|Home | About | Journals | Submit | Contact Us | Français|
Genetically encoded small molecule sensors are important tools for revealing the dynamics of metabolites and other small molecules in live cells over time. We recently developed RNA-based sensors that exhibit fluorescence in proportion to a small molecule ligand. One class of these RNA-based sensors are termed Spinach riboswitches. These are RNAs that are based on naturally occurring riboswitches, but have been fused to the Spinach aptamer. The resulting RNA is a fluorogenic riboswitch, producing fluorescence upon binding the cognate small molecule analyte. Here we describe how to design and optimize these sensors by adjusting critical sequence elements, guided by structural insights from the Spinach aptamer. We provide a step-wise procedure to characterize sensors in vitro and to express sensors in bacteria for live-cell imaging of metabolites. Spinach riboswitch sensors offer a simple method for fluorescence measurement of a wide range of metabolites for which riboswitches exist, including nucleotides and their derivatives, amino acids, cations, and anions.
Measuring the flux of cellular metabolites is important because metabolites exert a direct influence on cell physiology (Patti, Yanes, & Siuzdak, 2012). Metabolic pathways are highly linked to both signaling pathways and nutrient availability, and various lines of evidence show that alterations in cellular metabolite dynamics is a mediator of diverse diseases. In addition to imaging endogenously generated metabolites, there is considerable interest in imaging exogenously derived molecules such as antibiotics, drugs, and other xenobiotics.
Traditional methods for quantifying metabolite levels rely on liquid chromatography, NMR spectroscopy or mass spectrometry. However, these techniques require harvesting cells, and thus are not “continuous” assays. They do not enable quantification of metabolite levels in the same cells over time. Additionally, they do not reveal cell-to-cell variation in metabolite levels, which is important for understanding population heterogeneity.
Fluorescence imaging approaches can provide a continuous assay of metabolite concentrations in cells. There are two popular approaches for designing metabolite sensors. One approach is fluorescent protein (FP)-based sensors, which typically depend on the FRET signal generated when two fused FPs become closer together. This is regulated by a ligand-binding domain, such as calmodulin (Tian et al., 2009), that changes structure in response to binding the target ligand. The advantages of this approach are that it can be easily genetically encoded and that the fluorescence signal is ratiometric (Palmer, Qin, Park, & McCombs, 2011); however, the changes in signal are often not very pronounced. Another noteworthy approach is to derivatize organic dyes such that the dye fluorescence is activated by a metabolite-recognizing group, as is the case for Fura-2 and many reactivity-based sensors (Chan, Dodani, & Chang, 2012; Malgaroli, Milani, Meldolesi, & Pozzan, 1987; Valeur, 2000). Despite producing a robust signal, this approach can suffer from difficulty in delivering sensors intracellularly, toxicity problems, and the labor-intensive task of synthesizing novel dye-based sensors.
Recently a new type of sensor technology has been developed based on fluorogenic aptamers (Paige, Nguyen-Duc, Song, & Jaffrey, 2012; You, Litke, & Jaffrey, 2015) (Figure 1). These fluorescent RNA aptamers have been synthetically evolved to activate the fluorescence of an otherwise non-fluorescent dye that easily crosses membranes, shows minimal nonspecific fluorescence in cells, and is nontoxic. As an example, Spinach (Paige, Wu, & Jaffrey, 2011) binds to 3,5-difluoro-4-hydroxybenzylidene imidazolinone (DFHBI), an analogue of the chromophore formed in green fluorescent protein. Binding of the target metabolite is linked to the folding of Spinach, leading to fluorescence. This approach is genetically encodable since the RNA can be expressed within the cell.
Two types of approaches have been described: allosteric Spinach sensors (Paige et al., 2012), and Spinach riboswitches (You et al., 2015). The structure of allosteric Spinach sensors is shown in Figure 1A. The metabolite-binding aptamer is linked to a critical stem in Spinach via a short transducer region (Paige et al., 2012). In the absence of the metabolite target, the aptamer is unstructured and prevents Spinach from folding. However, metabolite binding results in hybridization of the transducer domain and folding of Spinach, leading to fluorescence. Optimal performance requires optimizing the transducer sequence. This approach has led to the development of sensors for adenosine diphosphate, cyclic-di-GMP, cyclic-di-AMP, etc. (Kellenberger, Wilson, Sales-Lee, & Hammond, 2013; Song, Strack, & Jaffrey, 2013; Strack, Song, & Jaffrey, 2013). These allosteric Spinach sensors allow rapid and dynamic monitoring of intracellular dynamics of metabolites. However, one key limitation is to identify a metabolite-binding aptamer that has suitable selectivity in a complex cellular environment.
