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

Matrix effects in cell-free diagnostics

With the ultimate goal of developing quantitative equipment-free diagnostics (regardless of matrix effects), we first sought to develop a clinically relevant test bed to demonstrate the potential of CFE to accomplish this goal. We chose to use zinc as the model analyte because of its global health relevance (zinc deficiency kills 100,000 children under the age of five every year) (14, 15), our group’s previous experience in developing whole-cell zinc biosensors (16–18), and the lack of field-deployable zinc diagnostics. Current clinical assessment of zinc levels requires high-pressure acid digestion and analysis with a mass spectrometer, a process that is costly and equipment intensive.

To begin, we tested whether zinc-responsive elements could properly control output in a CFE reaction. We constructed a regulator plasmid (pZntR) that constitutively expresses the zinc-responsive activator ZntR and a reporter plasmid (pGFP) that contains green fluorescent protein (GFP) under control of ZntR’s cognate promoter P_zntA (fig. S1A). When both plasmids are added to a CFE reaction, the system shows a dose-dependent zinc response between 0 and 8 μM (fig. S1B), demonstrating that zinc can be detected in a cell-free system. Notably, the system also shows a marked decrease in protein production at higher zinc concentrations. We attributed this trend to decreased expression from native Escherichia coli σ₇₀ promoters at high zinc concentrations, based on similar observations for constitutive E. coli σ₇₀ promoters (fig. S1C).

To better enable equipment-free testing, we replaced the GFP reporter with β-galactosidase to create a plasmid (pLacZ) that can mediate a colorimetric output (Fig. 1B). β-Galactosidase cleaves the yellow substrate chlorophenol red (CPR)–β-d-galactopyranoside (CPRG) to the purple product CPR; this enzyme/substrate pair has previously been used in the development of low-resource diagnostic tools (3). Colorimetric output can be quantified by measuring absorbance at 580 nm or assessed qualitatively by looking at reaction color (Fig. 1C). We characterized the system’s behavior in a bacterial extract with no β-galactosidase or ZntR background activity (fig. S2, A and B). All reactions are initially yellow (the color of CPRG), and at intermediate time points, tests are varying shades of yellow, orange, red, and purple based on zinc concentrations (Fig. 1D and fig. S3). Lyophilized systems show a similar colorimetric response to zinc, a critical requirement for field-friendly deployment (fig. S4).

To tune reaction timing for detection purposes, we modulated the concentration of reporter plasmid in the reaction. As expected, higher concentrations of the reporter plasmid (pLacZ) decrease the incubation time required for visible color, although they also decrease the time interval over which a dose-dependent zinc response (color difference) is detectable (fig. S5). Thus, plasmid dosage can be used to create tests that meet the necessary time parameters for a given application. For most experiments, we selected reporter plasmid concentrations that allowed readouts in less than an hour but maximized the time interval over which dose-dependent color differences were evident.

With a CFE zinc sensor in hand, we then sought to test the hypothesis that CFE is prone to substantial matrix effects. Whereas CFE reactions respond to zinc in high concentrations (25%) of serum, the serum matrix markedly affects test output, reducing protein expression by 25 to 50% (Fig. 1E). The addition of higher reporter plasmid concentrations enables reactions run in 25% serum and lyophilized reactions run in 100% serum to still produce a spectrum of colors across a large range of zinc concentrations (fig. S6). However, the markedly different expression levels of reactions run in serum (compared to reactions run without serum) preclude the use of chemical calibration standards, as they would not exhibit the same matrix effects. Because standards are required for a CFE quantitative diagnostic, a more sophisticated approach to quantification is necessary.

Test standardization in complex matrices

Since an ideal diagnostic would allow the user to match the test’s color output to a set of reference colors, accounting for all matrix effects caused by serum samples is critical. In addition, because the assay outputs change with time, those reference colors should also change over the course of the test. We next developed a standardization method that addresses both these issues so that the color of the test reaction can be matched to an array of standard reactions run in the exact same sample matrix as the test reaction (Fig. 1A).

