All organisms need to be able to react to the challenges thrown at them by their changing environment. Yeast, bacteria and other microbes have networks of genes that can give rise to many different traits and characteristics, which can also be referred to as phenotypes. A change in the environment can alter the activities' of the genes so that the microbes display a different phenotype. The point at which a small change in the environment can lead to a sudden switch in the phenotype is called a ‘critical transition’.

An individual microbe's history can influence the phenotype that it presents. However, it is not clear how microbes ‘remember’ their history, or how fluctuations in the environment might cause the microbe to lose the ability to store this memory and present a different phenotype instead. Here, Axelrod et al. studied phenotype memory in yeast cells grown in the laboratory. The experiments used cells that had been genetically modified to glow red in the presence of a molecule called anhydrotetracycline (or ATc) and to glow green in the absence of the molecule.

Axelrod et al. examined what effect altering the levels of this molecule would have on the phenotype produced by the cells. First, the cells were grown with no ATc present for several generations so that the cells glowed green. Next, Axelrod et al. added different amounts of ATc were added. For moderate levels of ATc the cells continued to glow green, illustrating that they ‘remembered’ their prior growth condition. However, cells exposed to higher levels of ATc lost this memory and changed color.

Next, Axelrod et al. carried out further experiments on cells exposed to ATc levels that were close to, or further away from the critical transition. At high levels of ATc (that is, close to the critical transition), many cells switched from green to red when exposed to high temperatures, salt and other changes in the environment. On the other hand, very few of the cells grown in low levels of ATc—and therefore further away from the critical transition—changed color in response to the same fluctuations in their environment.

These finding reveal that phenotype memory is less stable when yeast experience fluctuations in their environment close to a critical transition. Future work will seek to find out how salt or high temperatures can abolish phenotype memory.

Microbes such as yeast and bacteria often adopt a specific phenotype in response to an environmental stimulus, and in many cases this phenotype can be retained even after the stimulus has been removed (Ozbudak et al., 2004; Acar et al., 2005; Hu et al., 2012; Williams et al., 2013). Stably storing such a phenotype can be critical for survival during adverse environmental changes (Gasch et al., 2000; Gasch and Werner-Washburne, 2002; Chen et al., 2003; Brauer et al., 2008). Several well-studied natural gene networks can be found in multiple phenotypic states depending on the history that the cells have experienced, notably including the yeast galactose utilization network and sporulation commitment in Bacillus subtilis (Acar et al., 2005; Igoshin et al., 2006). Additionally, synthetic biology aims to create de novo phenotypic memory storage devices, which could be used in applications such as tracking dynamics of the gut microbiome (Ajo-Franklin et al., 2007; Kotula et al., 2014). In spite of the importance of phenotypic memory in natural and synthetic gene circuits, little is known about how environmental perturbations might disrupt a cell's ability to stably adopt a specific phenotype.

Given that phenotypic memory often results from feedback loops within the cell (Ferrell and Machleder, 1998; Pomerening et al., 2003; Xiong and Ferrell, 2003; Angeli et al., 2004; Ozbudak et al., 2004), it is possible that this phenotypic memory will display characteristics of other complex systems that exhibit bistability, memory, and associated critical transitions or ‘tipping points’ that lead to sudden changes in the state of the system in response to small changes in the environment. A phenomenon that may be especially relevant to cellular memory is that complex systems near a critical transition experience a loss of resilience of their stable states to external perturbations (Dai et al., 2012). In particular, a system far from a critical transition may return to its original state following a perturbation, whereas closer to the critical transition the same environmental perturbation may cause the system to switch to an alternative stable state (e.g., collapse of a population [Dai et al., 2012]). The dynamics in the vicinity of the critical point are very slow on both sides of the critical point. Therefore, it is worth noting that a short lived perturbation that pushes the system past the critical point may also not cause switching if the perturbation is short enough, if it pushes it to a position that is close to the critical point, or both. This aforementioned loss of resilience near a critical transition results from a shrinking basin of attraction in the stability landscape (Scheffer et al., 2009). In the context of a gene network, this loss of resilience would manifest as an increasing sensitivity of phenotypic memory against environmental perturbations approaching the environmental condition in which cells would (deterministically) switch to a different phenotype. Our initial goal was to observe memory in a model gene circuit and then characterize the resilience of the phenotypic state against perturbations in the extracellular environment.

