High-Throughput Screening of Isomeric Reaction Products by Droplet Microfluidics Coupled to Cyclic Ion Mobility-Mass Spectrometry

Abstract

Engineered biocatalysts enable highly selective chemical transformations with low environmental impact. Development of biocatalysts by directed evolution requires screening many enzyme variants for improved catalytic properties. The low throughput of commonly used label-free screening methods, e.g., liquid chromatography-mass spectrometry (LC-MS), becomes the rate-limiting step in biocatalyst development, limiting the coverage of protein sequence space explored. Direct MS methods have been applied to biocatalyst screening; however, these methods cannot be used to evaluate isomer selectivity. Ion mobility spectrometry is a separation technique readily combined with MS, facilitating isomer differentiation on the millisecond time scale. Here, we present a droplet microfluidic system coupled to cyclic ion mobility-mass spectrometry (cIM-MS) to enable the screening of isomeric products. The system was applied to the separation of biaryl benzofuran dimers formed by the fungal cytochrome P450 KtnC. The isomeric 5,7′-bibenzofuran and 7,7′-bibenzofuran products from KtnC variants were baseline-resolved within 32 ms by cIM. By infusing biocatalytic reaction mixtures as 5 nL droplets, an analysis throughput of 1.2 s/droplet was achieved using cIM-MS. Droplet cIM-MS was used to quantify standards in the reaction matrix with high agreement to actual concentrations of each isomer (i.e., R² of 0.97 and 0.98 for total bibenzofuran content and fractional 5,7′-bibenzofuran content, respectively). For samples with enzymatically formed product isomers, droplet cIM-MS identified the same active variants and had comparable reproducibility (RSD of 10–15%) to analysis by LC-MS and produced this data 128 times faster. The method is expected to be suitable for improving the rate of biocatalyst development for isomer-selective reactions.

Graphical Abstract

INTRODUCTION

Directed evolution is a powerful tool for tuning the chemistry and selectivity of enzymes.¹ In directed evolution, an array of enzyme variants are evaluated for favorable properties, with the top performing variants seeding the next round of evolution in this iterative process.¹ Given that thousands of variants must be tested in each round of evolution, screening throughput is often a bottleneck in enzyme engineering campaigns. Although optical high-throughput screening (HTS) platforms can achieve kHz rates in directed evolution screens, their application is limited to analytes with distinct optical spectra.^2–8 This restriction is further amplified in biotransformations where substrates, products, and byproducts largely conserve optically active functional groups. Thus, there is a need for technologies suitable for HTS of biochemical transformations.⁹ Liquid chromatography coupled to mass spectrometry (LC-MS) is a standard method for analysis of directed evolution samples. LC-MS is broadly applicable for directed evolution since the chromatographic separation reduces ionization suppression, improves quantification, and resolves isomers while the MS allows for selective, sensitive, and label-free detection of products.^10–13 However, typical LC-MS analyses require >60 s per sample, limiting its utility for high-throughput applications. In this work, we describe the use of droplet sampling interfaced to cyclic ion mobility MS (cIM-MS) to achieve rapid separation prior to MS analysis, enabling screening at 1.2 s/droplet for isomer products of enzyme variants.

High-throughput methods of direct MS analysis have been developed¹⁴ and are being investigated for enzyme engineering.^12,15–17 Formats for high-throughput direct MS include printing samples onto plates for matrix-assisted laser desorption/ionization (MALDI)-MS^16,17 and droplet sampling from well-plates to electrospray ionization (ESI)-MS.¹⁵ In addition, rapid solid-phase extraction online with ESI-MS (e.g., the Agilent RapidFire instrument) has also been applied in this way.¹⁸ However, direct-MS is not well suited for resolving isomers arising from convergent coupling reactions because their molecular ions possess identical m/z ratios and exhibit similar fragmentation pathways in tandem mass spectrometry (MS/MS).

The limitations of direct MS methods are significant in directed evolution campaigns that involve improving selectivity of isomer formation, an important application of biocatalysts. Examples include engineering aliphatic halogenases for site-selective halogenation of non-native substrates,¹⁹ rebeccamycin halogenase to halogenate indole C–H bonds,²⁰ and CYP450 enzymes for selective oxidative cross-coupling of phenolic substrates.¹⁰ In all cases, the formation of specific product isomers was measured using LC-MS to identify enzymes to advance in the evolution campaign. The low throughput of this approach constrains the number of variants that can be explored and subsequently flattens the trajectory of an evolution campaign. Overcoming this bottleneck necessitates a workflow that achieves high throughput while retaining the ability to differentiate isomers.

