Cascading signaling pathways improve the fidelity of a stochastically and deterministically simulated molecular RS latch
© Mills and Truong; licensee BioMed Central Ltd. 2009
Received: 17 February 2009
Accepted: 17 July 2009
Published: 17 July 2009
While biological systems have often been compared with digital systems, they differ by the strong effect of crosstalk between signals due to diffusivity in the medium, reaction kinetics and geometry. Memory elements have allowed the creation of autonomous digital systems and although biological systems have similar properties of autonomy, equivalent memory mechanisms remain elusive. Any such equivalent memory system, however, must silence the effect of crosstalk to maintain memory fidelity.
Here, we present a system of enzymatic reactions that behaves like an RS latch (a simple memory element in digital systems). Using both a stochastic molecular simulator and ordinary differential equation simulator, we showed that crosstalk between two latches operating in the same spatial localization disrupts the memory fidelity of both latches. Crosstalk was reduced or silenced when simple reaction loops were replaced with multiple step or cascading reactions, showing that cascading signaling pathways are less susceptible to crosstalk.
Thus, the common biological theme of cascading signaling pathways is advantageous for maintaining the fidelity of a memory latch in the presence of crosstalk. The experimental implementation of such a latch system will lead to novel approaches to cell control using synthetic proteins and will contribute to our understanding of why cells behave differently even when given the same stimulus.
Biological systems have been compared with digital systems, but such comparisons cannot be stretched too far as biological systems are affected by crosstalk between signals due to diffusivity in the medium, reaction kinetics and geometry [1–7]. Biological systems share some similarities with digital systems, such as basic logic functions and emergent network properties (i.e. memory and robustness) [1–4, 6, 7]. As demonstration, logic functions (AND, OR and NOT gates) have been constructed out of many biological molecules, such as DNA and proteins [1, 5, 6]. However, the analogy is limited by the basic issue of connectivity. Digital systems communicate between constituent parts using wires, providing direct connections between components with minimal interference. With biological systems, however, the biomolecule must diffuse through a common medium to find its binding partner, without confusing it with another similar partner. When biomolecular binding interactions are "confused" in this way, the result is crosstalk between distinct systems and pathways.
Crosstalk between proteins and signaling pathways introduces complexity and flexibility to cellular systems. However in the context of memory storage, there must be a high degree of fidelity between input signals and the various components of a system . Many enzymes have broad ranges of binding affinity for different substrates that allows them to affect a variety of cellular pathways. For example, receptor tyrosine kinases (RTKs) represent a family of enzymes that respond to different upstream signals (epidermal, fibroblast and platelet-derived growth factors), and yet can all result in the activation of the GTPase Ras to a varying degree . Consequently, at least three different input signals can all interact with the same effector protein producing crosstalk between various downstream pathways. However, it is undesirable for a memory system to be affected by molecules which are not being directly controlled by the desired, specific, upstream signal .
A stochastic biomolecular simulator based on previous models [12–22], as well as an ordinary differential equation simulator , was used to investigate a simple network of proteins that can replicate the functionality of an RS latch in the context of crosstalk. We chose to simulate crosstalk reactions using both methodologies because while classical deterministic modeling based on differential equations can efficiently simulate simple systems, their assumptions of spatial and temporal homogeneity are not always accurate in dynamic biological systems. Furthermore, in some systems the number of molecules being considered can be very low, thus resulting in a substantial degree of noise in a reaction network which cannot easily be handled with differential equations. In contrast, stochastic modeling can address both issues by allowing for the random movement of individual molecules in a particular location [1, 23]. Stochastic processes in biological systems modeling initially focused on gene expression [24–28], but have also been studied in other pathways such as metabolism and mitosis [29–32]. Thus, crosstalk was modeled as a set of enzymatic reactions between two RS latch systems. An enzyme from one RS latch system could react with the substrate of an analogous enzyme from the second system, but the reaction occurred at a reduced rate, usually 0.1 to 0.01 that of the analogous enzyme .
