- Email: [email protected]
Cell Lineage Trees Bear Fruit
Cell Systems
Previews of enzymes and pumps to maximize production while minimizing toxicity (Stevens and Carothers, 2015). With the addition of the work by Wang et al. (2016), RNA-based circuits are quickly matching the capabilities of protein-based circuitry. However, there are still challenges to overcome. One of the major challenges is the creation of ligand-switchable RNA regulators for ligands relevant to real-world applications such as metabolic engineering, diagnostics, and intracellular sensing. RNA aptamers are abundant in natural systems and can now be reliably evolved to sense a myriad of metabolites, cofactors, metals, ions, and macromolecules, with stunning specificity and sensitivity. Yet the vast majority of synthetic RNA switches have utilized only a handful of well-characterized aptamers, partly because like ribozymes, aptamers often make use of complex RNA-ligand interactions that are hard to measure and model computationally. While this challenge can be circumvented by high-throughput
screening of large variant libraries (Townshend et al., 2015), significant progress is being made in the quest for RNA design. For example, incorporating the thermodynamics of ligand binding into RNA design algorithms (Espah Borujeni et al., 2016) and using more versatile ribozymes (Felletti et al., 2016) hold promise for improving the reliability and efficacy of ligand-switchable RNA regulators. Moreover, given the inherent modularity of the innovations described by Wang et al. (2016), even if new RNA switches suffer from a low dynamic range, they can now be significantly improved, helping deliver on the promise of RNA circuitry.
Green, A.A., Silver, P.A., Collins, J.J., and Yin, P. (2014). Cell 159, 925–939.
REFERENCES
Wang, B., Barahona, M., and Buck, M. (2014). Nucleic Acids Res. 42, 9484–9492.
Chappell, J., Takahashi, M.K., and Lucks, J.B. (2015). Nat. Chem. Biol. 11, 214–220. Espah Borujeni, A., Mishler, D.M., Wang, J., Huso, W., and Salis, H.M. (2016). Nucleic Acids Res. 44, 1–13. Felletti, M., Stifel, J., Wurmthaler, L.A., Geiger, S., and Hartig, J.S. (2016). Nat. Commun. 7, 12834.
Isaacs, F.J., Dwyer, D.J., Ding, C., Pervouchine, D.D., Cantor, C.R., and Collins, J.J. (2004). Nat. Biotechnol. 22, 841–847. Stanton, B.C., Nielsen, A.A.K., Tamsir, A., Clancy, K., Peterson, T., and Voigt, C.A. (2014). Nat. Chem. Biol. 10, 99–105. Stevens, J.T., and Carothers, J.M. (2015). ACS Synth. Biol. 4, 107–115. Takahashi, M.K., Chappell, J., Hayes, C.A., Sun, Z.Z., Kim, J., Singhal, V., Spring, K.J., Al-Khabouri, S., Fall, C.P., Noireaux, V., et al. (2015). ACS Synth. Biol. 4, 503–515. Townshend, B., Kennedy, A.B., Xiang, J.S., and Smolke, C.D. (2015). Nat. Methods 12, 989–994.
Wang, Y.H., McKeague, M., Hsu, T.M., and Smolke, C.D. (2016). Cell Syst. 3, this issue, 549–562. Zadeh, J.N., Steenberg, C.D., Bois, J.S., Wolfe, B.R., Pierce, M.B., Khan, A.R., Dirks, R.M., and Pierce, N.A. (2011). J. Comput. Chem. 32, 170–173.
Cell Lineage Trees Bear Fruit Jordi Garcia-Ojalvo1,* 1Department of Experimental and Health Sciences, Universitat Pompeu Fabra, Barcelona Biomedical Research Park, Barcelona 08003, Spain *Correspondence: [email protected] http://dx.doi.org/10.1016/j.cels.2016.12.006
Two inference approaches harness the information present in cell lineage trees to better understand the dynamic transitions between cell states. Living cells are dynamic entities, with proteins and other biomolecules being produced and decaying constantly at rates that frequently vary in time. Many biological processes even involve cells switching between distinct cellular states as time goes by. This dynamic character, together with the unavoidable heterogeneity of cellular populations, requires monitoring the activity of gene regulatory circuits in time and at the level of single cells. Time-lapse microscopy of cells carrying fluorescent reporters is routinely used for this purpose (Locke and Elowitz, 2009), but this method suf-
fers from several shortcomings that limit its usefulness. First, only a handful of species (at most three or four) can be monitored simultaneously due to spectral overlap, which leads to a partial view of the process under study. Second, sampling frequency is limited due to photobleaching and phototoxicity of the fluorescent reporters, leaving time gaps during which the behavior of the system is unknown. Third, the combination of measurement errors and biological noise leads to significant uncertainty in the observations.
