Network calisthenics

Abstract

Stimulation of quiescent mammalian cells with mitogens induces an abrupt increase in E2F1–3 expression just prior to the onset of DNA synthesis, followed by a rapid decline as replication ceases. This temporal adaptation in E2F facilitates a transient pattern of gene expression that reflects the ordered nature of DNA replication. The challenge to understand how E2F dynamics coordinate molecular events required for high-fidelity DNA replication has great biological implications. Indeed, precocious, prolonged, elevated or reduced accumulation of E2F can generate replication stress that culminates in either arrest or death. Accordingly, temporal characteristics of E2F are regulated by several network modules that include feedforward and autoregulatory loops. In this review, we discuss how these network modules contribute to “shaping” E2F dynamics in the context of mammalian cell cycle entry.

Introduction

In response to mitogenic growth stimulation, quiescent mammalian cells can re-enter the cell cycle. Successful division requires faithful and complete duplication of genomic DNA within a narrow time frame of minutes to hours. To deal with the speed and fidelity demanded of this process, eukaryotes have evolved a parallel processing strategy: replication is asynchronously initiated from a subset of several thousand genomic locations called “origins of replication” (ORI). An organizing principle of this process is temporal ordering (Fig. 1A): helicase and accessory proteins forming the pre-replication complex (pre-RC) are synthesized, ORI are “licensed” by binding to pre-RC; replication initiation of licensed ORI is triggered by phosphorylation of the pre-RC components, and licenses are removed through a combination of phosphorylation-dependent degradation, inhibition and re-localization of pre-RC machinery.1,2 Temporal coordination ensures that DNA is faithfully duplicated “once and only once” during each cell cycle. Deregulation of this process commonly results in replication stress, i.e., aberrant re-initiation and DNA breakage resulting from uncoordinated progression of replication forks,3 which has been speculated to generate the genomic instability that underlies malignant transformation.4

E2F transcription factors play an integral role in coordination of DNA replication events. The first member, E2-factor 1 (E2F1), was identified through its physical association with the retinoblastoma (RB) tumor suppressor,5,6 which acts to sequester E2F1.7 Growth factor stimulation induces RB phosphorylation, permitting release and activation of E2F1 activity.8 E2F1, in association with DP1, behaves as a sequence-specific transcriptional activator of cellular genes, including those associated with growth and proliferation (e.g., c-Myc,9,10 Dihydrofolate reductase, c-Myb and Epidermal growth factor receptor8). This is consistent with observations that ectopic E2F1 stimulated DNA synthesis in quiescent cells.11–13 This early evidence supports the view of RB-E2F as a link between growth signals and cell cycle gene expression. Recent genome-scale measures of gene expression further revealed a role for E2F in activating not only genes at G₁/S that encode DNA replication proteins, but also genes at G₂/M that encode mitotic activities.

In the last two decades, eight E2F family members have been identified14,15 and divided into “activators” (E2F1–3) and “repressors” (E2F4–8). While this classification implies opposing roles by the two groups, it is increasingly clear that the activities of E2Fs are context-dependent.16,17 Regardless, the functions of various E2Fs in both normal18 and pathological circumstances19 have been extensively analyzed; this information has defined the “wiring diagram” of the wider RB-E2F regulatory network.20,21

Our group has shown that the RB-E2F pathway plays a central role in discriminating between different types of growth stimulation. Arthur Pardee coined the term restriction point (R-point) to describe the time at which cells commit to the cell cycle by discontinuing their dependence on mitogenic stimulation.22 The R-point can enforce one of two cell states (quiescence and proliferation) in accord with environmental conditions. Consistent with this notion, we have shown that the RB-E2F network acts as a bistable switch to convert graded growth inputs into an “all-or-none” response.23,24 Further, we have shown that the RB-E2F network can discriminate between normal and aberrant growth signaling from proto-oncogenes such as c-Myc (Myc). MYC is a critical mediator of physiological growth signals that facilitates E2F expression during cell cycle entry.25 The Myc locus is often amplified in human cancers, presumably to “short circuit” the need for external growth stimulation. In normal cells, however, overexpression of MYC fails to induce DNA replication or division,26,27 suggesting that cells can somehow respond specifically to MYC expressed in a physiological context. These differing responses are reconciled by the observation that E2F1 is only upregulated when MYC levels are within a narrow window comparable to levels achieved following growth factor stimulation.28

