We have studied EGF-induced signal transduction to ERK in single HEK293 cells, finding that the conversion of an analog signal at the single cell level to an apparent digital response at the population level can be mediated by a combination of cell-to-cell variability in protein expression and a pathway design that incorporates negative feedback (Figure ). A uniform step increase in EGF concentration causes a wide distribution of RasGTP levels due to cell-to-cell heterogeneity in protein expression. Cell-to-cell heterogeneity in protein expression also causes significant variability in the sigmoidal dose response relationship between RasGTP and ppERK, and in particular, in the ppERK activation threshold (Figure G and Figure ). Because cell-to-cell variability in RasGTP levels can span the range of ERK pathway activation thresholds, the pathway is activated to various degrees in individual cells. A distribution of ppERK levels ensues across the cell population. The mean of the ppERK distribution depends on EGF dosage and agrees with results obtained from Western blots. Despite the fact that the negative feedback smooths the RasGTP/ppERK dose–response relationship, a threshold for ppERK activation persists. This threshold element further enhances cell-to-cell variability in ppERK levels, and results in bimodal responses at the population level. Thus, the resulting bimodal distribution relies on a combination of a threshold behavior and a linear ppERK increase followed by saturation behavior with increasing EGF dose (Figure G). Surprisingly and counterintuitively, bimodality does not require switch-like behavior at the single-cell level, but can arise from cell-to-cell variability in protein expression and a pathway activation threshold. Thus, cells can retain the robustness benefits offered by negative feedback [37
], while generating on/off responses at the cell population level that are thought to be critical for cell fate decisions.
Figure 6 Conversion of Analog Inputs Into Bimodal Responses by a Negative Feedback System Combined with Protein Expression Noise. An analog EGF stimulus (blue, green, and red correspond respectively to small, medium, and large stimulation magnitudes) induces variable (more ...)
Our observations are unlikely to be caused by a fraction of cells simply not binding ligand. First, under our experimental conditions (~106 cells/mL), at the lowest ligand dose (0.01 nM), the ratio of EGF molecules to cells is approximately 1000, making it very unlikely that a cell does not encounter a ligand molecule. Second, for nearly all EGF doses, a significant fraction of cells is in the “ERK-on” population at some point in time, indicating that most cells have been activated and therefore had bound ligand.
How might cells still generate reliable signals despite protein expression noise? One possibility is that cells have a reliable fold-change response of ppERK from basal levels, and that downstream of ppERK cells employ systems that sense fold-changes rather than absolute levels. In fact this fold-change scenario has recently been shown to be the case. In cells stably expressing ERK2-YFP from the endogenous promoter, EGF stimulation led to widely varying maximum nuclear ERK2-YFP accumulation, with a coefficient of variation (CV) of approximately 0.3 [15
]. However, normalizing the maximum nuclear ERK2-YFP signal by the basal levels of ERK2-YFP in the same cell, which yields fold-change responses, lowers the CV by approximately 3-fold [15
]. This is consistent with our observed effects of total ERK abundance variability on the total variance of ppERK in the ERK-on population (Figure F-G). To sense these fold-changes, rather than absolute levels, a cell may use a type-1 incoherent feedforward loop (I1-FFL), where an input X activates both an intermediate Y and the output Z, but Y represses Z [45
]. Such a network structure may in principle be downstream of ppERK (X), which causes the immediate-early expression of multiple genes including c-fos
, which can mediate general transcriptional repression perhaps even of itself [46
Although protein expression noise is certainly a hindrance to some biological functions, and evolution has selected for mechanisms such as the I1-FFL that allow a cell to deal with this noise, there are potential benefits of and perhaps even essential functions for such noise. Tissue homeostasis may in fact require protein expression variability. Consider that there is no protein expression variability, and all cells that are involved with, for instance, hematopoiesis, respond identically to the various proliferation and differentiation cues. The body needs to produce, from the hematopoietic stem cells, a balance between the lymphoid and myeloid progenitors. If all the hematopoietic stem cells responded identically, then it would be nearly impossible for the body to maintain a finely tuned balance between the production of these two lineages. The same logic applies to the further differentiation of lymphoid and myeloid progenitors into various other downstream cell types, such as megakaryocytes, erythrocytes, B cells, T cells, and natural killer cells, where finely tuned control of differential cell-fate decisions is even more critical. Thus, it is likely that without protein expression noise-induced phenotypic variability, homeostasis of hematopoiesis, and probably other tissues, would not be possible. This logic argues for a conceptual model whereby growth factor concentration, in tissues, controls the probability a cell will choose a particular fate.