To study this predicted loss of resilience, we first employed a synthetic genetic switch in budding yeast that has previously been shown to exhibit hysteresis and bistability (Gardner et al., 2000; Blake et al., 2006; Ellis et al., 2009; Wu et al., 2013). This toggle switch is composed of two mutually inhibitory transcription factors, LacI and TetR (Figure 1A). To enable tracking of the state of the cell, different color fluorescent proteins are expressed depending upon which of these transcriptional factors is highly expressed (mCherry and eGFP, hereafter referred to as ‘RFP’ and ‘GFP’). As external knobs to control the state of the cell, the inducer IPTG modulates the strength of repression of LacI, whereas ATc modulates the strength of repression of TetR.

Figure 1 with 2 supplements see all

Download asset Open asset

As a demonstration of how cellular memory operates in this gene network, one population of cells was pre-grown in IPTG to initialize the cells in a high GFP state (Figure 1B). A separate population of cells was pre-grown in ATc to initialize the cells in a high RFP state. We then transferred the populations to a range of ATc concentrations and monitored the dynamics over several days using flow cytometry. All cultures were grown with a fixed concentration of IPTG (40 μM), which served as an orthogonal control variable for later experiments. For high and low concentrations of ATc, cells converged to a state that was independent of the history they were pre-grown in (Figure 1B,C). However, for intermediate concentrations of ATc there was little to no switching of the phenotypic state of the cells even after 260 hr of growth and over 80 cell divisions. The mutual repression in this toggle switch therefore indeed allows for remarkably stable cellular memory for intermediate ATc concentrations. As expected (Acar et al., 2005), the lifetime of the states decreases approaching the critical transition at which switching is deterministic (hereafter, we refer to this point as the ‘critical transition’ or ‘phenotypic switch’), although interestingly we observe non-exponential switching kinetics, potentially indicating the presence of metastable states in the gene network (Figure 1—figure supplements 1, 2). This decrease in lifetime is one manifestation of deteriorating cellular memory approaching the critical transition, but it is not obvious whether in ATc concentrations with long lifetimes (i.e., strong memory) the cells are also able to retain their memory in the face of environmental perturbations.

In analogy to other complex systems, we hypothesized that phenotypic states in this gene network might become increasingly sensitive to perturbations near a critical transition (Van Nes and Scheffer, 2007). This expectation arises because the state's basin of attraction shrinks as the stable and unstable fixed points approach one another (Figure 2A). Brief environmental perturbations will push the system out of equilibrium. If the system is far from the critical transition it will return to its original state after the perturbation is removed (Figure 2B). However, close to the critical transition, the same perturbation might cause the system to cross the basin boundary, thus causing the cell to switch phenotypic states. Therefore, phenotypic states are expected to lose resilience to environmental perturbations near a critical transition.

Figure 2 with 2 supplements see all

Download asset Open asset

To test this theoretically proposed loss of resilience we perturbed cells at different distances from the critical transition (i.e., different ATc concentrations) and measured how likely the cells were to switch into the alternative phenotypic state as a result of environmental perturbations. We examined two classes of perturbations, ‘directional’ (in which the perturbation increased the probability of switching into the alternative state by inhibiting one or another of the two transcription factors that form the toggle switch) and ‘generic’ (in which the interaction between the toggle switch and the perturbation was not immediately obvious).

