol Bioeng 108(9):2003–2010.

Overton et al. PNAS | Published online September 29, 2014 | E4393 Downloaded by guest on October 2, 2021