One of the most exciting fields opened in the last years is the new understanding of the existence of something that we could call as “biological heterogeneity”. This new field of study is focused in observing and understanding the differences between reactions of a same kind, between cells of a same kind, and ultimately, between members of the same species. Specially, cellular heterogeneity constitutes a major focus of interest among biologists today, because of its implications in development and diseases such as cancer (Altschuler and Wu, 2010).
Blinded by a deterministic view of the world, born with the discovery of the concept of the gene, biologists forgot, by many years, that not all the processes could be entirely specified by instructions encoded as a biological “bit of information”. And so, to study a specific cellular event, for example, the synthesis of an enzyme to produce a chemical reactions, biologists saw the process as a continuous event, in which all the cells started to synthesize the enzyme, at the same rate, leading to a single “state” which could be measured simply by averaging the measure of the enzyme concentration at the same time (Figure 1A). Hence, the particular amount of an enzyme at a specified time, meant that all the cells had the same amount, with small variations that can be represented as standard deviation.
However, by 1957, Novick and Weiner observed unexpected results from the study of the formation of β-galactosidase in the presence of low concentrations of an inducer. The linear growth of the production of the enzyme was unexpected, because it was known that the concentration of the inducer increases with the production of a permease which allowed the transport of the inducer. Novick and Weiner determined that, at low concentrations of the inducer, the population consisted “essentially of individual bacteria that are either making enzyme at full rate or not making it at all”. Such a rationale implies reconsider the usual thinking in terms of average in biochemical assays: an average can represent two discrete populations in terms of their behavior (Figure 1B).
Figure 1. Heterogeneity in a cellular response can lead to an “average response”, but the cellular population can either consists in cells showing an incremental and graded response (A) or an ON/OFF response (B). In the latter, drawing conclusions from the average can lead to a wrong interpretation of the biological phenomena.
Nowadays, there is a growing number of publications regarding the concepts of “noise”, “heterogeneity”, “stochasticity” and so on. One key difference between “noise” and “heterogeneity” is that heterogeneity is “a property of a cell population, not of individual cells” (S. Huang, 2009), whereas noise can be defined as a change in the distribution of amount of a measurable trait in a non-expected pattern. Depending on the author, a same word can mean similar (but not equal) things, but the key is to recognize the existence of variations between “biological units” (a cell, for example), even when these units have an equivalent genetic background. For example, individual cells derived from a clone can present different levels of expression for a same gene. These differences can have different origins. One example is heterogeneity between the cells. Two cells can respond differentially to a growth factor, due to a differential spacial localization. In the mouse embryo, at the 8-cell stage, starts the zygotic expression of Cdx2, but is has been reported that the initiation of CDX2 expression is not uniform, and this could be due to the specific locations of the blastomeres in the embryo (Zernicka-Goetz et al, 2009). This class of heterogeneity is referred as “extrinsic heterogeneity”, defined as “cell-to-cell variability in a population caused by non-uniform environmental factors that differentially affect individual cells” (S. Huang, 2009).
Opposed to the extrinsic heterogeneity, there is an “intrinsic heterogeneity”, which cannot be ascribed to environmental differences. In this case, the factors inducing heterogeneity are most probable intracellular. The most attractive source of intrinsic heterogeneity is the “noise” in biological processes. Noise is indeed a property of individual cells, and can be “temporal”, when the changes are observed across a time period, and also at the population levels, when the temporal noise in individual cells triggers different “states”.
Probably, the most interesting hypothesis to explain the presence of noise that transcriptional or translational bursts. Measuring single mRNAs (using FISH), it has been observed a particular mRNA can be transcribed in infrequent but potent bursts leading to cell-to-cell variations in mRNA number. A recent work reviews extensive evidence regarding the variations in mRNA/protein synthesis that can lead to noise (Raj and van Oudenaarden, 2008).
