Semisynthetic sensor proteins enable metabolic assays at the point of care

See allHide authors and affiliations

Science  14 Sep 2018:
Vol. 361, Issue 6407, pp. 1122-1126
DOI: 10.1126/science.aat7992

A protein designed to sense metabolites

Many diseases cause characteristic changes in blood metabolites. Yu et al. describe a paper-based assay in which a chosen metabolite can be oxidized to generate reduced nicotinamide adenine dinucleotide phosphate (NADPH). Color changes in a designed NADPH sensor protein are then quantified by a digital camera. The sensor system successfully generated point-of-care measurements of phenylalanine, glucose, and glutamate. Concentrations of phenylalanine in the blood of phenylketonuria patients were analyzed within minutes with only half a microliter of blood.

Science, this issue p. 1122


Monitoring metabolites at the point of care could improve the diagnosis and management of numerous diseases. Yet for most metabolites, such assays are not available. We introduce semisynthetic, light-emitting sensor proteins for use in paper-based metabolic assays. The metabolite is oxidized by nicotinamide adenine dinucleotide phosphate, and the sensor changes color in the presence of the reduced cofactor, enabling metabolite quantification with the use of a digital camera. The approach makes any metabolite that can be oxidized by the cofactor a candidate for quantitative point-of-care assays, as shown for phenylalanine, glucose, and glutamate. Phenylalanine blood levels of phenylketonuria patients were analyzed at the point of care within minutes with only 0.5 microliters of blood. Results were within 15% of those obtained with standard testing methods.

Diseases or injuries can result in large changes in metabolite blood concentrations. Examples include phenylalanine in phenylketonuria (PKU) (1), glutamate during ischemic strokes (2), galactose in galactosemia (3), leucine in maple syrup urine disease (4), and tyrosine in tyrosinemia (5). At present, these and numerous other medically relevant metabolites cannot be reliably quantified at the point of care (POC), hampering the diagnosis and management of the underlying medical conditions. To address this problem, we envisioned conducting paper-based enzymatic assays in which the metabolite is quantified through stoichiometric formation of the reduced redox cofactor nicotinamide adenine dinucleotide (NADH) or its phosphorylated form (NADPH). Such an approach would make all metabolites that can be specifically oxidized by the oxidized form of nicotinamide adenine dinucleotide (NAD+ and NADP+) accessible to POC analysis. Currently, NAD(P)H is measured by using a variety of different analytical methods (68), but the background and interference observed in whole-blood samples restrict the utility of these methods for quantitative POC assays. We therefore aimed at generating a biosensor that measures NADPH through a background-free bioluminescent output. Guided by our previous work on semisynthetic sensor proteins (9), we designed an NADPH sensor from the following three components: an NADPH-dependent receptor protein, the luciferase NanoLuc (NLuc) (10), and a fluorescently labeled ligand with NADPH-dependent affinity for the receptor (Fig. 1A). The fluorescent ligand is covalently tethered to the receptor protein through the self-labeling protein SNAP-tag (11). In the presence of NADPH, the ligand binds to the receptor and brings the fluorophore close to the luciferase, thereby increasing bioluminescence resonance energy transfer (BRET). The NADPH concentration is given by the ratio of the emission intensities of NLuc and the fluorophore, which can be measured by a digital camera in paper-based assays (9).

Fig. 1 Design of a BRET sensor for NADPH.

(A) Design principle: A protein chimera between a receptor and a luciferase (NLuc) is tethered via a SNAP-tag to a ligand derivatized with a fluorophore. NADPH triggers ligand binding to the receptor, thereby increasing BRET. (B) Structure of TMP-TMR and TMP-Cy3-BG. BG (gray) tethers the probe to the SNAP-tag. (C) Affinity between engineered receptors and TMP-TMR as measured by fluorescence polarization (FP) in the presence or absence of 100 μM NADPH. NLucInDHFR stands for cpNLuc inserted in eDHFR; cpDHFR stands for cpDHFR with NLuc fused to its new N terminus; eDHFR stands for wild-type eDHFR with NLuc fused to its N terminus. (D) Cartoon indicating the insertion of cpNLuc [Protein Data Bank (PBD) ID 5B0U] into eDHFR (PDB ID 4PDJ). The original N (blue) and C (red) termini of NLuc are linked by a (GGTGGS)2 linker. E, Glu; G, Gly. (E) Emission spectra and photograph of the sensor in the presence of various concentrations of NADPH. RLU, relative light units. (F) Titration of the sensor with different cofactors. (G) NADPH titrations of sensors carrying mutations in the receptor protein. R, Arg; A, Ala; H, His; Q, Gln. Values in (C), (F), and (G) are given as means ± SD for three independent measurements.