To explore the directional perturbation we prepared cells in the GFP state by pre-growing them in IPTG. We then transferred the populations to multiple ATc concentrations for 4 days, leaving the cells in the high GFP state but at varying distances from the critical transition. At the end of the fourth day, the cells were then perturbed by decreasing the concentration of IPTG for 1 day before returning the IPTG concentration to its original value. Far from the critical transition (2 ng/ml ATc), the IPTG perturbation caused almost no cells to switch their phenotypic state (Figure 2C). However, close to the critical transition (8 ng/ml ATc), the perturbation caused a great majority of the cells to switch into a high RFP state. Even after the perturbation was removed and the initial conditions were restored, many cells remained in a high RFP state. As expected, the severity of the environmental perturbation (i.e., the magnitude of reduction in IPTG) correlated with the fraction of cells that switched states in response to the perturbation (Figure 2D). Importantly, there was negligible switching (∼0.5%) into the high RFP state for control cells grown in constant IPTG concentrations, demonstrating that the phenotypic switching was indeed caused by the perturbation. Moreover, the lack of phenotypic switching in the absence of the perturbation also indicates that in all of these conditions the traditional measure of cellular memory—the lifetime of the state—would classify all of these conditions as being stable with a high degree of cellular memory. Similar results were observed in the other direction, as cells pre-grown in the RFP state approach the critical transition associated with sudden switching to the GFP state (Figure 2—figure supplement 1). Thus, phenotypic states in the toggle switch lose resilience to directional perturbations near a critical transition.

Cellular memory in development and cell cycle progression must be resilient against a wide range of different environmental perturbations. Given this, we wanted to explore whether memory in the toggle switch would lose resilience against generic perturbations approaching a critical transition. Cells from a high GFP history at different distances from the phenotypic switch (2, 4, and 8 ng/ml ATc) were perturbed in several different ways for 24 hr: heat stress, osmotic stress with sodium chloride, ethanol stress, and a glucose pulse. Remarkably, we observed a loss of resilience against all four of these generic perturbations (Figure 3A,B). Far from the critical transition (2 ng/ml ATc), there was little to no phenotypic switching in response to any of these ‘generic’ perturbations. However, close to the critical transition (8 ng/ml ATc) we observed nearly complete switching in all perturbations, despite the fact that there was essentially no switching (∼0.5%) in the absence of the perturbations. At a given distance from the phenotypic switch, increasing the strength of a generic perturbation increased the probability that cells would switch into the alternative state (Figure 3—figure supplement 1). The switching induced by the glucose perturbation can perhaps be understood by the fact that glucose shuts down expression of the entire system (LacI, TetR, GFP, and RFP) via catabolite repression of a GAL1 upstream activation sequence, thus pushing the cells toward a low GFP and low RFP state (Gardner et al., 2000; Ellis et al., 2009). The other three perturbations have much broader effects on the cell with no obvious connection to the toggle switch network being probed in our experiments. Cellular memory can therefore degrade near a critical transition for a wide range of different environmental perturbations (Figure 3C).

Figure 3 with 3 supplements see all

Download asset Open asset

Complex dynamic systems near critical transitions are predicted to lose resilience to a specific class of perturbations: those that push the system toward the alternative stable state. However, the effect of a salt shock or a heat shock on the GFP output of the toggle switch is challenging to predict a priori. Notably, it is possible that some generic perturbations will stabilize the state of the cells and reduce their probability of switching. Indeed, we found that none of the four generic perturbations led to a loss of resilience in the transition from the high RFP state (Figure 3—figure supplement 3). This result is unsurprising given that the perturbations did not strongly push the genetic network toward the high GFP state (Figure 3—figure supplement 3). For the salt, ethanol, and heat stresses, the perturbation increased the RFP output of the cells, thereby stabilizing the state. Thus, phenotypic states in the toggle switch only loss resilience to perturbations that push the system toward the alternative state.

We reasoned that our experimentally-observed loss of resilience could be caused by a shrinking basin of attraction of the phenotypic state. To test this hypothesis, we used our perturbation experiments to estimate the location of the basin boundary in GFP-RFP space and see how it shifted for different concentrations of ATc (Figure 4A). For a given concentration of ATc and a chosen perturbation, we measured the fraction of cells that switched into the alternative state at the end of the recovery period. Analysis of the GFP-RFP distribution at the end of the perturbation and before the recovery period revealed that many cells were in an indeterminate state between the two stable phenotypic states. We ranked the cells according to the ratio of GFP expression to RFP expression and assumed that a cell would switch phenotypic states when its ratio of GFP expression to RFP expression fell below a critical threshold. This is equivalent to assuming that the basin boundary between the two phenotypic states takes the form of a line on a log–log plot. To justify this assumption, we used a simple mathematical model of the toggle switch (see Figure 4—figure supplement 1 and ‘Materials and methods’). Our modeling indicated that this assumption would hold as long as the promoter strengths of the two transcription factors in the toggle switch were approximately equal.