The bistable and biphasic responses in the dose domain represent the culmination of successive temporal events initiated by growth signals. However, the temporal dynamics of E2F are equally important. In response to strong growth stimulation of quiescent cells (G₀), E2F1–3 expresion will rise and peak just prior to the onset of DNA synthesis (S phase) followed by inactivation just prior to the onset of mitosis (M) (Fig. 1A). This temporal program may be pivotal in coordinating the ordered molecular events required for high-fidelity DNA replication. First, E2F controls the expression of genes that constitute the pre-RC and licensing machinery which are absent in quiescent cells; perturbing their normal temporal pattern of expression induces double-strand breaks resulting from re-replication, followed by a p53-mediated checkpoint activation.29 Second, E2F is part of an intricate regulatory cascade that activates Cyclin A/Skp2, but in a delayed manner relative to Cyclin E. This differential temporal control presumably provides a “window of opportunity” between ORI licensing and initiation/delicensing, respectively.30–32 Third, persistent levels of E2F1 are unable to drive DNA synthesis to completion in quiescent fibroblasts but, rather, trigger a p53-mediated DNA damage checkpoint.33 Fourth, deletion of E2f1–3 in mice did not prevent cell cycling consistent with the existence of pathways parallel to RB-E2F;34,35 however, it did result in DNA damage attributed to replicative stress.17 Indeed, while E2F targets retain their overall expression pattern in the absence of E2f1–3, their dynamics are altered and unable to reach the same peak levels.36 Fifth, decoupling of E2F from control mechanisms that leads to either precocious37 or prolonged38 activity triggers DNA stress and a p53-mediated checkpoint. This is reminiscent of the impact of tumor-related disruptions of the RB-E2F pathway:39,40 deregulation of RB can lead to abnormal replication fork dynamics, DNA strand breakage and genomic instability.41

Underscoring the critical importance of the E2F temporal program, the RB-E2F network is governed by multiple layers of feedback and feedforward regulation (Fig. 1B). In this review, we summarize the regulatory mechanisms that may contribute to precise control of E2F expression and activity during cell cycle entry. We emphasize evidence from mammalian cells and the dynamics of E2F1–3 activators, since they positively correlate with replication in this context. In the future, coupling mathematical modeling and experiments will be essential for quantitative understanding of E2F temporal dynamics.

Delayed E2F: Derepression

Growth stimulation initiates a cascade of signaling events20 that generates an early peak in MYC (<60 min) along with a late peak (∼8 h) due to changes in stability mediated by ERK and PI3K, respectively.42–44 Though MYC is required for transcription of E2f1–3, the rise in E2f is delayed, occurring in concert with the second peak of MYC.25 This lag likely reflects the time required to remove complexes that otherwise silence E2F expression.

A critical aspect in E2F biology is negative regulation: E2F activators and many downstream target genes are repressed in quiescence but de-repressed during cell cycle entry (Fig. 2A). E2F1–3 are sequestered by several “pocket” proteins: retinoblastoma (RB), p107 and p130.45,46 RB is constitutively expressed, functioning as a bona fide tumor suppressor that is often disrupted during the genesis of many types of human cancers.47 In contrast, p107 and p130 are dominant in cycling and quiescent cells, respectively, and neither is altered in cancers despite their ability to compensate for aspects of RB function.48 Extensive work has shown that phosphorylation mediated by CDKs is a primary means to alleviate pocket protein inhibition of E2F activators.49–51

E2F transcription is also negatively regulated by pocket proteins. In quiescent fibroblasts, a protein complex of p130:E2F4/5 maintains low transcription of E2f1 as well as other E2F-regulated cell cycle genes (e.g., Cdc₆, Myb and Cyclin A)52 through direct binding to upstream regulatory sequences. Expression silencing is achieved in part through E2F4/5:p130-mediated recruitment of histone deacetylases (HDAC) that maintain a non-permissive chromatin state.53 Germline deletion of both p107 and p130 expressed higher basal levels of E2F and E2F-regulated targets and were constitutively acetylated,54 confirming the notion that continual HDAC activity is required to maintain low expression from these loci.