Heterogeneity, as a concept, is relevant when biologists select a scientific question and design experiments to answer that question. For example, studying the expression profile of a gene in time. Simply averaging the amount of the specific gene could lead to the wrong conclusion that the entire population produces a specific time, hiding the fact that one part of the population express high levels of the transcript, whereas the other part of the population express low levels, showing an ON/OFF behavior rather that a OFF-low-mid-high levels.
Can noise or heterogeneity at the cell level translate at the organism level? So far, the vast majority of the work has been focused in studying single cells, or characterizing cell-to-cell variability. But let us to make an exercise: if heterogeneity is a common property of cells, how does the organism to develop in such a patterned structure? For example, imagine a theoretical embryo (Figure 2A). This embryo has 32 cells (a 32-cell stage), and all its cells are performing biochemical and genetic processes influenced by extrinsic heterogeneity (such as the case of Cdx2), and intrinsic noise due to transcriptional bursts, chromatin remodeling and so on. As the embryo develops, three options remains: a) the stochasticity influences the development and the embryo grows in a stochastic pattern (Figure 2C); b) the embryo develops in an ordered pattern because, together with the stochasticity, the embryo harbors a system to buffer the noise (Figure 2A); c) the variability between cells makes a final average noise of zero (Figure 2B), as in a sum of two numbers with different sign (-2 + 2).
Figure 2. Control of noise in a theoretical embryo. Assuming the majority of the cells in the embryo display noise in their biological processes, specially transcription/translation, the embryo can buffer this noise by decreasing the influence of the noise through a biochemical network (A). Another possibility is that the noise becomes stabilized by a vectorial sum (B), where opposing effects of noise (for example, one cell express high levels of a ligand and another cell express low levels, leading to a physiological average”). Finally, if noise predominates, the embryo could display an unpredictable development (C).
Since many years, developmental biologists have been studying biochemical pathways with a special relevance to embryonic development. Any biochemical pathway is candidate to impose a buffer to heterogeneity and noise. Please consider that noise in embryo development is highly relevant, because initially everything start with one cell. It has been proposed that a mechanism to control noise is the negative feedback in circuits (Raj and van Oudenaarden, 2008). Fluctuations above and below the average are pushed back in those feedback loops. One classic example of negative feedback is provided by the Wnt pathway, which have a key role in embryonic development. The Wnt pathway comprises a family of secreted ligands, which can be divided into “canonical” ligands that activates the transcription factor β-catenin, and “non canonical” ligands that activate intracellular effectors which act in a β-catenin independent fashion (in a summarized view, because many data demonstrate that such classification is not always precise). In the canonical pathway, Axin2 is a target of the stabilized β-catenin, and acts as a negative regulator of the pathway, providing negative feedback. Theoretical and experimental evidence from the work of Lea Goentoro and Marc Kirschner (Goentoro and Kirschner, 2009) showed that the canonical Wnt pathway displays an interesting behavior: its activity is dictated by the fold-change in β-catenin levels and not to the absolute level of this transcription factor. The authors propose that, in such a system, noise os buffered, because simple variations in gene expression are not able to activate the pathway: it is required a precise threshold of variation in the components to variate the fold-change of β-catenin and, concomitantly, activate the pathway. Parallel work of Lea Goentoro (Goentoro et al, 2009, in the same issue of the Molecular Cell journal) explain how a circuit can provide fold-change detection, giving to a cell an advantage in a “noisy environment”. It is really interesting that many negative regulators of the Wnt pathway are, in fact, target of β-catenin: the pathway activates its own negative regulators, as evidenced by studying the transcriptional response of antagonists of the pathway in HEK293 cells (Gujral and MacBeath, 2010). It remains an open question whether the Wnt pathway can buffer noise and provide homogeneity to the embryo to allow development. Maybe the developmental patterning is encoded precisely in heterogeneity, but since heterogeneity is unpredictable, and since we can predict the development of the embryo, it seems more likely that the embryo display specific responses, in the form of pathways with negative feedback and response to fold-change of key elements (instead of responding to small changes in the amount of these elements), to buffer noise.
