key: cord-0315044-rs8m26rj
authors: Voyvodic, Peter L.; Conejero, Ismael; Mesmoudi, Khouloud; Renard, Eric; Courtet, Philippe; Cattoni, Diego I.; Bonnet, Jerome
title: Evaluating and mitigating clinical samples matrix effects on TX-TL cell-free performance
date: 2022-05-02
journal: bioRxiv
DOI: 10.1101/2022.05.02.489947
sha: 48005dc0f084c7591a3e75205f0796ac34089d99
doc_id: 315044
cord_uid: rs8m26rj

Cell-free biosensors are promising tools for medical diagnostics, yet their performance can be affected by matrix effects arising from the sample itself or from external components. Here we systematically evaluate the performance and robustness of cell-free systems in serum, plasma, urine, and saliva using two reporter systems, sGFP and luciferase. In all cases, clinical samples have a strong inhibitory effect. Of different inhibitors, only the RNase inhibitor mitigated matrix effects. However, we found that the recovery potential of RNase inhibitor was partially muted by interference from glycerol contained in the commercial buffer. We solved this issue by designing a strain producing an RNase inhibitor protein requiring no additional step in extract preparation. Furthermore, our new extract yielded higher reporter levels than previous conditions and tempered interpatient variability associated with matrix effects. This systematic evaluation and improvements of cell-free system robustness unified across many types of clinical samples is a significant step towards developing cell-free diagnostics for a wide range of conditions.

Biosensors are detection tools that integrate a biological recognition element from a sensor module and transduce the signal into a quick, measurable response 1 .

Rapid point-of-care testing applications are an alternative to complex technical procedures in clinical settings and reduce the need for time-consuming and equipment-dependent sample processing. Hence, portable biosensors allow close monitoring of chronic disease, as exemplified by portable glucose monitoring devices that have revolutionized diabetes care 2 . These tools are adapted to a real-time assessment of such clinical populations and more precisely determine clinical outcomes. In this way, less biased data are simply and repeatedly sampled outside the clinical setting while taking into account the patient's own environment and variability 3 . Moreover, biosensors are particularly promising tools for field diagnostics in low-resource settings 4 , especially in the context of transmissible infectious and endemic diseases such as HIV 5 , malaria 6 , or Zika 7 . The poor testing response of many countries to COVID-19 has highlighted the importance of rapid, low-cost, and easily distributable diagnostic devices worldwide 8 .

Among the vast array of available technologies, cell-free expression systems have recently emerged as promising candidates for versatile biosensor engineering 9,10 as they support the operation of sophisticated genetic circuits while requiring small reaction volumes 11 . Compared to whole-cell biosensors, cell-free systems can detect molecules like nucleic acids that do not cross cellular membranes, as well as ones typically toxic to living cells. In addition to being abiotic and non-replicating, with little need for biocontainment measures, cell-free expression systems endure no evolutionary pressure that can alter whole-cell biosensors. Reactions can occur at ambient or body temperature, by taping the sensor to the skin for example 12 , eliminating equipment like incubators at the point of detection. Performances are easily tunable by varying extract and plasmid composition, and the protein expression yield may be optimized through a large variety of methods 13 . Cell-free biosensors can be lyophilized and kept stable at room temperature for up to one year 14 and may provide rapid responses in as little as under one hour 15 . The combination of engineered cell-free transcription and translation (TX-TL) systems with electrochemical platforms further enables multiplexed biomarker detection 16 .