To create such a matrix-specific standardization approach, we designed standard reference reactions so that they have saturated biomarker concentrations and a variable, predetermined amount of the response regulator. We found that the zinc response for a given concentration of the regulator plasmid pZntR saturates above a certain concentration of zinc (fig. S7) such that additional zinc does not affect colorimetric output. Instead, colorimetric output is solely dependent on the concentration of the transcriptional regulator ZntR, and ZntR levels can be controlled by varying the concentration of the regulator plasmid pZntR. This means that we can then create reference reactions that have saturated concentrations of zinc and that are thus insensitive to the (unknown) zinc concentrations of whatever sample is added to the reference reactions. A sample to be tested can thus be added to both the test and reference reactions, and any matrix effects of serum on CFE machinery will be identical in both the sample and the reference standards. The output of each reference reaction can then be set by changing the level of ZntR to reproduce the output that the test reaction would yield (at its fixed level of ZntR) for any given zinc concentration (Fig. 1F). Together, this allows sample-specific calibration for the test reaction.

Specifically, we found that in a 25% serum matrix, colorimetric output is identical across a raw concentration range of 8 to 16 μM added zinc (fig. S7). If each standard reaction starts with 8 μM zinc and each raw sample added to a standard reaction has no more than 8 μM zinc, then the final standard reactions will have between 8 and 16 μM zinc, and thus have constant output. This enables tests run in 25% serum to differentiate between 0 and 8 μM raw zinc, corresponding to a clinically relevant serum zinc range of 0 to 32 μM.

To determine what concentrations of the regulator plasmid pZntR to use in predictive standards, we ran a large set of standard reactions (with saturated zinc and varied pZntR concentrations) and hypothetical test reactions (with a set pZntR concentration and varied zinc) in pooled serum (fig. S8) that was collected from three individual donors. For each concentration of zinc in the hypothetical test reactions, we identified a standard reaction with nearly identical colorimetric output (Fig. 1F). The relationship between regulator and zinc concentrations was consistent across experiments run on different days and in different batches of cell extract (fig. S9). From these runs, we determined the optimal regulator plasmid concentration to put in each reference reaction to create standards that correspond with different potential test results.

Accurate biomarker quantification in serum

We next used these optimal regulator plasmid concentrations to quantify zinc in single-donor serum samples. Different concentrations of zinc were spiked into individual (zinc-depleted) serum samples, and each of these serum samples was then added to (i) a test reaction containing a defined concentration of the pZntR plasmid with no added zinc and (ii) a set of standard reference reactions containing saturated zinc with the predetermined pZntR concentrations. Evaluation at a single time point within the 1-hour target time frame (56 min) shows accurate test results: The colorimetric output of each test reaction matched the appropriate reference reaction for all samples tested (including serum excluded from the pooled sample used for test calibration). The test accurately classified all serum zinc concentrations below 20 μM into concentration bins with a width of 4 μM (Fig. 2A). Although the test inaccurately classified some of the zinc concentrations above 20 μM (far above the deficiency threshold), the range of accurate predictions covers all clinically relevant serum zinc concentrations.

To quantitatively assess test prediction accuracy and specificity at all time points, we defined a quantification error metric (QEM) that captures the differences in absorbance, across all zinc levels, between test reactions and predictive standards (Supplementary Text). Ideally, the QEM of the correct predictions should be zero, and that of the incorrect predictions should be high. We set a QEM threshold that corresponds with the ability to differentiate between ¹/₁₀ increments of the color spectrum (Fig. 1C). In all serum samples, the QEM of the correct predictions is substantially lower than the set threshold, and that of the closest incorrect predictions is substantially higher than the threshold (Fig. 2B), indicating unambiguous interpretability of test results.

