Discussion
How the screen works
The principle of this cell-based screen for Rop function is the relatively simple idea that increasing plasmid copy number would increase the expression of a gene transcribed from that plasmid. It is worth noting that this implies, at least with these promoters and in the relevant copy number regime, that the amount of gene expression is related to the amount of DNA template. This seems to contrast with Rop expression from another plasmid created in this laboratory, pMR101, a low copy p15A plasmid used to over-express Rop from the T7lac promoter (Munson et al., 1994b
). The T7lac promoter is a very strong promoter used to over-express proteins from plasmids of low to high copy number. The exceptionally high level of Rop expression from pMR101 implies that the low copy number of the template is not limiting, and it may be that there is only a relationship between template copy number and expression level with relatively weak promoters.
In fact, the cellular fluorescence is only related to the copy number in the positive (PBAD–GFPuv) screen in a very narrow range of arabinose concentrations (near 0.0005%). At lower concentrations of arabinose the cells are dim, and at higher concentrations of arabinose, the cells are fluorescent, regardless of Rop activity. (By ‘cells’, we mean colonies or bulk culture; we did not examine individual cells microscopically.) More remarkable, of course, is that at this highly tuned arabinose concentration, cells with less ColE1 plasmid (i.e. active Rop resulting in less DNA template) make more GFP and are more fluorescent. This is especially peculiar in light of the Hu group’s recent observation that the relationship between arabinose concentration and PBAD expression is probably determined by how many cells express protein, not how much each one expresses (Siegele and Hu, 1997). Since the AraC level also presumably changes in response to Rop activity (it is expressed from the ColE1 plasmid), this implies that the probability of switching from repression to activation is controlled both by the concentration of arabinose and the concentration of AraC. Regardless, the effect is fortuitous: using the lac and PBAD promoters, we can screen for Rop activity either negatively or positively.
The dynamic range of the screen is limited by the difference in copy number of the ColE1 plasmid in the presence of wt Rop versus an inactive variant. Thus, maximizing this difference is critical to achieving the best level of signal. This required both using the pUC19 version of the ColE1 origin (instead of the pBR322 version) and screening at 42°C. The pUC19 origin differs from the pBR322 origin at a single nucleotide in RNA II, immediately before the RNA I inhibitor template, resulting in a higher copy number, and a much higher copy number at 42°C. Interestingly, Rop largely ameliorates the loss of plasmid copy number control at both 37 and 42°C, resulting in a larger copy number difference at the higher temperature. It is conceivable that a small library of randomized nucleotides near the beginning of the RNA I template might yield a ColE1 variant with even more desirable properties, and indeed it is known that other mutations in this region alter ColE1 copy number (Fitzwater et al., 1988).
The screen as an assay for Rop function
This in vivo screen has some advantages over the gel-shift assay that has recently been used to interrogate Rop function. It is technically simpler, since one need not purify the Rop variant or run a gel to know the activity. Moreover, one can directly relate the fluorescence level to plasmid copy number by extraction of the plasmid DNA from the cells used in the screen. Perhaps most importantly, the gel shift assay uses only small RNA stem–loops, while the screen involves the actual RNA substrates and demands functionality rather than merely binding. The result of this is clear: it is possible to find Rop variants with a Rop-like fold, that bind the small RNA stem–loops, that are nevertheless not active. This suggests that Rop function may indeed be more complicated than merely binding stem–loop RNAs, perhaps involving the observed ability to aid in the conversion of an unstable RNA I:RNA II complex into a more stable one (Tomizawa, 1990). While the model stem–loop complexes have been useful for elucidating structural details of Rop:RNA complexes, this screen may be useful in better elucidating Rop’s molecular mechanism. It should be noted that the screen is a de facto screen for Rop expression, as well, and it is conceivable that a negative phenotype from some Rop variants might be as a result of the lack of protein expression. However, all of the variants used in this study are produced in large quantities under conditions of T7 over-expression (in other expression plasmids), and we have no reason to believe that lack of expression is a problem here.
The gel-shift assay does offer the ability, at least in principle, to measure the dissociation constant of the Rop:RNA interaction. Interestingly, we have preliminarily observed that Rop variants with randomized core residues produce intermediate levels of fluorescence in some cases (T.J.Magliery, unpublished observations). It would be very interesting to investigate the relationship between the fluorescence level and parameters like the KD with respect to RNA binding or Rop dimerization.
Screening a library: technical considerations
When one is interested in the phenotype conferred by all of the molecules that are transformed from a particular cloning reaction, the efficiency of each step becomes a serious issue. With selections, ‘background’ (or ‘noise’ from erroneous cloning) is eliminated if the phenotype is null; with a screen, however, such background clones hamper our ability to derive useful information from screen ‘negatives’. Here, for example, extremely poor NdeI scission of the Rop expression vector required a means of physically separating singly cut vector from vector with the whole NdeI/BanI linker excised. The introduction of the CmR gene both allows for successful isolation of the empty vector and allows us to get a rough idea of the Rop expression level in the pAClac construct.
Rop is probably expressed at exceedingly low levels in its wild-type context in pBR322. Its cryptic promoter is likely to be quite weak, and its initiation codon is GUG (instead of the usual AUG), which is seen in only 14% of ORFs in the E.coli genome and requires decoding by tRNAfMet’s CAU anticodon (i.e. a G:U wobble-pair is required in the first codon position). Here, we have replaced the GUG start codon with an AUG codon, but the synthetic lac promoter nonetheless results in a very small amount of protein. One of the original reasons chloramphenicol resistance was engineered into vectors was because, unlike ß-lactamase, CAT is produced in quantities too low to observe on a gel in whole lysate (Martinez et al., 1988). This level of expression from pACYC184, for example, corresponds to a MIC of between 35 and 70 µg/ml for chloramphenicol. The MIC associated with pAClacCm is at least an order of magnitude lower than this, 
2–4 µg/ml, suggesting that the Rop variants are expressed at extremely low levels from pAClac. This is of some significance, since Rop presumably must act as a homodimer. If the monomer is present at vanishingly low levels in cytosol, which has a very high overall protein concentration, the screen therefore demands a relatively strong and specific interaction.
A new tool for combinatorial experiments in biophysics
We have demonstrated here the utility of a novel cell-based screen for the function of a relatively simple and well understood four-helix bundle protein. In combination with our ability to generate interesting libraries of Rop variants, this new tool gives us the ability to effectively sort extremely Rop-like, stable molecules from unstructured or unstable molecules. Moreover, Rop is easy to purify and examine in vitro by biophysical methods like CD, NMR and X-ray crystallography. We have begun screening libraries in which the two central layers of the Rop core are completely randomized, and we have found that there are multiple patterns that lead to stable proteins, including those deduced from systematic work and other novel modes (T.J.Magliery and L.Regan, manuscript in preparation). Using multiple libraries and screening technologies, we will be able to generate statistical models that will give further insight into the fundamental basis of protein stability and thereby allow the creation of more general and accurate potential functions for the computational analysis of protein variants.
Acknowledgements
The authors thank Jon Ness and Pim Stemmer (Maxygen) and Neal Cariello (GlaxoSmithKline) for supplying the pBAD-GFPuv vector. T.J.M. is an NIH Postdoctoral Fellow (GM065750-02). This work was supported by a grant to L.R. from the NIH (GM49146-09).
Answers to all your Biology Questions