The Spinach riboswitch approach takes advantage of the high specificity of naturally occurring riboswitches. Riboswitches are found in the 5′-UTR of certain mRNAs from many lower organisms, and undergo a conformational switch upon binding metabolites that affects mRNA translation or stability (Mandal & Breaker, 2004; Winkler, Nahvi, & Breaker, 2002). The key element of this regulation mechanism is the switching of a single strand, i.e., switching sequence, between binding to the riboswitch’s aptamer domain and to a semi-structured region adjacent that influences mRNA translation or stability, known as the expression platform.
Given the known structure of Spinach and several riboswitches, we modified several known riboswitches so that the switching triggered the formation of the DFHBI-binding pocket of Spinach (Figure 1B). This strategy takes advantage of the similarity between the structure of Spinach and the expression platform. The strategy also takes advantage of the similarity between sequences in the switching strand and the fluorophore-binding pocket in Spinach. The straightforward process of developing Spinach riboswitch sensors (Figure 2) allows the development of an array of sensors for many important cellular metabolites recognized by riboswitches. Here, we describe the procedure to design, optimize and apply Spinach riboswitch sensors for imaging of metabolite dynamics in bacteria cells.
For rational design of RNA Spinach riboswitch sensors, structural information is very useful in finding RNA sequences with sensor-like properties and optimizing their behavior, shown in Figure 3A. The 24 different classes of riboswitches are comprised of sequences that recognize nucleotides, amino acids, sugars, and cations/anions (Baker et al., 2012; Breaker, 2011). To date, the Nucleic Acid Database Project lists over 180 different crystal structures of known riboswitches (http://ndbserver.rutgers.edu/). Some riboswitches have structures in multiple conformations, which can help indicate which sequences might make suitable Spinach riboswitch sensor candidates.
Spinach forms a unique G-quadruplex structure (Warner et al., 2014), which along with a Hoogsteen U•A•U base triple, forms a binding pocket for DFHBI that activates the fluorophore’s fluorescence. In Spinach, the stem adjacent to the fluorophore-binding pocket (H1 stem) holds a 5′-UCC-3′ sequence (Figure 2 and Figure 3A). This sequence, called the transducer sequence, overlaps with the base triple that helps form the DFHBI-binding pocket and is critical for Spinach fluorescence (Paige et al., 2012; You et al., 2015). If this transducer sequence interacts with a complementary sequence from the riboswitch, then we would expect destabilization the U•A•U triplet and reduced DFHBI binding and fluorescence. If this complementary sequence also occupies different conformations in the riboswitch structure when the metabolite is bound or unbound, then we have the basis for a sensor.
Nature has already evolved the necessary complementarity in the switching sequence of many riboswitches. The key element of the riboswitch mechanism is the switching of this sequence between binding to the aptamer domain and to the expression platform, which, for example, contains a Shine-Dalgarno (SD) sequence. This switching is mediated by riboswitch conformation, such that metabolite binding to the riboswitch releases the SD sequence and enables ribosome binding, illustrated in Figure 3A. Furthermore, the switching sequence can be optimized to allow efficient switching between interaction with Spinach and the riboswitch in response to metabolite concentration.
Knowledge of riboswitch structure is immensely useful in identifying a switching sequence that will interact with Spinach’s transducer sequence. PyMOL is an invaluable open-source software tool for making structure-based hypotheses and is available for free from Schrodinger. A community-supported wiki (http://pymolwiki.org/) along with various online sources contain tutorials for the skills needed to observe and measure aspects of a riboswitch structure as in Figure 3B. An algorithmic prediction of RNA structure, such as mFold (http://unafold.rna.albany.edu/?q=mfold) can also help guide identification of a good switching sequence. Structural information has also been collected in the apo-states of certain riboswitches (Garst, Héroux, Rambo, & Batey, 2008; Huang, Serganov, & Patel, 2010) and many have been probed using 2′-hydroxyl probing (SHAPE) to identify bases that have altered conformations upon metabolite binding (Steen, Malhotra, & Weeks, 2010). In several riboswitch classes and putative classes, genomic alignment of homologous riboswitches (Barrick & Breaker, 2007; Weinberg et al., 2010) suggests which bases are more conserved and perhaps more essential for control of gene regulation. Without much prior knowledge of the aptamer domain, a stem that holds the 5′ and/or 3′ ends together in a riboswitch may also contain a good candidate for a switching sequence.
