RNA thermometers


  • Editor: Fritz Unden

Franz Narberhaus, Lehrstuhl für Biologie der Mikroorganismen, Ruhr-Universität Bochum, NDEF 06/783, D-44780 Bochum, Germany. Tel.: +49 234 322 3100; fax: +49 234 321 4620; e-mail: franz.narberhaus@rub.de


Temperature is an important parameter that free-living cells monitor constantly. The expression of heat-shock, cold-shock and some virulence genes is coordinated in response to temperature changes. Apart from protein-mediated transcriptional control mechanisms, translational control by RNA thermometers is a widely used regulatory strategy. RNA thermometers are complex RNA structures that change their conformation in response to temperature. Most, but not all, RNA thermometers are located in the 5′-untranslated region and mask ribosome-binding sites by base pairing at low temperatures. Melting of the structure at increasing temperature permits ribosome access and translation initiation. Different cis-acting RNA thermometers and a trans-acting thermometer will be presented.

Introduction: RNA sensors

Free-living microorganisms are frequently exposed to changing environmental conditions. Nutrient availability, osmolarity, pH, temperature and many other parameters are being surveyed constantly. To avoid hazardous consequences, bacteria have established an intricate network of protective mechanisms. Often, regulation of environmentally controlled genes is realized at the transcriptional level by the action of regulatory proteins. However, several post-transcriptional, RNA-based strategies have been discovered recently. It is becoming increasingly clear that certain messenger RNAs (mRNAs) are not simply a substrate for ribosomes but contain control elements that modulate their own expression in a condition-dependent fashion. Structural changes in such sensory RNAs are induced by specific environmental changes. Two principally different classes can be distinguished: cis-acting RNA elements that bear their regulatory potential embedded within the mRNA sequence and trans-acting small, noncoding RNAs that function by base-pairing with complementary mRNA sequences encoded elsewhere in the genome.

Unlike classical attenuators which adjust the RNA structure of a leader sequence, according to the position of the translating ribosome (Landick et al., 1996; Henkin & Yanofsky, 2002), cis-acting RNAs change their conformation in response to physical or chemical signals. So-called riboswitches monitor the metabolic state of a cell by binding to metabolites with high specificity and affinity. They are located in the 5′-UTR of genes coding for the biosynthesis, uptake or degradation of small metabolites and provide feedback control to these pathways. Their complex architecture is comprised of a receptor (aptamer) region characterized by a consensus sequence specifying the substrate molecule and an output (expression platform) region. Binding of a small molecule throws a conformational switch which alters gene expression by one of three possible mechanisms: premature transcription termination (Mironov et al., 2002; Winkler et al., 2002a; Epshtein et al., 2003; Grundy et al., 2003; Mandal et al., 2003; McDaniel et al., 2003; Sudarsan et al., 2003; Winkler et al., 2003), translation initiation (Nou & Kadner, 2000; Nahvi et al., 2002; Winkler et al., 2002a, b) or mRNA processing (Winkler et al., 2004) (Fig. 1a). Most riboswitches turn off expression in the bound state. However, a few ‘on’-switches have also been found (Mandal & Breaker, 2004a; Mandal et al., 2004). The presence of potential riboswitches in eukaryotes suggest a wide distribution of this presumably ancient mode of control (Kaempfer, 2003; Kubodera et al., 2003; Sudarsan et al., 2003).

Figure 1.

  Schematic illustration of regulatory principles of riboswitches (a) and RNA thermometers (b). The processes that are controlled are indicated in each case. Red circles represent metabolites. RNAP, RNA polymerase; SD, Shine–Dalgarno sequence; 30S and 50S, ribosomal subunits, ΔT, temperature change. AUG and UUUUUU; ribonucleotide sequences.

Because numerous excellent reviews have been published recently on this fascinating subject (Narberhaus, 2002; Lai, 2003; Winkler & Breaker, 2003; Grundy & Henkin, 2004; Mandal & Breaker, 2004b; Nudler & Mironov, 2004; Soukup & Soukup, 2004; Vitreschak et al., 2004), the present review will focus on a closely related type of sensory mRNAs, the RNA thermometers. In contrast to highly specific metabolite-binding riboswitches, they respond to a fairly global physical signal, i.e. the intracellular temperature, which is an important parameter under constant vigilance. It is a well-known characteristic of structured nucleic acid molecules that they melt as the temperature increases. Hence, it is easily conceivable that temperature shifts are able to modulate the conformation of regulatory RNAs. A collection of structurally and functionally diverse RNA thermosensors that control a variety of cellular processes will be presented in this review. All currently known cis- and trans-acting molecular thermometers bring about translational control by sequestration of the ribosome-binding region (Fig. 1b).

Control of various cellular processes by cis-acting RNA thermometers

Considering possible temperature-controlled genes, several processes come to mind. The most obvious are the heat- and cold-shock responses, i.e. the expression of genes that protect a living cell from deleterious consequences of sudden temperature up- or downshifts. Virulence gene expression is a possibility if one considers that many pathogenic micro-organisms spend most of their life outside the comfort of a warm-blooded host. All these processes are, at least in part, temperature-regulated and RNA thermometers play an important role. Much less intuitively, it is possible that regulatory circuits during infection of Escherichia coli with phage λ or of the starvation response are also modulated by RNA thermometers. However, the first RNA thermometer was discovered some 15 years ago in the λ cIII gene.

