macroMS: Image Guided Analysis of Random Objects by Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry

Abstract

Mass spectrometry imaging is well suited to characterizing sample surfaces for their chemical content in a spatially resolved manner. However, when the surface contains small objects with significant empty spaces between them, more efficient approaches to sample acquisition are possible. Image-guided mass spectrometry (MS) enables high-throughput analysis of a diverse range of sample types, such as microbial colonies, liquid microdroplets and others, by recognizing and analyzing selected location targets in an image. Here we describe an imaging protocol and macroMS, an online software suite that can be used to enhance MS measurements of macroscopic samples that are imaged by a camera or a flatbed scanner. The web-based tool enables users to find and filter targets from the optical images, correct optical distortion issues for improved spatial location of selected targets, input the custom geometry files into an MS device to acquire spectra at the selected locations, and finally, perform limited data analysis and use visualization tools to aid locating samples containing compounds of interest. Using the macroMS suite, an enzyme mutant library of Saccharomyces cerevisiae and nanoliter droplet arrays of Escherichia coli and Pseudomonas fluorescens have been assayed at a rate of ~2 s / sample.

Graphical Abstract

INTRODUCTION

Matrix-assisted laser desorption/ionization (MALDI)-time-of-flight (ToF)-mass spectrometry (MS) can be used to rapidly analyze large numbers of samples on a surface. MALDI-MS requires relatively straightforward sample preparation, has a high salt tolerance, and provides wide chemical coverage. Once the specific location on a target is located, the process of sample ionization and measurement takes ~1 s. Given the high-throughput nature of MALDI-MS, the approach is well suited to large-scale chemical screening projects. Perhaps its best-known use is mass spectrometry imaging (MSI) and related approaches that can probe tens of thousands of samples with fixed raster steps targeting sample arrays. ^1–10 However, there are examples involving imperfectly spaced sample arrays. For example, acoustic liquid handlers can array thousands of nanoliter sample spots, but each dispensing step introduces errors in the position of the spots, allowing the fixed raster MSI to miss several samples. ¹¹ One solution is to use highly dense fixed raster MSI to target all samples; while this can be effective, it is time consuming as the majority of MS samples are acquired on the spaces between the samples.

A more efficient solution is image-guided MS, which efficiently targets samples located on a surface. In this workflow, a whole-slide image containing the samples is visualized to locate the targets of interest, and MS analysis is performed only at these targeted locations. This approach has been used to perform histology-directed MALDI analysis of specific locations in human breast cancer and melanoma tumors.^12,13 Previously, our lab developed the microMS software suite¹⁴ for optically guided MS, which enables accurate targeting of cells scattered across a surface such as a microscope slide. The software performs feature recognition on microscope images to locate the cells and directs a MALDI-MS instrument to target the recognized cells for analysis. The system enabled high-throughput MS measurements of more than 30,000 individual neural cells of rodent brain revealing cell-to-cell lipid heterogeneity,¹⁵ individual cells of human pancreatic islets,¹⁶ and other single-cell analysis projects.^17,18 Because it can work with larger objects, it has also been used to screen microbial colonies for high-throughput metabolic engineering of multistep enzymatic reactions.¹⁹

Here we adapted the optically guided microMS workflow for macroscopic (>300 μm) objects with a simplified set of requirements. As opposed to the ~2 h of microscope imaging needed for locating single cells using microMS, macroscopic samples can be quickly imaged by a commonly available flatbed scanner or camera, making the workflow more adaptable and faster to implement. The overarching goal was to make the approach easier, more robust, and simpler for these larger samples. We demonstrate the method using a standard office flatbed scanner and cell phone camera to obtain images for the larger area samples. With the larger imaged area, a simpler fiducial training method was implemented, as well as interactive image thresholding and inversing. Moreover, new optical correction tools were added to resolve optical defects inherent to common imaging devices, including correction of moderate distortion and imperfect perspective. Also, for a rectangular grid of sample spots created by liquid handlers, macroMS records the identity for each sample spot and analyzes these selected spots. The program is built into a web application for more universal accessibility, equipped with interactive tools such as target recognition, target editing/filtering, a fiducial marking system, optical correction, and limited data analysis. With these enhancements, macroMS enables an efficient task workflow for high-throughput screening at a rate of 1 to 2 s per sample. While several instrument software suites are available, e.g., flexImaging (Bruker Corp., Billerica, MA), these tend not be designed for large numbers of samples, misaligned arrays and optical distortion, which are several of the areas addressed in the new software suite.