One way to accelerate isomer separations within a MS workflow is to integrate ion mobility spectrometry (IM), which adds a rapid, orthogonal separation dimension on a compatible time scale.²¹ In IM, analytes in the gas phase are accelerated through an electric field in a drift cell, colliding with neutral gas molecules. These collisions affect the time it takes an analyte to traverse a given space.²² The probability and extent of sustaining collisions with the drift gas are dictated by the collision cross section (CCS) of the molecule. IM has been implemented in the separation of constitutional isomers, isotopologues, isotopomers, and atropisomers.^23–26

IM has been used for reaction screening in a few pioneering works. Traveling wave IM-MS (TWIM-MS) was used to screen multisubstrate biotransformations from lipases and monooxygenases.²⁷ This method was later expanded to directed evolution samples by direct infusion of reaction mixtures, achieving a screening throughput of 40 s per sample.²⁸ In another work, infrared matrix-assisted laser desorption ionization (IR-MALDI) coupled to drift tube IM was used to screen the photocatalytic products of the ethyl-3-(pyridine-3-yl)but-2-enoate E/Z isomerization from a microreactor device within a few min/sample.²⁹ cIM-MS has been used to screen isomeric products of carbohydrate synthases at a minute per sample.³⁰ Trapped ion mobility spectrometry (TIMS) coupled to MS has also been used for screening the asymmetric α-alkylation of aldehydes by batch derivatizing the enantiomeric products of the catalytic reactions.³¹ For this work, the reported throughput was 90 s per sample, which included sample injection, IM-MS analysis, and loop cleaning, requiring 450 μL of solvent per sample. Most recently, IM coupled to matrix-assisted laser desorption ionization (MALDI)-MS was used to screen isomeric kainoid natural products from biotransformations.³²

While the above assays demonstrate the potential of IM-MS for analyzing isomeric reaction products, the throughput is not substantially better than LC-MS. With fast IM-MS analysis, sample injection time can become a limiting step for throughput. Systems with smaller injection volumes and higher throughput are available.^33,34 For example, acoustic ejection MS uses acoustic waves to eject nanoliter samples from a well-plate into a 30 cm transfer capillary for infusion to a ESI source.³⁵ This method has been coupled to differential mobility spectrometry to screen for enzyme inhibitors using isomeric substrates and products. The method used 2.5 nL of sample at 1.5 s/sample to screen a 1536 well-plate.³⁶

Another approach to high-throughput MS analysis uses droplet microfluidics.^15,37–39 Droplet microfluidics uses discrete, small volume samples (pL-nL) segmented by an immiscible fluid within a microfluidic channel. Droplets can be directly infused through ESI,^40–46 sheath flow ESI,^15,47,48 and nESI^38,49 sources for analysis, achieving a throughput as high as 30 samples/s,⁵⁰ limited by MS acquisition rate. Droplets can also be coupled to MALDI-MS.^51–53 Droplet microfluidics also enables high-throughput sample manipulation such as splitting,⁵⁴ reagent addition,⁵⁵ merging,⁵⁶ and sorting,⁵⁷ thus facilitating complete workflows on the picoliter to nanoliter scale.⁵⁸ The capability of droplet microfluidics for directed evolution has been demonstrated with fluorescence² and absorbance⁵⁹ detection. Droplet microfluidic workflows for directed evolution have also been developed for ESI-MS detection.^60,61

Examples of droplet microfluidics coupled with IM or IM-MS are limited. We have previously reported on coupling droplets to TWIM-MS to study protein structure for drug discovery.⁶² Droplet-based high-throughput IM has been developed and demonstrated for small molecule reaction screening without MS^63,64 with throughput up to 120 samples/s. These prior works suggest that hyphenating droplet infusion and IM or IM-MS yields versatile workflows for high-throughput chemical separation and MS analysis.