Lysis-lysogeny decision of phage λ

Pioneering studies showed that alternative mRNA structures determine the translation rate of the λ cIII gene in a temperature-dependent manner (Altuvia et al., 1989). The cIII gene product plays a role in the decision of whether phage λ enters the lytic or lysogenic cycle. It serves as an alternative substrate for the FtsH protease, which degrades the cII protein, a central regulator in the lysogenic pathway (Herman et al., 1997; Shotland et al., 1997). High concentrations of the cIII protein thus result in the stabilization of cII, which in turn promotes lysogeny. Several point mutations affecting translation of cIII were mapped in a region covering approximately 85 nucleotides upstream of the AUG start codon and 45 nucleotides in the coding region (Altuvia & Oppenheim, 1986; Kornitzer et al., 1989). Two alternative mRNA structures were predicted and verified experimentally by structure probing in vitro and in vivo (Altuvia et al., 1989, 1991). One structure sequesters part of the Shine–Dalgarno (SD) sequence and the AUG start codon, whereas the other structure leaves the translation initiation region accessible to the ribosome (Fig. 2). Single-point mutations are able to lock the RNA into either one or the other conformation. The equilibrium between both structures is temperature-dependent. High temperatures (45°C) favour the energetically more stable conformation in which the ribosome-binding site is occluded. Blocked translation of cIII results in decreased cII concentration and initiates the lytic pathway. Physiological temperatures (37°C) shift the equilibrium towards the structure in which the ribosome binding is available and cIII protein is produced.

Figure 2.

  Temperature-dependent sequestration of the λ cIII ribosome-binding site. Grey ovals, ribosome; AAGGAG, Shine–Dalgarno (SD) sequence; AUG, start codon.

Interestingly, the switch region is preceded by two additional stem-loop structures which are identical in either conformation (Fig. 2). The integrity of the second stem loop is required for efficient ribosome binding (Altuvia et al., 1991). It has been proposed that this region is involved in tertiary RNA–RNA interactions (Altuvia et al., 1989, 1991). It has also been shown that RNaseIII binds to this stem loop and it is assumed that binding (but not RNA cleavage) introduces a structural transition that stimulates translation (Altuvia et al., 1987). Based on this finding, the cIII mRNA might be considered a protein-dependent riboswitch. However, the temperature-induced structural alterations shown in vitro (Altuvia et al., 1989) demonstrate that it is a bona fide RNA thermometer.

What might be the physiological reason for temperature control of the cIII gene? The lysis-lysogeny decision of phage λ is regulated by a complex network of transcriptional and post-transcriptional mechanisms in response to multiple environmental signals, among them nutritional status and ambient temperature. The phage tends to proliferate effectively when the host cells are healthy and it is more likely to integrate into the chromosome when growth conditions are poor. However, under life-threatening conditions (e.g. ultraviolet (UV) damage, severe heat shock) it might be beneficial for the phage to escape from the host. The amount of cII protein is critical to the fate of the phage. Enhanced degradation of cII by the heat-inducible FtsH protease in combination with reduced production of the competitor cIII will stimulate the lytic pathway under severe heat stress conditions.

Heat-shock gene expression

A sudden temperature upshift poses a serious threat to the integrity of almost all cellular macromolecules. The structure of membrane lipids, DNA, RNA and proteins is altered as the temperature rises. The expression of heat-shock proteins is a universal response found in all living cells. The induced chaperones or proteases constitute a protein quality-control system that either rescues misfolded proteins or promotes their degradation (Gross, 1996; Yura et al., 2000). The cellular concentration of heat-shock proteins is regulated primarily at the level of transcription and involves positive control by alternative sigma factors or negative control by repressor proteins (Narberhaus, 1999; Yura et al., 2000; Servant & Mazodier, 2001; Schumann, 2003). An additional level of control is often brought about by built-in feedback loops. A subset of heat-induced chaperones or proteases is commonly recruited to control the activity or stability of the regulatory factors involved. For example, the activity and stability of the E. coli heat-shock σ factor σ32 (RpoH) is tightly controlled by the DnaK chaperone and the FtsH protease (Straus et al., 1990; Herman et al., 1995; Tomoyasu et al., 1995). While this mechanism acts on the already established cellular pool of the transcription factor, an RNA thermometer in the rpoH gene ensures that translation efficiency of the master regulator of the heat-shock response correlates with the growth temperature.

