The Price of Tags in Protein Localization Studies

TEXT

These continue to be heady times for bacterial cell biology, both for experimentalists and for modelers. The ability to follow the dynamics of fluorescently labeled proteins has had a huge impact in understanding the organization of DNA, RNA, and protein complexes in all cells, particularly in bacteria, which do not have many features visible by electron microscopic methods (17). Yet the fluorescent protein tags used for live cell imaging, as well as small epitope tags used for purification and/or immunolocalization, are a double-edged sword. There have always been suspicions that tags could cause localization artifacts, but these suspicions usually remain just that, as publishing a paper that claims a previous localization pattern as an artifact requires an extra measure of proof, not just negative data.

Swulius and Jensen, in an article in this issue (26), have obtained precisely the strong confirmatory evidence needed to refute some previous published localization patterns. By using cryo-electron tomography (cryo-ET) on Escherichia coli cells producing a tagged and untagged protein, they show that MreB, the bacterial actin homolog required for cylindrical cell wall growth, does not localize as a helix unless it is tagged with yellow fluorescent protein (YFP) at its N terminus. This is important because several papers had shown beautiful helical patterns of fluorescence using YFP-MreB in E. coli (21, 33–35). Images from immunofluorescence microscopy of native MreB proteins using anti-MreB did not differ significantly from this helical pattern, so the live cell data were not questioned. Finally, it was very satisfying esthetically that a relatively stable actin cable network was the organized scaffold needed to direct and orient cell wall biosynthesis. It was hard to believe that these strong patterns were not real, and consequently, these patterns, including MreB rings that flanked the E. coli septal ring, were assumed to be physiologically relevant by many. In Bacillus subtilis, functional green fluorescent protein (GFP)-tagged MreB homologs also formed helix-like filaments and moved along helical tracks (4), consistent with immunofluorescence data (13), in support of the general helical model.

MreB's helical façade began to crack last year with the publication of three papers showing that MreB and its homologs localized to dynamic and distributed patches in several bacterial species, with no helical cables in sight (5, 8, 32). These patches, which may consist of a network of short MreB polymers, move circumferentially around the cell, and their movement depends on active cell wall biosynthesis. These findings cast serious doubt on the existence of long-range helical cables of native MreB and the idea of a bacterial actin filament scaffold. To track their dynamics, these MreB proteins were fused with fluorescent protein tags, usually at their N termini. In some cases, these fusions could replace the resident native copy of MreB (or the homolog Mbl or MreBH in B. subtilis) without any notable perturbation of cell shape or patch dynamics, indicating that these fusions were functional and thus could be trusted to report physiologically relevant localization. At about the same time, Swulius and Jensen also published a brief report showing that no MreB cables were detectable in cryo-ET images of rod-shaped bacteria containing MreB (25). Assuming that cryo-ET preserves specimens in a state as close to the native state as possible, these data, along with the fluorescence studies, suggested that the helical cables reported previously, particularly those made by YFP-MreB, were an artifact of high expression of the fusion or the YFP tag itself (6).

From the present study by Swulius and Jensen, it is clear from cryo-ET that the N-terminal YFP tag induces the accumulation of large helical bundles of MreB in E. coli, even at low expression levels, whereas high expression levels of the untagged protein produce no such bundles. As implicated by the recent studies above, it is not simply the presence of a fluorescent protein tag attached to MreB that causes problems, as N-terminal GFP tags of B. subtilis MreB homologs seem to be fully functional and fail to form helices. It could be, however, that E. coli MreB is more sensitive to the presence of an N-terminal tag. A version of E. coli MreB with an mCherry tag inserted in an internal loop of MreB is fully functional and does not form a helical pattern by fluorescence (2, 32) or bundles visible by cryo-ET (26). Moreover, a GFP fusion to RodZ, which partners with MreB to coordinate cell shape and is required for normal MreB localization patterns, colocalizes with patches of untagged MreB, supporting the MreB localization data (2, 23), although some GFP-RodZ localization can also be interpreted to be helical.

Why would the YFP tag induce formation of helical bundles, at least for E. coli MreB? One possibility is that an N-terminal extension interferes with the function of E. coli MreB's N-terminal amphipathic helix, which is normally involved in tethering the cytoplasmic MreB protein to the membrane as a membrane targeting sequence (MTS) inserted into the lipid bilayer (20). B. subtilis MreB homologs may be less sensitive to this perturbation, probably because they bind to the membrane via an internal membrane insertion loop instead of a terminal MTS (20). This specific mechanism may be applicable to other membrane-associated bacterial actins, because removing a similar C-terminal MTS in the actin homolog FtsA strongly enhances formation of FtsA polymers and polymer bundles (27). As the FtsA bundles are formed longitudinally in the cytoplasm instead of circling the cell division septum where they normally localize, they are artifacts of perturbed membrane binding by a protein that tends to form polymers. C-terminal GFP fusions to FtsA are not fully functional (although they localize to the septum), probably because the tag disrupts the MTS. The mechanism underlying this enhancement of polymerization upon perturbation of membrane binding is not clear, but may result from competition between protein binding to the phospholipid bilayer and binding to another protein monomer. More binding to the membrane would therefore suppress the tendency to self-associate.