The situation would seem to worsen in populations of proliferating cells due to the amplification in the number of variables to measure, resulting from the exponential increase in cell number. However, as shown in the November issue of Cell Systems (Feigelman et al., 2016; Hormoz et al., 2016), the genealogical structure of proliferating cell populations, reflected in their lineage trees, provides constraints that can in fact relieve some of the limitations of timelapse fluorescence microscopy, substantially enhancing its ability to shed light on the molecular and functional
Cell Systems 3, December 21, 2016 ª 2016 Elsevier Inc. 511
Cell Systems
Previews
Figure 1. Cell Lineages Impose Correlations between Related Cells (Top) that Are Reflected in the Concentrations of Key Molecules across Time (Bottom)
mechanisms regulating cellular processes. Coincidentally, both studies demonstrate the power of lineage trees in the same model system, namely embryonic stem (ES) cells. The molecular circuitry regulating the establishment and maintenance of pluripotency in ES cells is still under investigation. One of the core pluripotency factors is Nanog, a protein whose expression is markedly heterogeneous and dynamical (Kalmar et al., 2009; Singer et al., 2014). Nanog is known to autoregulate its own expression, but the exact nature of this feedback (in particular, whether it is positive or negative) is still under debate. Feigelman et al. (2016) set out to address this issue by applying Bayesian inference on time-lapse data of Nanog expression. Going beyond previous work in which stochastic inference was applied to singlecell trajectories (Zechner et al., 2011), Feigelman et al. (2016) use the genealogic organization of the cellular population (Figure 1, top) to limit the set of plausible model parameters that are compatible with the full dynamics of the lineage tree. The approach is based on the appearance of temporal cross-correlations in the dynamics of cells related by a common ancestry (Figure 1, bottom), resulting from daughter cells inheriting the cellular state of their mothers. To represent those constraints, Feigelman et al. (2016) extend a recursive particle filter (Zechner et al., 2011) with a model of cell division. First, the authors validate their algorithm (which they term Stochastic Inference on Lineage Trees, or STILT) using synthetic data generated 512 Cell Systems 3, December 21, 2016
by three simple models of Nanog expression with different feedback structures (positive, negative, and no feedback). Next, they apply their method to previously published time-lapse data (Filipczyk et al., 2015). A goodness-of-fit test allows them to reject positive feedback as a possible mechanism of Nanog autoregulation due to its inability to explain the experimental data. In order to distinguish between the remaining two alternatives, negative feedback and no regulation, the authors use the inferred parameters of those models to predict the response of the system to exogenous Nanog. Flow cytometry reveals a downregulation of endogenous Nanog production for increasing exogenous Nanog expression, which can only be reproduced quantitatively by the negative feedback model. This result contrasts with the prevalent view of pluripotency regulation, according to which Nanog activates its own expression. The study thus offers a novel approach to using lineage information for resolving the molecular circuitry underlying dynamic phenotypes in populations of proliferating cells. Time-lapse fluorescence microscopy is also limited by the difficulty of generating cell lines with the appropriate reporters, which can still be a fairly timeconsuming task. Interestingly, Hormoz et al. (2016) show that lineage information, combined with end-point measurements of cellular states throughout the colony (which does not require fluorescent reporters), can be sufficient to characterize the dynamical properties of cellular processes. In particular, they study the transition dynamics between multiple phenotypically distinct ES cells in normal growth conditions (LIF+serum). While these states are known to underlie pluripotency in ES cells, the transitions among them are not well understood, since measuring switching rates seemingly requires tracking the state of individual cells over time. But Hormoz et al. (2016) show that this is not necessary: switching rates can be calculated by identifying the proliferating cells through direct time-lapse imaging and measuring the final state of the colony with multicolor single-molecule RNA-FISH, with no fluorescent reporters being necessarily involved at any time. Their approach, which they term kin correlation analysis (KCA), builds upon