The parallelized calibration approach is robust to both assay conditions and multiple assay design parameters. Although optimal standards were chosen by correlations determined in freshly assembled reactions run at 37°C, the test accurately quantifies zinc when run at both 34° and 40°C and following lyophilization (Fig. 2C). Moreover, the method can reliably quantify zinc in samples without serum (fig. S10) and in assay designs where the transcriptional regulator is added to the reaction rather than being expressed during the reaction (figs. S11 and S12).

To further characterize our standardization approach, we created an ordinary differential equation model that qualitatively recapitulates our experimental results (Supplementary Text). On the basis of previously published approaches and parameters (19–23), sensitivity analysis was used to guide an iterative parameter fitting method that minimized the error between computational predictions and data collected from reactions run in 25% serum with a range of zinc concentrations (fig. S13); this approach provided improved parameter estimates compared to simultaneous optimization of all parameter values against all experimental data. When using these optimal parameter estimates, the model qualitatively reproduced the trend in the relationship between low zinc concentrations in the test reactions and regulator concentrations in the standard reactions and also reproduced the effect of incubation time on test interpretability and accuracy (fig. S14). The predicted regulatory concentrations were usually within a factor of 2 of the experimentally observed results (fig. S9), although there were differences in the exact quantitative values due to the limitations of our model’s assumptions (including binding kinetics and parameter lumping). Of the identified relevant parameters, the predictive ability of the test is most sensitive to changes in the binding constant of zinc to ZntR, the transcriptional rate of ZntR, the degradation rate of RNA, the translational rate of proteins, and the cleavage rate of CPRG to CPR (fig. S15).

Equipment-free test interpretation in serum

Ideally, for an equipment-free test, users could easily match the color of the test reaction to the color of standard reference reactions, similarly to how one reads pH strips. The diverse array of colors that correspond with intermediate reaction stages (Fig. 1C) readily enables this interpretation. However, when the test is run in serum, the distinct colored intermediates (orange and red) are eliminated, and all intermediate reaction states are varying degrees of yellow and purple (Fig. 3A). While this has minimal effect when quantifying test output via absorbance, the lack of distinct intermediates makes quantitation by eye much more challenging.

Elimination of colored intermediates corresponds with a 10-nm shift in the absorption spectrum of CPR (fig. S16). Dyes with similar structures to CPR are known to change color upon interaction with serum albumin (24, 25), and we demonstrated that human serum albumin reproduces the color change seen in serum samples (fig. S17), which is consistent with previous work using albumin-coated paper (3). To reverse this color change, we added different compounds known to bind serum albumin (26), with the aim of displacing CPR from albumin. Naproxen addition elicited the most marked effects, nearly restoring reaction color to its appearance without added serum (Fig. 3B and fig. S18) and thus enabling equipment-free biomarker quantification (Fig. 3C).

Generalizability of quantification approach

To demonstrate that our approach can be used to quantify multiple classes of biomarkers using various molecular-level sensing mechanisms, we applied the same framework to quantification of RNA via an established toehold switch (Fig. 4A) (3). We constitutively expressed a toehold switch from the plasmid pSwitch in the CFE reaction to measure trigger RNA that was directly added to CFE reactions. In this nucleic acid sensing implementation of our calibration approach, the pSwitch plasmid serves as the response regulator (analogous to the pZntR plasmid in the zinc sensing application), and the trigger RNA is the biomarker. Thus, the test reaction has a defined amount of the pSwitch plasmid with no initial trigger RNA, and standard reference reactions have saturated levels of trigger RNA (fig. S19) and variable amounts of the pSwitch plasmid. Standard reference reactions should reproduce the response of test reactions with different measured levels of trigger RNA. For each trigger concentration tested, we identified a standard reaction that shows a nearly identical colorimetric output (fig. S20); the relationship between trigger concentration in the test reactions and switch concentration in standard reactions is consistent across experiments (fig. S21). From these data, we determined the optimal set of switch concentrations that match each desired trigger concentration for a calibration curve.