The E. coli rpoH gene

Expression of the rpoH gene in E. coli is modulated by four different promoters which respond to different environmental signals (Yura et al., 2000) and by two intragenic segments (regions A and B) in the coding sequence (Fig. 3a). As rpoH transcription is only moderately altered in response to heat stress, translational control was responsible for accumulation of the sigma factor at high temperatures (Tilly et al., 1986; Straus et al., 1987). Using a large set of rpoHlacZ fusions it was found that two distinct 5′-proximal regions in the coding sequence of rpoH are responsible for translational control (Nagai et al., 1991). Secondary structure formation between these two regions was postulated to repress translation initiation at low temperatures (Fig. 3a). Disruption of that structure by a temperature upshift would liberate the ribosome-binding site. Similar structures were predicted to occur in the rpoH genes from other Gram-negative bacteria (Nakahigashi et al., 1995). Experimental evidence for such an RNA structure was generated by deletion analysis, point mutagenesis and structure probing of E. coli rpoH (Morita et al., 1999a). The structure in stem I appears to be weak as several nucleotides were modified by chemical probes, although they are predicted to base pair. A certain instability of the RNA structure might be required to permit residual ribosome entry and basal production of the σ factor at low temperatures. Starting from a lacZ reporter fusion downstream of box B, a minimal RNA thermometer was created (Morita et al., 1999a). Its secondary structure was trimmed down to the SD region followed by stem loops I and IV (Fig. 3c). In the minimal fusion construct, mismatches had to be introduced into stem I in order to maintain heat inducibility, indicating that imperfect base pairing is required for proper thermosensing. It is notable that stem loops II and III were dispensable only when certain destabilizing mismatches were introduced into the remaining sequence. It thus seems that each individual hairpin plays a distinct role in the complex architecture of the wild-type structure. Stem loop III was suggested to function as a ‘wedge’ between stems I and II, conferring the instability to the mRNA structure required for melting in the physiological temperature range (Morita et al., 1999a). Whether the region downstream of stem I opens up entirely after heat shock or whether movement of the ribosome during translation disrupts the remaining base pairs is not known.

Figure 3.

  Regulatory elements of the Escherichia coli rpoH gene (a) and schematic drawing of the secondary structures involved in translational control (b, c). The entire rpoH gene is comprised of 852 nucleotides and regions A and B extend from 6 to 20 and 112 to 208, respectively. The actual secondary structure of the region represented by dotted lines is more extensive than shown in (b). The minimal, functional thermoregulator is illustrated in (c). AAGGAG, Shine–Dalgarno (SD) sequence; AUG, start codon.

Further support for the importance of secondary structure formation was obtained by an extensive mutagenesis study (Morita et al., 1999b). Point mutations predicted to loosen the RNA structure increased reporter fusion expression, whereas mutations predicted to increase RNA stability reduced expression. Compensatory mutations recovered almost normal expression demonstrating that the mRNA structure, rather than the primary nucleotide sequence, is critical for thermoregulation. In agreement, circular dichroism (CD) spectroscopy of synthetic rpoH RNA fragments revealed a strong negative correlation between the RNA thermostability in vitro and corresponding expression levels in vivo.

The key prediction of the RNA thermometer model is that ribosomes have access to the rpoH transcript only at high temperatures. Indeed, toeprinting experiments (primer extension inhibition assays) showed that high temperatures were required for ribosome loading (Morita et al., 1999b). Full-length reverse transcripts initiated from a primer complementary to the coding region of rpoH were obtained at 30°C. At 42°C, however, primer extension was blocked by the assembly of a ternary complex between the rpoH transcript, 30S ribosome and tRNAfMet. Experiments with RNA sequences carrying mismatches or stabilizing point mutations in the thermometer region were used to establish that ribosome binding is dependent upon the stability of the RNA structure. Evidence for the contribution of additional cellular factors was not found in these in vitro assays, suggesting that the RNA structure itself suffices as molecular thermometer.

Repression Of heat-Shock gene Expression (ROSE)

While the rpoH thermosensor reaches far into the coding region, most other thermometers are located primarily upstream of the translation start codon. A conserved sequence of about 100 nucleotides in the 5′-UTR of multiple small heat-shock genes in Bradyrhizobium japonicum was strongly indicative of a novel regulatory element (Narberhaus et al., 1996, 1998; Münchbach et al., 1999). The expression of a translational lacZ fusion to the hspA gene downstream of such a sequence was very low at 30°C, but increased dramatically when large internal regions were deleted. Based on this inhibitory effect, the cis-acting regulatory element was designated ROSE for Repression Of heat-Shock gene Expression (Narberhaus et al., 1998). The absence of ROSE-containing transcripts at low temperatures was interpreted initially as evidence for transcriptional control, presumably by a repressor protein binding to this conserved region. However, all attempts to isolate a regulatory factor binding to ROSE-containing DNA failed (Nocker et al., 2001a). The first evidence that the mRNA itself might comprise the regulatory entity originated from computer calculations using the mfold program (Zuker et al., 1999; Zuker, 2003). The ROSE1 RNA from B. japonicum was predicted to fold into a complex structure occluding both the SD sequence and AUG start codon (Fig. 4a). ROSE elements from related Rhizobiaceae such as Mesorhizobium loti, Rhizobium sp. NGR 234 and Agrobacterium tumefaciens (Fig. 4b) are assumed to fold into very similar structures (Nocker et al., 2001a, b; Balsiger et al., 2004). The regulatory potential of these structures to act as RNA thermometers was examined by mutational studies with ROSE1 preceding the hspArpoH1 operon of B. japonicum (Fig. 4a). Introduction of mismatches in the paired regions of stem-loop IV increased expression of translational lacZ fusions, compensatory mutations restored repression and alterations in unpaired nucleotides were neutral, demonstrating that a structured RNA rather than a particular RNA sequence is critical for thermosensing (Nocker et al., 2001a; Chowdhury et al., 2003).