previous work in bacterial colonies (Hormoz et al., 2015) and is originally inspired by a cosmological multiverse model (Harlow et al., 2012). In contrast with the approach of Feigelman et al. (2016), which exploits the temporal cross-correlations between cells that share a common ancestry, the method of Hormoz et al. (2016) uses the endpoint correlations between those cells (Figure 1, top), which can be calculated from the transition matrix linking the different states. This relationship can be inverted, which enables the transition rates to be inferred from the end-point correlations. Hormoz et al. (2016) validate their approach using time-lapse fluorescence data on the promoter activity of the Esrrb gene, which exhibits a bistable profile. The switching rates obtained with the KCA approach are consistent with direct measurements using the fluorescent reporter. Next, they extend their method to five distinct cellular states defined by the levels of three proteins, Tbx3, Zscan4, and the above-mentioned Esrrb. After verifying that these states have different morphological and functional (as measured by RNA-seq) properties, the authors use KCA to uncover a state transition network in the form of a linear chain, in which cells switch stochastically and reversibly between adjacent states. This chain orders the states in a sequence ranging from a totipotent-like state to the more differentiated epiblastlike state. The finding is relevant for our understanding of pluripotency, since it provides us with a highly structured perspective of ES heterogeneity, which contrasts with the widespread view of this regime as a random walk among all ES cell states. This result highlights the potential of using lineage information to characterize the functional aspects of cellular regulation. While the method was originally devised for reversible transitions and stationary dynamics, the authors generalize it to the analysis of irreversible and non-stationary dynamics. Notably, they exclude the possibility of cell-cell communication playing a role in the transition dynamics. It would be interesting to extend this method to situations in which cell-cell signaling modulates the switching rates, and thus the kin correlations.
Cell Systems
Previews Together, these two papers open up new avenues in our ability to measure and interpret the dynamical behavior of regulatory networks at the single-cell level at a time in which the roles of dynamics and heterogeneity in cells are becoming increasingly more evident. It is time to climb up the cell lineage tree.
REFERENCES Feigelman, J., Ganscha, S., Hastreiter, S., Schwarzfischer, M., Filipczyk, A., Schroeder, T.,
Theis, F.J., Marr, C., and Claassen, M. (2016). Cell Syst. 3, 480–490.e13.
Kalmar, T., Lim, C., Hayward, P., Mun˜oz-Descalzo, S., Nichols, J., Garcia-Ojalvo, J., and Martinez Arias, A. (2009). PLoS Biol. 7, e1000149.
Filipczyk, A., Marr, C., Hastreiter, S., Feigelman, J., Schwarzfischer, M., Hoppe, P.S., Loeffler, D., Kokkaliaris, K.D., Endele, M., Schauberger, B., et al. (2015). Nat. Cell Biol. 17, 1235–1246.
Locke, J.C., and Elowitz, M.B. (2009). Nat. Rev. Microbiol. 7, 383–392.
Harlow, D., Shenker, S.H., Stanford, D., and Susskind, L. (2012). Phys. Rev. D Part. Fields Gravit. Cosmol. 85, 063516.
Singer, Z.S., Yong, J., Tischler, J., Hackett, J.A., Altinok, A., Surani, M.A., Cai, L., and Elowitz, M.B. (2014). Mol. Cell 55, 319–331.
Hormoz, S., Desprat, N., and Shraiman, B.I. (2015). Proc. Natl. Acad. Sci. USA 112, E2281–E2289.
Zechner, C., Pelet, S., Peter, M., and Koeppl, H. (2011). Recursive Bayesian estimation of stochastic rate constants from heterogeneous cell populations. In 2011 50th IEEE Conference on Decision and Control and European Control Conference, pp. 5837–5843.
Hormoz, S., Singer, Z.S., Linton, J.M., Antebi, Y.E., Shraiman, B.I., and Elowitz, M.B. (2016). Cell Syst. 3, 419–433.e8.