Even if a riboswitch does not contain a clear candidate for a switching sequence in the native conformation, there are additional steps to consider that may lead to an effective sensor. Critically, the U bases of the Spinach H1 stem can base pair with A or G in RNA, and Spinach tolerates some mutations in the H1 stem, as long of it retains a stem structure, forming A-U, G-C, or G-U base pairs. There is evidence that certain modifications of the 3′ end of this transducer sequence do not impair the function of Spinach-based metabolite sensors, as shown in Figure 3C. Thus, there is flexibility to change the “native” 5′-UCC-3′ of Spinach into other sequences that can act as the transducing sequence that will interact with the switching sequence. Changing the sequence of the riboswitch to adjust transduction of the sensor is also possible, but may impair sensitivity of the sensor for the metabolite target. These principles allow us to design sensors from a wider variety of riboswitches for detection of many different metabolites.
The way in which the riboswitch and Spinach are linked can change the behavior of the sensor. Previous sensors have placed the riboswitch on the 5′ end of Spinach, but in principle the order of domains in the 5′-to-3′ sequence is not restricted. Once putative switching sequences in the riboswitch and transducer sequences in Spinach have been determined, the linker sequence between the riboswitch and Spinach should be considered, which also benefits greatly from a structural approach.
The purpose of adjusting the linker sequence is to optimize positioning of the transducer sequence relative to the switching sequence. The crystal structure of Spinach demonstrates that the distance between its 5′ end and the CUU within the transducer sequence and the base triplet is roughly 3.4 nm (Warner et al., 2014). The distance between the riboswitch’s 3′ end and switching sequence normally cannot be changed (Serganov, Polonskaia, Phan, Breaker, & Patel, 2006), but the linker length can be controlled to fine tune a sensor, as illustrated in Figure 3A. The optimal distance from the linker to the switching sequence should be around this 3.4 nm distance. For example, as shown in the structural model of the optimized thiamine pyrophosphate (TPP) Spinach riboswitch sensor (Figure 3A), the predicted distance from the linker to the switching sequence (3.0 nm) is nearly the same as the distance to the corresponding U base of the transducer sequence (3.4 nm).
Optimized length and flexibility of the linker can also enhance sensor function. Long lengths of single-stranded linkers allow flexibility that lets the sensor explore different orientations between the domains. On the other hand, a long linker may allow “too much flexibility” if the domains of the sensor are already in an ideal position. Less flexible double-stranded linkers that extend from the stem of the riboswitch or Spinach can also be tested. One consideration is that double-stranded linkers will both extend and rotate the Spinach or riboswitch helixes, thereby inducing steric effects that may facilitate or disturb the interaction between the transducer and switching sequences. Once again, structural information is a critical guide for optimizing linker types and the behavior of RNA-based sensors.
Once a series of putative sensors have been conceived, the next step towards imaging cellular metabolites is to characterize these RNA sequences in vitro. As discussed above, there are several structure-guided design approaches that may lead to discovery of an optimized Spinach riboswitch sensor. To identify which designs of switching sequence, transducer sequence, and linker sequence are most successful, one must enzymatically synthesize each RNA and then observe fluorescence changes with different levels of metabolite concentration. It is also important to measure other sensor characteristics, such as kinetic response and specificity of recognition. All of these in vitro tests are critical to identifying potential Spinach riboswitch sensors for intracellular imaging.
Testing the fluorescence characteristics of each sensor begins with measurements of fluorescence response to physiological amounts of metabolite. Sensors that exhibit the highest fold-increase in fluorescence after adding metabolite and lowest fluorescence in the absence of metabolite are tested further for dose response and kinetics. Fluorescent aptamers like Spinach, which activate dye fluorescence should be measured with fluorophore in excess because of the low background of fluorophore, when free in solution. In this way, the fluorescence observed will be determined exclusively by the percent of the RNA sensor that adopts a folded conformation. If the RNA is in excess and the fluorophore is limiting, any unfolded sensor will not be detectable since the maximal fluorescence level would be determined by the concentration of the fluorophore. The suitable fluorophore for these sensors include 3,5-difluoro-4-hydroxybenzylidene imidazolinone (DFHBI, Lucerna) or its trifluoroethyl derivative, DFHBI-1T (Lucerna). Other measurement conditions may vary from sensor to sensor, however, when using Spinach as a fluorescing domain, sufficient magnesium concentration is necessary for proper folding and fluorescence. Previous Spinach-based sensors are most effective at 2–5 mM magnesium concentration in vitro, which resembles bacterial levels, with fluorophore concentrations roughly 100-fold higher than RNA concentrations.