Here, we present droplet microfluidics coupled to cIM-MS, a modality of TWIM-MS, as a tool for label-free high-throughput screening in enzyme engineering. This method achieves isomer separation at a sample injection rate of 1.2 s/droplet. Using KtnC variants as a model system, we hypothesized that microfluidics coupled to cIM-MS could accelerate the screening of biocatalysts for yield and site-selective production of 5,7′-bibenzofuran over 7,7′-bibenzofuran (Figure 1). In the oxidative dimerization of hydroxydi-benzofuran, multiple positions could be a site of bond formation, yielding potential for multiple constitutional isomers to be formed. As a starting point, wtKtnC afforded only trace amounts of dimer product. Investigating a previously generated KtnC variant, LxC4, revealed an enzyme which yielded the isomeric products in improved yield and a 95:5 ratio favoring the 7,7′-bibenzofuran over the 5,7′-bibenzofuran. This reaction serves as a test case for engineering a biocatalyst to access constitutional isomers that are otherwise synthetically intractable. Screening for such a reaction requires determining both the total and fractional content of each isomer formed. In previous work with analogous enzyme mediated cross-couplings, a two-tier screening approach was employed for enzyme variants.¹⁰ Reactions were first analyzed via rapid solid-phase extraction (SPE) coupled to MS to determine the total dimer yield through a 40 s method. Subsequently, high-yielding reactions were screened by LC-MS for isomer selectivity utilizing a 3 min method. In this work, we evaluate the use of cIM-MS for obtaining both the total product formation and the fractional isomer content in a single, rapid assay that is appropriate for high-throughput directed evolution workflows.

Figure 1.

Workflow for using droplet microfluidics coupled to cyclic ion mobility-mass spectrometry (cIM-MS) to screen isomers from enzymatic reactions. Top panel: enzymatic cross-coupling reaction of interest that produces both 7,7′-bibenzofuran (blue) and 5,7′-bibenzofuran (yellow), the latter being the desired isomer for the enzyme engineering campaign. Reactions are carried out in a well-plate format. Middle panel: Droplets generated from the well-plate are infused into the nESI source for cIM-MS analysis at ~1 droplet/s. Bottom panel: isomers are rapidly separated by cIM-MS and resulting signals for each isomer plotted as a function of infusion time for analysis.

EXPERIMENTAL SECTION

Materials

Perfluorodecalin was purchased from Oakwood Chemical (Colombia Hwy, Estill, SC). Acetonitrile, OmniSolv LC-MS grade and formic acid was purchased from Millipore Sigma (Burlington, MA). nESI emitters were purchased from CoAnn Technologies (Richland, WA). Marine epoxy was purchased from Loctite (Westlake, OH). Well-plates (384 format) were purchased from Greiner Bio-One (Monroe, NC). Teflon PFA HP+ resin tubing was purchased from IDEX (Lake Forest, IL). Water used for reactions was purified to 18 MΩ using a Thermo Scientific Barnstead E-Pure Ultrapure Water Purification system (Hampton, NH). Yeast Nitrogen Base, dropout supplements, and all amino acids were purchased from Millipore Sigma (Burlington, MA). Glucose and galactose were purchased from Fisher Scientific (Hampton, NH). Filters (0.2 μm PTFE) were purchased from Chrom Tech (Apple Valley, MN).

Small Molecule Standards and Whole Cell Matrix

Isomer standards were prepared from 10 mM stocks of diethyl 6,6′-dihydroxy-4,4′-dimethyl-[7,7′-bibenzofuran]-2,2′-dicarboxylate, abbreviated 7,7′-bibenzofuran, and diethyl 6,6′-dihydroxy-4,4′-dimethyl-[5,7′-bibenzofuran]-2,2′-dicarboxylate, abbreviated 5,7′-bibenzofuran, in DMSO (SI 1). Solutions of isomers were prepared by diluting stocks (SI 1) in 1:1 acetonitrile: 0.1% aqueous formic acid (FA), which were then pipetted into a 384 well-plate for droplet generation. In some experiments, 1 μM of 9H,9′H-[4,4′-bicarbazole]-3,3′-diol (SI 2), abbreviated 4,4′-bicarbazole, was added as an internal standard. In some experiments, 3,3′-dimethyl-9H,9′H-[1,1′-bicarbazole]-2,2′-diol (SI 3), abbreviated 1,1′-bicarbazole, was used as a marker to ensure registry of droplet signals with droplet identity. To simulate solution conditions for method validation, a biotransformation reaction matrix was prepared with cells not producing KtnC (SI 4).