Cell Systems 3, December 21, 2016 513
Previews of enzymes and pumps to maximize production while minimizing toxicity (Stevens and Carothers, 2015). With the addition of the work by Wang et al. (2016), RNA-based circuits are quickly matching the capabilities of protein-based circuitry. However, there are still challenges to overcome. One of the major challenges is the creation of ligand-switchable RNA regulators for ligands relevant to real-world applications such as metabolic engineering, diagnostics, and intracellular sensing. RNA aptamers are abundant in natural systems and can now be reliably evolved to sense a myriad of metabolites, cofactors, metals, ions, and macromolecules, with stunning specificity and sensitivity. Yet the vast majority of synthetic RNA switches have utilized only a handful of well-characterized aptamers, partly because like ribozymes, aptamers often make use of complex RNA-ligand interactions that are hard to measure and model computationally. While this challenge can be circumvented by high-throughput
screening of large variant libraries (Townshend et al., 2015), significant progress is being made in the quest for RNA design. For example, incorporating the thermodynamics of ligand binding into RNA design algorithms (Espah Borujeni et al., 2016) and using more versatile ribozymes (Felletti et al., 2016) hold promise for improving the reliability and efficacy of ligand-switchable RNA regulators. Moreover, given the inherent modularity of the innovations described by Wang et al. (2016), even if new RNA switches suffer from a low dynamic range, they can now be significantly improved, helping deliver on the promise of RNA circuitry.
Green, A.A., Silver, P.A., Collins, J.J., and Yin, P. (2014). Cell 159, 925–939.
REFERENCES
Wang, B., Barahona, M., and Buck, M. (2014). Nucleic Acids Res. 42, 9484–9492.
Chappell, J., Takahashi, M.K., and Lucks, J.B. (2015). Nat. Chem. Biol. 11, 214–220. Espah Borujeni, A., Mishler, D.M., Wang, J., Huso, W., and Salis, H.M. (2016). Nucleic Acids Res. 44, 1–13. Felletti, M., Stifel, J., Wurmthaler, L.A., Geiger, S., and Hartig, J.S. (2016). Nat. Commun. 7, 12834.
Isaacs, F.J., Dwyer, D.J., Ding, C., Pervouchine, D.D., Cantor, C.R., and Collins, J.J. (2004). Nat. Biotechnol. 22, 841–847. Stanton, B.C., Nielsen, A.A.K., Tamsir, A., Clancy, K., Peterson, T., and Voigt, C.A. (2014). Nat. Chem. Biol. 10, 99–105. Stevens, J.T., and Carothers, J.M. (2015). ACS Synth. Biol. 4, 107–115. Takahashi, M.K., Chappell, J., Hayes, C.A., Sun, Z.Z., Kim, J., Singhal, V., Spring, K.J., Al-Khabouri, S., Fall, C.P., Noireaux, V., et al. (2015). ACS Synth. Biol. 4, 503–515. Townshend, B., Kennedy, A.B., Xiang, J.S., and Smolke, C.D. (2015). Nat. Methods 12, 989–994.
Wang, Y.H., McKeague, M., Hsu, T.M., and Smolke, C.D. (2016). Cell Syst. 3, this issue, 549–562. Zadeh, J.N., Steenberg, C.D., Bois, J.S., Wolfe, B.R., Pierce, M.B., Khan, A.R., Dirks, R.M., and Pierce, N.A. (2011). J. Comput. Chem. 32, 170–173.
Cell Lineage Trees Bear Fruit Jordi Garcia-Ojalvo1,* 1Department of Experimental and Health Sciences, Universitat Pompeu Fabra, Barcelona Biomedical Research Park, Barcelona 08003, Spain *Correspondence: [email protected] http://dx.doi.org/10.1016/j.cels.2016.12.006
Two inference approaches harness the information present in cell lineage trees to better understand the dynamic transitions between cell states. Living cells are dynamic entities, with proteins and other biomolecules being produced and decaying constantly at rates that frequently vary in time. Many biological processes even involve cells switching between distinct cellular states as time goes by. This dynamic character, together with the unavoidable heterogeneity of cellular populations, requires monitoring the activity of gene regulatory circuits in time and at the level of single cells. Time-lapse microscopy of cells carrying fluorescent reporters is routinely used for this purpose (Locke and Elowitz, 2009), but this method suf-
fers from several shortcomings that limit its usefulness. First, only a handful of species (at most three or four) can be monitored simultaneously due to spectral overlap, which leads to a partial view of the process under study. Second, sampling frequency is limited due to photobleaching and phototoxicity of the fluorescent reporters, leaving time gaps during which the behavior of the system is unknown. Third, the combination of measurement errors and biological noise leads to significant uncertainty in the observations.
