Quantifying phenotypic diversity first requires quantifying phenotypes, which can be accomplished by measuring functional and behavioral parameters in individual cells. For
E. coli chemotaxis, such parameters include the following: the tumble bias—the fraction of time the cell spends tumbling; the switching frequency—the number of switches per unit of time between the run and tumble states; the pathway gain, or magnitude of the response relative to a stimulus change both at the level of the kinase and of the motor; the adaptation time, or time to return to the original behavioral pattern after a stimulus; and the rotational diffusion, all of which affect chemotactic performance
[10][18][25][26][27][28][29][30]. Alternatively, the phenotype of an individual cell can also be quantified without making assumptions about the set of possible behavior states a bacterium can be in (e.g., run-and-tumble). In one approach, each individual cell trajectory was represented as a scatter plot of time points in the space of translational and rotational velocity
[31]. Viewed as a two-dimensional probability distribution, the scatter plot defines the motility states the bacterium exhibit over time and the fraction of time it spends in each
[32][33].
An important consideration is that phenotypic parameters may vary on multiple time-scales. For example, a cell might exhibit the same tumble bias over its entire life span, or its tumble bias might change from the moment it is born until it divides. At any given instant, multiple stochastic processes operate on different time scales and contribute to phenotypic variability (Figure 1). Below, we review recent advances in current understanding of these various mechanisms and how they affect E. coli chemotaxis.
Figure 1. Phenotypic diversity arises from processes operating at different timescales. (
a) Unequal partitioning of proteins and organelles upon cell division, (
b) stochastic gene expression, and (
c) stochasticity in the chemical reactions of signaling pathways all contribute to phenotypic diversity. (
d) The number and size of chemoreceptor clusters varies from cell to cell due to random partitioning. This variation may lead to diverse sensitivities to stimuli. (
e) The expression of class II motility genes such as
fliA occurs in pulses lasting multiple generations. These pulses are stochastic, likely generating cells with very different sensory and swimming capabilities. (
f) Fluctuations in the kinase activity of the chemotaxis network. In red are cells containing the adaptation enzymes CheR and CheB, while in blue are
cheRB deletants. The cycle of methylation and demethylation, as well as receptor cooperativity both contribute to fluctuations. Panels d, e, and f are adapted from Koler et al., 2018
[34]; Kim et al., 2020
[35]; and Keegstra et al., 2017
[36].
2.1. Variation Arising at Cell Division
One mechanism likely to contribute to long-lasting differences between cells is partitioning noise—the unequal distribution of biomolecules between daughter cells during cell division (
Figure 1a). General features of partitioning noise and its quantitative treatment have been subject to review elsewhere
[13][37]. For chemosensing and motility, the partitioning of organelles, such as molecular motors and large receptor clusters, is likely to contribute to phenotypic variability. However, so far, only few studies have examined this aspect of phenotypic variability.
Recent studies have determined the location and size distributions of chemoreceptor clusters in growing cells
[38][39]. At early growth stages, cells contain multiple small receptor clusters, which are positioned along the length of the cell (
Figure 1d)
[34]. However, as the culture approaches stationary phase, individual cells typically contain only one or two polar clusters. The resulting cell-to-cell differences in cluster size and number are likely to translate into differences in sensitivity to chemical signals. While a direct link between cluster positioning and portioning noise has not been demonstrated directly, an uneven distribution of chemoreceptor complexes could contribute to long-lasting diversity, since some individuals will inherit multiple clusters, while others will inherit none.
2.2. Stochastic Pulses of Motility Gene Expression
By far, the most-studied source of phenotypic variability is stochastic gene expression (
Figure 1b), where it is understood that transcription and translation events occur in bursts with variable frequency and size, leading to broad distributions of protein expression in isogenic cell populations
[6][11][12][14]. Different regulatory schemes can lead to fluctuations in gene expression with various magnitudes and timescales
[40][41]. In
E. coli, the arrangement of chemotaxis genes on multicistronic operons has been shown to buffer the effect of some of these fluctuations in the signaling output by ensuring gene products that participate in the same step of signal transduction are kept at an acceptable ratio
[42]. Furthermore, a temperature-sensitive secondary RNA structure upstream of
cheR was suggested to compensate for temperature-induced differences in the rate of adaptation of cells to signals
[43][44]. While these compensatory mechanisms ensure that individual phenotypes remain functional, gene expression noise gives rise to large cell-to-cell variability within this range of functional phenotypic space
[45]. For example, the expression ratio of different receptor species varies greatly
[46], leading to variation in the sensitivity for different chemoeffectors
[36][47].
Recently, a study by the Cluzel lab revealed how regulated stochasticity in gene expression plays an important role in the transcription network governing
E. coli chemotaxis
[35]. In
E. coli, there are 14 operons collectively encoding the flagellar components and the chemotaxis machinery. These operons contain over 50 genes that are regulated in a three-tier hierarchy. At the top is the class I operon,
flhDC, which encodes the master regulator of flagella and chemotaxis. Then, FlhDC activates the class II genes encoding the basal body and flagellar hook, as well as the alternative sigma factor FliA, which subsequently activates the class III genes encoding the flagellar filament and chemotaxis network
[48].
Using a microfluidic device called the “mother machine” where a single cell’s lineage can be monitored over multiple division events, the Cluzel lab quantified the expression of class I, II, and III genes over multiple generations using fluorescent reporters
[35]. While the class I genes were expressed constitutively, class II and III gene expression occurred in pulses lasting multiple generations interspersed with long periods of inactivity (
Figure 1e). Interestingly, these pulses did not depend on the regulation of
flhDC transcription but were instead shown to require interactions between FlhDC and its post-translational regulator YdiV, which amplified transcriptional noise, allowing small temporal fluctuations in
flhDC transcription to generate large pulses of class II expression. In short, transcriptional noise, coupled with a post-translational circuit, converts constitutive expression of class I genes into pulsatile expression of class II and III genes.