Figure 4.

Repression Of heat-Shock gene Expression (ROSE) elements from Rhizobiaceae. The predicted secondary structures of Bradyrhizobium japonicum ROSE1 (a) and Agrobacterium tumefaciens ROSEAT2 (b) is shown. Transcriptional start sites (+1), Shine–Dalgarno sequence (SD) and the translational start site (START) are indicated. A conserved bulged G is marked with an arrow. The four hairpin structures of ROSE1 are labelled I, II, III and IV.

A common feature of all ROSE elements is the presence of a G residue opposite to the SD sequence that appears to be bulged in all the predicted structures (Fig. 4). This nucleotide was found to be functionally important, as its elimination made the thermometer unresponsive to high temperatures (Nocker et al., 2001a). Presumably, optimized base pairing with the SD sequence in this mutant prevents opening of the folded structure in the temperature range between 30 and 40°C. It is currently unknown whether the entire ROSE RNA melts as the temperature increases, or whether local perturbations around the ribosome-binding site are sufficient to initiate translation. Only subtle structural changes in a synthetic ROSE transcript were recorded by UV and CD spectroscopy, suggesting that the latter might be the case (Chowdhury et al., 2003). In vitro, thermally induced alterations were reversible provided that the temperature was not raised far beyond the physiological temperature range.

Recent nuclear magnetic resonance (NMR) studies with a portion of the fourth stem loop containing the SD sequence showed that opening of the thermometer begins at an internal loop adjacent to the SD sequence, which appears to be the key temperature-sensing element (Chowdhury et al., unpublished results). The conserved G residue opposite the SD sequence closes the loop by a weak GG base pair which breaks at higher temperatures, making the ribosome-binding site accessible. Deletion of the G leads to the formation of a more stable hairpin that fails to open up within the physiological temperature range.

Similar to the cIII and rpoH thermometers, all ROSE structures consist of more than a single stem loop sequestering the SD sequence and AUG codon. In each case, two or three additional structural elements are present (Figs 2–4). Truncation of stem loops I–III of ROSE1 retaining only hairpin IV did not abolish repression at 30°C but reduced heat induction in vivo (Chowdhury et al., 2003). Mutations that destabilized the top of stem loop III resulted in enhanced expression at low temperatures. Compensatory mutations in the same region restored repression (Nocker et al., 2001a). Proper temperature sensing might require the entire and intact structure of the ROSE mRNA. The hairpins upstream of the one that sequesters the ribosome-binding site might have several important functions. (i) They might serve as a zipper enforcing correct folding of the thermometer as the RNA is being transcribed. It is noteworthy that the initial nucleotide (+1) of the ROSE transcript is paired (Fig. 4). Folding from the very beginning of the transcript might provide a scaffold preventing the ultimate stem loop engaging in fortuitous interactions when it emerges from the RNA polymerase. The observation that synthetic ROSE RNA did not return to its original conformation after it was heat-denatured at 80°C supports the importance of cotranscriptional folding (Chowdhury et al., 2003). (ii) In analogy to the proposed wedge function of stem III in the rpoH thermometer (Morita et al., 1999a), it is conceivable that structures I–III in the ROSE-containing transcript might be involved in tertiary contacts with the SD region. With increasing temperature, these tertiary interactions might act in a cooperative fashion once the melting commences at the internal loop and help to open up the entire ribosome-binding site to facilitate 30S binding and translation initiation. (iii) The complex architecture might also play a role in another aspect of ROSE-mediated regulation. In vivo, ROSE-containing mRNA is barely detectable at low temperatures, suggesting that the transcript is rapidly degraded when it is in the folded conformation. After a shift to 42°C, however, the translation-competent transcript is extremely abundant (Narberhaus et al., 1996, 1998; Nocker, 2001b). It is possible that the folded RNA displays recognition motifs to cellular RNases, which eliminate the untranslatable transcript. The presence of an instability element early in the transcript would make economic sense. Stabilization of the RNA at higher temperature could be achieved by melting of recognition motif(s) and/or by shielding of the translated transcript by the ribosome.

All currently known ROSE elements derive from Rhizobiaceae (Münchbach et al., 1999; Nocker et al., 2001b; Balsiger et al., 2004). A search for ROSE-type sequences in completed bacterial genome sequences using the basic linear alignment sequence tool (blast) program (Altschul et al., 1997) did not provide new candidates, either because they do not exist or because the sequence conservation is too low. The latter seems to apply, as a closer bioinformatic inspection of the upstream region of genes coding for small heat-shock proteins revealed a number of ROSE-type sequences (Fig. 5a; T. Waldminghaus et al., unpublished results). Some were found in Alphaproteobacteria belonging, or related closely, to Rhizobiaceae such as Brucella mellitensis and Caulobacter crescentus. More surprisingly, good candidates also exist in Gammaproteobacteria; for example, upstream of small heat-shock genes in E. coli, Salmonella or Vibrio. The predicted secondary structures of C. crescentus and E. coli ROSE elements are shown in Fig. 5b and c, respectively.

Figure 5.

