GreA itself is a transcription regulator that can interact with RNA polymerase at two levels, relief of elongational pausing (13
) and promoter-specific facilitation of initiation (27
). These features raise the possibility that GreA functions could be recruited to feedback regulate its own synthesis. Indeed, reporter lacZ
fusions indicate that GreA can auto-regulate its expression; deletion of greA
derepresses, and overproduction of GreA represses the reporter activity 2–3-fold. However, this phenomenon could not be reproduced with the pure in vitro
transcription system. Thus we conclude that the auto-regulation is due to indirect effects or additional factors are necessary to observe it in vitro
. We were instead rewarded with a different discovery. The greA
leader region contains two strong promoters and an unusual terminator that gives rise to two clusters of small RNA chains that we call GraL.
Despite the primer extension evidence (A), the existence of the P3 and P4 promoters is questionable. First, they lack clear consensus sequences (C). Second, we are unable to find in vitro transcription conditions that detect their activity (with RNAPσ70 or σE; data not shown). Third, lacZ reporter fusions yield barely measurable activities. It seems likely P3 and P4 primer extension bands are generated post-transcriptionally. Also, note that the primer anneals to sequences downstream of the terminator and only the P1 and P2 transcripts that read through the terminator are detected.
The intrinsic terminator accounts for the differences in lacZ reporter activities between greApA-all and P1 or P1P2 (B). Also, auto-regulation by GreA, mentioned above, seems to rely on the presence of this terminator as the activity of the greAP1P2+T fusion is derepressed in the absence of GreA, similar to the Pall fusion. On the other hand, greAP1 as well as greAP1P2 fusion strains did not exhibit this effect (data not shown).
The extent of termination in vivo can be calculated as 66% efficient from the difference between P1, P1P2 (~750 U) and P1P2 + T or pA-all (~250 U). The slopes of this differential plot seem constant over a wide range of cellular densities. In vivo termination efficiency is in good agreement with termination in vitro: ~60% ( and ).
Termination at the greA intrinsic terminator is exceptionally imprecise. The GraL RNA chains comprise a population whose ends range over ten nucleotides. The occurrence of these frayed ends in vivo is demonstrable by northern blots (C). This could be explained by post-transcriptional endonuclease attack, followed by exonucleolytic nibbling. However, the cluster of RNA chains is visualized in a pure in vitro transcription system (A and B). The purity of the in vitro system is validated because more precise termination is accomplished by swapping three base pairs at the base of the stem, preserving the structure of the terminator (). Nevertheless, we tested the effects of rnc and rne mutants by northern blots, and found no significant effect (C and data not shown).
We attempted to identify the ends of the short RNA chains by direct RNA sequencing (using 3′deoxy-NTPs). However, compression due to the GC-rich sequence of stem II prevented a clear identification of the first termination site, despite substituting ITP for GTP in sequencing reactions (data not shown). Because of this caveat, we can pinpoint each terminated position ±2 nt in the UAU7GCU2 sequence.
The predicted structure with the 11 bp stem is rare; terminators with such long stems comprise only 12% among 148 natural terminators surveyed (42
). To our knowledge, imprecise termination over a range of 10 sites is not reported previously in prokaryotes or eukaryotes. Rho-dependent termination, which is distinctly different from intrinsic termination, is known for its imprecision but the stop points are often spread over a 100 bp of DNA (43
). Therefore, we decided the basis of this behavior warrants further investigation.
A recent allosteric termination model proposes that the terminator hairpin forms within RNAP and induces extensive conformation changes across polymerase (7
). In this model, the 7–8 bp lambda tR2 hairpin ultimately clashes with the G(trigger)-loop to shorten the RNA–DNA hybrid that leads to the collapse of the transcription bubble and transcript release. It could be imagined that the 11 bp GraL stem might enhance the conformational constraints within RNAP. When the GraL hairpin is shortened without changing the 7 bp sequence at the bottom of the stem the termination efficiency is greatly reduced (). This could be viewed as favoring or disfavoring the model. On one hand it could be expected that the GraL terminator should be very efficient while having just a 7 bp hairpin. On the other, if the base line value for the termination efficiency is applied from the GraL 7 bp hairpin construct, it is consistent that the longer the stem length the better termination. Without crosslinking studies and modeling it is difficult at this point to predict what would happen with an 11 bp stem.
It has been proposed that formation of 2–3 bp at the bottom of the stem is critical for termination through shearing of the RNA–DNA hybrid (4
). Because of the space constraint, basepairing at the bottom of the stem might be limited for longer stems. Thus it can be imagined that the GraL hairpin does not form completely. Yet, if pairing is disrupted at the base of the stem (pAd1), termination efficiency is reduced (), as if pairing of the wild-type GraL stem is complete.
On the other hand, the 3 nt sequence just upstream of the U-tract is also important for termination, independent of basepairing. If this sequence is changed but basepairing is maintained, as for pAswap, termination efficiency increases relative to wild-type. If the same 3 nt sequence is present without basepairing (pAd2), the efficiency of termination is restored to wild-type levels ().
Still, the imprecision of the GraL terminator is yet another issue. Because of its uniqueness it is not addressed by current models, which are derived from studies of precise terminators. Our evidence is that basepairing and sequence at the bottom of the stem, not stem length, are important determinants of precision because imprecision persists when the stem is shortened from the top ().
sequence of the GraL U-rich tract is also unique, as the first three residues of the tract are typically UUU (41
). However, the second A residue in this region of GraL does not seem to be evolutionarily conserved (Supplementary Figure S1
), whereas the bottom basepair GU is. The significance of this is uncertain without knowing the precision of the other terminators. Still, the precision is greatly improved when the GU basepair is replaced by UG (pAswap, ).