Enzyme Variants Used for 96 Well-Plate of Variable Total and Fractional Isomer Content

To demonstrate the cIM-MS method as an alternative for LC-MS analysis, variants from rounds of directed evolution known to produce a variety of total bibenzofuran content and isomer ratios were selected for analysis. Variants were streaked from glycerol stocks onto histidine dropout yeast growth plates and single colonies were inoculated into glucose media lacking histidine. Biotransformations were carried out as outlined in SI 5. After quenching the biotransformations with methanol (1:1), samples were diluted 25-fold in 1:1 acetonitrile:0.1% FA, for a total dilution factor of 50. Analysis was conducted on both LC-MS and droplet-cIM-MS.

Droplet Generation

Droplets were generated by alternately withdrawing sample and perfluorodecalin into a 1.5 m long segment 100 μm ID × 360 μm OD PFA tubing (IDEX, Northbrook, IL) at 800 nL/min using a Chemyx Fusion 200 syringe pump as described previously.^15,39,40 The open (inlet) end of the tubing was mounted on an XYZ-positioner above a Grenier 384-well-plate (Monroe, NC) containing samples. The outlet end was connected to a 25 μL Hamilton glass syringe (Reno, NV) prefilled with perfluorodecalin. Prior to droplet generation, the PFA tubing was primed with perfluorodecalin, and the syringe barrel was backfilled with 5 μL of perfluorodecalin, leaving 20 μL available volume for droplet withdrawal. Samples were overlaid with a layer perfluorodecalin, unmodified by surfactants, which was contained by a wall made from Loctite marine epoxy (Rocky Hill, Connecticut) around the outside edge of the well-plate.^15,39,40 The tube inlet was positioned for 0.2 s in the oil phase and 0.6 s in the sample well to generate 5.2 ± 0.2 (n = 23) nL droplets spaced by 4.5 nL of carrier fluid (Figure S1) so that 48 droplets were generated per minute. Droplet formation was monitored and recorded with an Adonstar AD203S camera (Shenzhen, China).

Coupling Droplets to Waters NanoLockSpray Source

nESI emitters (100 μm ID × 360 μm OD, 10 μm emitter) purchased from CoAnn Technologies (Richland, WA) were surface-treated by submerging the blunt end in Rain-X for 10 min, then rinsed with deionized water and oven-dried at 60 °C for 30 min. Subsequently, they were sputter-coated with gold through four 45 s cycles at 40 mA, with 2 min intervals between cycles using a Cressington 108 auto sputter coater (Watford, UK).

For analysis, the PFA tubing containing the generated droplets was coupled to the gold coated capillary via a New Objective Pico-Clear union (Littleton, MA). The capillary was threaded through a Waters Universal Sprayer (Milford, MA) which was mounted on a Waters NanoLockSpray source (Milford MA) and connected to a supply of nebulizing sheath gas (N₂). The syringe pump infused droplets at 800 nL/min.

Cyclic Ion Mobility-Mass Spectrometry

cIM-MS was carried out on a Waters Select Series Cyclic Ion Mobility-Mass Spectrometer (Milford, MA). For ESI, the capillary voltage = −1600 V, cone voltage = 40 V, and cone gas flow rate = 50 L/h. The cyclic separation region was pressurized to 1.77 mbar. The traveling wave velocity was 400 m/s and the wave height was 23 V. A separation time of 5 ms was used for the first pass, while additional passes required an additional ~9 ms of separation each. The time-of-flight detection was operated in ‘V’ mode and acquisition rate for all experiments was 10 Hz.

Measuring CCS on Drift Tube IM

An Agilent 6560C drift tube IM-MS platform (Santa Clara, CA) was used for CCS determination. Modifications to the instrument have been described previously, and all measurements were performed under high purity N₂ in negative mode.^65,66 Samples were introduced by direct infusion at a capillary voltage of 4000 V, and multifield CCS calibration was performed for individual isomers by ramping the drift tube entrance voltage from 1450 to 1700 V in 50 V steps. Resolving power (R_P) for cIM-MS was calculated based on the CCS values determined from the drift tube IM-MS as described previously.⁶⁷

Data Analysis

TWIM Extract⁶⁸ was used to generate ion chromatograms of the selected molecules by integrating the monoisotopic molecular ion ([M – H]⁻ signal (±0.05 m/z) within the specified drift time window. A custom-authored Python script, wrapped in a graphic user interface (described in SI 6), was developed to automate data analysis.⁶⁹ The code was used to automatically pair time-resolved intensity clusters (i.e., droplets) with their corresponding well by aligning the marker trace with the marker column of the well-plate at predicted intervals. The output of this script is a spreadsheet containing a peak list for each droplet, including the averaged signals of the selected molecules and labeled with the originating well. For both the calibration matrix and the biotransformation screen, the extracted ion counts for both isomers were normalized to the internal standard.