We validated the approach by screening median chain fatty acid (MCFA)-producing variants from a site saturation mutant (SSM) library for site Gly 1250 in the ketoacyl synthase domain of fatty acid synthase (FASII) of Saccharomyces cerevisiae. Next, macroMS analysis was performed on a nanoliter droplet array of Escherichia coli and Pseudomonas fluorescens cultures to demonstrate the utility for testing samples from high-throughput liquid handlers by showing differences in their fatty acyl profiles.

EXPERIMENTAL

Software.

Programs were written in Python 3.5 (Python Software Foundation, Wilmington, DE) to enable running several microMS tasks in the cloud environment, such as locating targets from images and target patterning. Programs for optically correcting target coordinates and producing MALDI-MS coordinate files were written using the Python libraries Numpy, Scipy, and OpenCV-Python. The user interface for accessing the functionalities was written in JavaScript implementing the OpenSeadragon viewer library. This viewer enables interactive visualization of the image and highlighting the identified targets, editing targets, and plotting data. The Plotly JavaScript library was used to plot the measured optical distortion as well as plotting mass spectra for samples. Informatics pipeline processing of the submitted MALDI-MS spectra was written in Python combining CompassXport (Bruker Corp., Billerica, MA) and the pymzml library. Django was used as a web framework for connecting the user interface to the server operations for the image analysis, and all job requests submitted by users enter a job queue that is sequentially processed by Celery workers through the RabbitMQ broker.

The Amazon Web Service (AWS) (Seattle, WA), a commercial cloud computing platform, is used to host the web application. The AWS Lambda service is used for serverless processing of time-consuming jobs such as MALDI-MS target path optimization and calculating distance to the nearest neighbor for the distance filter. The mass spectra data for analysis are uploaded directly to the AWS Simple Storage Service (S3) using the AWS Software Development Kit in JavaScript. A separate nano node is launched in AWS Elastic Compute Cloud (EC2) for each data file submitted to AWS S3, and MALDI-MS data analysis is performed in the EC2 node, which generates output files and an interactive map for matching the data analysis results to the targets in the image. The overall architecture of macroMS is shown in Figure 1, and it designed for versions of Chrome and Firefox web browsers that are made after 2017. The web address of the application is https://macroms.scs.illinois.edu.

Figure 1.

Architecture of the macroMS web application. The macroMS web page collects user input for running image analysis and delivers the data to the server. The RabbitMQ / Celery job queue sequentially processes user input by running the core macroMS programs. The web browser displays the result received from the server. Uploading acquired data activates a computer that processes the data and sends results.

Chemicals.

2,5-Dihydroxybenzoic acid (DHB) and N-Phenyl-2-naphthylamine (PNA) were purchased from MilliporeSigma (St. Louis, MO). Acetonitrile was purchased from Thermo Scientific (Waltham, MA). Deoxyribonucleic acid (DNA) extraction and purification kits were purchased from Qiagen (Valencia, CA). DNA Q5 polymerase and all restriction enzymes were purchased from New England Biolabs (Ipswich, MA). All oligonucleotides for polymerase chain reactions (PCRs) and sequencing, e.g., primers with customized degenerate codons, were synthesized by Integrated DNA Technologies (Coralville, IA).

Sample Preparation.