The situation would seem to worsen in populations of proliferating cells due to the amplification in the number of variables to measure, resulting from the exponential increase in cell number. However, as shown in the November issue of Cell Systems (Feigelman et al., 2016; Hormoz et al., 2016), the genealogical structure of proliferating cell populations, reflected in their lineage trees, provides constraints that can in fact relieve some of the limitations of timelapse fluorescence microscopy, substantially enhancing its ability to shed light on the molecular and functional
Cell Systems 3, December 21, 2016 ª 2016 Elsevier Inc. 511
Cell Systems
Previews
Figure 1. Cell Lineages Impose Correlations between Related Cells (Top) that Are Reflected in the Concentrations of Key Molecules across Time (Bottom)
mechanisms regulating cellular processes. Coincidentally, both studies demonstrate the power of lineage trees in the same model system, namely embryonic stem (ES) cells. The molecular circuitry regulating the establishment and maintenance of pluripotency in ES cells is still under investigation. One of the core pluripotency factors is Nanog, a protein whose expression is markedly heterogeneous and dynamical (Kalmar et al., 2009; Singer et al., 2014). Nanog is known to autoregulate its own expression, but the exact nature of this feedback (in particular, whether it is positive or negative) is still under debate. Feigelman et al. (2016) set out to address this issue by applying Bayesian inference on time-lapse data of Nanog expression. Going beyond previous work in which stochastic inference was applied to singlecell trajectories (Zechner et al., 2011), Feigelman et al. (2016) use the genealogic organization of the cellular population (Figure 1, top) to limit the set of plausible model parameters that are compatible with the full dynamics of the lineage tree. The approach is based on the appearance of temporal cross-correlations in the dynamics of cells related by a common ancestry (Figure 1, bottom), resulting from daughter cells inheriting the cellular state of their mothers. To represent those constraints, Feigelman et al. (2016) extend a recursive particle filter (Zechner et al., 2011) with a model of cell division. First, the authors validate their algorithm (which they term Stochastic Inference on Lineage Trees, or STILT) using synthetic data generated 512 Cell Systems 3, December 21, 2016
by three simple models of Nanog expression with different feedback structures (positive, negative, and no feedback). Next, they apply their method to previously published time-lapse data (Filipczyk et al., 2015). A goodness-of-fit test allows them to reject positive feedback as a possible mechanism of Nanog autoregulation due to its inability to explain the experimental data. In order to distinguish between the remaining two alternatives, negative feedback and no regulation, the authors use the inferred parameters of those models to predict the response of the system to exogenous Nanog. Flow cytometry reveals a downregulation of endogenous Nanog production for increasing exogenous Nanog expression, which can only be reproduced quantitatively by the negative feedback model. This result contrasts with the prevalent view of pluripotency regulation, according to which Nanog activates its own expression. The study thus offers a novel approach to using lineage information for resolving the molecular circuitry underlying dynamic phenotypes in populations of proliferating cells. Time-lapse fluorescence microscopy is also limited by the difficulty of generating cell lines with the appropriate reporters, which can still be a fairly timeconsuming task. Interestingly, Hormoz et al. (2016) show that lineage information, combined with end-point measurements of cellular states throughout the colony (which does not require fluorescent reporters), can be sufficient to characterize the dynamical properties of cellular processes. In particular, they study the transition dynamics between multiple phenotypically distinct ES cells in normal growth conditions (LIF+serum). While these states are known to underlie pluripotency in ES cells, the transitions among them are not well understood, since measuring switching rates seemingly requires tracking the state of individual cells over time. But Hormoz et al. (2016) show that this is not necessary: switching rates can be calculated by identifying the proliferating cells through direct time-lapse imaging and measuring the final state of the colony with multicolor single-molecule RNA-FISH, with no fluorescent reporters being necessarily involved at any time. Their approach, which they term kin correlation analysis (KCA), builds upon