Figure 4 with 4 supplements see all

Download asset Open asset

Using the assumption described above, we estimated the location of the basin boundary consistent with our experimentally observed fractions at the end of the recovery period. Encouragingly, we find that for a given concentration of ATc, the switching fraction for the eight different perturbations (weak and strong IPTG, weak and strong ethanol, weak and strong salt, heat, glucose) is well fit by a single separatrix (Figure 4B). Moreover, as ATc increases from 2 to 8 ng/ml, the location of the basin boundary gets closer to the location of the high GFP stable fixed point (Figure 4C). Knowing the location of the basin boundary allowed us to map the bifurcation (Figure 4D). Thus, the loss of resilience appears to be driven by a shrinking basin of attraction of the phenotypic state near the critical transition.

Given that the phenotypic states associated with our toggle switch lose resilience to perturbations near the critical transition, a natural question is whether, just as in other examples of critical transitions in complex systems such as ecosystem collapse, it is possible to develop warning indicators that this loss of resilience is taking place (Dai et al., 2012). For example, one might expect that the mean GFP of cells in the high GFP state would decrease with increasing ATc, thus potentially signaling that the critical transition is approaching (Isaacs et al., 2003). However, we find that the mean GFP of unswitched cells is approximately constant over the range of ATc concentrations in which we observe a loss of resilience (Figure 1B and Figure 5A). Researchers in a number of fields have explored early warning indicators based on a loss of stability near a critical transition (broadening of the effective potential as illustrated in Figure 2A) (Scheffer et al., 2009). This loss of stability would manifest in our experiments as an increase in the variation of GFP fluorescence among the population of (unswitched) cells. However, we find experimentally that there is no increase in variation within the population approaching the critical transition, even very close to the transition where the GFP state is metastable (Figure 5B,C). The theoretically proposed early warning indicators based on local stability therefore fail to predict the critical transition in this gene network (Menck et al., 2013).

Figure 5

Download asset Open asset

To explore the generality of our results, we chose to study the yeast GAL network, which displays an ‘all-or-none’ response in some sugar environments containing galactose (Biggar and Crabtree, 2001; Song et al., 2010). However, the wild-type GAL network exhibits only weak memory (Song et al., 2010). To expand the range of galactose concentrations for which the system has memory (Acar et al., 2005), we used a strain of yeast constitutively expressing the repressor GAL80 (which codes for a transcriptional repressor involved in one of the many feedback loops that stabilize memory in this network). To assess the state of the network, yellow fluorescent protein (YFP) expression was driven by a GAL1 promoter. Cells were pre-grown in high galactose (GAL ON) and then grown for a day in a range of galactose concentrations (Figure 6A). We then examined the resilience of the GAL network to perturbations at different distances from its critical transition, similar to our experiments with the toggle switch. For the GAL network, a glucose pulse served as a directional perturbation (due to catabolite repression [Gancedo, 1998]), and we again used heat, salt, and ethanol as generic perturbations. We found that the ‘GAL ON’ phenotypic state lost resilience to glucose and ethanol perturbations but not salt or heat (Figure 6B). Similarly, we performed experiments probing the resilience of the ON state in the Escherichia coli lac operon, where we again observed a loss of resilience to both directional perturbations and to the generic perturbation ethanol (Figure 6—figure supplement 1). These three genetic switches have widely different architectures, and operate in two different species. Our results thus indicate that a loss of resilience approaching a phenotypic switch could be a general property of multistable gene networks and that ‘generic’ or global perturbations, such as temperature or salt shocks, can cause widespread loss of cellular memory.