Overall, the GraL terminator is an exception in terms of both structure and multiple termination sites. Its study can provide additional insights and constraints into mechanisms of termination derived from models based on typical terminators. Unusual chimeric terminators have contributed useful information (44
), and yet there are naturally occuring terminators that lack dyad symmetry stems and poly-U tracts that are unexplored (42
It is a puzzle why transcription from the two strong promoters (P1 and P2) is terminated so that greA
expression depends largely on read-through. To maintain a constant amount of greA
) it would seem simpler to have a single constitutive promoter. The complexity of this region may be explained by additional role for GraL, independent of GreA.
An additional role for GraL is suggested by the conservation of its sequences and/or predicted structures among related enteric bacteria. The biological importance of GraL is experimentally verified by the fitness experiments described in . Cells with constitutive overexpression of GraL do enhance biological fitness as judged by their 95% predominance over cells with wild-type GraL expression during repeated cycles of overnight growth on rich LB media. This could be due to effects during exponential growth, stationary-phase or outgrowth after dilution into the fresh medium. Since overexpression of an RNA with scrambled GraL sequence did not show similar effect, we conclude these effects are GraL specific.
In parallel with wild-type strains, the fitness experiment was attempted in a ppGpp0
host. However, several malT
alleles themselves had effects on fitness in this background and this prevented assessment of specific effects of GraL. We suspect this is because so many cell processes are affected by ppGpp (21
) that the malT
mutation is likely to give multiple indirect effects.
Profiles of global transcription activity were performed in search of presumably subtle changes due to GraL overproduction. Contrary to our expectation acute GraL induction did not produce the expected smaller subset of genes when compared to long-term exposure. Strikingly, short-term changes appear only in a ppGpp0 background. We attribute this to the likelihood that gene expression in a ppGpp0 strain is already imbalanced and a subtle perturbation caused by GraL overexpression can have more pronounced effects.
Many genes repressed by Fur are activated by GraL in ppGpp0
background during short-term exposure. A simple explanation would be that GraL affects Fur levels, instead of regulating expression of each gene individually. On the other hand, this does not explain why only some, and not all Fur-dependent genes are derepressed [specifically, the suf
genes which are invariably affected by iron starvation (45
)]. Since a western blot analysis reveals that Fur levels change by only ~2-fold (), perhaps there is a hierarchy of sensitivity of Fur operators, i.e. changes in Fur levels can titrate out operators with weak Fur binding. The mechanism by which GraL affects Fur levels remains to be determined.
Figure 8. Effects of GraL overexpression on Fur levels and iron assimilation phenotypes. (A) Example of a western blot with anti-Fur and anti-RpoA (control) antibodies. The numbers above each band represent densities in arbitrary units. Numbers below the line give (more ...)
Does GraL nevertheless induce an unusual form of iron starvation or perhaps change the cell tolerance to limiting iron concentration? GraL activates some genes involved in the synthesis of enterochelin (entA, B, C and E) as well as the ent-Fe complex uptake (fepC, D) and its use (fes). The uptake of other forms of iron involves fhu, cirA and fiu. This reinforces the possibility that GraL overproduction causes or mimics an iron starvation.
We pursued this with the use of CAS plates, where the formation of orange color on blue plates is an indication of excreted enterochelin and thus demonstrates the cell’s need for iron. As evident from B, there is no significant difference between strains bearing the GraL overproducing plasmid or the vector control in ppGpp+ background. However, a GraL-dependent slight color variation is detected in ppGpp0 strains that could be enhanced by color swapping software when black is substituted for orange (B, lower panel). This might indicate perturbance of iron metabolism in the absence of ppGpp and presence of GraL.
On the other hand, GraL represses 25 genes involved in flagella synthesis and four chemotaxis genes during short-term exposure (). Analysis of the data from four way comparisons (Supplementary Table S4
) reveals that these genes are actually induced in the ppGpp0
strain and GraL cancels this effect. These effects are evident for genes in early, middle and late operons for flagella synthesis, and include those regulated by σ70
, and also apply to transcription of flagellar regulators flgM
(but not fliA
that encodes σF
itself). Exponential growth in LB is known to involve continuous switching between many different nutrients (46
). There is also a report that carbon source starvation generally derepresses flagella and chemotaxis genes (47
). Perhaps when GraL is overproduced in the absence of ppGpp, either starvation is not sensed, or there is not enough starvation to induce flagella. In support of this hypothesis, in a ppGpp0
strain, GraL represses 17 genes involved in amino acid metabolism, including arginine, glycine, histidine, lysine, tryptophan and tyrosine.
Clear biological effects can be attributed to GraL. Yet, despite regulatory effects on over 100 genes, we are unable to identify the complementarity with target mRNAs that is usually expected of many small RNAs (39
). It is possible that GraL does not function through basepairing, but instead through binding of proteins, similarly to CsrA or 6S RNA (39
It is interesting to note that in eukaryotes a chaperone that protects RNA polymerase III transcripts from 3′ exonucleases and promotes their maturation, La protein, binds RNA depending on the number of U residues at the 3′ termini (48
). Products of PolIII that do not possess long enough 3′oligo U ends, undergo maturation in a different pathway. Perhaps something similar is happening in E. coli
and there is a different function associated with different members of the GraL cluster, rather than all of the species acting similarly.