S. cerevisiae (BY4741 strain) was cultured at 30°C in a liquid YPD medium (10 g/L yeast extract, 20 g/L peptone, 20 g/L glucose). All yeast transformants were then cultured and selected in an agar plate of Synthetic Complete (SC) dropout medium lacking uracil (MP Biomedicals, Irvine, CA). The agar plate was stored at 4°C for > 24 h to increase detection of fatty acyls by MALDI-MS. For more details about the cell cultivation and selection conditions, refer to our previous publication.²⁰

To obtain enhanced MS signals for fatty acids, a 75 mm × 50 mm × 1.1 mm indium tin oxide (ITO) coated glass slide with R_s = 70–100 Ω (Delta Technologies LLC, Loveland, CO) was coated with gold nanoparticle (AuNP) using a Desk II TSC sputter coater (50 s, 65 mTorr, 40% power) (Denton Vacuum, Moorestown, NJ). For transferring colonies, a Durapore membrane filter (EMD Millipore, Burlington, MA) was laid on the yeast colonies in the Petri dish. The colonies transferred to the filter paper were imprinted onto the ITO slide by applying gentle pressure with a foam sponge. After imaging by a flatbed scanner at 1200 dpi, the slide was coated with 11 mL of a matrix solution (PNA dissolved in acetonitrile at 5 mg/mL) at a rate of 410 mL/h using a modified version of a spraying device²¹ with the nitrogen gas pressure set to 40 psi.

For the bacterial droplet array, colonies of E. coli and P. fluorescens grown on Lysogeny broth agar plates were collected and mixed with water. Using an acoustic liquid dispenser (Echo 550, Labcyte, San Jose, CA), 2.5 nL droplets of the mixtures were spotted into two 5 × 50 arrays on an ITO slide freshly coated with ~50 s of AuNP sputter coating, followed by a another sputter coating of AuNP (~25 s, 65 mTorr, 40% power) after spotting. Application of the AuNP coating before spotting resulted in uniform spot shapes and sizes of ~300 μm, and the second coating was for laser desorption/ionization-MS.

For the DHB array, water was saturated with DHB by >10 min of sonication. Then 2.5 nL of the DHB solution was dispensed into a 10 × 50 array on an ITO glass slide that was fresh sputter coated with AuNP for ~50 s.

Imaging.

ITO slides with imprinted colonies, droplet arrays of bacterial sample, and a 51 × 51 reference grid array used to measure optical distortion (FA127, Max Levy Autograph, Philadelphia, PA) were imaged by a flatbed scanner (EPSON V300 Perfection, Nagano, Japan) at 1200 dpi. A Prespotted AnchorChip II (PACII) target, prespotted with α-Cyano-4- hydroxycinnamic acid (HCCA) matrix (Bruker Corp.), was scanned at 800 dpi. A black background was achieved by scanning in a dark room while the cover of the scanner was fully lifted. A 13 M-pixel cell phone camera (K30, LG, Seoul, Korea) was used to photograph the DHB spot array on the ITO slide, the grid distortion target, and the PACII target. Steps for generating target lists from the images and the method for imaging and targeting the DHB spot array by cell phone imaging are described in the Supporting Information.

Target Accuracy Measurement.

For measuring the accuracy of sample targeting by macroMS with the different imaging tools and optical correction modes, the HCCA matrix spots on the PACII plate were targeted using the macroMS output. The accumulation of 2,000 laser shots, with the laser size set to ‘Medium’ at 80% power, created visible ablation marks (~50 μm) in the HCCA spots, and the μm scale X and Y coordinates of the ablation marks were obtained from the MALDI-MS control software flexControl 3 (Bruker Corp.) when the laser target point of the video feed was moved to the center of the mark. The X and Y coordinates of the center of the target HCCA spot were measured by averaging the X-axis coordinates of the left and the right edges of the spot, and averaging the Y-axis coordinates of the top and the bottom edges of the spot. Accuracy was measured by calculating the distance between the ablation mark and the center of the target HCCA spot using the Pythagorean theorem where the X-axis and Y-axis distances between the two points were used as the sides of the right triangle. The smaller HCCA spots (~460 μm) at the four corners of the PACII were chosen as fiducials for all images used for accuracy testing.

MALDI-ToF-MS Analysis.