The pulsatile expression pattern of class II genes was further explored in theoretical work that proposed a possible molecular mechanism
[49]. Simple stochastic models of class I and class II gene expression successfully capture the expected distributions of class II expression. By fitting these models to single-cell data
[35] of class II expression with and without YdiV, Sassi et al. found that YdiV’s functions were twofold: first, it makes class II gene expression ultrasensitive to changes in FlhDC concentration. Second, it allows the system to integrate FlhDC expression over time to filter away small fluctuations. As such, they proposed a molecular mechanism where YdiV sequesters FlhDC, such that small changes in FlhDC expression lead to ultrasensitive increases in class II promoter activity
[49]. This simple sequestration mechanism allows the expression of motility genes to follow a filter-and-integrate method of pulse generation that is qualitatively distinct from other known biological pulse generators that rely on bistability in transcriptional feedback loops
[50][51].
What is the end result of such an expression pattern? Ultimately, stochastic gene expression generates a wide diversity of sensory and motile phenotypes that may contribute to phenotypic changes over the course of cell growth as observed in behavioral studies
[33]. An interesting hypothesis is that cells in generations following a valley of little expression could have different chemotactic phenotypes than those following a pulse of expression. Such an expression pattern could also play a role in how individual cells within one population sort themselves in space during collective migration, as described in later sections.
2.3. Spontaneous Temporal Fluctuations in Pathway Activity
In addition to partitioning and gene expression noise, the behavior of individual cells is also affected by temporal fluctuations that arise due to the inherent stochasticity of the chemical reactions of the chemotaxis signal transduction pathway (
Figure 1c). The first study to characterize these fluctuations measured the rotational direction of latex beads attached to flagella in cells that were adapted to a motility buffer devoid of any signal
[52]. Analyzing the power spectra of these time-series revealed surprisingly large fluctuations in tumble bias with characteristic times scales on the order of 10s of seconds. Blocking adaptation by receptor methylation eliminated these fluctuations, while titrating the amount of methyltransferase CheR changed their time scale and amplitude
[52].
These data can be recapitulated with a coarse-grained model of the signaling pathway in which the rates of methylations and demethylations follow Michaelis–Menten kinetics with a sub-linear dependency on the activity of the receptors
[45][52][53][54][55]. This model assumes that in vivo, CheR and CheB work near saturation
[53][54], which is consistent with population average FRET measurements of adaption kinetics
[56] and with stoichiometric measurements—CheRB are expressed orders of magnitude less than the receptors
[57]. A consequence of the saturated enzyme kinetics hypothesis is that the steady-state activity of the system becomes ultra-sensitive
[58] to the ratio of CheR and CheB abundances, which can contribute to the amplification of spontaneous fluctuations in the activity of the receptors
[45][52][54][55].
Recently, the wide availability of high-sensitivity cameras allowed the classic CheZ/CheY FRET method
[59]—originally developed to measure kinase activity in a population of cells—to be adapted for quantifying kinase activity in single cells
[36][60][61][62] (
Figure 1f). Measurements of the kinase activity in individual unstimulated cells revealed large fluctuations in the kinase activity that could be attributed to two different sources. The first source was traced back to the activity of CheR and CheB. Similar to previous results
[52], these fluctuations have characteristic time scales of tens of seconds and were strongly suppressed in
cheRB-. Moreover, perturbations to the CheB phosphorylation feedback loop caused changes in steady-state kinase activity that were consistent with an ultrasensitive dependency of the activity of receptor-kinase complexes on the ratio of CheR/CheB abundances, and the role of the CheB phosphorylation feedback loop in reducing such ultra-sensitivity
[36][54].
Interestingly, the magnitude of the fluctuations measured at the kinase were larger than previous estimates from fluctuations measured at the motor output
[36], suggesting that a second source of fluctuations might be present. Indeed, measurements in
cheRB- cells stimulated with a constant sub-saturating stimulus revealed methylation-independent fluctuations with time scales of 100s of seconds
[36][60]. These fluctuations were eliminated when the scaffolding protein that mediates allosteric interactions in the receptor-kinase complexes, CheW, was replaced with the mutant CheW-X2
[60], which is a mutant that abolishes allosteric interactions but maintains signaling capabilities
[22][63]. These large fluctuations also depended on the composition of receptor clusters and became almost switch-like between a few discrete states in cells that only expressed Tsr
[36]. Together, these data suggest that the second source of fluctuations is caused by thermal fluctuations in highly cooperative receptor clusters
[36][60].
While a full biophysical model is still missing, the emerging picture is that receptor clustering and adaptation kinetics both contribute to temporal fluctuations in the chemotaxis pathway. While thermal fluctuations and allosteric interactions in receptor clusters can introduce large fluctuations over very long timescales (up to 100s of seconds), the stochasticity of methylation events contributes smaller fluctuations on time scales of 10s of seconds. Interestingly, receptor clustering is required for fluctuations to be observed even in the presence of methylation and demethylation, which is probably because of the role of “assistance neighborhoods”—in which CheR (or CheB) tether to one receptor and methylate (or demethylate neighboring receptors—in signal amplification and adaptation kinetics; see
[60][63][64][65][66]. The ultra-sensitivity of the kinase activity with respect to the ratio of CheR/CheB is also a source of phenotypic variability, which becomes more apparent when the buffering effect of the CheB phosphorylation feedback is removed
[36].