Optimized standard reactions accurately quantify trigger concentrations in all test reactions (Fig. 4B). QEM quantification (Fig. 4C) shows nearly ideal test behavior across time points, and the colors of the test reactions accurately match the color of the correct corresponding standard reactions (Fig. 4D). We also created an ordinary differential equation model recapitulating toehold switch results from experiments (Supplementary Text and fig. S22). Parameter values obtained from the estimation procedure mentioned before qualitatively reproduced the logarithmic relationship between trigger concentration in test reactions and switch concentration in standard reactions (fig. S23). Of the identified parameters, the mapping of test and standard reaction is most sensitive to changes in the maturation rate of trigger RNA to its active form and the transcriptional rate of the switch (fig. S24).

The resulting RNA quantification platform could be used to directly quantify RNA in biological samples. It could also be used for more sensitive RNA assessment by processing samples through a pipeline that includes reverse transcription and transcription if the processing pipeline yields quantitatively reproducible amplification (the calibration approach can compensate for matrix effects, but not quantitative variation introduced by upstream processing).

Construction of zinc-responsive circuit

Three plasmids were assembled through Gibson assembly: one in which the transcription factor ZntR was under the control of a standard T7 promoter, one in which P_ZntA controlled sfGFP production, and one in which P_ZntA controlled β-galactosidase production. ZntR and P_ZntA were amplified from a previously described plasmid (17). β-Galactosidase was amplified from BL21(DE3) genomic DNA, and sfGFP was amplified from the plasmid pJL1. All operons were cloned into a plasmid containing a ColE1 origin and a kanamycin resistance cassette. All primers and plasmid sequences can be found in the Supplementary Materials. All plasmids were purified with EZNA midiprep columns and sequence-confirmed.

Construction of knockout strains

Lambda red recombination was used to make all knockout strains. The kanamycin resistance cassette was amplified from the plasmid pKD4 using primers that contained the P1 and P2 priming sequences specified by Datsenko and Wanner (34) and an additional 50 nucleotides of the appropriate upstream (P1) or downstream (P2) homologous sequence for the specific knockout. The polymerase chain reaction (PCR) product was gel-purified and transformed via electroporation into cells that contained the plasmid pDK46, which expresses the lambda red recombinase under control of the P_Bad promoter. Two knockout strains were constructed: A strain with the full Lac operon deleted was constructed as previously described (35) to make BL21(DE3) ΔlacIZYA, and a dual knockout strain with zntR deleted was also constructed to make BL21(DE3) ΔlacIZYA ΔzntR. Successful knockouts were selected on kanamycin plates, and replacement of the target gene with the kanamycin cassette was confirmed via PCR. The kanamycin selection marker was not excised. For the dual knockout strain, zntR was first knocked out and selected via antibiotic resistance. Then, lacIZYA was knocked out, and blue-white screening and PCR were used to confirm successful knockouts.

Preparation of cellular lysate

Cellular lysate for all experiments, excluding toehold switch experiments, was prepared as previously described (36, 37). BL21(DE3) ΔlacIZYA ΔzntR cells were grown in 2× YTPG medium at 37°C and 180 rpm to an optical density (OD) of 2.0, which corresponded with the mid-exponential growth phase. Cells were then centrifuged at 2700 rcf and washed three times with S30 buffer. S30 buffer contains 10 mM tris-acetate (pH 8.2), 14 mM magnesium acetate, 60 mM potassium acetate, and 2 mM dithiothreitol. After the final centrifugation, the wet cell mass was determined, and cells were resuspended in 1 ml of S30 buffer per 1 g of wet cell mass. The cellular resuspension was divided into 1-ml aliquots. Cells were lysed using a Q125 sonicator (Qsonica, Newton, CT) at a frequency of 20 kHz and at 50% of amplitude. Cells were sonicated on ice with three cycles of 10 s on and 10 s off, delivering approximately 180 J, at which point the cells appeared visibly lysed. An additional 4 mM dithiothreitol was added to each tube, and the sonicated mixture was then centrifuged at 12,000 rcf and 4°C for 10 min. The supernatant was removed, aliquoted, and stored at −80°C for future use.