An ultrafleXtreme mass spectrometer (Bruker Corp.) was used for sample analysis as well as target accuracy testing. For yeast colony screening, 1,500 laser shots were collected at 1000 Hz with the laser size set to ‘Ultra’ (~70 μm footprint) at 65% power in negative ion reflectron mode, and the operating parameters were 170 ns for pulsed ion extraction time, 20 kV for accelerating voltage, 17.85 kV for extraction voltage, 5 kV for lens voltage, and 26.4 kV for reflector voltage. An 800-μm random walk was used with 50 shots per raster point. Instrument settings for testing the microbial sample arrays and the DHB droplets array are described in the Supporting Information.

Data Analysis for MALDI-ToF-MS of Microbial Samples.

The macroMS data analysis program was used to analyze the generated mass spectra. First, the m/z values for the upper and lower edges of peaks corresponding to C16, C16:1, and C18:1 fatty acids (m/z 253, m/z 255, or m/z 281, respectively) were obtained using flexAnalysis 3 (Bruker Corp.) from ~5 randomly sampled mass spectra that were plotted in overlaid view mode. The text file listing the m/z ranges for the peaks, the original image used for generating the targets, and the zipped file containing the acquired mass spectra were submitted to the macroMS platform for data analysis and data visualization. The resulting Excel spreadsheet lists ion counts summed over the peak m/z ranges for all mass spectra. The output data were used to sort colonies by the C16:1/C18:1 ratio, and to distinguish the droplet samples of E. coli and P. fluorescens by the C16/C16:1 ratio.

PCR Sequencing.

Five S. cerevisiae colonies were identified for high C16:1 fatty acid and PCR-amplified using Q5 High Fidelity polymerase from New England Biolabs (Ipswich, MA), and purified with a PCR purification kit from Qiagen (Valencia, CA). The PCR products were sequenced by Sanger sequencing through ACGT, Inc. (Wheeling, IL).

RESULTS AND DISCUSSION

macroMS Workflow.

The overall objective of the macroMS software suite is to determine the location of analytes in a sample, even if irregularly distributed, to guide the acquisition of MS spectra from the desired locations, and aid data analysis for locating analytes of interest. The software requires the user to obtain an image so that it can identify analyte locations. Our goal was to develop a robust platform that will work with most images acquired either by a camera or a flatbed scanner. Accordingly, we needed to account for image distortion or an imperfect viewpoint. Figure 2 presents an overview of the workflow for the macroMS software.

Figure 2.

The macroMS workflow for optically guided MS. First, the image can be thresholded or inverted for target finding. Interactive tools are provided for filtering the target list and for marking fiducial points. Locations of targets and fiducials can be corrected for image distortion and/or an imperfect viewpoint. Optionally, the software performs limited data analysis.

Upon submission of an image, macroMS provides a web page with the functionalities needed for performing each step of the workflow (Figure S1). These steps include (1) preprocessing the image, (2) target finding, (3) filtering and editing target lists, (4) marking fiducial points, (5) optical correction of target images, (5) generating a geometry file, and (6) data analysis (Figure 2). The web application has an accompanying Gitbooks website that describes the procedure for imaging, using the web application, and setting up the MALDI-ToF-MS instrument for analysis.

Target List Generation.

For image preprocessing, macroMS provides an image thresholding filter to maximize the contrast between the sample and the background, enabling more robust discovery of targets. Another preprocessing tool is image inversing for identifying samples that are darker or brighter than the background (Figure S2). Feature finding is performed using the microMS¹⁴ source code, which implements the scikit-image Python library. When the microMS feature-finding parameters are set, such as color, pixel size, and intensity, clicking the run button results in green boxes highlighting the identified targets in the image (Figure S2). Targets can be edited by removing them in the box remove mode and by clicking on a location to add a box.

For population-level filtering of the targets, histogram graphs can be opened that show the distribution of size, circularity, or distance to the nearest neighbor (Figure S3). The histogram plot enables the user to interactively set upper and lower bounds for population-level filtering. Size filtering can be useful for removing artifacts such as small specks of dusts or including samples of a specific size. Filtering by circularity can be useful for removing individual circular samples that are merged; the merged samples have lower circularity than individual spots.