Overall, with short reaction times, long-term storage, and the compatibility with nano-electrode interfaces 16 , cell-free systems are good candidates toward point-of-care testing of pathological biomarkers in clinical settings and remote locations. The emergence of decentralized, portable diagnostics may improve the acquisition of epidemiological data worldwide, which is proving critical in planning global healthcare measures to mitigate the effects of non-communicable diseases and recent outbreaks, such as Zika or the COVID-19 pandemic 17 . Yet, some issues are currently limiting the translation of cell-free diagnostic platforms to clinical use, such as cross detection of non-identified markers, or most importantly, the interference from biological samples. Variability in the components of complex samples can affect the readouts of even sophisticated analytical instrumentation, a phenomenon known as matrix effects that can yield inaccurate results 18 . These effects were reported in several types of biosensors produced in E. coli extracts that have been tested using human samples ( Table 1) . coli TX-TL cell-free extract across four types of clinical samples taken from patients in hospital settings through minimally invasive methods. Samples were not processed before testing except in their basic preparation (i.e. blood centrifugation for serum and plasma after collection in appropriate vacuum tubes). We tested matrix effects of the samples on both constitutively-produced superfolder Green Fluorescent Protein (sfGFP) and firefly luciferase 15 . In addition, we systematically examined the improvement of the E. coli TX-TL cell-free system by adding RNase and protease inhibitors. Importantly, we found that the glycerol present in commercial RNase inhibitors reduces cell-free production. As such, we developed an E. coli strain that can produce its own RNase inhibitor in situ during extract production without any additional procedure steps, simultaneously eliminating the cost of commercial inhibitors and increasing protein production in the presence of clinical samples over that obtained when they are used. Finally, we examined the interpatient variability effects through testing ten serum, plasma, and urine patient 

We first measured the matrix effects of various clinical samples (serum, plasma, urine, and saliva) on cell-free activity. To do so, we monitored the production of two constitutively expressed reporters, superfolder GFP (sfGFP) and firefly luciferase (Luc) in the presence or absence of clinical samples. We chose sfGFP and luciferase because they are both common reporters widely used for signal quantification. Plasmids constitutively expressing sfGFP or luciferase were mixed with E. coli TX-TL extract prepared using a French press as previously described 15 ,

and an optimized buffer containing the necessary building blocks, salts, and energy source for transcription and translation. Finally, as these core reaction components take up 80-90% of the available reaction volume, non-processed clinical samples were added to the reaction mix as 10% of the final reaction volume ( Figure 1A) .

We quantified the matrix effects of different clinical samples on constitutive reporter expression in the absence or presence of RNase inhibitor and two protease inhibitors (bacterial and mammalian) relative to a positive control with neither clinical sample nor inhibitor ( Figure 1B-C) . We found that all clinical samples had an inhibitory effect on reporter production, albeit to a different extent. Without any inhibitors, serum and plasma almost completely impeded reporter production (>98% inhibition with respect to no sample addition). Urine inhibited more than 90% of reporter production for both sfGFP and Luc with respect to no sample addition, whereas saliva produced the least interference for both reporters (70% inhibition on luciferase and 40% on sfGFP, with respect to control). 

We tested two categories of additives inhibiting enzymatic activities that could affect the reactions: RNases and proteases. RNAse inhibitor was previously shown to improve the efficiency of cell-free reactions 19 in some types of clinical samples or were systematically added to cell-free reactions with biological fluids 15, 18 .

Also, endogenous proteases from E. coli extracts have been suggested to affect the yield of protein synthesis 24, 25 . We choose to test both bacterial and mammalian protease inhibitors to account for proteases found in both the E. coli extract and in the human clinical samples. While the use of RNase inhibitors has been previously tested 19 , no previous study has evaluated neither protease inhibitors nor all four of the clinical samples described here.

The addition of RNase inhibitor improved sfGFP production by about 70% in urine, 20% in serum, and 40 % in plasma, while bacterial and mammalian protease inhibition failed to improve cell-free reaction performance in any of the clinical samples ( Figure 1B) . Results were comparable in all conditions when using the firefly luciferase reporter. The addition of RNase inhibitor restored luciferase signals in saliva, plasma, serum, and to a lesser extent in urine, reaching 50% of the luciferase production from the absence of a clinical sample ( Figure 1C ). As with sfGFP, bacterial and mammalian protease inhibitors did not produce significant improvement on cell-free protein synthesis for the Luc reporter.