The second possibility is that dimerization of the attached YFP tag, and potentially other tags, enhances oligomerization of MreB. Therefore, monomeric variants of YFP (as well as mCherry, which is monomeric) may be safer fluorescent protein tags to use in these situations. The full functionality of MreB with the internal mCherry fusion is consistent with this suggestion, although the placement of this tag in an internal loop may also be the key to its success. N-terminally myc-tagged MreB forms helices in E. coli (7), arguing against the tag dimerization idea, indicating that any tag at the N terminus of E. coli MreB may perturb its localization.

What can we learn from this study, and what about other localization results that might be artifacts? One long-standing example of a fluorescent protein fusion that is not fully functional is FtsZ (15). Although these fusions cannot fully complement an ftsZ mutant, there is no compelling evidence to date that the overall localization patterns or dynamics of FtsZ are perturbed by tagging it with fluorescent proteins in the presence of native FtsZ. For example, immunofluorescence of FtsZ in fixed cells, as well as localization of FtsZ-binding proteins such as ZipA, is consistent with established FtsZ localization patterns from fluorescent protein fusions in live cells (10). These localization patterns include the Z ring as well as helix-like patterns along the sidewall. The fact that these patterns are visible from FtsZ fluorescent protein fusions in live cells as well as immunostaining, including superresolution imaging of immunostained FtsZ in fixed cells, suggests that the patterns are real (12, 18, 30). However, small changes at the FtsZ C terminus result in significant alteration of its assembly properties (3), suggesting that, at least, FtsZ-GFP may not entirely reflect the behavior of native FtsZ. Moreover, overproduction of FtsZ-GFP or even untagged FtsZ can induce formation of strong FtsZ helical structures, which may reflect a static version of a previously dynamic helix or an artifact of overproduction (16).

Another example of heavy reliance on fluorescent protein fusions in live bacterial cells are the Min proteins of E. coli, whose pole-to-pole oscillations could not have been discovered without such fusions (19). Like MreB or FtsA, E. coli MinD and MinE can self-associate and bind to the membrane via terminal amphipathic helices, but most of the fluorescent protein fusions have been to the opposite termini of the proteins, decreasing the chances of perturbing this activity significantly. It helps that MinC and even FtsZ have been shown to oscillate from pole to pole in a manner dependent on MinD and MinE, with similar kinetics (11, 30). Yet, similar to MreB, fluorescent protein fusions to MinD have been reported to form helix-like coils that extend the length of the E. coli cell (22). Given that this system is the subject of numerous mathematical simulations and that MinD coils are not universally observed, it will be important to confirm the presence or absence of MinD coils using the Swulius and Jensen approach as a guide.

Other fluorescently tagged proteins form convincing-looking, long-range helical structures, including chemoreceptors, secretion proteins, RNase E, RNA helicase B, and the chromosome partitioning proteins SetB/C (7, 24, 28). How many of these are real? In many of these studies, the helical patterns were also observed with immunofluorescence, which make the results more convincing. Once again, the ideas of a spatially organized protein secretion system, RNA degradosome, and chromosome partitioning system are attractive. However, it is difficult to know which of these helical patterns are real and which may be artifacts. Even with immunofluorescence, c-myc or hemagglutinin (HA) epitope tags were sometimes used (e.g., with RNase E and SetB/C), and their localization behavior may be altered by these tags. Taking another look at some of the reported localization patterns, they could also be consistent with a distribution of patches along the membrane that could be imagined as a helical pattern, like much of the recent MreB patch localization data. As with MreB, the ability of some of these proteins to polymerize in vitro, such as RhlB (RNA helicase B) (29), may mean either that large in vivo polymers are physiologically relevant or that such polymers are artifacts of protein tags or altered stoichiometries.

Twenty years ago, most bacteria were not thought to have much intracellular organization apart from the nucleoid. Fluorescent tags such as GFP as well as immunofluorescence changed all that, but perhaps in our excitement we have gone too far in the other direction and ascribed cytoskeletal organization to components that, while not necessarily randomly diffuse, do not really form a polymeric eukaryotic-cell-like cytoskeleton. It is also difficult to rationalize now how so many proteins at the membrane can form long-range helical structures, especially if MreB itself does not form such structures. Nevertheless, other evidence, independent of using protein tags, does point to some type of underlying helical organization of the bacterial envelope (1, 9), possibly at the level of phospholipids, that could form a track for membrane protein binding. Superresolution light microscopy, in combination with confirmation by cryo-ET, may permit greater certainty about these localization behaviors, although artifacts of those methods will need to be understood as well. Ultimately, each protein needs to be tested for functionality at near-native levels with a variety of tags, in case some of the tags perturb its behavior (14). Inserting tags at relatively innocuous sites, such as the internal loop in E. coli MreB, should also be a key to reducing these artifacts. Finally, once the biology is done, better quantitation of localization patterns using improved image analysis methods will help to increase data integrity (31).