Furthermore, physically thick samples can result in reduced sensitivity. In this case, the patterning function can generate four targets around the identified target area, which enables probing at the edge side of the samples (Figure S4). The edge region of a thick sample is often thinner than the center area and can result in a better MALDI-MS signal. However, quadrupling the number of targets to above ~10,000 resulted in a slower response by the viewer to user activities, indicating the upper limit of manageable target numbers by macroMS is 10,000 (data not shown).

Once the targets have been located, the list needs to be transferred to the MS instrument control software. Before this is done, macroMS includes an option for minimizing the stage travel distances via the traveling salesmen problem (TSP) algorithm, which reduces MS analysis time. When the number of samples is greater than 1000, the samples are divided by subregions that are individually TSP optimized. If TSP optimization is not selected, travel distance is reduced by grouping targets by 100 subregions and processing targets by group, resulting in targeting closely located samples first.

Conversion of Coordinates into the Instrument System.

When the target list is generated, image pixel coordinates are converted to the physical coordinates used by the mass spectrometer (i.e., the sample-stage coordinate system). macroMS uses a point-based similarity registration method where fiducial points are three of the four corner edges of the rectangular area holding the samples. The corner edge points can be the corner edge of an ITO slide or dots positioned at the corner edges of a rectangular area, such as small DHB droplets or MALDI laser ablation marks. The points are indicated to the macroMS software by clicking at the spots in the fiducial edit mode (Figure S5). The fiducial points are also used to define a rectangular area for a relative coordinate system to convert the pixel coordinates. The final output file lists the targets and the fiducials in the relative coordinates system, which is interpretable by the instrument. The file is downloaded from macroMS and opened by the instrument’s device control software on the local computer. The physical coordinates for the fiducial points are indicated to the instrument by using the Teach Positions functionality of the device control software (Figure S5). Using the physical coordinates and the relative coordinates of the fiducial points, the device software generates a mathematical transformation function for generating physical coordinates from the relative coordinates of all targets. Use of three fiducial points for coordinate registration simplifies the coordinate transformation procedure compared to microMS, which uses many more fiducial points to allow smaller objects to be accurately targeted. The possible disadvantage is that any errors in the three points can cause targeting issues, but we did not find this to be an issue in our experiments.

Optical Correction.

macroMS provides tools for correcting two common errors caused by the imaging tools. First, using a camera may result in an imperfect viewpoint where the rectangular ITO slide holding the samples appears as a parallelogram, a trapezoid, or a kite. This can be problematic because geometrically correct representation of targets in the image is required for accurate target finding. The correction functionality performs viewpoint correction using the OpenCV library to derive the coordinates of the targets from the corrected ideal camera angle (Figure 3a). For this computation, the pixel coordinates of the four corner edge points of the rectangular ITO glass, or any rectangular area containing the samples, are indicated to the macroMS software by clicking on these four points in the fiducial edit mode. By using the measured and the expected shapes of the rectangular area, macroMS performs affine transformation for generating the corrected target coordinates.

Figure 3.

Demonstration of perspective correction and distortion correction using the macroMS tools. (a) An image of the PACII plate with an imperfect perspective and angle was processed in the macroMS platform resulting in a trapezoidal array. The perspective transformation function generates a geometrically corrected target list. (b) An image of the 51 × 51 reference grid acquired by a cell phone camera was processed in macroMS resulting in 2601 recognized points. macroMS then performs array registration by using the coordinates for the grid and measures the optical distortion. This distortion is subtracted from the target coordinates, resulting in an improved target accuracy for the 460-μm HCCA spots by cell phone imaging. The examples shown for the corrected spot and uncorrected spot are target R1C5 of Figure 4g and Target R2C5 of Figure 4f, respectively.