While RNase inhibitors generally led to increased protein production in the presence of clinical samples, in all cases full signal (i.e. when no clinical sample was present) was never recovered. Additionally, we tested the effect of inhibitors in cell-free reactions without any clinical sample (Supplementary Figure 1) and observed that all of them degraded reporter production, with RNase inhibitor being the most detrimental (~50% reduction of signal with respect to no inhibitor). We wondered if the buffer composition of the commercial RNAse inhibitor could be responsible for this phenomenon. First, we compared cell-free reaction efficiency, with and without RNase inhibitor, vs. buffer with identical components as listed by the manufacturer (50 mM KCl, 20 mM HEPES, 8 mM DTT, 50% glycerol) (Figure 2A) . Indeed, there was a marked decrease in protein production from adding RNase inhibitor that was identical to that of adding the buffer alone. This result was consistent over a range of plasmid concentrations (Supplementary Figure 2) . To disentangle which part of the buffer was responsible, we separately added each component of the commercial buffer to the cell-free reaction, individually and in all possible combinations ( Figure   2B ). We found that glycerol alone (at 1% final reaction concentration) accounted in all cases for the sfGFP production decrease independently of the presence of any other component of the buffer. These data demonstrate that the decrease of cell-free reaction performance observed when adding RNase inhibitor is exclusively due to the glycerol contained in the buffer solution.

In situ RNase inhibitor production allows equivalent protection to that of commercial inhibitor while allowing higher reporter production

We then hypothesized that we could avoid glycerol inhibition by expressing RNase inhibitor in E. coli while growing the cell-free extract. This would prevent the need for glycerol and have the added benefit of reducing the overall reaction cost.

After cloning a codon-optimized version of the murine RNase inhibitor (mRI) gene into a plasmid under a T7 promoter, we transformed it into competent E. coli for extract production. Because adding IPTG during the growth process is a common cell-free practice for producing T7 RNA polymerase in situ, having the mRI gene under a T7 promoter would allow both high mRI production and potentially require no additional steps in the process ( Figure 2C ). To examine whether our mRI-doped strain could indeed protect against RNases as efficiently as the commercial product, we evaluated reporter signal vs.

increasing concentrations of RNaseA for the standard extract with and without commercial mRI against our mRI-doped extract ( Figure 2D) . While protein production with no inhibitor sharply drops to almost no production with less than 0.1 µg/mL RNaseA, both commercial and doped mRI extracts show no decrease in signal up to 0.3 µg/mL of RNaseA and can still produce a measurable signal when confronted with up to 3 µg/mL. Importantly, when examining the absolute signal of each condition, the mRI-doped extract shows more than ten times higher levels of protein production than the glycerol-inhibited commercial mRI reactions ( Figure 2E) .

Thus, our mRI-doped extract provided equivalent levels of relative RNase protection to that of the commercial inhibitor while allowing for much higher absolute levels of protein production.

We sought to compare the performance in clinical samples of our doped extract relative to extract supplemented with commercial RNAse inhibitor. We tested serum, plasma, urine, and saliva using standard extract, standard extract with commercial mRI, and doped extract for constitutive sfGFP expression (Figure 3A-B) . While matrix effects still reduced overall protein expression, the doped extract showed marked improvement in fluorescent signal over standard extract supplemented with commercial inhibitor (Figure 3C ). This demonstrates that the effects observed in reducing RNase activity observed with extract-expressed RNase translate to an improvement with clinical samples as well, while avoiding the costs and glycerol effects of commercial inhibitors. Most studies examining the development and optimization of cell-free based biosensors employ pooled clinical samples or artificial equivalents of clinical samples. We thought that our system, which did not directly measure the presence of any biomarker, represented an optimal benchmark opportunity to study matrix variability from patient to patient. Furthermore, we wished to test if in situ production of mRI further mitigated interpatient matrix effects. We examined, over three consecutive days, the ten different patient samples of serum, plasma, and urine, ensuring independence between measurements (Figure 3D) populations with a broad range of pathological conditions. This step is crucial to achieve their potential as next-generation platforms for rapid, low-cost, field diagnostics. Our procedure is straightforward, reproducible, and may be applied in real clinical settings to validate the use of TX-TL cell-free biosensors on a large scale.

One of the strengths of our workflow is the absence of sample preprocessing that would require heavy technical methods, such as analyte extraction or protein phase separation 20 . The only sample processing was the standard centrifugation procedure applied to obtain serum and plasma from whole blood tubes. Our data support the expansion of cell-free biosensors to a variety of fresh human samples in remote settings without the need for complex technical sample processing. In assays where blood plasma or serum are required, these biosensors could be coupled with low-tech centrifuges, like the paperfuge or egg-beater centrifuge, for sample preparation 26, 27 .