Results and discussion
RS latch behaviour resembles enzymatic reactions
An RS latch implemented in this way can be a model of a pair of inverse enzymatic reactions. That is, the enzyme R can be a kinase that phosphorylates the substrate Q to phospho-Q, or substrate P. The enzyme S is then a phosphorylase that converts substrate P back to substrate Q. In this way, the simplest biological memory unit is the functional state (in this model, the phosphorylation state specifically) of a particular molecule. More complex states, such as phenotypes, can be built up of many molecular states in the same way an FSM is built of many RS latches. In the biological context, it could be possible for both a phosphorylase and kinase to be available to act on a substrate at the same time. However, this conflicts with the invalid input R = S = 1 for an RS latch. Going forward, we will assume that upstream signals that control phosphorylation and dephosphorylation will suppress this possibility thereby maintaining a clearly defined input state for our model.
Crosstalk requires modification to latch design
To evaluate the effect of crosstalk between latches and our attempts to silence it, simulations were performed throughout the paper under three scenarios, each compared with the ideal output (Figure 4A): no crosstalk (Figure 4B), minimal crosstalk with kcrosstalk = 0.01kprimary (Figure 4C) and moderate crosstalk with kcrosstalk = 0.1kprimary (Figure 4D). Under the no crosstalk scenario, each latch behaved ideally (Figure 4B). The continued presence of a non-zero time constant is a physical reality and cannot be eliminated. The initial conditions were set to a random state, represented by each output being 50% of its maximal potential output. When the first input was provided (S = 1 and B = 1), the outputs quickly approached their expected values of P = 0 and C = 1. Once a value was achieved, it was maintained until the next input signal arrived at 1000 ms.
Under the minimal crosstalk scenario, there were three important changes in the system's behaviour, which were revealed by one or both simulation methods (Figure 4C). First, the output signals P and C did not reach their new values as quickly (both methods) or as smoothly (stochastic method only) as with no crosstalk. Second, once the steady state value for P or C was achieved, it was not as complete a signal as with no crosstalk (both methods) and there was fluctuation in the output (stochastic method only). With both simulation methods, the desired output for the latches from time points 0 to 1000 ms were no longer P = 0 and C = 1 but rather approximately P = 0.02 and C = 0.98. The fluctuations in output were evident in only the stochastic simulations: for example, from time points 0 ms to 2000 ms. During this time period, the latches were assuming opposite values. As a result, once one latch achieved close to 100% of its desired output, an opposing enzyme from the other latch could temporarily alter a substrate into the undesired state. The temporal nature of this interaction was lost in the deterministic modeling. A third consequence of crosstalk was a partial loss of the substrate C signal between 4000 ms and 5000 ms because one system is passively holding its output while the other is actively outputting the opposite value (both methods). In this period one system, SRPQ, was being actively set to the substrate P = 0 state while the ABCD system was in the "hold output" state for substrate C = 1. Essentially, the SRPQ system was overriding the information that was stored in the ABCD system.
Under the moderate crosstalk scenario, the latches showed an exaggeration of the trends established in the minimal crosstalk example, such as reduced steady state signal (now P = 0.10 and C = 0.90) and output fluctuations (Figure 4D). Notably, the loss of substrate C signal from 4000 ms to 5000 ms increased. In this situation, substrate C dropped below 0.5 and so the ABCD system outputted substrate D = 1 by the end of the relevant time period. Hereafter, this phenomenon will be referred to as "forced state switching". This is a serious flaw in the latch system and must be corrected to ensure a high fidelity memory system. To improve the memory storage fidelity of the two latch systems, any solution must reduce forced state switching and the fluctuations in output values.
Reaction modifications to silence the effect of crosstalk
The crosstalk in a pair of two-step latches was implemented similarly to the original one step latch (Figure 5). Now, there was additional crosstalk between enzymes S' and A', shown in rounded rectangles, and the enzymes R' and B', shown in ellipses. Crosstalk is biologically motivated using the same principles as with a one-step loop: phosphorylation and dephosphorylation can be catalyzed by a range of kinases and phosphatases on a variety of substrates with different kinetics. This idea is easily extrapolated for a three-step latch.