LC-MS Analysis of Enzyme Reactions

LC analysis was performed using a Vanquish high performance liquid chromatography system (Thermo Fisher Scientific, USA) equipped with a Waters Acquity UPLC HSS T3 column (2.1 mm ID × 50 mm length, 1.8 μm, 100 Å) at 40 °C. Mobile phase A was 18 MΩ water and mobile phase B was acetonitrile. One μL of sample was injected and separated using a linear gradient from 50 to 55% B over the course of 1.5 min with 0.5 min for column re-equilibration at 0.7 mL/min. The LC was coupled to a TSQ Quantis triple quadrupole mass spectrometer (Thermo Fisher Scientific). ESI was operated at −3500 V. Xcalibur and Freestyle (Thermo Fisher Scientific) were used for instrument control and data processing, respectively.

RESULTS AND DISCUSSION

Cyclic Ion Mobility for Multipass Isomer Separations

cIM-MS allows separation resolution to be increased by passing selected ions through the cyclic path multiple times prior to MS analysis. The CCS values obtained from drift-tube IM measurements for the 5,7′-bibenzofuran and 7,7′-bibenzofuran dimers were 214.2 and 202.2 Ų, respectively, suggesting feasibility⁷⁰ for rapid cIM separation. Figure 2 shows ion mobilograms by cIM with passes from 0 to 5, corresponding to a 0 to 5 m separation path. The first pass allowed separation of the isomers in 22 ms with a resolving power (R_P) of 47 (CCS/ΔCCS). The fifth pass allowed for separation of isomers in 60 ms with a R_P of 126. Although cIM allows for enhanced resolving power at higher pass counts, the ion transmission decreases as a function of the passes, resulting in decreased overall signal.⁷¹ Since the isomer pair was baseline-resolved within 35 ms using a 3-pass method (R_P of 88), this method was used for further analysis to avoid unnecessary reductions in signal or increases in analysis time with more passes.

Figure 2.

Separation of bibenzofuran isomers as a function of passes in the cIM instrument. 7,7′-bibenzofuran (CCS = 202 Ų) has a faster arrival time than 5,7′-bibenzofuran (CCS = 214 Ų). Resolving power (R_P) is shown above the peaks for each pass.

Benzofuran Droplet nESI and Analyte Containment

To obtain discrete droplet-MS traces for each droplet, the ionization threshold of the analyte must be below that of the carrier phase.⁷² The capillary voltage was ramped from 0 to −2000 V in intervals of −100 V to identify the window of electrical potential under which analytes could be ionized without interference from the carrier phase (Figure S2). The maximal signal was found at 1.6 kV, leaving an 800 V offset to the onset potential of perfluorodecalin (2.4 kV). Matching the IDs of the PFA and fused-silica emitter capillaries facilitated stable transfer of droplets from the tube to the emitter, in contrast to mismatched IDs which led to problems such as droplet breakage during the transfer. Keeping the flow under 1 μL/min allowed us to robustly transfer at least 500 droplets with no droplet breaking at the union. The gold coating remained stable for at least the time of infusion of 500 droplets (about 15 min) and longer times were not tested.

A challenge in droplet microfluidic workflows is cross talk resulting from mass transfer between droplets.⁷³ Such mass transfer may be mediated by intermolecular interactions of the analyte with different components in the system such as carrier fluid and tubing.⁷⁴ For many biological assays, cross talk tends to be low since analytes that are charged or highly polar are often less likely to be transported between droplets; however, transfer of nonpolar synthetic products can be greater.⁷⁵ To evaluate analyte transfer, a train of droplets was generated in alternating quintuples, each containing 10 μM of individual isomeric species. For a given isomer, the quintuples intended to contain it were termed its “α-droplets”, while the succeeding quintuples containing the other isomer, and therefore not intended to contain it, were termed its “β-droplets” (Figure 3). Carryover was quantified as the trace isomer detected over the five β-droplets. The first β-droplet had up to 15% residual signal, but all subsequent β-droplets generated only baseline signal (Figure S3). Therefore, for all further screening, the first droplet of a set was excluded from quantification to avoid contamination from the prior sequence (typically 4 droplets were used per sample). The last droplet may also be affected by loss of material; however, this effect appeared to be relatively small. For example, in a screen where widely different concentrations were detected (discussed below), we found that the average RSD when using 3 of 4 droplets was 14% and when using just the middle 2 droplets was 12%. It may also be possible to include wash droplets between sequences, larger sequences, or other modifications to control carryover. In general compromises are made between complexity, throughput, and accuracy if carryover is present. In principle it may be useful to explore other tubing materials, such as modified fused silica, or ID to further reduce carry-over; however, for these experiments the convenience of using commercial tubing was an important factor.⁷⁵

Figure 3.