Cellular lysate for toehold switch experiments was prepared as previously described by Sun et al. (38). BL21 Star (DE3) ΔlacIZYA cells were grown in 2× YTP medium at 37°C and 220 rpm to an OD of 2.0. Cells were centrifuged at 2700 rcf and washed three times with S30A buffer [50 mM tris, 14 mM magnesium glutamate, 60 mM potassium glutamate, 2 mM dithiothreitol (pH-corrected to 7.7 with acetic acid)]. Wet cell mass was determined after the third centrifugation step, and the cells were resuspended in 1 ml of S30A buffer per 1 g of wet cell mass and sonicated using the same method described above, although with a sonication output of approximately 400 J. Following the centrifugation of the sonicated cellular mixture, the supernatant was removed, divided into 0.5-ml aliquots, and incubated at 37°C and 220 rpm for 80 min. After this runoff reaction, the cellular lysate was centrifuged at 12,000 rcf and 4°C for 10 min. The supernatant was removed and loaded into a 10-kDa molecular weight cutoff dialysis cassette (Thermo Fisher). Lysate was dialyzed in 1 liter of S30B buffer [14 mM magnesium glutamate, 60 mM potassium glutamate, 1 mM dithiothreitol (pH-corrected to 8.2 with tris)] at 4°C for 3 hours. Dialyzed lysate was removed and centrifuged at 12,000 rcf and 4°C for 10 min. The supernatant was removed, aliquoted, flash-frozen in liquid nitrogen, and stored at −80°C for future use.

Cell-free reactions: Constitutive expression and zinc sensing

Cell-free reactions for all experiments, excluding toehold switch reactions, were run as previously described (39). Each cell-free reactions contained 0.85 mM each of guanosine triphosphate (GTP), uridine triphosphate (UTP), and cytidine triphosphate (CTP), in addition to 1.2 mM adenosine triphosphate (ATP), folinic acid (34 μg/ml), E. coli transfer RNA (tRNA) mixture (170 μg/ml), 130 mM potassium glutamate, 10 mM ammonium glutamate, 12 mM magnesium glutamate, 2 mM each of the 20 standard amino acids, 0.33 mM nicotine adenine dinucleotide (NAD), 0.27 mM coenzyme A (CoA), 1.5 mM spermidine, 1 mM putrescine, 4 mM sodium oxalate, 33 mM phosphoenol pyruvate, 27% cell extract, and the specified concentration of plasmids. In all reactions producing β-galactosidase, CPRG was added to a final concentration of 0.6 mg/ml. All chemicals were purchased from Sigma-Aldrich. In reactions containing human serum, ribonuclease (RNase) inhibitor (New England BioLabs) was added to a concentration of 0.6 U/μl.

In all reactions producing GFP, plasmids were added to reactions at a final concentration of 20 ng/μl. Plasmid dosage of β-galactosidase and ZntR expression plasmids (pLacZ and pZntR) was optimized so that color change would be observable at approximately 60 min. In reactions without serum, pLacZ was added at a concentration of 1.0 ng/μl when ZntR was preexpressed and at a concentration of 1.5 ng/μl when ZntR was coexpressed. When ZntR was coexpressed in reactions without serum, pZntR was added to reactions at a concentration of 0.5 ng/μl. In all reactions with serum, pLacZ was added at a concentration of 15 ng/μl, and pZntR was added at a concentration of 2.5 ng/μl.