Second, moderate distortions can be made to the image due to an imperfect camera lens or a mechanical error in the flatbed scanner.²² Such image distortions may result in inaccurate target coordinates, preventing the analysis of smaller samples (~300 μm). macroMS measures optical distortion by using a reference grid array with precise dot positions. By comparing the measured and the expected positions of the dots in the grid, a spatial map of optical distortion is calculated. The distortion expected for the location of the targets is subtracted from the target coordinate to correct for the distortion (Figure 3b).

Lastly, flatbed scanners require the sample side of the ITO slide to face down while imaging. This is problematic when using a nonstandard ITO slide that is larger than 75mm × 50 mm; available slide holders are incompatible with the larger size and result in a small gap between the slide and the scanner glass. However, without a slide holder, the sample on a nonstandard ITO slide will make physical contact with the scanner glass. macroMS provides an option for creating a target coordinate that is a horizontal mirror image of the measured target list. This enables imaging of ITO slides with the sample side facing away from the scanner, resolving the sample contact problem.

Together, these simplified imaging tools, simple coordinate registration, and immediate target finding using the web tool combine to provide a rapid workflow for high-throughput analysis. Excluding MALDI matrix application, the preparatory steps needed for the microMS workflow for the analysis of single cells on a 75 mm × 25 mm microscope slide take several hours, and include fiducial marking, imaging, target finding, and instrument coordinate setup. All preparatory steps for macroMS analysis, including MALDI matrix application, for microbial colonies on a 75 mm × 110 mm ITO slide typically take about 35 min.

Data Analysis and Visualization.

macroMS provides limited data analysis for mass spectra acquired from the target points that are listed in the output of the macroMS target finding operation. To aid screening for specific compounds of interest, macroMS reports ion counts summed over defined m/z values for the compounds. The analysis functionality accepts short MALDI-ToF-MS mass spectra in an m/z range of 1,000. Using CompassXport 4.0, the analysis converts the Bruker proprietary fid file format into the mzML format. Then the pymzml package is used to read the mass spectra from the mzML file and the ion counts for these m/z ranges are summed in the Python environment. Finally, an Excel file listing ion counts for the m/z ranges for all of the mass spectra is generated, and users can use Excel functions to find samples and perform further data analysis (Figure S6).

Another output available is the web page for interactive viewer plotting data analysis results to the input image. The page is equipped with selection tools for locating samples of interest, and the mass spectra near the peak m/z ranges can be plotted for the selected targets. Lastly, the program can generate a heatmap scatter plot of data for samples by placing a box at the sample spot that is colored in the heatmap scale, which aids discovery of spatial patterns in the data. The web addresses for downloading the Excel file and accessing the image viewer are sent to users by email through the AWS Simple Email Service. Finally, for a rectangular array of sample spots created by liquid handlers, macroMS provides an array registration function to enable matching sample names to each spot in the array (Figure S7). To confirm the array registration with the user, rows and columns of the recognized array are plotted within the image of the array. After confirming, the user submits a CSV file listing sample names, which will be visualized next to each dot in the array. The matched sample names are added to sample data in the Excel file for the data analysis output and also to the file names for the acquired MS spectra for the samples.

Demonstration of the Accuracy of Sample Targeting.

An important performance parameter for optically guided MS is target accuracy of the computational workflow. Targeting errors were measured from targeting arrays of MALDI matrix spots of a specific size. Targeting the 460-μm spots in the PACII array using a scanned image at 800 dpi resolution (Figure 4a) resulted in an average accuracy of 115 ± 40 μm among the 24 sampled spots (Figure 4b). Targeting accuracy for larger spots (~790 μm) by a cell phone photograph (Figure 4c) with perspective correction was 115 ± 65 μm among the 48 sampled spots (Figure 4d). All sampled spots were correctly targeted without optical distortion correction, and the upper bounds of errors measured for both spot sizes were well within the radius of the target spots. The data indicate the target spot sizes are within the working range of the target size for the imaging modes. The results also show that the perspective transformation function of macroMS can be used to target samples that are imaged at non-ideal camera angles so long as the samples are within a rectangular area with visible corner edge points.

Figure 4.