Under the minimal crosstalk scenario, a pair of two-step (Figure 6A) or three-step latches (Figure 6B) eliminated the effect of crosstalk by reducing fluctuations and minimizing forced state switching (previously observed in Figure 4B). However, the time constant increased considerably when compared with the no crosstalk scenario using a one-step latch (Figure 4A). These trends were noticeable in both the stochastic and deterministic simulations. Essentially, the second and third steps slowed the reaction by requiring more collisions between substrates and enzymes.
Under the moderate crosstalk scenario, the two-step latch showed some improvement in fidelity (Figure 6C), however the improvements were more noticeable with three steps (Figure 6D). In this scenario, significant improvement in fluctuations, steady state output and forced state switching was not seen until a third step was added, suggesting that cascading reaction pathways are able to transmit signals with a higher fidelity. Furthermore, the forced state switching from substrate C = 1 to substrate D = 1 between 4000 ms and 5000 ms was prevented (Figure 6D). However, there were two discrepancies between the simulation methods. First, the deterministic simulations showed an overshoot past the steady state value for both the two step and three steps loops that is not present with the stochastic simulations. Second, the deterministic simulations suggest that the three step loop is not as successful at restoring the steady state output as was suggested by the stochastic simulations (deterministic predicts C = 0.92 versus C = 0.96 for stochastic from time 0 to 1000 ms). These discrepancies could be due to the fluctuations present in the stochastic simulations that are not present in the deterministic simulations.
The effective output state of the system was re-defined for the multiple step latches: for the two-step loops, the outputs were the sum of substrates P + P', and similarly for the other substrates Q, C, and D; for three-step loops, the sum of substrates P + P' + P". Essentially, the additional steps in the latch buffered the output molecules and limited the effect of the crosstalk not on any one molecule, but the reaction pathway as a whole. Thus, each of the molecular species was summed because it was part of the same pathway and had a similar functionality. To see a biological basis for this, consider the MAPK example again. Three distinct molecules make one half of the reaction pathway (the Q, Q' and Q" half): unphosphorylated MAPK, p-Tyr MAPK and p-Thr MAPK. All three molecules represent inactive forms of MAPK, and so summing them represents an overall state rather than a specific molecule. The summing operation is a way of reconstructing the state from a collection of distinct molecules. The second, activated, state of MAPK is only achieved when MAPK is doubly phosphorylated [34, 35], which is not included in the sum, and resides in the other half of the reaction loop. Given the model of a memory state as a molecule switching between its phosphorylated and de-phosphorylated states, it is likely that a molecule can only exist in two functional states (on or off). However, such a model could only apply to situations where the functionalities of each of the intermediate molecules in one half of the reaction loop are the same and would preclude a molecule being in a transition state where it was either partially active or had some functionality unrelated to either end state.
Both simulation methods showed that the time constant of the switch was longer for reactions with more steps (Figure 6). This suggested that cascading pathways had two competing factors: longer cascading pathways transmitted higher fidelity signals, but required more time to reach a steady state. This effect was more exaggerated in the stochastic simulations compared to the deterministic simulations even though the steady state values were similar. Given that the stochastic simulations provided more information on temporal details in previous simulations (Figure 4), it is likely that the stochastic simulation is showing a more accurate representation of the system in its early stages. The compromise between speed and other biological networks parameters has been noted in previous work such as negative feedback loops that reduce rise times at the expense of steady state signal  and reaction conditions that improve signal fidelity and specificity at the expense of speed using compartmentalization  or low-affinity scaffolding . While speed is sacrificed in our simulations to improve fidelity, biological mechanisms exist that can speed the reaction network without loss of fidelity such as high-affinity scaffolding in the case of the MAPK cascade .
Overall, this work provides a basis for interpreting biochemical signaling cascades in terms of biological memory and the limitations placed on this by crosstalk. First, we have shown that it is possible to consider a very simple enzymatic system (any two complementary enzymes such as a kinase and phosphatase) as a memory storage unit based on the functional state of their common substrate. However, without modification, the initial memory storage system failed under conditions of crosstalk. Crosstalk, which is used by cells to both reduce the number of different proteins necessary for signaling cascades as well as to add complexity to signal regulation, has a significant parasitic effect on the storage of memory with even two systems present. We have shown that the parasitic effects of crosstalk can be silenced by cascading several components to create one long multistep cascade. This finding provides insight into the length of biological systems as well as the tradeoff between network parameters that we have discussed above. The ability of cells to store memory allows them to interpret future signals appropriately by combining these inputs with past information . A model of biological memory will improve our ability to interpret how cells respond to signals as they do, and may provide insights into their rational manipulation.