A starting sample buffer for testing Spinach-based RNA sensors may contain 40 mM HEPES pH 7.4, 100 mM KCl, and 5 mM MgCl2 with 10 μM DFHBI-1T and 0.1 μM of RNA. Assessment of sensors begins with measurement of fluorescence before and after incubation with physiological concentration of metabolite (e.g., 100 μM TPP) at 37 °C for 30 minutes. Any fluorescence that is detected after incubation should also be compared to fluorescence of Spinach without the target-binding aptamer domain (Figure 3C), and any modifications to the Spinach transducer region must be reflected in this control. Measurement of fluorescence with a fluorimeter should be made with 470 nm excitation and 503 nm emission.
The best candidate sensors normally have the highest fold-change upon adding metabolite, and it is critical to see which candidates have a suitable affinity for the metabolite. A suitable affinity (EC50) should fall well within the physiological concentration range of the target, and preferably be slightly higher than the physiological concentration range. In this case, a doubling of the metabolite concentration should result in approximately a doubling in the binding of the metabolite to the sensor and a corresponding increase in fluorescence. Thus, the sensor will produce a fluorescence in a nearly linear manner compared to the metabolite concentration. Titration curves are measured as the fold-change measurement for separate samples with a range of metabolite concentrations or for a single sample with progressive addition of metabolite. These fluorescence measurements form a curve that can be fit by a nonlinear regression to a model for substrate binding.
Kinetic measurements of high-affinity sensors should be made as well to help pick the best sensor and to get a sense of the time course for sensor function in bacteria. Continuous or time-lapsed measurements of fluorescence should be conducted to identify sequences with the highest rate of response to metabolite. An acceptable sensor will have rapid metabolite-induced fluorescence signal change relative to the timescale of physiological dynamics of metabolite. For example, we developed a TPP-targeting Spinach riboswitch sensor that reaches 90% of its maximal fluorescence signal within 10 min. This time frame was suitable for experiments in which we monitored TPP biosynthesis kinetics that occurred over several hours in cells (You et al., 2015).
Furthermore, the specificity of RNA sensors can be measured by determining the fluorescence response to metabolites that are known to occur in cells and which are molecularly similar to the metabolite, e.g., thiamine and thiamine monophosphate for TPP sensors. For sensors designed from naturally occurring riboswitches, this step is still important, even though riboswitches tend to have high specificity for their intended metabolites (Mandal & Breaker, 2004).
After in vitro characterization of RNA-based sensors, it is critical to validate these sensors in live cells. RNA-based metabolite sensors are still difficult to use in mammalian cells, mainly due to the limited stability and abundance of these short and structured RNAs. Despite this, these RNAs typically maintain a steady state level in cells due to a balance of new transcription and degradation. RNA-based sensors with fast kinetic response and readily detectable fluorescence signal are becoming popular for metabolite analysis in bacterial cells (Kellenberger et al., 2013; You et al., 2015). Shown below is a protocol using Spinach riboswitch sensor for TPP imaging in live E. coli cells.
Versatile RNA-based sensor technologies have been developed to monitor dynamics of metabolites inside live cells. Here we describe protocols for designing, optimizing, and implementing modular RNA-based metabolite sensors. We have illustrated a structure-guided strategy to identify riboswitches that can be fashioned into genetically encoded fluorogenic riboswitch sensors, using both their metabolite sensing ability and their structure-switching capacity. Our example of imaging bacterial TPP dynamics based on the thiM TPP riboswitch demonstrates that one can ultimately use a sensor to discern different individual cellular levels of metabolite.
This approach to sensor design could be greatly improved by a number of anticipated advancements in RNA imaging (Filonov, Kam, Song, & Jaffrey, 2015) and new types of metabolite-recognizing riboswitches (Lynch & Gallivan, 2009; Weigand & Suess, 2007). By exploiting the diversity of naturally occurring riboswitches, it should be possible to image a large number of intracellular metabolites. Additionally, improvements in fluorogenic RNA aptamers may lead to enhanced sensitivity and future extension of this technology for dynamic imaging of mammalian metabolites and proteins.
We thank Grigory S. Filonov for helpful suggestions and comments. Support was provided by the Tri-Institutional PhD Program in Chemical Biology (MSKCC, Weill-Cornell Medical College, and Rockefeller University). This work was also supported by NIH Grants R01 NS064516 and R01 EB010249 (to S.R.J.)