In the presence of growth factors, p130 levels decrease sharply between 6–10 h, coincident with the increase in E2F activator mRNA levels.55 Decreased p130 is dependent upon CDK_4,6 phosphorylation, which signals SCF^SKP2-mediated ubiquitination, reducing the p130 half-life to ∼1 h.56,57 In the absence of p130, which normally tethers E2F4/5 in the nucleus, E2F4 is found predominantly in the cytoplasm, thus restricting its association with gene regulatory sequences.53,58,59 It is likely that the dynamics of p130 degradation (∼5 h after CDK increase) are rate limiting for subsequent stages of E2F regulation.

From OFF to ON: Positive Feedback

Johnson et al.60 showed that a growth regulated region of the human E2f1 gene is activated by E2F1–3, which physically associate with two consensus E2F binding sites (TTTSSCGC, where S is either a G or a C) situated in the proximal promoter.52–54 Mutation of E2F binding sites simultaneously abrogated E2F protein binding and resulted in constitutively high promoter activity, consistent with a role in mediating repression by p130:E2F4/5 complexes. Further dissection of each binding site revealed subtle differences: The upstream site mediates repression, whereas the proximal site activation.59 The functional distinction between sites is reflected in their association with different protein complexes.53 The specific role that E2F transactivation (vs. de-repression) plays in the transcriptional dynamics of E2f1–3 or other downstream targets remains to be seen.

E2f2 and E2F3a employ similar regulatory mechanisms to modulate their expression.61–63 In vivo evidence suggests cross-association between E2F activators at promoter binding sites during the exit from quiescence.52,53 It remains unclear how individual E2F activators contribute to overall E2F expression. At least in terms of development, a single E2F activator can suffice.64

Another major player in positive autoregulation is Cyclin E, a regulatory subunit for CDK₂, which has similar dynamics as E2F1.65,66 The activities of RB, CYCLIN E and E2F are deeply interconnected (Fig. 2A). First, as for E2f1–3, pocket proteins negatively regulate Cyclin E transcription, and their disruption leads to increased Cyclin E expression even in quiescent cells.54,67–69 Second, pocket proteins are phosphorylated by CYCLIN E:CDK₂ complexes at the conclusion of G₁, leading to disruption of their E2F binding.5 Third, Cyclin E is a direct transcriptional target of E2F1; mutation of two canonical, promoter-proximal E2F binding sites results in qualitatively similar temporal dynamics in response to serum but with a premature peak and overall elevated levels.70 Cyclin E is essential for exit from quiescence, likely owing to its role in promoting assembly and licensing of pre-RC.71,72 Moreover, constitutive expression of Cyclin E results in cell cycle arrest and chromosomal instability,73 underscoring the interwoven nature of positive autoregulation and DNA replication.

Another source of positive feedback involves p53. Growth stimulation and E2F activation are required to eliminate a p53-p21^WAF1-mediated block in cell cycle entry in late G₁.74 p21^WAF1 is a transcriptional target of p53 75 and promotes RB activity by inhibiting its phosphorylation by CDKs. One proposed link between E2F and p53 is Sirt1, which is induced by E2F1 and encodes a deacetylase that inactivates p53 activity.76,77 Another possible link is the E2F-mediated induction of the Arf tumor suppressor that inactivates MDM2, a ubiquitin ligase of p53. However, it remains unclear if modulation of Arf expression and activity is cell cycle-dependent.36,78,79 Regardless of specific mechanisms, suppression of the p53-p21^WAF1 axis in an E2F-dependent fashion represents an additional means to alleviate E2F sequestration by pocket proteins.