Profile of targeting errors for PACII spots from the acquired images on the left, with corresponding heatmap error plots on the right. C#s and R#s in the heatmaps refer to column and row numbers of targets in the corresponding arrays, respectively. (a) Image acquired by a flatbed scanner at 800 dpi and (b) localization errors in μm, targeting a 460-μm spot array. (c) Cell phone image and (d) localization errors in μm, targeting a 790-μm spot array. (e) Cell phone image and (f) localization errors in μm, targeting an even row of a 460-μm spot array. (g) μm errors for the odd row of the array using image (e) after distortion correction. Perspective correction was applied for measuring target accuracy for (d), (f), and (g).

To demonstrate the optical distortion correction functionalities, smaller 460-μm spots were targeted using cell phone imaging (Figure 4e). With perspective transformation only, average accuracy was measured to be 165 ± 95 μm for 46 sampled spots, and ablation marks for 15 of the 46 of the sampled spots were off-target or near miss (Figure 4f). After perspective transformation and distortion correction of the same cell phone image, average accuracy improved to 80 ± 15 μm, correctly targeting all 46 sampled targets (Figure 4g). The image of the reference grid and the measured optical distortion using the reference grid are shown in Figure 3b. With perspective transformation and distortion correction, 2.5 nL nanoliter droplets of DHB solution in the 50 × 10 array could be correctly targeted using a cell phone image, as demonstrated by detection of DHB [+H] signal at all of the 300-μm spots shown in Figure S8.

Reducing the target size is important as the maximum number of samples per slide increases according to the squared value of the reciprocal of target size. Thus, we investigated approaches to decrease sample sizes. For our distortion correction experiments using a cell phone camera, the laser was slightly out of focus, resulting in a 70-μm diagonal shift for the ablation mark from the target mark. When the measured coordinates of the ablation marks were corrected by subtracting the horizontal and vertical components of the laser shifts, the accuracy improved from 80 ± 15 μm to 30 ± 15 μm (Figure S9). The majority of the residual error after distortion correction is from our shifted laser focus. Therefore, to further improve target accuracy, it may be helpful to perform fiducial spot registration using the actual laser ablation point by better aligning the laser ablation spot with the targeting mark. For samples smaller than 500 μm, further improvements in targeting accuracy may be achieved by using smaller fiducial points and higher imaging resolution.

Different imaging devices resulted in different targeting accuracy and precision. However, several generalizations can be made. Microbial colonies and droplet samples created by liquid droplets are mostly larger than 1000 μm, which is significantly larger than the size of the droplets that could be successfully targeted by both flatbed scanner and cell phone imaging. Therefore, both imaging modes can be used for these samples. For samples ranging from 500 μm to 1000 μm, a scanner should be used due to its higher accuracy relative to cell phone imaging. For smaller samples that range from 300 μm to 500 μm, distortion correction is needed. As an alternative to distortion correction, a random walk mode can be considered to probe the smaller samples that are screened by the minimum distance filter. For samples smaller than 300 μm, microMS is more effective.

Colony Screening of a Yeast Fatty Acid Synthase Mutant Library.

We next validated the ability of macroMS to locate hundreds of microbial colonies on a surface using a simple office scanner and then probe them for their chemical contents. Previously, the Gly1250 site of the FASII enzyme was found to impact production of MCFAs where a glycine to serine variant showed an increased production of MCFA.²⁰ The variant was found by screening colonies of the SSM library for the Gly1250 site for an increased level of phosphatidylcholine (PC) with shorter chain fatty acyl chains PC(32:2) versus longer chains PC(34:2) as a proxy for MCFA production. The expectation is that colonies with the G1250S mutation will show a greater level of shorter chain fatty acyl fragments from PCs when exposed to a MALDI laser at high power. Therefore, macroMS-based colony screening was performed to find colonies with shorter chain fatty acyl fragments.