Figure 6 with 1 supplement see all

Download asset Open asset

Here we have shown that brief perturbations in the extracellular environment can dramatically increase the rate of phenotypic switching from a highly stable memory state into an alternative state. We argue that this loss of resilience to perturbations near the bifurcation can be explained by a shrinking of the basin of attraction of the phenotypic state. By observing the GFP-RFP distribution at the end of the perturbation period and applying a simple threshold, we were able to accurately predict the fraction of cells that would return to the pre-perturbation memory state at the end of the recovery period. At a given distance from the bifurcation, the same basin boundary was able to accurately predict the switching fraction in response to eight different perturbations. The predictive success of this threshold is particularly impressive given that the different perturbations have dramatically different effects on the GFP-RFP distribution (see Figure 4A).

There is an interesting question of the precise mechanism by which the perturbations act on the system to push the cells to a new location in GFP-RFP space. Part of the answer is certainly that the perturbations cause a shift in the stability landscape underlying the phenotypic states. It is also possible that the perturbations cause an increase in noisy gene expression. Understanding what is occurring during the perturbation period, both biochemically and from a dynamical systems perspective, will be an area of investigation for future research.

The perturbations (salt stress, heat stress, etc) lasted for a significant duration: 24 hr, or roughly 3 to 8 cell divisions. It is interesting to ask whether we would have seen similar results if we had perturbed the system for a shorter duration. To address this question, we performed Gillespie simulations using a very simple phenomenological model of the toggle switch (see Source code 1). Similar to our test tube experiments, we initialized the system in a high Lac state. The system was allowed to equilibrate for several generations, before we suddenly changed one of the parameters of the system (either the disassociation constant between the transcription factor and the DNA or the promoter strength). We performed this perturbation for a variable duration ranging from 0.2 generations to 20 generations. The system was then allowed to recover (by restoring the parameter to its initial value) for several generations, and we then determined what fraction of cells remained in a high Lac state. Two key observations emerged. (1) The degree of switching is set by both the intensity and duration of the perturbation. For dramatic perturbations, the minimum duration of perturbation to induce switching is approximately set by the cell division time (Figure 4—figure supplement 3). (2) Not all perturbations induced phenotypic switching. Of note, perturbations which stabilized the high Lac state induced negligible switching into the high Tet state (Figure 4—figure supplement 4). These observations held regardless of whether the perturbation was achieved by changing the disassociation constant or the promoter strength.

Understanding the stability of phenotypic states in gene networks remains an important challenge in biology. Here we have demonstrated that for robust cellular memory in natural contexts subject to environmental noise, low rates of stochastic switching from the phenotypic state is not sufficient. This is because the cellular memory must also be robust against environmental fluctuations and perturbations that are impossible to avoid. Here we found experimentally that several phenotypic states lost resilience to multiple environmental perturbations near a critical transition. Given that over time there will often be multiple kinds of environmental fluctuations and perturbations, our results argue that many forms of cellular memory will become destabilized near a critical transition. Our results provide a roadmap for exploring cellular memory and phenotypic switching in other contexts, from development to cancer progression.

1. 1. Acar M
   2. Becskei A
   3. van Oudenaarden A

   (2005) Enhancement of cellular memory by reducing stochastic transitions

   Nature 435:228–232.

   https://doi.org/10.1038/nature03524

   • Google Scholar
2. 1. Ajo-Franklin CM
   2. Drubin DA
   3. Eskin JA
   4. Gee EP
   5. Landgraf D
   6. Phillips I
   7. Silver PA

   (2007) Rational design of memory in eukaryotic cells

   Genes & Development 21:2271–2276.

   https://doi.org/10.1101/gad.1586107

   • Google Scholar
3. 1. Angeli D
   2. Ferrell JE
   3. Sontag ED

   (2004) Detection of multistability, bifurcations, and hysteresis in a large class of biological positive-feedback systems

   Proceedings of the National Academy of Sciences of USA 101:1822–1827.

   https://doi.org/10.1073/pnas.0308265100

   • Google Scholar
4. 1. Biggar SR
   2. Crabtree GR

   (2001) Cell signaling can direct either binary or graded transcriptional responses

   The EMBO Journal 20:3167–3176.

