Efficient biological signal processing often requires complex spatial organization of the signaling machinery. Understanding how this spatial organization is generated, maintained, and repaired inside cells is a fundamental theme of biology. A well-understood signaling network with complex spatial organization is the bacterial chemotaxis system, which directs the movement of cells towards or away from sugars, amino acids, and many other soluble molecules 
. In Escherichia coli
, five types of transmembrane chemoreceptors form trimers of dimers 
, which cluster into large complexes containing tens of thousands of proteins 
. Receptor clustering enables cooperative interactions between receptors 
, contributing to a bacterium's ability to sense nanomolar concentrations of chemicals and small fractional changes in chemical concentrations over a wide range 
. Chemotaxis clusters are stabilized by the adaptor protein CheW and the histidine kinase CheA, which bind receptors in a ternary complex. CheA transduces signals from membrane receptors to the cytoplasmic response regulator CheY, which diffuses to flagellar motors and modulates their direction of rotation (; for review see 
Membrane receptor clusters transduce chemotatic signals.
A variety of imaging studies have advanced our understanding of how the spatial organization of the chemotaxis network arises and contributes to function 
. Time-lapse fluorescence microscopy suggests that receptors are inserted randomly into the lateral membrane via the general protein translocation machinery and then diffuse to existing clusters 
. Immunoelectron and fluorescence microscopy have shown that receptor clusters are found at the cell poles 
and future division sites 
Despite much research, the fundamental mechanisms responsible for positioning chemotaxis clusters at specific sites in the membrane remain unclear 
. Perhaps cells possess intracellular structures that anchor clusters to periodic sites along cell length 
. However, fluorescence microscopy of cells overexpressing all chemotaxis proteins showed that the number of clusters per cell saturates well below the number of proposed cluster anchoring sites. Furthermore, the distance between chemotaxis clusters varies broadly within cells 
. Based on those observations, Thiem and Sourjik 
proposed that cluster nucleation and growth is a stochastic self-assembly process in which receptors freely diffuse in the membrane and then join existing clusters or nucleate new clusters. In their model, clusters nucleate anywhere in the membrane and later become attached to anchoring sites. Shortly thereafter, it was reported that anchoring sites may not be required for periodic positioning; surprisingly, simulations reveal that periodic positioning of clusters can emerge spontaneously in growing cells 
Direct tests of these stochastic nucleation models involve measuring, as accurately as possible, the relative spatial positioning of clusters and the distribution of cluster sizes. This requires (1) the high specificity of genetically encoded fluorescent tags and (2) spatial resolutions sufficient to count and localize single proteins, even when these proteins are densely packed. Electron microscopy has the required spatial resolution, but the density of immunogold labeling is too low to visualize a significant fraction of receptors 
. Cryo-electron microscopy tomography has provided detailed information on large polar clusters 
, but identification of individual receptors is not yet possible. Fluorescence microscopy does not have the required spatial resolution to observe individual receptors in dense clusters. Single-cell Förster resonance energy transfer (FRET) studies have been instrumental in measuring the dynamics of signaling within the chemotaxis network 
, but cannot obtain the distribution of receptors inside cells.
The optical super-resolution technique photoactivated localization microscopy (PALM) combines high specificity with high resolution. In PALM, target proteins are genetically labeled with photoactivatable proteins, thus rendering them nonfluorescent until activated by near-UV light. By employing near-UV light of sufficiently low intensity, only one protein per diffraction-limited region (~250 nm) is activated at a time. Following activation, each individual protein is then excited and imaged. Since only one protein is imaged at a time in each diffraction-limited region, the center of each molecular point spread function indicates the location of each protein 
. Serial cycles of activation and excitation are repeated until all fusion proteins are bleached. Since individual proteins are imaged, we can count the number of proteins and computationally assemble the locations of all proteins into a composite, high-resolution image. The location of each protein can be determined to a precision of 2–25 nm, or approximately 10–100× better than the diffraction limit 
. The localization error in each protein location depends on the number of photons collected for that protein, as well as background noise, pixel size, sample drift, and whether cells are live or chemically fixed 
. The highest optical resolution is obtained with chemically fixed cells 
. Several other optical techniques, including FPALM 
, STORM 
, STED 
, and SSIM 
, also image below the diffraction limit.
Here, we use PALM images to directly test stochastic nucleation models of chemotaxis cluster self-assembly in E. coli. We show that many receptors are part of small clusters not previously observed in electron microscopy or fluorescence microscopy, and that these small clusters provide direct evidence for a stochastic nucleation mechanism without anchoring sites.