previous work in bacterial colonies (Hormoz et al., 2015) and is originally inspired by a cosmological multiverse model (Harlow et al., 2012). In contrast with the approach of Feigelman et al. (2016), which exploits the temporal cross-correlations between cells that share a common ancestry, the method of Hormoz et al. (2016) uses the endpoint correlations between those cells (Figure 1, top), which can be calculated from the transition matrix linking the different states. This relationship can be inverted, which enables the transition rates to be inferred from the end-point correlations. Hormoz et al. (2016) validate their approach using time-lapse fluorescence data on the promoter activity of the Esrrb gene, which exhibits a bistable profile. The switching rates obtained with the KCA approach are consistent with direct measurements using the fluorescent reporter. Next, they extend their method to five distinct cellular states defined by the levels of three proteins, Tbx3, Zscan4, and the above-mentioned Esrrb. After verifying that these states have different morphological and functional (as measured by RNA-seq) properties, the authors use KCA to uncover a state transition network in the form of a linear chain, in which cells switch stochastically and reversibly between adjacent states. This chain orders the states in a sequence ranging from a totipotent-like state to the more differentiated epiblastlike state. The finding is relevant for our understanding of pluripotency, since it provides us with a highly structured perspective of ES heterogeneity, which contrasts with the widespread view of this regime as a random walk among all ES cell states. This result highlights the potential of using lineage information to characterize the functional aspects of cellular regulation. While the method was originally devised for reversible transitions and stationary dynamics, the authors generalize it to the analysis of irreversible and non-stationary dynamics. Notably, they exclude the possibility of cell-cell communication playing a role in the transition dynamics. It would be interesting to extend this method to situations in which cell-cell signaling modulates the switching rates, and thus the kin correlations.
Cell Systems
Previews Together, these two papers open up new avenues in our ability to measure and interpret the dynamical behavior of regulatory networks at the single-cell level at a time in which the roles of dynamics and heterogeneity in cells are becoming increasingly more evident. It is time to climb up the cell lineage tree.
REFERENCES Feigelman, J., Ganscha, S., Hastreiter, S., Schwarzfischer, M., Filipczyk, A., Schroeder, T.,
Theis, F.J., Marr, C., and Claassen, M. (2016). Cell Syst. 3, 480–490.e13.
Kalmar, T., Lim, C., Hayward, P., Mun˜oz-Descalzo, S., Nichols, J., Garcia-Ojalvo, J., and Martinez Arias, A. (2009). PLoS Biol. 7, e1000149.
Filipczyk, A., Marr, C., Hastreiter, S., Feigelman, J., Schwarzfischer, M., Hoppe, P.S., Loeffler, D., Kokkaliaris, K.D., Endele, M., Schauberger, B., et al. (2015). Nat. Cell Biol. 17, 1235–1246.
Locke, J.C., and Elowitz, M.B. (2009). Nat. Rev. Microbiol. 7, 383–392.
Harlow, D., Shenker, S.H., Stanford, D., and Susskind, L. (2012). Phys. Rev. D Part. Fields Gravit. Cosmol. 85, 063516.
Singer, Z.S., Yong, J., Tischler, J., Hackett, J.A., Altinok, A., Surani, M.A., Cai, L., and Elowitz, M.B. (2014). Mol. Cell 55, 319–331.
Hormoz, S., Desprat, N., and Shraiman, B.I. (2015). Proc. Natl. Acad. Sci. USA 112, E2281–E2289.
Zechner, C., Pelet, S., Peter, M., and Koeppl, H. (2011). Recursive Bayesian estimation of stochastic rate constants from heterogeneous cell populations. In 2011 50th IEEE Conference on Decision and Control and European Control Conference, pp. 5837–5843.
Hormoz, S., Singer, Z.S., Linton, J.M., Antebi, Y.E., Shraiman, B.I., and Elowitz, M.B. (2016). Cell Syst. 3, 419–433.e8.
Cell Systems 3, December 21, 2016 513