We identified Escherichia coli dihydrofolate reductase (eDHFR) as a potential receptor protein because it has an NADPH-dependent affinity for its ligand trimethoprim (TMP) (12) and has been shown to bind to derivatized ligands (9). The affinity of eDHFR for a fluorescent TMP derivative [TMP-tetramethylrhodamine (TMR)] (Fig. 1B) increases by a factor of 23 in the presence of NADPH (Fig. 1C and table S1), but this modest cooperativity is not sufficient to generate an NADPH sensor (fig. S1). We hypothesized that the cooperativity of ligand binding could be increased through partial unfolding of the binding site that could be reversed by NADPH to induce the binding of the tethered ligand. We first analyzed a circular permutated eDHFR (cpDHFR) in which new termini were created in a loop (between residues Asn23 and Leu24) that bridges the binding sites for NADPH and the ligand. The resulting cpDHFR showed increased cooperativity in ligand binding relative to the wild type: adding saturating concentrations of NADPH increased the affinity of cpDHFR for TMP-TMR by a factor of 150 (Fig. 1C and table S1). This cooperativity was further increased by inserting a circular permutated variant of NLuc (cpNLuc) (13) in the same binding site loop of eDHFR (Fig. 1D): The affinity of the resulting protein chimera for TMP-TMR increased by a factor of 1400 upon the addition of saturating concentrations of NADPH (Fig. 1C and table S1). cpNLuc was used instead of regular NLuc because the distance between the N and C termini of the latter (23 Å) was considered to be too large for the loop insertion. To create a functional sensor from the cpNLuc-inserted eDHFR, we fused it via a proline-30 (P30) linker to a SNAP-tag (11) (Fig. 1A) and labeled the SNAP-tag with a fluorescent TMP derivative [TMP-Cy3-benzylguanine (BG)] (Fig. 1B). The resulting BRET sensor displayed a 15-fold change in the emission ratio (NLuc/Cy3) when titrated with NADPH (Fig. 1E). The c50 of the sensor (the NADPH concentration resulting in 50% of the maximum sensor response) was measured to be 5.1 ± 0.4 nM (mean ± SD). In addition, the sensor showed selectivity for NADPH over NADP+, NADH, and NAD+ by a factor of at least 8000 (Fig. 1F). Furthermore, by introducing point mutations in the NADPH binding site of the sensor, we generated a family of sensors with c50 values for NADPH ranging from 5 nM to 6 μM (Fig. 1G). An important feature of the sensor is its large BRET ratio change. We attribute this large change to (i) the insertion of cpNLuc into the binding site loop of eDHFR, which brings the luciferase in close proximity to the fluorophore in the closed state of the sensor, and (ii) the P30 linker between the SNAP-tag and the engineered eDHFR, which separates the luciferase and the fluorophore in the open state of the sensor (14).