Regulation of E2F by multiple positive feedback loops is critical for the control of cell cycle entry. Positive feedback is a hallmark of bistable responses80 and may underlie the self-sustaining behavior of the RB-E2F network. Furthermore, coupled slow and fast positive feedback loops can generate rapid transit from the OFF to ON state yet remain noise-resistant when ON.81 Using a synthetic mammalian gene circuit, Longo et al. demonstrated that positive feedback can reduce both the time needed to surpass basal expression levels and cell-cell variability in gene expression (Fig. 2B). These findings are similar to those gathered by our group using single-cell measurements of E2f1 expression coupled with stochastic simulations of an RB-E2F network model:83 feedback mediated by CDK₂ reduced both the minimum time for E2F to surpass a basal threshold (“time delay”) and the variability of ON-switching across a population (“transition rate”). An intriguing result of this work is the correspondence between variability in E2F activation and cell-cell variability in the time between G₀ and division previously described using phenomenological models.84,85 This suggests that E2F dynamics may have direct consequence on both DNA replication and the rate of cell proliferation. Whether positive autoregulation arising through cross-regulation or p53 has an effect on time delay, coherence and noise in E2F or the relative contribution of each feedback loop is unclear. Importantly, how positive feedback may constrain the temporal pattern of pre-RC synthesis, assembly, loading and licensing remains to be seen.

Amplitude Modulation: Rapid Negative Feedback/Feedforward

There is a growing appreciation for the role of micro RNA (miRNA) in modulating RB-E2F activity, target genes and replication stress.86–89 The miR-17 cluster is regulated by both MYC and E2F. Two mature products of this locus, miR-17-5p and miR-20a, target E2F mRNA and downregulate translation.90 The relationship between MYC, E2F and miRNA represents an incoherent feedforward loop (I1-FFL), while that between E2F and miRNA represents negative feedback91,92 (Fig. 3A).

I1-FFL can generate distinct network behaviors,93 including adaptation,94 fold-change detection95 and biphasic dose-response.96–98 Negative feedback can generate adaptation, oscillations99 and expand the dynamic range of a dose-response.100 Both I1-FFL and negative feedback can increase response speed:101–103 repression of steady state levels by either module can be offset through an increased production rate, which reduces the time required to achieve half-maximal levels (Fig. 3B).

Several lines of evidence suggest that repression due to miR-17 may act as a mechanism to facilitate accelerated E2F induction. First, theoretical work demonstrates that miRNA downregulates steady state E2F output.104 Second, miR-17 expression is rapid relative to E2F1–3, peaking within 1 h following growth stimulation.90 Third, once induced, levels of miR-17 are relatively stable throughout the cell cycle. These properties suggest that the miRNA functions to attenuate overall E2F levels rather than playing a role in turning E2F OFF following DNA replication. Indeed, inhibition of miR-17-5p and miR-20a in human fibroblasts by antisense RNA led to a reduction in de novo DNA synthesis arising from engagement of a p53-dependent DNA damage checkpoint.37 While the overall pattern of E2F1 expression was maintained, there was precocious E2F1 induction (3–6 h) along with an increase in maximal E2F1 expression.

Like miR-17, the Arf tumor suppressor is regulated by E2F and MYC, and ARF protein can enhance proteasomal-mediated degradation of E2F.105–111 However, Arf expression during the cell cycle may be context-dependent. Arf expression was described as constitutive throughout the cell cycle in rat and human cell lines stimulated by serum.78 In contrast, serum-stimulated MEFs showed a decrease in Arf transcription during the period coincident with S phase.79 Our work has shown that Arf is, in fact, induced in rat fibroblasts with kinetics similar to E2f1.28 Thus, the qualitative and quantitative contribution of Arf toward E2F dynamics is unclear. It is possible that the influence of ARF may be context-dependent112 or involve post-translational, cell cycle-dependent modifications rather than changes in its expression.

ON to OFF: Delayed Negative Feedback

Concurrent with the entry into S phase, E2F activity is downregulated as persistent expression of E2F1 induces apoptosis.11–13,33,113 Intuitively, abrupt suppression of E2F and licensing proteins may prevent aberrant re-initiation of replication, which otherwise triggers a cell cycle checkpoint. As discussed, rapid activation of moderate negative feedback/incoherent feedforward (e.g., miRNA) can modulate steady state levels of gene expression. On the other hand, strong delayed negative feedback permits levels of an upstream node to overshoot before it is repressed.114 Multiple sources of delayed negative feedback may play a role in quenching E2F activity following S-phase entry.