We compared the signal output from two well-known reporter modules, Some of the specific effects of the RNase inhibitor were likely muted, impairing full signal recovery. We found that glycerol contained in the RNase inhibitor commercial buffer was responsible for the reduced efficiency of cell-free reactions.

While evidence of glycerol inhibition of cell-free reactions has been previously reported 34 , the lack of current glycerol-free RNase inhibitors precludes another viable commercial option. We decided to sidestep the glycerol effects by designing an E.

coli plasmid that could produce mRI in situ during the extract preparation process.

We found that by producing mRI under the control of a T7 promoter during the cell growth phase, we created a cell extract that could protect against RNaseA as well as a commercial RNase inhibitor without the overall signal reducing effects of buffer glycerol. Additionally, due to the absence of glycerol, this extract exhibited much higher signal in the presence of clinical samples. While recent publications have similarly produced mRI to help protect against clinical matrix effects, they required multiple reactions or the combination of multiple extracts grown at different temperatures with additional folding chaperones 22, 23 . In contrast, our strain does not require any additional steps, aside from adding IPTG during cell growth, which is already commonly done to induce T7 polymerase production in cell-free extracts.

Finally, we examined the interpatient variability of three different clinical samples: serum, plasma, and urine using a commercial inhibitor and our doped mRI extract. Using ten individual samples of each, we found that the matrix effects produced less variability between patients when using the doped extract, particularly for plasma and urine. This kind of validation, using a large number of individual patients, is essential for the development of new biosensors, as we show that matrix effects can be so significant as to completely abolish protein production in one patient's urine sample. As concentrations of different ions and metabolites can vary widely in urine depending on patient hydration 35 , it is perhaps not surprising to see large variability; however, these findings highlight that significant robustness remains to be engineered into cell-free diagnostics before they can be reliably used in a point-of-care setting. Interestingly, blood plasma showed the lowest interpatient variability. While blood plasma and serum can frequently be used interchangeably, the coagulation process in acquiring serum can lead to changes in certain platelet components, such as aspartate aminotransferase, lactate dehydrogenase, fibrin/fibrinogen, and serotonin 36 . Further studies could investigate which components specifically influence cell-free expression, but our work indicates that plasma may be a good clinical sample candidate for future applications requiring a high degree of reproducibility as in chronic conditions or long-term follow-up studies or diagnostics.

Overall, these results represent a comprehensive overview of some of the promise and potential challenges of using cell-free systems for clinical diagnostics with unprocessed samples. In addition to providing a dataset that compares four types of commonly used clinical samples across a unified set of experiments, it represents, to the best of our knowledge, the first time that blood plasma has been used in cell-free reactions. Additionally, saliva is of particular interest since it involves a pain-free, non-invasive collection process and can be potentially used to monitor the advancement of neurodegenerative and neuropsychiatric disorders like autism, Alzheimer's, Parkinson's, and Huntington's disease, in addition to the oral microbiome, which is involved in dental and periodontal health 37, 38 . Opening cell-free biosensors to a broader range of clinical samples and ensuring their reliability through systematic optimization will help provide portable, low-cost diagnostics tools to address global health challenges.

The reporter plasmid for sfGFP (pBEAST2-sfGFP) was previously characterized 15 and is available on Addgene (#126575). The plasmid for luciferase (pBEST-Luc) was obtained from Promega. DNA for cell-free reactions was prepared from overnight bacterial cultures using Maxiprep kits (Macherey-Nagel).

To create the plasmid to induce murine RNase inhibitor expression in E. coli extract, the amino acid sequence was obtained from Uniprot (#Q91VI7) and

codon-optimized for E. coli and synthesized by Twist Bioscience. It was then cloned by Gibson assembly into the pBEAST backbone under the control of a T7 promoter and transformed into either BL21 Star or BL21 Star (DE3)::RF1-CBD3 E. coli for extract production. The final plasmid, pPLV_C1, will be available from Addgene.