In experiments in which ZntR was preexpressed and added to the reaction (figs. S11 and S12), ZntR was first produced overnight in a cell-free reaction. Reactions were carried out in 1.5-ml microcentrifuge tubes, and the reaction volume in each tube was 50 μl. A plasmid expressing ZntR (pZntR) was added at 20 ng/μl, and a plasmid expressing a ZntR-GFP fusion protein was added at 1 ng/μl as a way to assess the output of the reaction. Reactions were incubated at 30°C for 16 hours. Reactions were then centrifuged at 12,000 rcf and 4°C for 10 min, and the supernatant was removed and subsequently used. On the basis of the assumptions that the transcriptional rates of both ZntR and the ZntR-GFP fusion protein are identical (because they have the same promoter) and that the translational rates of both proteins are identical (because they have the same 5′ untranslated region and initial coding sequence), fluorescence was used as a proxy measurement for total ZntR production. Although these fluorescent measurements give an arbitrary approximation of final ZntR production and not an absolute concentration, they enable easy and fast comparison across different CFE reactions. To measure fluorescence, each reaction was diluted by a factor of 10 in nuclease-free ultrapure water. Ten microliters was added to a well of a 384-well plate, and the fluorescence was measured with a plate reader (Synergy4, BioTek). Excitation and emission for sfGFP were 485 and 528 nm, respectively. Fluorescence values were used to account for batch-to-batch variations so that a constant amount of ZntR was added to each reaction. Unless otherwise specified, the overnight ZntR reaction was added to fresh reactions at a concentration of 1% (± variations of less than 0.05% determined through fluorescent quantification measurements).

CFE reactions were run in 8 μl of volumes in 384-well small volume plates (Greiner Bio-One), and a clear adhesive film was used to cover the plate and prevent evaporation. Fluorescence of GFP was quantified as described above. β-Galactosidase activity was monitored by measuring absorbance at 580 nm. For time course experiments, plates were incubated at 37°C, and either fluorescence or absorbance was measured every 2 min. For endpoint experiments, plates were incubated at 37°C, and fluorescence was measured after 22 hours.

Cell-free reactions: Toehold switches

For experiments involving toehold switches, cell-free reactions were run according to the protocol described by Sun et al. (38). Each reaction contained 1.5 mM ATP and GTP, 0.9 mM CTP and UTP, E. coli tRNA mixture (0.2 mg/ml), 0.33 mM NAD, 0.26 mM CoA, 0.75 mM cyclic adenosine monophosphate, 0.068 mM folinic acid, 1 mM spermidine, 30 mM 3-phosphoglyceric acid, 50 mM Hepes, 1.5 mM each of the 20 standard amino acids, 4 mM magnesium glutamate, 100 mM potassium glutamate, 2% PEG-8000 (polyethylene glycol, molecular weight 8000), 33% cell extract, CPRG (0.6 mg/ml), and DNA expression plasmids. Toehold switch was expressed from the plasmid pSwitch. Unless otherwise specified, the concentration of pSwitch in reactions was 2.5 nM. Trigger RNA was produced from linear DNA template (40 ng/μl) in an in vitro transcription reaction using T7 RNA Polymerase (New England BioLabs), following the manufacturer’s protocol. Synthesized RNA was treated with deoxyribonuclease to remove the DNA template and then purified using a Zymo RNA Clean and Concentrator kit (R1014), according to the manufacturer’s protocol. Time course reactions were run as described above.

Determination of best-fit reactions and predictive standard reactions

In determining which standard reference reaction best fit each zinc concentration, sum of squared error (SSE) minimization was used. The difference between each standard and test reaction was calculated at all time points between 30 and 90 min. The SSE was calculated over this time frame, and the standard reaction that had the lowest SSE was determined to be the best-fit reaction and thus the optimal regulator for that experimental run. When choosing what regulator concentration best predicts each potential biomarker concentration, the overall optimal regulator concentration was determined to be the one closest to the average of the optimal matching regulator concentration across three runs.

Preparation of human serum

Pooled human serum was purchased from Corning. Chelex 100 (Bio-Rad) was used to deplete zinc from the serum. One gram of resin was added to 100 ml of serum, and serum was stirred for 2 hours. Resin was removed from serum through filtration.