CYCLIN A:CDK₂ activity is essential for DNA replication.115 Similar to Cyclin E, the transcription of Cyclin A is growth regulated and under negative control through E2F binding sites.116 Importantly, Cyclin A expression is delayed relative to E2F and Cyclin E,36,117 and temporal staggering is enforced at both the transcriptional30 and post-translational levels.31 These successive interactions have a functional role in allowing pre-RC assembly (Cyclin E) to precede replication initiation and delicensing (Cyclin A).32 CYCLIN A also downregulates E2F DNA binding by phosphorylating and inhibiting the obligate DNA binding partner, DP.118,119 Prolonged DNA binding activity of an E2F1 mutant resistant to CYCLIN A:CDK₂ triggers a DNA damage checkpoint in conjunction with apoptosis.38 A subtle observation is that both E2F1 and E2F3 are required for cell cycle entry,120 but in subsequent cell cycles, only E2F3 binding activity is required.121 The mechanism and significance of this selectivity is unclear. It will be important to understand how precisely prolonged or unscheduled E2F DNA binding activity (i.e., E2F1 during subsequent cell cycles) impacts the operation of the DNA replication machinery.

E2F protein stability is modulated through ubiquitin-mediated proteasomal degradation.122–124 E2F1 directly activates transcription of Skp2 gene,125 which encodes a subunit of the SCF^SKP2 ubiquitin ligase that targets E2Fs for destruction.126 Changes in SKP2 levels are cell cycle-dependent and, importantly, are delayed with respect to E2F1 through a mechanism similar to the one leading to delayed increase in Cyclin A.127 Moreover, SKP2 exists in a protein complex with CYCLIN A:CDK₂ and is required for its ability to promote DNA replication.127,128 This physical coupling may represent a way to integrate the initiation and delicensing machinery, potentially minimizing the window of time in which aberrant re-initiation may occur. Thus, in addition to its role in replication, delayed negative feedback from CYCLIN A/SKP2 downregulates E2F, underscoring the inextricable coupling of negative feedback and DNA replication events.

Another source of negative feedback involves E2F7 and E2F8, the most distantly related members of the E2F family.129,130 Although they can form homo- and heterodimers on E2F DNA binding sites, E2F7/8 do not interact with DP proteins, and their expression is delayed, rising at the conclusion of S phase. Work by Li et al.131 found that these genes are direct targets of E2F1, and germline disruption of E2f7/8 led to both higher and prolonged levels of E2f1 mRNA beginning at S phase. Deletion of E2f7/8 was accompanied by massive apoptosis that was dependent upon the presence of intact E2f1 and p53. These observations were initially surprising, because downregulation of E2f1 was fully dependent upon E2f7/8 despite the presence of Cyclin A and Skp2.132 However, this could be expected in light of the fact that E2F7/8 target transcription, while CYCLIN A/SKP2 act at the post-translational level, although the presence of positive feedback complicates this interpretation. These sorts of discrepancies emphasize the need to understand how different regulatory modules impact E2F at both the transcriptional and post-transcriptional level.

A Quantitative Framework of E2F Dynamics

We have presented evidence indicating that E2F dynamics encode information from growth signals, enabling the coordinated activity of cell cycle modules involved in DNA replication. A framework to describe the quantitative relationship between E2F dynamics and the replication machinery would aid in determining how coordination is specifically achieved and ways it can become deregulated. Two important challenges lie ahead in this regard. First is the development of an experimental platform sensitive enough to detect endogenous levels of multiple genes in individual cells with high temporal resolution. For the most part, observations of E2F dynamics have been made using population-average methods that mask cell-cell differences likely to have profound phenotypic consequences. A second challenge is the development of appropriate modeling tools, such as stochastic differential equations (SDEs), that can both describe overall network behavior and capture the cell-cell variability in gene expression. A development cycle involving modeling and quantitative experiments provides a synergistic platform for both refining model parameters (based upon experimental measurements) and making testable predictions (using model simulations) about how genetic or environmental perturbations may deregulate network function. Quantitative frameworks will be invaluable for the systematic investigation of E2F function in normal and pathological circumstances. Ultimately, it may provide opportunities for the rational design of targeted cancer therapeutics aimed at quantitative modulation of network behavior.