   https://doi.org/10.1093/emboj/20.12.3167

   • Google Scholar
5. 1. Blake WJ
   2. Balázsi G
   3. Kohanski MA
   4. Isaacs FJ
   5. Murphy KF
   6. Kuang Y
   7. Cantor CR
   8. Walt DR
   9. Collins JJ

   (2006) Phenotypic consequences of promoter-mediated transcriptional noise

   Molecular Cell 24:853–865.

   https://doi.org/10.1016/j.molcel.2006.11.003

   • Google Scholar
6. 1. Brauer MJ
   2. Huttenhower C
   3. Airoldi EM
   4. Rosenstein R
   5. Matese JC
   6. Gresham D
   7. Boer VM
   8. Troyanskaya OG
   9. Botstein D

   (2008) Coordination of growth rate, cell cycle, stress response, and metabolic activity in yeast

   Molecular Biology of the Cell 19:352–367.

   https://doi.org/10.1091/mbc.E07-08-0779

   • Google Scholar
7. 1. Chen D
   2. Toone WM
   3. Mata J
   4. Lyne R
   5. Burns G
   6. Kivinen K
   7. Brazma A
   8. Jones N
   9. Bähler J

   (2003) Global transcriptional responses of fission yeast to environmental stress

   Molecular Biology of the Cell 14:214–229.

   https://doi.org/10.1091/mbc.E02-08-0499

   • Google Scholar
8. 1. Choi PJ
   2. Cai L
   3. Frieda K
   4. Xie XS

   (2008) A stochastic single-molecule event triggers phenotype switching of a bacterial cell

   Science 322:442–446.

   https://doi.org/10.1126/science.1161427

   • Google Scholar
9. 1. Dai L
   2. Vorselen D
   3. Korolev KS
   4. Gore J

   (2012) Generic indicators for loss of resilience before a tipping point leading to population collapse

   Science 336:1175–1177.

   https://doi.org/10.1126/science.1219805

   • Google Scholar
10. 1. Ellis T
    2. Wang X
    3. Collins JJ

    (2009) Diversity-based, model-guided construction of synthetic gene networks with predicted functions

    Nature Biotechnology 27:465–471.

    https://doi.org/10.1038/nbt.1536

    • Google Scholar
11. 1. Ferrell JE
    2. Machleder EM

    (1998) The biochemical basis of an all-or-none cell fate switch in Xenopus oocytes

    Science 280:895–898.

    https://doi.org/10.1126/science.280.5365.895

    • Google Scholar
12. 1. Gancedo JM

    (1998)

    Yeast carbon catabolite repression

    Microbiology and Molecular Biology Reviews 62:334–361.

    • Google Scholar
13. 1. Gardner TS
    2. Cantor CR
    3. Collins JJ

    (2000) Construction of a genetic toggle switch in Escherichia coli

    Nature 403:339–342.

    https://doi.org/10.1038/35002131

    • Google Scholar
14. 1. Gasch AP
    2. Werner-Washburne M

    (2002) The genomics of yeast responses to environmental stress and starvation

    Functional & Integrative Genomics 2:181–192.

    https://doi.org/10.1007/s10142-002-0058-2

    • Google Scholar
15. 1. Gasch AP
    2. Spellman PT
    3. Kao CM
    4. Carmel-Harel O
    5. Eisen MB
    6. Storz G
    7. Botstein D
    8. Brown PO

    (2000) Genomic expression programs in the response of yeast cells to environmental changes

    Molecular Biology of the Cell 11:4241–4257.

    https://doi.org/10.1091/mbc.11.12.4241

    • Google Scholar
16. 1. Hu J
    2. Qin KR
    3. Xiang C
    4. Lee TH

    (2012) Modeling of hysteresis in gene regulatory networks

    Bulletin of Mathematical Biology 74:1727–1753.

    https://doi.org/10.1007/s11538-012-9733-1

    • Google Scholar
17. 1. Igoshin OA
    2. Price CW
    3. Savageau MA

    (2006) Signalling network with a bistable hysteretic switch controls developmental activation of the σF transcription factor in Bacillus subtilis

    Molecular Microbiology 61:165–184.