Test for carryover between droplets. (a) Extracted ion chromatogram (EIC) for the monoisotopic [M – H]⁻ of bibenzofuran (437.12 m/z) showing 50 droplets containing 10 μM of a single bibenzofuran isomer with alternation of the isomer every five droplets. (b) Contour plot showing drift time for each droplet during same experiment. Droplets containing the intended isomer are termed α-droplets while those in which the isomer should be absent are termed β-droplets. Stars (*) indicate carryover of each isomer in their respective β- droplets.

Determining Dilution Factor Allowing for Linear Quantification of Bibenzofuran Isomers in Whole Cell Matrix

Enzymes are produced in a complex mixture of cells (bacteria or yeast) and culture media. Buffers, salts, and cellular components in the expression and biotransformation matrices can contribute to ionization suppression during ESI. Diluting the crude reaction matrix both ameliorates the effects of salts on ionization and reduces ESI source fouling. Considering that the concentration of bibenzofurans produced by KtnC variants was ~50 μM and that the limit of quantification for droplet-MS for bibenzofuran detection in 1:1 acetonitrile:0.1% FA was 30 nM, we estimated that a 50-fold dilution would yield bibenzofuran concentrations within the lower-middle quantification range. To assess the effects of ionization suppression caused by the reaction matrix in the dilute sample, two sets of calibration standards (0–5 μM bibenzofuran) were analyzed by droplet-MS, one diluted in 1:1 acetonitrile:0.1% FA, and the other with biotransformation reaction matrix added to 2% (v/v) the diluted sample, mimicking the amount of matrix present following 50-fold dilution of reaction mixtures (Figure 4). The presence of the reaction matrix reduced the signal (EIC of 437.12 m/z) by 20-fold. Despite the reduction in signal, Figure 4b shows that the standards yield a linear signal response (R² = 0.991) to the bibenzofuran isomers from 50 to 5000 nM in the biotransformation matrix. The nonspiked calibration curve displayed some departure from linearity at higher concentrations, likely due to detector response compression or ionization suppression. The analytical sensitivity (standard deviation of signal divided by slope) in the range of 0–750 nM (significant for screening) was 28 nM based on a slope of 0.635 counts/nM and average signal standard deviation of 18 counts. From this, we estimate that a concentration change of 56 nM can be detected with 95% confidence, or 15% change, within the middle of the concentration range. This discrimination would be adequate for detecting relatively small changes in enzyme activity among variants.

Figure 4.

Effect of matrix on ESI-MS signal for bibenzofuran. (a) Extracted ion chromatogram (EIC) for the monoisotopic [M – H]⁻ of bibenzofuran (437.12 m/z) for droplet samples in 2% reaction matrix in 1:1 acetonitrile:0.1%FA with 2% reaction matrix (termed 2% Rxn Matrix; red) and the same solution with no reaction matrix added (termed 0% Rxn Matrix; blue). Droplets contain increasing concentrations from 0 to 5000 nM bibenzofuran. (b) Calibration curves generated from droplet-MS. Each point is an average obtained from 3 droplets per concentration, and error bars are ±1 standard deviation. R² = 0.993 for the 2% Rxn Matrix and R² = 0.982 for 0% Rxn matrix.