Cell-free E. coli extracts were produced using a modified version of existing protocols as previously detailed 15 Luciferase Assay Reagent (Promega) was added and mixed by manual orbital agitation. The plates were sealed and luciferase levels were measured in a plate reader 5 min after the addition of the reagent. The subsequent data were processed and graphs were created using Python scripts.

Cell extract and buffer conditions were maintained from those used in optimization reactions. Saliva, urine, serum, and plasma were pooled from three individual patient 

Wearable biosensors for healthcare monitoring

Detection of pathological biomarkers in human clinical samples via amplifying genetic switches and logic gates

Ecological momentary assessment

Point-of-Care Diagnostics in Low Resource Settings: Present Status and Future Role of Microfluidics

An Alternative Medical Diagnosis Method: Biosensors for Virus Detection

Advances in biosensors and optical assays for diagnosis and detection of malaria

Low-Cost Detection of Zika Virus Using Programmable Biomolecular Components

Diagnosing COVID-19: The Disease and Tools for Detection

Cell-free biosensors for biomedical applications

Cell-free gene expression: an expanded repertoire of applications

The All E. coli TX-TL Toolbox 2.0: A Platform for Cell-Free Synthetic Biology

Point-of-Use Detection of Environmental Fluoride via a Cell-Free Riboswitch-Based Biosensor

Custom-made transcriptional biosensors for metabolic engineering

Paper-based synthetic gene networks

Plug-and-play metabolic transducers expand the chemical detection space of cell-free biosensors

A multiplexed, electrochemical interface for gene-circuit-based sensors

Aspects of nano-enabling biosensing systems for intelligent healthcare; towards COVID-19 management

Point-of-care biomarker quantification enabled by sample-specific calibration

Biosensing estrogenic endocrine disruptors in human blood and urine: A RAPID cell-free protein synthesis approach

A Cell-Free Biosensor for Detecting Quorum Sensing Molecules in P. aeruginosa-Infected Respiratory Samples

Cell-Free Protein Synthesis Approach to Biosensing hTRβ-Specific Endocrine Disruptors

Streamlining cell-free protein synthesis biosensors for use in human fluids: In situ RNase inhibitor production during extract preparation

Rapid RNase inhibitor production to enable low-cost, on-demand cell-free protein synthesis biosensor use in human body fluids

Reduction of protein degradation by use of protease-deficient mutants in cell-free protein synthesis system of Escherichia coli

Improvements in the cell-free production of functional antibodies using cell extract from protease-deficient Escherichia coli mutant

Paperfuge: An ultralow-cost, hand-powered paper-centrifuge inspired by the mechanics of a whirligig toy

Egg beater as centrifuge: isolating human blood plasma from whole blood in resource-poor settings

Human endothelial cells selectively express large amounts of pancreatic-type ribonuclease (RNase 1)

Purification and properties of ribonucleases from human urine

Ribonuclease in different types of saliva from cystic fibrosis patients

Cell-free biosensors for rapid detection of water contaminants

Exploiting NanoLuc luciferase for smartphone-based bioluminescence cell biosensor for (anti)-inflammatory activity and toxicity

Smartphone-based biosensors: A critical review and perspectives

Linear DNA for rapid prototyping of synthetic biological circuits in an Escherichia coli based TX-TL cell-free system

Adjusting urinary chemical biomarkers for hydration status during pregnancy

Use of anticoagulants in diagnostic laboratory investigations

Salivary biomarkers for the diagnosis and monitoring of neurological diseases

Salivary biomarkers: toward future clinical and diagnostic utilities

Differentially optimized cell-free buffer enables robust expression from unprotected linear DNA in exonuclease-deficient extracts

We thank the Bonnet lab members for fruitful discussions. We are grateful to Pau Bernado, Annika Urbanek, and Anna Morato for helpful discussions on cell-free systems and extract creation, Gottfried Otting for BL21 (DE3) Star::RF1-CBD3 strain, and the patients for providing their samples. This work was supported by an ERC 

P.L.V. is a shareholder at Twist Bioscience (NYSE: TWST), which was used to synthesize the murine RNase inhibitor.