Abbreviations

pre-RC	=	pre-replication complex
ORI	=	origin of replication
E2F	=	E2-factor
RB	=	retinoblastoma
miRNA	=	micro RNA
I1-FFL	=	incoherent feedforward loop

Figures and Tables

Figure 1 Temporal correspondence between DNA replication and E2F. (A) (Top) Overview of successive temporal events in DNA replication. Gene products regulated by E2Fs are shown in blue. The licensed pre-replication complex (pre-RC) contains CYCLIN E, ORC, CDC₆, CDT1 and MCM situated at origins of replication (ORI). Initiation involves MYC and CYCLIN A:CDK-mediated activation of pre-RC helicase activity. Delicensing occurs through inhibition of CDT1 by protein sequestration by GEMININ along with SKP2 and PCNA-mediated ubiquitination. A temporal delay between CYCLIN E- and CYCLIN A-associated CDK activity is mediated by APC/C^CDH1. APC/C, anaphase-promoting complex/cyclosome with CDH1; GEM, GEMININ; ORC, origin recognition complex; CYC, CYCLIN complexed with cyclin-dependent kinase (CDK); MCM, minichromsome maintenance proteins 2–7; Pol, DNA polymerase. (Bottom) Typical temporal pattern for E2F activators (E2F1–3) as cells re-enter the cell cycle from quiescence (G₀) following growth factor stimulation. The temporal dynamics summarized in this review are indicated: (1) Delayed E2F increase relative to immediate early genes; (2) switching OFF to ON; (3) amplitude modulation and (4) switching ON to OFF. (B) (Left) Genes induced by E2F1–3 and their associated network modules. (Right) Overview of network logic involving modules of the RB-E2F network. IN, upstream signals originating from growth factor signaling and MYC.

Figure 2 Switching OFF to ON: Derepression and positive feedback on E2F. (A) Events following growth factor receptor engagement that increases E2F expression. In the exit from quiescence, a repressive p130:E2F4/5 complex (designated by RB_p:E2FR_Nuc) situated on E2F target genes is removed via phosphorylation and degradation of p130; E2F4/5 is exported from the nucleus (Nuc) to the cytoplasm (Cyt) in the absence of p130. RB, on the other hand, cycles through states of hypo- and hyperphosphorylation during the cell cycle. Repression of p53 is achieved through E2FA-mediated upregulation of the SIRT1 deacetylase. Green arrows indicate positive feedback. GF, growth factors; E2FA, activators E2F1–3; E2FR, repressors E2F4–5; RB_p, hypophosphorylated pocket proteins; RB_pp, hyper-phosphorylated pocket proteins; p21^WAF1, cyclin-dependent kinase inhibitor; (B) Effect of positive feedback on gene expression. Cell-cell variability in isogenic cells (noise) can manifest in differences in expression dynamics. Depicted are time courses from distinct cells with a gene subject to weak (gray) or strong (black) positive feedback. Positive feedback can decrease the time needed to exceed basal expression (time to cross horizontal dotted line; τ_Delay) and the coherence across a population (difference in τ_Delay between cells).

Figure 3 Switching ON to OFF—negative feedback and feedforward. (A) Negative feedback and incoherent feedforward (I1-FFL) modules. Direct regulation of miR-17–92 cluster of miRNAs along with the tumor suppressor Arf by both MYC and E2F represent incoherent feedforward/negative feedback loops. CYCLIN A and SKP2 are involved in suppression of DNA binding and stability of E2F1–3, respectively. E2f7/8 are transcriptional targets of E2F that bind to and suppress E2f1–3 transcription in the late phases of the cell cycle. (B) Effect of rapid incoherent feedforward (I1-FFL) and negative feedback (NF). Repression provided by I1-FFL/NF can reduce steady state but when offset by increased production rate, it can allow a shorter rise time (τ_rise) compared with a circuit without IFF/NF.