Evaluating Quantitative Measurement of the Bibenzofuran Isomers by Droplet cIM-MS

Factors like ionization suppression between analytes⁷⁶ as well as loss in ion transmission due to space charge effects⁷⁷ may affect the relative quantification of isomers. To examine the accuracy of both the measured total and fractional bibenzofuran isomer content by droplet-cIM-MS across a range of concentrations, we used a calibration matrix made by preparing solutions at 10 different concentrations ranging from 0 nM to 2.5 μM for 5,7′- and 7,7′-bibenzofuran. These solutions were combined 1:1 at all possible combinations yielding 100 calibration samples. Samples were prepared in 1:1 acetonitrile:0.1% FA with 2% reaction matrix to mimic conditions used for screening. Figure 5a shows that the throughput for the cIM-MS analysis was such that all 430 droplets (4 replicates of each solution plus 3 replicates of a marker in each row) were infused within 10 min or 1.4 s/droplet. The average RSD for the total bibenzofuran content of droplet triplicates from the same well was 9.0%. Droplet cIM-MS signals had expected patterns. For example, Figure 5b shows the individual droplet-cIM-MS trace over a single row where 5,7′-bibenzofuran is held constant at 375 nM while 7,7′-bibenzofuran was reduced from 2500 to 0 nM. The signal at 31 ms, corresponding to 5,7′-bibenzofuran, is constant while the signal at 29 ms, corresponding to 7,7′-bibenzofuran, steps down as expected. Figure 5c,d shows the resulting heatmap for total bibenzofuran content measured (sum of signal for each isomer divided by internal standard signal) and fractional 5,7′-bibenzofuran content (signal of 5,7′-bibenzofuran divided by signal for both isomers) across the well-plate, further highlighting the expected pattern of signals for the array of samples. The total signal had good linear relationship with total concentration of bibenzofuran (R² = 0.97). Likewise, the fractional signal for the 5,7′ isomer had a good linear relationship (R² = 0.98) with the fractional 5,7′ content over all 100 samples (Figure 5e,f).

Figure 5.

Cross-calibration of isomers by droplet cIM-MS. Ten concentrations of 5,7′- and 7,7′-bibenzofuran solutions from 0 to 2500 nM were mixed 1:1 at all combinations. All solutions used 1:1 acetonitrile:0.1% FA with 2% reaction matrix as solvent. Samples were generated from a well-plate in a programmed sequence. (a) Droplet cIM-MS temporal contour map showing log₁₀ of extracted intensity for the monoisotopic [M – H]⁻ of bibenzofuran (437.12 m/z) as a function of drift time. (b) Inset of trace from panel (a) showing the droplet cIM-MS trace of Row F, corresponding to 375 nM of 5,7′-bibenzofuran and descending concentrations of 7,7′-bibenzofuran. (c) Heat map for signal intensity of total isomer across the well-plate. Axis shows the actual concentration, and color indicates signal intensity for the total isomer content. (d) Same as c with heat map showing the fractional content of 5,7′-bibenzofuran. Axis shows the known concentration, and color indicates signal intensity for the fractional content of 5,7′-bibenzofuran. (e) Plot of sum of dimer concentration versus the measured signal (S) of both isomers over the internal standard (R² = 0.975). (f) Plot of the actual fractional 5,7′-bibenzofuran content against the measured signal response (R² = 0.984).

To further evaluate possible interference of one isomer on the other, we plotted the signal for each isomer at a given concentration as a function of changing the second isomer concentration (Figures S4a,c). In the absence of cross-interference, the slope for target isomer concentration should be zero. An analysis of covariance (ANCOVA) confirmed that increasing concentrations of background 5,7′-bibenzofuran had no detectable effect on the signal for 7,7′-bibenzofuran, as all slopes were statistically equivalent to zero (Figure S4b; p > 0.05). Increasing the background concentration of 7,7′-bibenzofuran had no significant effect on the 5,7-bibenzofuran isomer signal at low target concentrations. Similarly, ANCOVA suggests there is no detectable effect from background 7,7′-bibenzofuran on the signal of 5,7′-bibenzofuran at concentrations below 1250 nM (Figure S 4d). However, the signal for 2500 nM 5,7′-bibenzofuran increases with increasing concentrations of 7,7′-bibenzofuran (slope is nonzero with p < 0.05). This signal interference is possibly related to band broadening of peaks causing overlap in signal. Since such high concentrations are unlikely to be encountered in screens, these results show that factors like space charge and suppression of one isomer over the other do not impair the ability to screen within acceptable concentration ranges.