    https://doi.org/10.1111/j.1365-2958.2006.05212.x

    • Google Scholar
18. 1. Isaacs FJ
    2. Hasty J
    3. Cantor CR
    4. Collins JJ

    (2003) Prediction and measurement of an autoregulatory genetic module

    Proceedings of the National Academy of Sciences of USA 100:7714–7719.

    https://doi.org/10.1073/pnas.1332628100

    • Google Scholar
19. 1. Kotula JW
    2. Kerns SJ
    3. Shaket LA
    4. Siraj L
    5. Collins JJ
    6. Way JC
    7. Silver PA

    (2014) Programmable bacteria detect and record an environmental signal in the mammalian gut

    Proceedings of the National Academy of Sciences of USA 111:4838–4843.

    https://doi.org/10.1073/pnas.1321321111

    • Google Scholar
20. 1. Menck PJ
    2. Heitzig J
    3. Marwan N
    4. Kurths J

    (2013) How basin stability complements the linear-stability paradigm

    Nature Physics 9:89–92.

    https://doi.org/10.1038/nphys2516

    • Google Scholar
21. 1. Ozbudak EM
    2. Thattai M
    3. Lim HN
    4. Shraiman BI
    5. Van Oudenaarden A

    (2004) Multistability in the lactose utilization network of Escherichia coli

    Nature 427:737–740.

    https://doi.org/10.1038/nature02298

    • Google Scholar
22. 1. Pomerening JR
    2. Sontag ED
    3. Ferrell JE

    (2003) Building a cell cycle oscillator: hysteresis and bistability in the activation of Cdc2

    Nature Cell Biology 5:346–351.

    https://doi.org/10.1038/ncb954

    • Google Scholar
23. 1. Scheffer M
    2. Bascompte J
    3. Brock WA
    4. Brovkin V
    5. Carpenter SR
    6. Dakos V
    7. Held H
    8. van Nes EH
    9. Rietkerk M
    10. Sugihara G

    (2009) Early-warning signals for critical transitions

    Nature 461:53–59.

    https://doi.org/10.1038/nature08227

    • Google Scholar
24. 1. Song C
    2. Phenix H
    3. Abedi V
    4. Scott M
    5. Ingalls BP
    6. Kaern M
    7. Perkins TJ

    (2010) Estimating the stochastic bifurcation structure of cellular networks

    PLOS Computational Biology 6:e1000699.

    https://doi.org/10.1371/journal.pcbi.1000699

    • Google Scholar
25. 1. Van Nes EH
    2. Scheffer M

    (2007) Slow recovery from perturbations as a generic indicator of a nearby catastrophic shift

    The American Naturalist 169:738–747.

    https://doi.org/10.1086/516845

    • Google Scholar
26. 1. Williams K
    2. Savageau MA
    3. Blumenthal RM

    (2013) A bistable hysteretic switch in an activator–repressor regulated restriction–modification system

    Nucleic Acids Research 41:6045–6057.

    https://doi.org/10.1093/nar/gkt324

    • Google Scholar
27. 1. Wu M
    2. Su RQ
    3. Li X
    4. Ellis T
    5. Lai YC
    6. Wang X

    (2013) Engineering of regulated stochastic cell fate determination

    Proceedings of the National Academy of Sciences of USA 110:10610–10615.

    https://doi.org/10.1073/pnas.1305423110

    • Google Scholar
28. 1. Xiong W
    2. Ferrell JE

    (2003) A positive-feedback-based bistable ‘memory module’that governs a cell fate decision

    Nature 426:460–465.

    https://doi.org/10.1038/nature02089

    • Google Scholar

Author details

1. Kevin Axelrod

   Harvard University, Graduate Program in Biophysics, Cambridge, United States

   Contribution

   KA, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

2. Alvaro Sanchez

   Rowland Institute, Harvard University, Cambridge, United States

   Contribution

   AS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

3. Jeff Gore

   Department of Physics, Massachusetts Institute of Technology, Cambridge, United States

   Contribution

   JG, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   gore@mit.edu

   Competing interests

   The authors declare that no competing interests exist.

• 2,315

  views

• 465

  downloads

• 18

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.