  Predicted Repression Of heat Shock gene Expression (ROSE)-like sequences and structures from Gram-negative bacteria. The principal modules of ROSE-type thermometers with the promoters (−35 and −10), the transcription start site (+1), several hairpin structures (I–IV), the Shine–Dalgarno sequence (SD) and the AUG codon (START) is drawn at the top of panel (a). An alignment of region IV of predicted ROSE elements from Alpha- and Gammaproteobacteria is shown below. The predicted secondary structures of two candidate RNA thermometers are presented in (b) and (c). Characteristic features such as the predicted transcription start (+1), the SD sequence and a conserved bulged G (arrow) are labelled.

According to their amino acid sequence, bacterial small heat-shock proteins can be divided into two distinct classes, A and B (Münchbach et al., 1999; Narberhaus, 2002b). Class A proteins are related closely to inclusion body-binding proteins A and B (IbpA and IbpB), the two E. coli small heat-shock proteins. Class B proteins comprise a very heterogeneous family. Many rhizobia contain both class A and class B proteins. All ROSE elements found so far are linked to genes coding for class A small heat-shock proteins. Neither Gram-positives nor Archaea seem to contain ROSE-like sequences, probably because their small heat-shock genes belong to class B.

The presence of four hairpin structures might be restricted to Bradyrhizobium sp. (Fig. 4a). Other ROSE elements seem to consist of three stem loops equivalent to the bradyrhizobial stems II, III and IV (Fig. 5a). It is important to note, however, that these predictions are preliminary, as the transcriptional start sites of most of the candidate thermometers are unknown. Sequence conservation in the first hairpins is only marginal and hence not included in the alignment (Fig. 5a). Stem loop IV, on the other hand, is characterized by several highly conserved residues, including the SD region and the complementary sequence blocking it at low temperatures. It is evident from all the studies conducted so far that the G, appearing as bulged from the predicted secondary structures just opposite to the SD sequence, is universally conserved and its deletion leads to severe compromise of the thermosensitivity of ROSE. Interestingly, exchanging it for an A in ROSE1 did not alter expression (Nocker et al., 2001a).

In summary, the bioinformatic search uncovered many new ROSE-type RNA thermometers in the 5′-UTR of class A small heat-shock genes from Alpha- and Gamma-proteobacteria. They are predicted to fold into complex secondary structures which have an evolutionarily conserved hairpin in common, including the SD sequence. ROSE elements are clearly the most widely distributed RNA thermometers known to date.

Other cis-acting heat-shock sensors in prokaryotes

That rpoH- and ROSE-type thermometers were found only in Alpha- and Gammaproteobacteria does not exclude the fact that other organisms make use of RNA thermometers with entirely different sequence and structure. Given that this simple mechanism was predicted to originate from an early RNA world (Narberhaus, 2002a; Lai, 2003; Mandal & Breaker, 2004b; Vitreschak et al., 2004), the presence of RNA thermometers in other bacteria and in Archaea is entirely conceivable. In fact, potential RNA thermometers have been described in various organisms. The C. crescentus dnaKJ transcript comprises an extended 5′-UTR, in which the SD sequence is trapped by complementary base pairs (Avedissian et al., 1995). Elimination of most of the UTR stimulated expression. Temperature-controlled accessibility of the SD sequence was also postulated for the Haemophilus ducreyi dnaK transcript (Parsons et al., 1999). Translation of the small heat-shock gene hsp18 in the Gram-positive organism Streptomyces albus is temperature dependent. Here, too, a predicted secondary structure in the 5′-UTR might interfere with ribosome binding at low temperatures (Servant & Mazodier, 2001). Although a great deal more work is required to demonstrate that these candidates qualify as RNA thermometers, the findings already suggest that many new RNA-based heat-shock sensors remain to be discovered in phylogenetically diverse prokaryotes.

RNA thermometers in eukaryotes?

There is accumulating evidence that mRNA folding also might be an important parameter in translational control of some eukaryotic heat-shock genes. As in prokaryotes, thermal stress in eukaryotes triggers a heat-shock response (Morimoto, 1998). In addition to transcriptional control, expression of heat-shock genes can also be modulated at the level of mRNA processing (Yost & Lindquist, 1991) and by control of translation initiation. Here we discuss some of the diverse modes of translational control adapted by different eukaryotic systems to combat thermal stress.

Translation initiation in eukaryotes differs substantially from the prokaryotic process and proceeds by two principal pathways. The first, termed ‘cap-dependent translation initiation’, involves the interaction of a 5′-cap with the cap-binding protein complex that mediates recruitment of the small ribosomal subunit at the extreme 5′-end of the mRNA. Cap-dependent translation initiation is severely compromised under heat-shock conditions (Lamphear & Panniers, 1991; Cuesta et al., 2000). A second pathway, which is emerging as the potential choice for eukaryotic cells under stress, uses complex secondary structure elements in the mRNA termed ‘internal ribosomal entry segments’ (IRES). They are present in the mRNAs of several survival factors and act directly by RNA–ribosome contacts or indirectly via initiation factors that can bind both the IRES and the ribosome (Holcik et al., 2000; Von der Haar et al., 2004).