Finally, we must consider that not all the noise is a bad thing during development. Two examples show us that the heterogeneity is necessary during development. In the Drosophila eye, the optical units that compose the eye contains photoreceptors, which specification is provided by a stochastic process, where in a specific cell type (R7), the stochastic expression of the spineless gene dictates the Rh4 gene expression, which in turn is permissive to the expression of the Rh6 gene in the R8 cells. Failing to express spineless above a specific threshold, is permissive to Rh3 expression in R7 cells, which in turn instruct the expression of the Rh5 gene in R8 cells (reviewed in Samoilov et al, 2006). The second example is presented by the work of Raj and coworkers in a recent paper in Nature (Raj et al, 2010), in which they study the effect of modify a genetic network that controls intestinal differentiation in C. elegans. Mutant conditions increase transcriptional noise, which leads to an ON/OFF state of a master regulatory gene. One direct implication is that the incomplete penetrance of some mutations can be explained by stochasticity. Another implication of this work (and underestimated by the authors) is that there are natural buffering systems in organisms that control noise in gene expression, and altering these buffers triggers developmental responses.
Buffering noise is an unexplored field with great implications in development and disease. For example, disruption of buffers in adult humans may lead to cancer development due to stochastic expression of oncogenes. This field is in its infancy, despite the concept of variability relies in the foundations of modern biology, since Darwin studied variations and similarities between species and recognizing that variations are relevant properties of living systems.
Altschuler, S., & Wu, L. (2010). Cellular Heterogeneity: Do Differences Make a Difference? Cell, 141 (4), 559-563 DOI: 10.1016/j.cell.2010.04.033
Huang, S. (2009). Non-genetic heterogeneity of cells in development: more than just noise Development, 136 (23), 3853-3862 DOI: 10.1242/dev.035139
Raj, A., & Vanoudenaarden, A. (2008). Nature, Nurture, or Chance: Stochastic Gene Expression and Its Consequences Cell, 135 (2), 216-226 DOI: 10.1016/j.cell.2008.09.050
Samoilov, M., Price, G., & Arkin, A. (2006). From Fluctuations to Phenotypes: The Physiology of Noise Science’s STKE, 2006 (366) DOI: 10.1126/stke.3662006re17
Novick, A. (1957). Enzyme Induction as an All-or-None Phenomenon Proceedings of the National Academy of Sciences, 43 (7), 553-566 DOI: 10.1073/pnas.43.7.553
Zernicka-Goetz, M., Morris, S., & Bruce, A. (2009). Making a firm decision: multifaceted regulation of cell fate in the early mouse embryo Nature Reviews Genetics, 10 (7), 467-477 DOI: 10.1038/nrg2564
Goentoro, L., & Kirschner, M. (2009). Evidence that Fold-Change, and Not Absolute Level, of β-Catenin Dictates Wnt Signaling Molecular Cell, 36 (5), 872-884 DOI: 10.1016/j.molcel.2009.11.017
Goentoro, L., Shoval, O., Kirschner, M., & Alon, U. (2009). The Incoherent Feedforward Loop Can Provide Fold-Change Detection in Gene Regulation Molecular Cell, 36 (5), 894-899 DOI: 10.1016/j.molcel.2009.11.018
Gujral TS, & MacBeath G (2010). A system-wide investigation of the dynamics of Wnt signaling reveals novel phases of transcriptional regulation. PloS one, 5 (4) PMID: 20383323
Johnston Jr., R., & Desplan, C. (2010). A Penetrating Look at Stochasticity in Development Cell, 140 (5), 610-612 DOI: 10.1016/j.cell.2010.02.018
Raj, A., Rifkin, S., Andersen, E., & van Oudenaarden, A. (2010). Variability in gene expression underlies incomplete penetrance Nature, 463 (7283), 913-918 DOI: 10.1038/nature08781