This study used stochastic and deterministic modeling to show that a system of two reciprocal enzymes that toggles a substrate between two states can resemble a memory element, specifically an RS latch. However in a cellular environment, crosstalk between similar enzymes receiving different upstream signals will likely interfere with any signal transduction or memory storage system. This was shown to be the case when only two latch systems were operated in the same spatial localization with a moderate degree of crosstalk. Increasing the number of steps in the latches improved the fidelity of two latch systems. However, additional steps also increased the switching time constant thus slowing the system response. This may provide a basis for explaining the length of biological reaction cascades as a compromise between demands for fidelity and speed of response. Further modeling of biological systems' outputs based on known inputs as well as current states stored in molecular memory will aid in understanding why cells behave differently even when given the same stimulus. Implementing such a latch system in vitro to prove its feasibility may also lead to novel approaches to cell control using synthetic proteins that can be made to behave as a memory system.
Stochastic and deterministic simulators used
A stochastic simulator was developed for this study based on previous stochastic simulators [12–22]. The stochastic simulator tracks the location of each molecule created and allows it to perform a random walk (diffuse) through space. Collisions between molecules are the basis of mass action reactions. The stochastic simulator accepts user scripts defining a set of molecules with a particular size and diffusion coefficient. Then, a set of reactions is modelled as some combination of association between two molecules with a rate, kf, or dissociation of one molecule into others with a rate, kr. The user can then specify time points for the addition or removal of molecules.
Deterministic simulations were carried out using Dynetica .
Crosstalk was incorporated into the simulation as follows, based on previous definitions [4, 7, 8]. Crosstalk, essentially a type of interference, is only relevant when two or more systems are present. Assume that there are two systems, one termed SRPQ with four elements, S, R, P and Q and an identical one termed ABCD with elements A, B, C and D. Assume that in SRPQ there is one reaction, which is S + P → SP → S + Q at a rate kprimary = 104. In the second system there is an equivalent reaction A + C → AC → A + D at the same rate. Crosstalk is then incorporated as an additional layer between the two systems. The exact way that the crosstalk reactions are modeled depends on the nature of the model. For example, if S and A are enzymes, Q and D are substrates and P and C are reaction products, then the cross-talk reactions would be S + C → SC → S + D at a rate kcrosstalk = 103 and A + P → AP → A + Q at the same reduced rate [4, 7]. Usually kcrosstalk = 0.1 kprimary (moderate crosstalk) or 0.01 kprimary (minimal crosstalk).
This work was supported by a fellowship to EM from Natural Science and Engineering Research Council (NSERC) and grants to KT from the Canadian Foundation of Innovation (#10296), Canadian Institutes of Health Research (#81262), Heart and Stroke Foundation (#NA6241) and the Natural Science and Engineering Research Council (#283170).