Early investigations with Hsp70, an abundant protein under heat-shock conditions in eukaryotes, showed that the 5′-UTR of its mRNA is required for efficient translation in Drosophila cells subjected to heat shock (Klemenz et al., 1985; McGarry & Lindquist, 1985). Evidence for specific RNA-binding factors was not found (Hess & Duncan, 1994). It emerged that minimal secondary structure is the key element of preferential translation during heat stress in most invertebrates. Introduction of structural elements into the 5′-UTR of hsp70 mRNA inhibited translation at higher temperatures without affecting it at unstressed temperatures (Hess & Duncan, 1996). It has been suggested that under thermal stress, the capacity of the translational machinery to unwind the 5′-UTR secondary structure of the mRNA is reduced, which might result from shortage of one or more translation initiation factor activities that are absent under thermal stress in Drosophila. Under such circumstances, hsp70 expression is mediated by the IRES-dependent pathway (Hernandez et al., 2004).

Hsp 90 is an abundant protein in Drosophila cells grown at normal temperatures, with its mRNA present at significant levels (Lindquist & Craig, 1988). The 5′-UTR of hsp90 mRNA is the only Hsp mRNA identified so far in Drosophila that possesses significant secondary structural elements typical to non-heat-shock mRNA (Ahmed & Duncan, 2004). Two modes of translation of the hsp90 mRNA were postulated. Under normal temperature conditions, translation proceeds through the cap-dependent scanning mechanism. This process is relatively inefficient because of secondary structures. During heat shock, translation is assumed to switch to another mechanism that involves a more direct recruitment of the small ribosomal subunit to the AUG initiation codon promoted by mRNA–rRNA base pairing. Deletion of nucleotides close to the AUG severely compromised heat-dependent translation but did not substantially affect translation under non-heat-shock conditions. In analogy to bacterial RNA thermometers, it was postulated that the preferential translation of hsp90 mRNA under thermal stress might rely on temperature-controlled alterations of the mRNA structure (Ahmed & Duncan, 2004). A related mechanism might be responsible for selective translation of human hsp70 mRNA during heat stress (Yueh & Schneider, 2000).

The IRES-dependent translation of mRNA has also been shown to be crucial for survival and proliferation of cells under stress such as therapeutic radiation and hypoxia-induced DNA damage. Although it might be a preferred route when the cap-dependant pathway is inactive, there have been few studies on its role in heat-shock gene translation. It was reported to play a role in translation control of the heat-shock response of BiP, a Hsp 70 family member that facilitates protein folding, assembly of nascent proteins and scavenging of misfolded proteins in the endoplasmic reticulum (Kim & Jang, 2002). Enhanced translation of BiP mRNA mediated by its IRES was registered in HeLa cells under heat-shock conditions, suggesting that this mode of translation might be critical to the survival and proliferation of cells under thermal stress.

In trypanosomatids extreme environmental changes, e.g. alterations of temperature and extracellular pH, induce developmental gene expression. Hsp70 plays an important role in cell survival during developmental changes and is regulated exclusively at the post-transcriptional level (Vanhamme & Pays, 1995). Although Hsp70 and Hsp83 are expressed constitutively throughout the life cycle of Leishmania, translation of the heat-shock transcripts increases at least 10-fold upon temperature elevation (Shapira et al., 1988). Unlike in other higher eukaryotes, the responsible regulatory element of Leishmania and Trypanosoma brucei hsp70 resides in the 3′-UTR (Lee, 1998; Quijada et al., 2000). Deletion analysis confirmed that sequences within the 3′-UTR of hsp83 mRNA are essential in stabilization and preferential translation of this heat-shock transcript in Leishmania (Zilka et al., 2001). The 5′-UTR could not generate this effect alone but had synergistic effects on increasing translation at higher temperatures when combined with the 3′-UTR. This result indicates a possible long-range interaction between the 5′-UTR and the 3′-UTR, which has also been reported in other eukaryotic systems (Jarzembowski et al., 1999).

Virulence gene expression

Bacteria living in the soil, in water or in food usually experience a temperature upshift when swallowed by mammals. The increase in temperature might not be severe enough to induce a heat-shock response, but it comprises an important signal for pathogenic micro-organisms that a warm-blooded host has been entered successfully. As a consequence, many virulence genes are under temperature control (Konkel & Tilly, 2000). The regulatory mechanisms involved are diverse. Several pathogens control transcription of virulence genes by temperature-induced changes in DNA supercoiling, which modulates the binding of regulatory proteins. Others use protein thermometers, i.e. regulatory proteins that alter their conformation in response to temperature (Hurme & Rhen, 1998). That a thermosensor might also reside in the mRNA of a virulence gene was first postulated in the early 1990s. The cellular level of the Yersinia pestis virulence gene activator LcrF was found to be much higher at 37°C than at 26°C, although transcription was insensitive to temperature (Hoe et al., 1992; Hoe & Goguen, 1993). Temperature-dependent LcrF production was also observed in E. coli when transcription was driven from a T7 promoter, suggesting a post-transcriptional expression signal in the lcrF message itself (Hoe & Goguen, 1993). Although degradation of lcrF mRNA might play a role, differential translation of the message was suggested to be the main reason for the increase of the activator protein at high temperatures. A secondary structure in the lcrF mRNA was predicted to sequester the translation initiation region. Thermal destabilization of the stem loop allowing translation initiation was calculated to occur in the physiologically relevant temperature range between 33 and 37°C (Hoe & Goguen, 1993).