In single-donor experiments, blood was collected from donors as approved in Institutional Review Board protocol number H17489. Venous blood was collected in 6 ml of BD Vacutainer collection tubes for trace element testing, and tubes were left on ice for 30 min to clot. Blood was then transferred to a 50-ml conical tube and centrifuged at 2700 rcf, 4°C for 30 min. The serum was removed and either immediately frozen or treated with Chelex 100 resin. Eight milligrams of resin was added to 8 ml of serum, and the mixture was vigorously stirred for 2 hours. Resin was isolated from the samples through centrifugation, and serum was syringe-filtered. All serum samples were aliquoted to minimize freeze-thaw cycles and stored at −20°C.

Zinc measurements

Zinc concentration of pooled serum was analyzed using an x-ray fluorescence (XRF) spectrometer. Two microliters of serum was spiked with 1 ppm (part per million) of a gallium internal standard and pipetted onto a quartz disc, and atomic fluorescence emission spectra were collected. Zinc concentration in individual donor serum samples was measured at the University of Georgia Laboratory for Environmental Analysis. Samples were digested with concentrated acid and analyzed with inductively coupled plasma–mass spectrometry (ICP-MS) according to EPA method 3052.

Spectrum analysis

To analyze the spectrum of CPR and reaction intermediates, a large batch of concentrated CPR was made by adding CPRG to a small amount of extract made from standard BL21(DE3) cells (which have high baseline β-galactosidase activity). The spectrum of CPR was analyzed to ensure that all CPRG reacted to CPR by screening for the absence of a detectable peak at 410 nm. Different combinations of CPRG and CPR were then added to solutions containing 27% bacterial cell extract [made from BL21(DE3) ΔlacIZYA cells], with or without serum. The final concentration of dye in each solution analyzed was 1.02 μM, which corresponds with the addition of CPRG (0.6 mg/ml).

When adding small molecules to the reaction, 10× stocks of each molecule were made. Penicillin, dicloxacillin, and naproxen were dissolved in water; ibuprofen was dissolved in ethanol; and sulfisoxazole was dissolved in chloroform. Three microliters of each stock was added to PCR tubes, and tubes were left open in a fume hood overnight so that solvents could evaporate. A 30-μl solution containing the specified amount of dye, protein extract, and serum was used to redissolve the small molecule. Concentrations reported are the final concentration of small molecule in the analyzed solutions.

The spectra of all solutions reported were measured in triplicate with a NanoDrop 2000 spectrophotometer. When analyzing reaction intermediates, spectrum height was normalized to a peak at 280 nm. When further analyzing the peak corresponding with CPR, each spectrum was first height-normalized so that the maximum of the peak corresponding with CPR was the same for all spectra. Then, the wavelengths that correspond with four different absorbance intensities were determined and subtracted from the values of the control spectrum. The four differences were averaged to compute the overall wavelength shift. Reported averages are the average of the overall wavelength shift of the three initial replicates.

Color imaging and processing

All pictures were taken with a Panasonic Lumix camera in a light-controlled setting. All pictures in figures showing color results are of 8-μl reactions in 384-well plates. The centers of selected reaction wells were cropped using Adobe Photoshop and combined to make color arrays. A brightness filter was uniformly applied to photos to make them better resemble actual appearance.

Lyophilization

Thirty-microliter reactions containing all components of the cell-free reaction (including lysate, small-molecule mix, DNA templates, and CPRG) at a 1× concentration were prepared in PCR tubes and flash-frozen in liquid nitrogen. Frozen samples were removed from liquid nitrogen and added to a Labconco Fast-Freeze flask that contained a small amount of liquid nitrogen. Care was taken to transfer samples quickly and keep samples cold throughout the transfer process. Flasks were connected to a Labconco benchtop freeze dryer and lyophilized at −50°C and 0.05 mbar for 3 hours. Samples were then removed and rehydrated on ice.