Conserved sequences associated with myriad antibiotic resistance genes are integron components and not aminoglycoside riboswitches

In a recent publication in Cell (Jia et al., 2013), the authors conclude from bioinformatics, genetic, and biochemical analyses that a conserved sequence element associated with antibiotic resistance genes represents a novel widespread riboswitch class for certain aminoglycosides. Below, we point out previously reported functions for this element that were not discussed by the authors, and we note other issues with the validation experiments that raise significant concerns about the conclusions of the study.

Most riboswitches contain highly conserved ligand-binding domains, and therefore the identification of consensus sequence and structure models is critical in providing support for proposed riboswitch classes. Jia et al. describe a conserved sequence element that is frequently associated with aminoglycoside acetyl transferase (aac) and aminoglycoside adenyl transferase (aad) genes, and ultimately conclude that it corresponds to mRNA leader sequences with riboswitch functions. However, the consensus sequence they describe was already known to be a key component of integrons, serving as a recombination site (attI) that facilitates the exchange of gene cassettes sometimes involved in antibiotic resistance (Mazel, 2006).

Sequences in their alignment (Data S1) correspond specifically to the attI1 class of recombination sites (Partridge et al., 2000), and the locations of these sites – within class 1 integrons, between the intI1 gene encoding the integrase and the integron cassettes – are consistent with a function in recombination. Importantly, the most highly conserved segments in the alignment, which contain the sequence defined by the authors as the riboswitch aptamer domain, precisely encompass the regions of the attI1 DNA that interact with the integrase and that are necessary for recombination to occur (Recchia et al., 1994; Collis et al., 1998; Gravel et al., 1998). Moreover, there are no segments other than those corresponding to the attI1 site that are universally conserved among the aligned sequences. Thus, if these sequences also function at the level of RNA as aminoglycoside riboswitches (in addition to their known functions at the DNA level as specialized recombination sites), the conserved domains responsible for each independent function would have to be perfectly coincident.

Although Jia et al. mention in the Discussion that their reported sequences are “constituent(s) of the integron cassette system”, they never state that these sites have biochemical functions that were already well understood. Given the previously established function of this element, it seems unlikely that the conserved nucleotides are also important for riboswitch function. This concern is reinforced by the fact that the locations of attI1 sequences are not limited to regions near aminoglycoside resistance genes as implied by the data depicted in Figure 1C. Integrons containing attI1 sites commonly contain gene cassettes involved in resistance to a wide range of antibiotics including trimethoprim, rifampicin, chloramphenicol, tetracycline, erythromycin, sulfonamides, quinolones, quaternary ammonium compounds, and various beta-lactams (Moura et al., 2009). It seems counterintuitive for the optimal expression of these myriad antibiotic resistance genes to be contingent on the function of a riboswitch that responds only to a limited number of aminoglycoside derivatives.

In addition, there are numerous instances of aminoglycoside resistance genes occurring in contexts apart from integrons, many examples of which are listed by the authors in Table S1A. In such instances, the aminoglycoside resistance genes are never observed to be associated with attI1 sites. It is only when these genes are contained within integrons that attI1 sites are present. This strict association of attI1 sequences with integrons, rather than with aminoglycoside resistance genes, provides further evidence that these elements function exclusively in their known capacity as integron components, and not as aminoglycoside riboswitches.

In light of the previously defined role of attI1 sites, interpretations of the authors’ biochemical and genetic data should be assessed more cautiously. As the authors note, the cationic amino groups of aminoglycosides result in a propensity for binding interactions with polyanionic RNA. Several reports have demonstrated that specific aminoglycosides have pronounced effects on the activities and structures of various RNAs (e.g. see Hermann and Westhof, 1999). In many of these examples, however, the interaction between the antibiotic and the target RNA is presumed not to be biologically relevant. Such studies underscore the hazards of imputing biological significance to interactions in vitro between RNA and aminoglycosides, and suggest that the binding assays of Jia et al. likely reflect spurious interactions. Consistent with this possibility is the fact that the most prominent site of aminoglycoside-dependent structure modulation observed by the authors (SD2; Figure 3C) occurs outside of the attI1 sequence, in one of the least conserved regions of their sequence alignment.