A similar mechanism was shown to control virulence gene expression in the food-borne pathogen Listeria monocytogenes. The virulence gene activator PrfA is not present at low temperatures, although the prfA transcript is being made (Renzoni et al., 1997). An extended stem-loop structure was predicted in the leader sequence of about 120 nucleotides (Fig. 6). Chemical probing experiments and the introduction of a series of mismatches, compensatory and stabilizing mutations suggested strongly that this structure is responsible for thermoregulation of prfA expression (Johansson et al., 2002). Listeria mutants engineered to produce PrfA at 30°C owing to destabilized leader sequences were able to invade cultured kidney cells. Translational control of the prfA gene was retained when it was transferred to E.coli, although L.monocytogenes (Gram-positive) and E.coli (Gram-negative) are only distantly related bacteria using quite different gene expression mechanisms. This observation provides strong evidence that no Listeria-specific factor is involved in translational control and that the mRNA alone functions as a thermometer.

Figure 6.

  Schematic drawing of the prfA thermometer from Listeria monocytogenes. The dotted loop indicates that the top of the extensive hairpin structure is not drawn to scale.

In contrast to the heat-shock sensors discussed above, the prfA structure is not branched and consists of a single stem loop. Base-pairing with the SD sequence is weak and the AUG codon is not masked (Fig. 6). It is thus conceivable that the virulence thermometer is more sensitive than the others and functions at slightly lower temperatures. However, a detailed comparative expression analysis would be required to address this question. Unlike ROSE-containing transcripts, the prfA mRNA does not seem to be processed at low temperatures. Northern blot analysis revealed equivalent amounts of the monocistronic transcript at 20 and 37°C (Johansson et al., 2002).

Cold-shock gene expression

Like heat, cold is a physical parameter that severely affects multiple processes in a living cell. A temperature drop influences membrane fluidity as well as the conformation and functionality of nucleic acids and proteins. As a protective measure, most bacteria are able to mount a cold-shock response inducing so-called cold-shock proteins (CSPs), which have diverse and partly unknown functions (Graumann & Marahiel, 1996; Phadtare et al., 2000). Regulation of CSP genes occurs primarily at the post-transcriptional level. Low stability of CSP mRNAs at normal temperatures plays a major role, but various layers of translational control are also involved (Gualerzi et al., 2003). Certain proteins, including initiation factors and CSPs, some of which act as RNA chaperones, might contribute to the preferential translation of cold-shock genes at low temperatures (Phadtare et al., 2000; Gualerzi et al., 2003). In the context of this review, it is interesting that a conspicuously long 5′-UTR consisting of 159 nucleotides in the E. coli cspA mRNA has been postulated to act as an RNA thermometer (Yamanaka et al., 1999). A deletion analysis in the 5′-UTR revealed that cspA expression did not correlate with the amount of transcript, suggesting that the leader sequence controls translation efficiency. Alternative RNA structures at high and low temperatures are thought to modulate the accessibility of the ribosome-binding site (Yamanaka et al., 1999). Stability control of the cspA mRNA might also be because of alternative RNA structures. Rapid degradation of the transcript by RNase E occurs at high temperatures when the recognition site is predicted to be single-stranded and accessible (Fang et al., 1997).

Thermoregulation by the small noncoding RNA DsrA

So far, we have discussed inbuilt cis-acting thermosensors that control mRNA translation upon thermal signals. A segment of RNA complementary to the ribosome-binding site is located in more or less close proximity, allowing intramolecular base pairing. A different regulatory mode is carried out by trans-encoded antisense RNAs. Here, a small noncoding RNA is expressed as an independent transcription unit from its target. Often these antisense RNAs have multiple targets, and some targets are regulated by multiple noncoding RNAs, thereby generating a network of global regulation (Wassarman et al., 1999; Gottesman, 2002; Storz, 2002).

The only documented example of temperature-regulated translation by a trans-encoded antisense RNA is found in the regulation of RpoS. To combat environmental challenges, E. coli triggers expression of many genes involved, both in temporary emergency responses and in long-term adaptation processes. The transcription of many of these genes is dependent on the stationary phase or starvation σ factor, RpoS (σ38). RpoS is more abundant at 25°C than at higher temperatures, and RpoS-dependent promoters are expressed efficiently at low temperatures (Sledjeski et al., 1996). The abundance of RpoS at low temperatures results from thermal regulation of its translation by an 85-nucleotide small noncoding RNA, the DsrA RNA. Temperature is one of the environmental factors that determine the amount of DsrA present in the cell at two stages: transcription initiation and stability of the antisense RNA. As a consequence, the amounts of DsrA are higher at lower temperatures (Sledjeski et al., 1996; Repoila & Gottesman, 2001).