- Pham E, Li I, Truong K: Computational modeling approaches for studying of synthetic biological networks. Current Bioinformatics. 2008, 3: 130-141. 10.2174/157489308784340667.View ArticleGoogle Scholar
- McAdams HH, Shapiro L: Circuit simulation of genetic networks. Science. 1995, 269: 650-656. 10.1126/science.7624793View ArticlePubMedGoogle Scholar
- Milo R, Shen-Orr S, Itzkovitz S, Kashtan N, Chklovskii D, Alon U: Network motifs: simple building blocks of complex networks. Science. 2002, 298: 824-827. 10.1126/science.298.5594.824View ArticlePubMedGoogle Scholar
- Bhalla US, Iyengar R: Emergent properties of networks of biological signaling pathways. Science. 1999, 283: 381-387. 10.1126/science.283.5400.381View ArticlePubMedGoogle Scholar
- Unger R, Moult J: Towards computing with proteins. Proteins. 2006, 63: 53-64. 10.1002/prot.20886View ArticlePubMedGoogle Scholar
- Bray D: Protein molecules as computational elements in living cells. Nature. 1995, 376: 307-312. 10.1038/376307a0View ArticlePubMedGoogle Scholar
- Weng G, Bhalla US, Iyengar R: Complexity in biological signaling systems. Science. 1999, 284: 92-96. 10.1126/science.284.5411.92PubMed CentralView ArticlePubMedGoogle Scholar
- Komarova NL, Zou X, Nie Q, Bardwell L: A theoretical framework for specificity in cell signaling. Mol Syst Biol. 2005, 1: 2005.0023 10.1038/msb4100031PubMed CentralView ArticlePubMedGoogle Scholar
- Novak B, Tyson JJ: A model for restriction point control of the mammalian cell cycle. J Theor Biol. 2004, 230: 563-579. 10.1016/j.jtbi.2004.04.039View ArticlePubMedGoogle Scholar
- Ramakrishnan N, Bhalla US: Memory switches in chemical reaction space. PLoS Comput Biol. 2008, 4: e1000122- 10.1371/journal.pcbi.1000122PubMed CentralView ArticlePubMedGoogle Scholar
- Simon MA: Receptor tyrosine kinases: specific outcomes from general signals. Cell. 2000, 103: 13-15. 10.1016/S0092-8674(00)00100-8View ArticlePubMedGoogle Scholar
- Sedwards S, Mazza T: Cyto-Sim: a formal language model and stochastic simulator of membrane-enclosed biochemical processes. Bioinformatics. 2007, 23: 2800-2802. 10.1093/bioinformatics/btm416View ArticlePubMedGoogle Scholar
- You L, Hoonlor A, Yin J: Modeling biological systems using Dynetica – a simulator of dynamic networks. Bioinformatics. 2003, 19: 435-436. 10.1093/bioinformatics/btg009View ArticlePubMedGoogle Scholar
- Le Novere N, Shimizu TS: STOCHSIM: modelling of stochastic biomolecular processes. Bioinformatics. 2001, 17: 575-576. 10.1093/bioinformatics/17.6.575View ArticlePubMedGoogle Scholar
- Lok L, Brent R: Automatic generation of cellular reaction networks with Moleculizer 1.0. Nat Biotechnol. 2005, 23: 131-136. 10.1038/nbt1054View ArticlePubMedGoogle Scholar
- Hoops S, Sahle S, Gauges R, Lee C, Pahle J, Simus N, Singhal M, Xu L, Mendes P, Kummer U: COPASI – a COmplex PAthway SImulator. Bioinformatics. 2006, 22: 3067-3074. 10.1093/bioinformatics/btl485View ArticlePubMedGoogle Scholar
- Boulianne L, Al Assaad S, Dumontier M, Gross WJ: GridCell: a stochastic particle-based biological system simulator. BMC Syst Biol. 2008, 2: 66- 10.1186/1752-0509-2-66PubMed CentralView ArticlePubMedGoogle Scholar
- Mao L, Resat H: Probabilistic representation of gene regulatory networks. Bioinformatics. 2004, 20: 2258-2269. 10.1093/bioinformatics/bth236View ArticlePubMedGoogle Scholar
- Meng TC, Somani S, Dhar P: Modeling and simulation of biological systems with stochasticity. In Silico Biol. 2004, 4: 293-309.PubMedGoogle Scholar
- Sanford C, Yip ML, White C, Parkinson J: Cell++ – simulating biochemical pathways. Bioinformatics. 2006, 22: 2918-2925. 10.1093/bioinformatics/btl497View ArticlePubMedGoogle Scholar