As the first application of our NADPH sensor, we developed a POC assay for PKU. PKU patients are classified according to their whole-blood phenylalanine levels before treatment: levels of 120 to 600 μM denote mild hyperphenylalaninaemia; levels of 600 to 1200 μM, mild PKU; and levels above 1200 μM, classic PKU (1). Controlling phenylalanine blood levels is critical for pediatric and pregnant PKU patients (1, 15). For example, maternal phenylalanine levels above 360 μM risk reducing the cognitive ability of the offspring (1). In order to avoid neuropsychological complications in pediatric PKU patients, target phenylalanine blood levels in the first decade of life are 120 to 360 μM (1). Numerous assays for phenylalanine quantification at the POC have been proposed (1618). However, none of these are simple and accurate enough to be suitable for patient self-testing. To quantify phenylalanine with our sensor, we envisioned using the following reaction for the sensing scheme (fig. S2): Embedded Image(1)The conversion of phenylalanine is quantitative under suitable reaction conditions (19). A candidate enzyme to catalyze the reaction is phenylalanine dehydrogenase (PDH) from Rhodococcus sp. M4, which has a high specific activity toward phenylalanine (19). As the enzyme is specific for NAD+ and NADH whereas our sensor binds NADPH, we engineered its cofactor specificity such that it accepts NADP+ instead of NAD+ (fig. S3). Current protocols measure phenylalanine levels in whole blood. To perform POC phenylalanine measurements that match this standard, we designed an assay in which whole blood is diluted in a solution containing PDH, NADP+, and the luciferase substrate furimazine. We lyophilized the bioluminescent sensor onto test paper in the presence of a surfactant, resulting in the lysis of blood cells once they are applied to the paper (fig. S4) to release intracellular phenylalanine for analysis. A drop of the blood solution was added to the test paper containing the lyophilized sensor and analyzed by using a digital camera mounted on a cardboard box (Fig. 2A and movie S1). For the phenylalanine measurement, we chose a biosensor with a c50 of 1.01 ± 0.06 μM for NADPH under the assay conditions (Fig. 2B and table S2). The threshold for abnormal phenylalanine concentrations is 120 μM (1). We therefore diluted blood samples by a factor of 50 to achieve maximal sensor response around this threshold. The dilution step and the analysis of samples as a thin film on the paper reduce the matrix effect from the whole blood (9). The calibration of the test paper was performed with the whole blood spiked with defined concentrations of NADPH (Fig. 2B). In this way, the normal endogenous NADPH level (20 μM) in blood (20) represents the baseline, whereas the effect of interpatient variations in whole-blood NADPH levels was considered negligible (table S3). Phenylalanine concentrations were then calculated from the emission ratios of the test paper. The whole measurement takes less than 15 min and requires only 0.5 μl of whole blood. The ratiometric nature of the sensor facilitates the POC application, as neither the exact concentration of PDH, NADP+, or furimazine nor the precise volume of the drop applied to the paper should affect the result. The measured ratio is stable over minutes (fig. S5), indicating that neither sample evaporation nor cofactor degradation is substantial under these conditions. Furthermore, the sensor lyophilized on the test paper proved stable at room temperature over a period of 42 days (fig. S6).

Fig. 2 Paper-based phenylalanine assay.

(A) Procedure for measuring phenylalanine: 0.5 μl of whole blood from a finger prick is diluted by a factor of 50 in reaction buffer. After a 10-min incubation period, 5 μl of this mixture is added onto a test paper and subsequently analyzed with a digital camera. (B) Calibration of test paper by spiking whole-blood samples with NADPH. R2, coefficient of determination. (C) Quantification of phenylalanine spiked in whole blood. Test paper results are plotted against results obtained by LC-MS. r, correlation coefficient. (D) Quantification of phenylalanine in 40 patient plasma samples. Test paper results are plotted against results obtained by LC-MS; the underlying data are listed in table S4. (E) Bland-Altmann analysis for 40 patient plasma samples measured by test paper and LC-MS. (F) Comparison between test paper and MS-MS methods for measuring whole-blood samples from four patients. Red boxes represent MS-MS results ± an allowable error of 15%. (G) Bland-Altmann analysis for four patient whole-blood samples measured by test paper and MS-MS. Error bars represent SD for three independent measurements.