The full-length 85-nucleotide long DsrA primary transcript (F form) is processed to a 60- or 61-nucleotide form (T form) lacking the 5′-region which interacts with the rpoS mRNA. The F form is more stable at 25°C than at 37 or 42°C, thus outscoring the T form at low temperatures (Repoila & Gottesman, 2001; Repoila et al., 2003). Interaction of the F form with the leader sequence of the rpoS transcript relieves translational repression by disrupting complementary base pairing of the region flanking the SD sequence (Fig. 1B). Removal of the 5′-end of the F form might occur at an RNase E cleavage site between two stem loops in the DsrA RNA (Moll et al., 2003). Exactly how temperature influences processing of the small RNA is unclear. However, because temperature reportedly controls the abundance and stability of the active conformation of DsrA it has been considered an RNA thermometer of the RpoS regulon (Repoila et al., 2003).

Conclusions and perspectives

The chemical nature of RNA provides a vast repertoire of options for intra- and intermolecular interactions which can be used for regulatory purposes. RNA can utilize reversible temperature-dependent conformational alterations to tune gene expression. Also, the inherent short half-life of mRNA makes it an attractive regulatory candidate for transient temperature changes.

In terms of a rapid response, translational control is one step ahead of transcriptional control as it acts on already-existing RNA. Economically, it may not be the cheapest solution to produce a transcript that is not translated (or even degraded). However, the option to induce stress responses or virulence genes at appropriate temperatures very rapidly might be worth the expense. To avoid further biosynthetic costs caused by undesired cotranscriptional translation, most thermometers are placed in the 5′-UTR rather than in the coding sequence.

Although most RNA thermometers function by a conceptually simple melting mechanism, the underlying molecular details are far from being understood. Some of the open questions that remain to be addressed in the future are as follows:

  • 1Structure of known RNA thermometers. Computer-aided secondary structure predictions have been extremely helpful in establishing the RNA thermometer concept. However, it appears from preliminary NMR studies that the actual ROSE structure around the SD sequence deviates, at least in part, from the computer-calculated secondary structure. Hence, a detailed analysis of a number of structurally and functionally different RNA thermometers by structure probing techniques, NMR and/or X-ray crystallography is required in order to comprehend fully their mode of action.
  • 2The melting process. Matters are complicated by the fact that most RNA thermometers exhibit a complex architecture consisting of several hairpins whose molecular interplay is not yet understood. As described in previous sections, mutational studies have implied that the modular design of the rpoH and ROSE thermometers is functionally important. The calculated melting temperature of entire thermometer sequences is usually far above the relevant physiological temperature. It is possible that translation initiation requires only local destabilization around the ribosome-binding site and/or depends on the coordinated action of individual stem loops.
  •   Another complication is that RNA thermometers probably do not have two distinct alternative conformations and therefore, in contrast to riboswitches, they are unlikely to act as an on/off switch. Instead, RNA folding–unfolding dynamics seems to play an important role in this case. RNA thermometers presumably use a dimmer-type mechanism progressing through multiple subtly different structures along a temperature gradient. Resolving the exact secondary and tertiary interactions and their dynamic response to temperature will be a major challenge for future studies.
  •   Another important task will be to provide direct experimental proof for the presumed translational control mechanism. Except for the rpoH mRNA (Morita et al., 1999b), it remains to be demonstrated that binding of RNA thermometers to the ribosome correlates directly with temperature. To establish whether accessory factors other than the transcript and the 30S particle are required for thermosensing, toe-printing studies with purified components are needed.
  • 3Stability of thermometer containing transcripts. Some RNA thermometers are postulated to decay when they are not translated (Nocker et al., 2001a); why the cell produces a transcript simply to degrade it is unknown. Interestingly, disposal of unused RNA sensors seems to be a common principle, as several riboswitch-containing transcripts also are intrinsically unstable (Hoe & Goguen, 1993; Nou & Kadner, 1998; Winkler et al., 2004). Neither the identity of the RNAses in action nor the decay mechanism is known.
  • 4Novel RNA thermometers. The most speculative aspect is whether there are new RNA thermometers to be discovered. A particularly interesting question relates to the existence of RNA thermometers in psychrophilic or thermophilic micro-organisms. Entirely new structures are to be expected. Also, we still do not know whether regulation of genes not coding for heat-shock, cold-shock or virulence proteins can be mediated by similar RNA elements.
  •   The low sequence and structure conservation among the known RNA thermometers poses a principal challenge, as it provides little prospect for homology-based searches. To this end, the ongoing development of an advanced bioinformatic analysis tool using improved structure modeling algorithms is promising (Mathews et al., 1999; Zuker et al., 1999; Hofacker, 2003; Gardner & Giegerich, 2004; Voss et al., 2004).
  • 5Novel mechanisms. It will also be interesting to discover whether the regulatory capacity of RNA thermometers extends beyond translational control. All known thermosensors act by controlling access to the ribosome-binding site. So far, examples for control of transcription termination or ribozyme activity as is used in some riboswitches (see Fig. 1a) have not been found.

The concept of gene regulation by RNA sensors has emerged only very recently. It is surprising that 2% of all Bacillus genes have been predicted to be regulated by riboswitches, most of them controlling fundamental metabolic processes (Mandal et al., 2003). New RNA sensors are being discovered in rapid succession, demonstrating that our knowledge on the regulatory potential of RNA is just beginning to unfold. Evidently, there is a message in many untranslated regions. We just need to read it.


Previous support by the ETH Zürich (TH 0-20037-02) and present funding by the German Research Foundation (DFG NA240/4-1) is gratefully acknowledged.