- Adalsteinsson D, McMillen D, Elston TC: Biochemical Network Stochastic Simulator (BioNetS): software for stochastic modeling of biochemical networks. BMC Bioinformatics. 2004, 5: 24- 10.1186/1471-2105-5-24PubMed CentralView ArticlePubMedGoogle Scholar
- Ohki N, Hagiwara M: Bio-Object, a stochastic simulator for post-transcriptional regulation. Bioinformatics. 2005, 21: 2478-2487. 10.1093/bioinformatics/bti316View ArticlePubMedGoogle Scholar
- Gillespie DT: Stochastic simulation of chemical kinetics. Annu Rev Phys Chem. 2007, 58: 35-55. 10.1146/annurev.physchem.58.032806.104637View ArticlePubMedGoogle Scholar
- Elowitz MB, Leibler S: A synthetic oscillatory network of transcriptional regulators. Nature. 2000, 403: 335-338. 10.1038/35002125View ArticlePubMedGoogle Scholar
- Blake WJ, M KA, Cantor CR, Collins JJ: Noise in eukaryotic gene expression. Nature. 2003, 422: 633-637. 10.1038/nature01546View ArticlePubMedGoogle Scholar
- Elowitz MB, Levine AJ, Siggia ED, Swain PS: Stochastic gene expression in a single cell. Science. 2002, 297: 1183-1186. 10.1126/science.1070919View ArticlePubMedGoogle Scholar
- McAdams HH, Arkin A: Stochastic mechanisms in gene expression. Proc Natl Acad Sci USA. 1997, 94: 814-819. 10.1073/pnas.94.3.814PubMed CentralView ArticlePubMedGoogle Scholar
- Weinberger LS, Burnett JC, Toettcher JE, Arkin AP, Schaffer DV: Stochastic gene expression in a lentiviral positive-feedback loop: HIV-1 Tat fluctuations drive phenotypic diversity. Cell. 2005, 122: 169-182. 10.1016/j.cell.2005.06.006View ArticlePubMedGoogle Scholar
- Arkin A, Ross J, McAdams HH: Stochastic kinetic analysis of developmental pathway bifurcation in phage lambda-infected Escherichia coli cells. Genetics. 1998, 149: 1633-1648.PubMed CentralPubMedGoogle Scholar
- Borisuk MT, Tyson JJ: Bifurcation analysis of a model of mitotic control in frog eggs. J Theor Biol. 1998, 195: 69-85. 10.1006/jtbi.1998.0781View ArticlePubMedGoogle Scholar
- Chen KC, Csikasz-Nagy A, Gyorffy B, Val J, Novak B, Tyson JJ: Kinetic analysis of a molecular model of the budding yeast cell cycle. Mol Biol Cell. 2000, 11: 369-391.PubMed CentralView ArticlePubMedGoogle Scholar
- Edwards JS, Ibarra RU, Palsson BO: In silico predictions of Escherichia coli metabolic capabilities are consistent with experimental data. Nat Biotechnol. 2001, 19: 125-130. 10.1038/84379View ArticlePubMedGoogle Scholar
- Sivakumaran S, Hariharaputran S, Mishra J, Bhalla US: The Database of Quantitative Cellular Signaling: management and analysis of chemical kinetic models of signaling networks. Bioinformatics. 2003, 19: 408-415. 10.1093/bioinformatics/btf860View ArticlePubMedGoogle Scholar
- Zhang YY, Mei ZQ, Wu JW, Wang ZX: Enzymatic activity and substrate specificity of mitogen-activated protein kinase p38 alpha in different phosphorylation states. Journal of Biological Chemistry. 2008, 283: 26591-26601. 10.1074/jbc.M801703200PubMed CentralView ArticlePubMedGoogle Scholar
- Camps M, Nichols A, Arkinstall S: Dual specificity phosphatases: a gene family for control of MAP kinase function. FASEB J. 2000, 14: 6-16.PubMedGoogle Scholar
- Rosenfeld N, Elowitz MB, Alon U: Negative autoregulation speeds the response times of transcription networks. J Mol Biol. 2002, 323: 785-793. 10.1016/S0022-2836(02)00994-4View ArticlePubMedGoogle Scholar
- Bardwell L, Zou X, Nie Q, Komarova NL: Mathematical models of specificity in cell signaling. Biophys J. 2007, 92: 3425-3441. 10.1529/biophysj.106.090084PubMed CentralView ArticlePubMedGoogle Scholar
- Zou X, Peng T, Pan Z: Modeling specificity in the yeast MAPK signaling networks. J Theor Biol. 2008, 250: 139-155. 10.1016/j.jtbi.2007.09.024View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.