To validate our POC assay, we spiked whole blood with phenylalanine and analyzed the samples. The results showed excellent correlation with the values measured in parallel by liquid chromatography–mass spectrometry (LC-MS) (Fig. 2C). We furthermore measured the phenylalanine concentrations in 40 patient plasma samples. Again, the results showed very good correlation with two independent reference methods currently used in clinical laboratories (Fig. 2, D and E; fig. S7; and table S4). The test papers also demonstrated very good reproducibility, with an average coefficient of variation of 7% ± 4% for technical triplicates. Lastly, we measured phenylalanine levels in venous blood samples freshly obtained from four different PKU patients (Fig. 2, F and G). Our results showed very good overlap with MS-MS–based dried-blood-spot measurements performed in parallel, with an averaged total analytical error of 17% ± 4%. The average coefficient of variation for the test paper was 6% ± 3%. It is instructive to compare the performance of our test paper with that of a conventional NAD-dependent colorimetric assay using the same enzymatic reaction (21). When measuring whole-blood samples spiked with 200 μM phenylalanine, the test paper showed a 232% ratio change, whereas the conventional assay showed only a 2% increase in absorbance over background (fig. S8). Furthermore, the sensor works in a “one-pot reaction” in solution, and it can be lyophilized together with the reaction buffer, including the luciferase substrate, on the test paper (fig. S9). This setting allows all the steps of the assay, with the exception of the dilution step, to be performed on paper, further simplifying the procedure. We also demonstrated that a smartphone can be used as the camera (fig. S10), facilitating future POC applications. The simplicity of the assay procedure should eventually enable patient self-testing, even though some modifications of the assay would be needed. At present, our assay requires a 10- to 15-min incubation. A more immediate result could be achieved by increasing the activity of PDH. Furthermore, the user currently needs to dilute a defined volume of blood, and a more automated assay format could be developed on the basis of microfluidics devices (7, 22). Finally, potential interferences with our assay by endogenous or exogenous substances will have to be kept in mind (table S3). For example, PKU patients receiving the antibiotic TMP would not be able to use the assay, as TMP directly binds to the sensor. Resolving these technical problems could put a much-needed tool in the hands of PKU patients.

In principle, any metabolite of clinical importance that can be oxidized with NAD(P)+ could be analyzed with our paper-based assays at the POC. Using enzymatic reactions for glucose and glutamate (23, 24), we accurately analyzed commercial plasma samples spiked with various glucose or glutamate concentrations (Fig. 3, A and B). As blood concentrations of glucose are in the millimolar range (25), in principle only 10 nl of capillary blood would be needed for such an assay.

Fig. 3 Generalization of assay and scale-up.

(A) Measurement of glucose spiked in plasma samples. (B) Measurement of glutamate spiked in plasma samples. (C) Photograph of test paper for analyzing 96 samples for phenylalanine in parallel (32 different blood samples measured in triplicates). (D) Quantification of phenylalanine for 32 blood samples in triplicates. Test paper results are plotted against concentrations of spiked phenylalanine. Error bars represent SD for three independent measurements. The y axis represents the measured analyte concentration after subtraction of the background concentration of the corresponding analyte in the sample.

Forty-two medically relevant metabolites that are potentially suitable for quantification in our assay are listed in table S5. These include carbohydrates such as galactose (26), alcohols such as ethanol (27), lipids such as triglycerides (28) and cholesterol (29), and other amino acids such as branched-chain amino acids (30). For analytes with lower concentrations, possible interpatient variations of the endogenous NADPH level would have to be considered (table S3). Such individual background corrections could be readily achieved by analyzing the sample in both the presence and absence of the analyte-converting enzyme (fig. S11). Furthermore, for some of these analytes, only NAD-specific dehydrogenases have been described. However, as demonstrated for PDH and other dehydrogenases (31), the conserved nature of the NAD(P) binding sites of dehydrogenases should make it possible to engineer their cofactor specificities.

The use of a digital camera enables the simultaneous analysis of multiple metabolic assays for either the same or different analytes on a single piece of paper. As a proof-of-principle for neonatal PKU screening applications in resource-limited settings, we analyzed 96 whole-blood samples spiked with various concentrations of phenylalanine with one camera shot (Fig. 3, C and D). This experiment demonstrates the scalability of our metabolic assays.

In summary, we combined synthetic chemistry and protein engineering to create a biosensor for the accurate quantification of metabolites in blood by a paper-based assay. The assay should enable the POC diagnosis and management of numerous diseases.

Supplementary Materials

Materials and Methods

Figs. S1 to S12

Tables S1 to S5

References (3237)

Movie S1

References and Notes

Acknowledgments: We thank K. Schwarz for experimental assistance and M. C. Colombo for discussions. Funding: This work was supported by the Max Planck Society, the Swiss National Science Foundation, and École Polytechnique Fédérale de Lausanne. Author contributions: Q.Y. and K.J. conceived this study. Q.Y., L.X., J.H., R.G., S.F., C.R., P.-A.B., D.H., and J.G.O. performed experiments and contributed to data analysis. Q.Y., K.J., and L.X. wrote the manuscript. Competing interests: Q.Y. and K.J. are inventors on a patent application on the sensor filed by the Max Planck Society. Data and materials availability: All data are available in the main text or the supplementary materials.

Stay Connected to Science

Navigate This Article