Automated design of protein-binding riboswitches for sensing human biomarkers in a cell-free expression system

Abstract

Cell-free genetically encoded biosensors have been developed to detect small molecules and nucleic acids, but they have yet to be reliably engineered to detect proteins. Here we develop an automated platform to convert protein-binding RNA aptamers into riboswitch sensors that operate within low-cost cell-free assays. We demonstrate the platform by engineering 35 protein-sensing riboswitches for human monomeric C-reactive protein, human interleukin-32γ, and phage MS2 coat protein. The riboswitch sensors regulate output expression levels by up to 16-fold with input protein concentrations within the human serum range. We identify two distinct mechanisms governing riboswitch-mediated regulation of translation rates and leverage computational analysis to refine the protein-binding aptamer regions, improving design accuracy. Overall, we expand the cell-free sensor toolbox and demonstrate how computational design is used to develop protein-sensing riboswitches with future applications as low-cost medical diagnostics.

Introduction

Synthetic biologists have created a wide variety of sensor systems to detect small molecules and nucleic acids^{1,2,3,4,5,6,7,8,9,10,11,12,13}. Several of these sensors have been developed for usage in cell-free expression systems, where there is no barrier between the expression machinery and exogenously added bulky macromolecules that would otherwise be unable to pass through a cellular membrane^14,15,16. Cell-free sensors are particularly useful as low-cost, portable diagnostic and field assays as they genetically encode their own detection machinery and do not require a cold chain during storage and distribution^{14,15,17,18,19,20,21,22}. However, even though protein detection is a cornerstone of both modern medical diagnostics and biological research²³, there are only a few cell-free sensors that utilize gene regulation to detect proteins of interest^24,25,26.

Currently, measuring protein titers is widely carried out using immunoassays (e.g., enzyme-linked immunosorbent assays) or liquid chromatography–mass spectrometry analytics, which can offer high sensitivity and specificity across a diverse range of protein targets^27,28. More recently, another class of nucleic acid-based recognition elements, called aptamers, have been harnessed for protein detection and diagnostics^29,30. Protein-binding aptamers are now available for specific binding to a wide variety of targets, including human proteins^31,32,33,34, HIV viral proteins^35,36, and bacterial toxins³⁷. However, these assays require expensive detection reagents (e.g., purified antibodies or synthesized aptamers), expensive and bulky instruments, sample cold chain storage and distribution, and trained personnel. Instead, it is possible to utilize RNA-based aptamers to develop low-cost, genetically encoded riboswitch biosensors that carry out in situ protein detection within cell-free expression systems (TX-TL)³⁸. Past efforts to engineer such riboswitch sensors have largely relied on trial-and-error experimentation, for example, constructing and characterizing large random libraries to identify riboswitch variants that work best. Instead, it is possible to apply biophysical modeling and computational design to engineer riboswitch sensors to directly couple protein binding to gene regulation, thereby creating a sense-and-respond capability without trial-and-error experimentation.

Here, we applied biophysical modeling and computational design to engineer protein-detecting riboswitches that directly regulate the expression of a desired output protein within the TX-TL cell-free expression system, utilizing our Riboswitch Calculator algorithm to automatically convert RNA aptamers into designed riboswitch sequences³⁹. We initially engineered riboswitches to detect the phage MS2 coat protein as a proof-of-principle, followed by engineering riboswitches to detect human monomeric C-reactive protein (mCRP) and interleukin-32 gamma (IL-32γ) as examples of medically relevant biomarkers. The best riboswitch sensors regulated reporter expression levels by 13.8, 15.9, and 2.5-fold when sensing the MS2, mCRP, and IL-32γ proteins, respectively, at biomarker concentrations of 1.25 μM mCRP and 0.78 μM IL-32γ. We demonstrated that these riboswitches controlled gene expression levels via two distinct mechanisms: (i) protein-induced conformational changes to RNA structure, which modifies the ribosome’s ability to initiate translation; and (ii) protein-dependent steric repression, which blocks the ribosome from binding to the 5′ untranslated region. We critically tested the accuracy of the Riboswitch Calculator model predictions and found that improving the specification of the protein-aptamer interaction led to higher model accuracy. Overall, our automated design approach can be applied to convert any protein-binding RNA aptamer into a protein-detecting, cell-free biosensor with potential applications as portable, low-cost diagnostics.

Results

Riboswitch design and characterization platform

We created protein-binding riboswitch sequences using a biophysical model of translation-regulating riboswitches called the Riboswitch Calculator, which combines statistical thermodynamics with computational optimization to design synthetic riboswitches according to inputted specifications³⁹. The design specifications include (i) the sequence of an RNA aptamer that binds to the protein of interest; (ii) the secondary structure of the RNA aptamer when it is bound by the protein; (iii) the protein’s binding free energy (or binding affinity) to the RNA aptamer; and (iv) the coding sequence of the protein whose expression level is regulated by the riboswitch (Fig. 1A). The Riboswitch Calculator then identifies synthetic pre-aptamer and post-aptamer sequences that maximize the riboswitch’s dynamic range, utilizing a genetic algorithm to carry out computational multi-objective sequence optimization. Together, these pre-aptamer and post-aptamer sequences vary in length from 44 to 55 nucleotides, creating an overall searchable sequence space of 10²⁶ to 10³³ sequences. When designing riboswitches that activate translation (ON switches), the activation ratio is R_max = TIR_bound/TIR_unbound, where TIR_bound and TIR_unbound are the mRNA’s translation initiation rates in the protein-bound and unbound states, respectively. When designing riboswitches that repress the translation rate (OFF switches), the repression ratio is R_max = TIR_unbound/TIR_bound. Multiple equally optimal riboswitch sequences are plausible. The Riboswitch Calculator identifies the Pareto-optimal set of synthetic riboswitch sequences that are all predicted to maximize R_max with varying magnitudes of TIR_bound and TIR_unbound.