From the scanned image of the ITO slide with imprinted S. cerevisiae colonies (Figure 5a), 749 colonies were detected and targeted by MALDI-ToF-MS following macroMS (Figure 5b). The average C16:1 over 18:1 ratio for the colonies was 3.8 ± 2.9. Using the Excel output file, five of the colonies showing the highest C16:1 over 18:1 ratio were found, with the average value of these colonies being 18.1±1.1 (Figure 5c). DNA sequencing of these colonies demonstrates that all of the five isolated colonies were G1250S mutants (Figure 5d). Thus, our results demonstrate that the macroMS-based screening of enzyme mutant libraries is effective when testing imprinted microbial colonies.

Figure 5.

Yeast colony screening for G1250 site saturation mutagenesis library. (a) A scanned image (1200 dpi) of a 50 mm × 75 mm ITO slide with imprinted S. cerevisiae colonies. (b) Result of target finding by macroMS. Green marks are the found targets. (c) Scatter plot of the per colony ratio of C16:1 fatty acid over C18:1 fatty acid. The red arrows indicate colonies selected for sequencing, and the horizontal line shows a median value of 2.7. (d) DNA sequencing showed all colonies had the G1250S genotype.

Imprint-based colony screening is a promising method for testing colonies in biofuel research and microbial identification. Instead of using colony-picking robotics to transfer colonies to a MALDI plate one at a time, this method uses simple filter papers to transfer hundreds of colonies onto an ITO slide by imprinting them directly from Petri dishes. macroMS enables the use of simple imaging methods, a fast fiducial training procedure, and rapid data analysis for efficient analysis of the imprinted colonies. The colony screening method, combining the imprinting technique and macroMS, can take as little as 30 min from Petri dish to the start of MALDI spectra acquisition, which continues at a rate of 1–2 s per colony. For the purpose of screening multiple samples, this method is an improvement over other MS techniques, such as gas chromatography-MS, which runs 30 min per sample in addition to sample preparation time.

Analysis of Nanoliter Droplets of Bacterial Colonies Solution.

As a demonstration for high-throughput testing of densely packed samples created by liquid handlers, the macroMS workflow was used to test 5 × 50 arrays of 2.5 nL droplets of solutions for E. coli and P. fluorescens (Figure 6a) using distortion correction (Figure S10). The average ratios of the C16:1 over C16 fatty acids for the array for P. fluorescens and the array for E. coli were 1.11 ± 0.04 and 0.45 ± 0.06, respectively (Figure 6c). Using the heatmap plotting functionality of macroMS, the ratios were plotted showing the distinct and reproducible differences between the two samples (Figure 6b). This result indicates that the macroMS workflow can be used for high-throughput screening of nanoliter sample arrays created by liquid handlers. For example, fatty acid fragments and proteins may serve as signatures for high-throughput identification of different microbial species for medical diagnostics and food microbiology by combining liquid handling robots and macroMS. A limitation of the workflow is that liquid samples of nanoliter volumes can be invisible to imaging devices and macroMS analysis when dried, due to the small volume and low concentration of salt that remain after drying. In this case, the concentration of salt or MALDI matrix can be increased for visibility.

Figure 6.

Testing of nanoliter droplets of microbial samples. (a) A 5 × 50 array of 2.5 nL droplets of cultures of P. fluorescens and E. coli. The height of the image is approximately 5 cm. (b) Corresponding heatmaps of the arrays in (a) showing the ratios for C16:1/C16 fatty acids. (c) Scatter plot of the ratios for C16:1/C16 fatty acids.

CONCLUSIONS

macroMS is a web platform that enables the use of commonly available imaging tools for optically guided MS analysis of arbitrary macroscopic samples that are visible in an image. The platform provides functionalities for sample finding, target filtering, and device coordinate generation through a simple web interface. macroMS also provides correction tools for obtaining accurate locations of samples from images with moderate optical distortion and imperfect viewpoints. Equipped with several data analysis and visualization tools for the mass spectra generated for the samples, macroMS provides a pipeline for completing MS screens to search for specific compounds. Using the full workflow, samples of 300 μm were targeted reliably, and the upper limit for target number was found to be ~10,000. For samples smaller than 300 μm, microMS is more effective. The workflow can be used in diverse situations to include directed evolution of enzymes, compound screening, or discovery of antibiotics producing microbes.