Jia et al. also prepared genetic fusions between the attI1 DNA sequence and a β-galactosidase gene and observe 1.5- to 2.5-fold increases in gene expression in response to antibiotics such as kanamycin B and sisomycin. However, these increases are exceedingly modest compared to the more typical 10- to 1000-fold dynamic ranges for gene regulation achieved by known riboswitches. Moreover, these aminoglycoside antibiotics function by targeting ribosomes, which will inherently alter global protein production and therefore could easily cause small changes in reporter gene expression by mechanisms that are independent of a riboswitch. Although the authors conducted experiments intended to rule out a ribosome-based mechanism for the gene expression changes they observe, the resulting data from these experiments are more easily explained by antibiotics binding to ribosomes instead of a riboswitch.

The authors propose a model in which an RNA containing the attI1 sequence undergoes a conformational change in response to specific aminoglycosides, resulting in increased levels of translation from the unmasked Shine-Dalgarno sequence (SD2). They also note (Figure 7B) that an ORF potentially encoding a short leader peptide overlaps the attI1 sequence. Although it is not stated by the authors, this ORF is a feature of attI1 sites that was reported previously (Hanau-Berçot et al., 2002). This coding region, termed ORF-11, functions to enhance the expression at the translational level of antibiotic resistance genes within class 1 integrons, probably by recruiting ribosomes to cassette sequences with deficient translation initiation regions. This earlier work also demonstrated that the efficacy of ORF-11 depends on its translation, but not on the sequence of the leader peptide.

Indeed, the mutational analysis of Jia et al. (Figure 5A,B) is entirely consistent with the gene control mechanism proposed by Hanau-Berçot et al. (2002). Changes in the attI1 sequence that interfere with translation of the leader peptide (M1, M2, M3) result in four- to six-fold decreases in expression of the β-galactosidase reporter gene. Moreover, sequences containing deletions that maintain the reading frame of the leader peptide (M9, M14, M15, M16) generate only two-fold reductions in reporter gene expression, even when these deletions result in the elimination of as many as nine nucleotides from a highly conserved domain of the attI1 RNA. In contrast, a deletion in the same region that alters the reading frame (M19) yields a 20-fold reduction in β-galactosidase expression. Collectively, these data offer considerably more support for the model of Hanau-Berçot et al. (2002), in which the enhancement in gene expression is derived from translation of the leader peptide, than for the model of Jia et al., which posits a riboswitch mechanism.

The authors also designed three mutant attI1 sequences (M23, M24, M25), each containing several conservative mutations that maintain the sequence of the leader peptide (Figure 7B). Constructs bearing these mutations generated substantial reductions in reporter gene expression as compared with levels generated by the wild-type sequence (Figure 7C). Jia et al. conclude from these experiments that high levels of gene expression are driven by the sequence of the attI1 RNA rather than by that of the leader peptide. The interpretation of this experiment is confounded, however, because each of the three mutants bears an additional point mutation, not present in the wild-type control, occurring two nucleotides upstream of the leader peptide initiation codon. This mutation (C11A), which is not discussed by the authors, creates a run of five consecutive purine residues that could potentially function as a decoy SD sequence and disrupt normal expression of the leader peptide. Even if this is not the case, it is possible that a mutation in this region could result in decreased translation of the leader peptide, and therefore might account for the reduced β-galactosidase levels in a manner consistent with the model of Hanau-Berçot et al. (2002).

Considering that the roles of attI1 sequences (both as class 1 integron recombination sites and as translational enhancers) have been previously established, that the data in Jia et al. are consistent with these functions, and that there are hundreds of known examples of attI1 sites positioned immediately upstream of genes conferring resistance to structurally diverse classes of antibiotics, readers should be circumspect in evaluating the authors’ claim that these sequences function as components of aminoglycoside-specific riboswitches.