96 Well-Plate Screen with Directed Evolution Enzymes

96 enzyme variants selected from a library generated for a KtnC engineering campaign to develop a biocatalyst for production of 5,7′-bibenzofuran were screened by droplet-cIM-MS to test the system against actual reaction mixtures. The reaction mixtures were diluted 50-fold in 1:1 acetonitrile:0.1% FA and deposited into a 384 well-plate with each row on the well-plate containing variants from a given round of KtnC engineering. For calibration, the final row contained standards from 50 nM to 5 μM. The throughput for measurement was 1.2 s/droplet as illustrated in the extracted ion trace of the screen (Figure 6a). The variants exhibited a wide range of total product and selective isomer formation. The resulting signals by droplet cIM-MS correlated well with LC-MS analysis for both total product (R² = 0.938) and fraction of 5,7′-bibenzofuran (R² = 0.975) formed (Figure 6b,c). We did observe a noisier correlation for higher total isomer content (see data above 2 on the x-axis of Figure 6b). We did not investigate the cause of this variability, but it may be due to ionization competition, space charge effects, or a suboptimal LC method. Nevertheless, a Bland-Altman analysis on the data (Figure S5) shows that the droplet cIM-MS method is biased to undercount the total isomer content with respect to LC-MS data; but 96% of the paired samples lie within the limits of agreement. Given that the bias is systematically negative and proportional to the concentration of the analytes, this should not present a risk within the context of screening. The average RSD for the measured total bibenzofuran content from droplets of the same well was 10% while the RSD for LC-MS triplicates was 15% showing comparable precision. The cIM-MS data allows a quick readout of the overall performance of the different variants by plotting total bibenzofuran output against the fractional 5,7′-bibenzofuran (Figure 6d). An ideal variant would be in the upper right quadrant indicating high yield and high selectivity for 5,7′-bibenzofuran. This experiment validates the practical application of droplet cIM-MS for isomer screening in directed evolution campaigns.

Figure 6.

Screening of KntC variants from different rounds of directed evolution by droplet cIM-MS. (a) Droplet cIM-MS temporal trace showing log₁₀ of extracted intensity of the monoisotopic [M – H]⁻ of bibenzofuran (437.12 m/z). The trace, composed of 468 droplets, includes 96 samples, 9 sets of the marker, and 12 calibration standards. (b) Plot of the total bibenzofuran signal (S) as measured by LC-MS against the total bibenzofuran signal as measured by droplet cIM-MS (R² = 0.938). (c) Plot of the fractional 5,7′-bibenzofuran signal as measured by LC-MS against the fractional 5,7′-bibenzofuran signal as measured by droplet-cIM-MS (R² = 0.975). (d) Plot of the total bibenzofuran signal against the signal fraction due to 5,7′-bibenzofuran.

The results illustrate that droplet cIM-MS can screen isomeric reaction products at high throughput (~1 s/droplet). While demonstrating the potential of droplet cIM-MS for screening, improvements are desirable prior to routine use in directed evolution. Presently, it requires about 10 min to fill a tube (~470 droplets) and the same amount of time to analyze so in a series assays one tube could be filled while the other was analyzed to maintain throughput; however, it required about 3 min to connect the tube to MS which reduces the throughput by 30%. Greater automation of droplet generation to analysis would be useful for screening large libraries of >10⁴ variants. One approach may be to use Venturi-based sampling from well-plates to allow more facile automation.^78–80 Alternatively, coupling droplet cIM-MS to entire droplet-based directed evolution workflows would leverage the full capability of sample reduction and automation possible with droplet microfluidics.⁶⁰ Higher throughput in the cIM-MS would also be valuable. Direct MS has been reported at 33 Hz,⁵⁰ so such improvements may be feasible for IM-MS.

CONCLUSIONS

We have developed and validated a droplet cIM-MS platform for high-throughput screening of isomers from enzymatic reactions using benzofuran dimers resulting from the directed evolution of the cytochrome P450 KtnC as a test case. The method is 128 times faster than the 3 min LC-MS method used to screen the reaction while providing comparable results. This throughput advantage is expected to yield substantial decreases in time to develop biocatalysts by multiround directed evolution. For example, in a 5-round directed evolution campaign with 5000 variants per round, the analysis time by droplet MS would be 14 h (including duplicate sampling with a wash droplet) whereas LC-MS (1 replicate per variant) would require 179 days (assuming 10 h operation time per day). Thus, the presented method enables broader and deeper exploration of protein sequence landscapes compared to the reliance on LC-MS, thereby accelerating the development of biocatalysts.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.6c02164.

Synthetic protocols for noncommercial small molecules; reaction matrix protocol; droplet images; nESI capillary voltage ramping data; carryover evaluation expanded data; custom data analytics software description; ANCOVA analysis of calibration matrix; and Bland–Altman comparison of LC-MS vs cIM-MS quantitation (PDF)