A Statistical Approach to System Suitability Testing (SST) for Mass Spectrometry Imaging

Abstract

RATIONALE:

Mass spectrometry imaging (MSI) elevates the power of conventional mass spectrometry (MS) to multi-dimensional space, elucidating both chemical composition and localization. However, the field lacks any robust quality control (QC) and/or system suitability testing (SST) protocols to monitor inconsistencies during data acquisition, both of which are integral to ensure the validity of experimental results. To satisfy this demand in the community, we propose an adaptable QC/SST approach with five analyte options amendable to various ionization MSI platforms (e.g., DESI, MALDI, MALDI-2, IR-MALDESI).

METHODS:

A novel QC mix was sprayed across glass slides to collect QC/SST regions-of-interest (ROIs). Data was collected under optimal conditions and on a compromised instrument to construct and refine the principal component analysis (PCA) model in R. Metrics including mass measurement accuracy (MMA) and spectral accuracy (SA) were evaluated, yielding an individual suitability score for each compound. The average of these scores is utilized to inform if troubleshooting is necessary.

RESULTS:

The PCA-based SST model was applied to data collected when the instrument was compromised. The resultant SST scores were used to determine a statistically significant threshold, which was defined as 0.93 for IR-MALDESI-MSI analyses. This minimizes the Type-I error rate, where the QC/SST would report the platform to be in working condition when cleaning is actually necessary. Further, data scored after a partial cleaning demonstrates the importance of quality control and frequent full instrument cleanings.

CONCLUSIONS:

This study is the starting point for addressing an important issue and will undergo future development to improve the efficiency of the protocol. Ultimately, this work is the first of its kind and proposes this approach as a proof-of-concept to develop and implement universal QC/SST protocols for a variety of MSI platforms.

1. INTRODUCTION

Mass spectrometry (MS) is a powerful analytical technique that has been utilized for a multitude of biological applications such as proteomics, metabolomics, lipidomics, and many more. The advent of mass spectrometry imaging (MSI) has expanded the capabilities of conventional mass spectrometric techniques by reporting the chemical composition of a sample with respect to analyte spatial localization. Mass spectrometry analyses, regardless of the platform, are complex and inherently variable between experiment days, samples, and scans^1–3; nonetheless, MS-based approaches offer the sensitivity and specificity needed to investigate complex biological questions. Therefore, the variability within and between platforms requires characterization and monitoring to ensure the reproducibility, reliability, and validity of the analytical measurements obtained. In response, quality control (QC) and system suitability testing (SST) protocols have been developed and implemented to evaluate the performance of an analytical platform^4–6.

QC procedures are used to measure and control the existing variability in an experimental platform⁷ at different steps in an experimental workflow. Provided that most MS experiments involve coupling liquid chromatography (LC) and electrospray ionization (ESI), QC standards may be used to monitor variability in sample preparation (e.g., protein digestion), during chromatographic separation (e.g., retention time, peak abundances), the MS performance (e.g., spray stability, mass measurement accuracy, calibration, peak abundances), and post-processing tools (e.g., data dimensionality reduction, peak picking, database annotation)^7,8. Such processes are intended to provide a holistic evaluation of the system to ensure proper working condition prior to sample analysis⁹ and may include a previously characterized standard, isotopically labeled compound, and/or commercially available mixture.^6,10 For instance, a wide variety of standards, such as a whole-cell lysate or peptide mixture, could be employed as a QC in proteomics⁷ and are accessible through commercial vendors. Untargeted metabolomics platforms may utilize a diverse small molecule mixture that contains a combination of native and isotopically labeled species^9,10.

Despite commonalities with other MS techniques and its growing influence, the field of MSI lacks any universal QC/SST protocols and standards. Regardless of application, MSI experiments commonly consist of hundreds to thousands of scans per sample across several days, ultimately increasing the likelihood of introducing platform-derived variability. To compare data collected from different experimental groups over time, it is imperative to minimize variability throughout the duration of the experiment and ensure the instrument is collecting reliable data. Furthermore, MSI experiments predominately involve tissue imaging that is commonly sourced from a model organism or disease state cohort. If sub-optimal instrument performance were to result in the loss of samples, this could compromise an entire study. Whole tissue sections are precious, particularly in the case of serial sections, and cannot be sacrificed for troubleshooting or quality control. Developing a robust, community-based QC standard is particularly important due to the utility of MSI in applied biological and clinical studies^11–14 to ensure the validity of experimental results when studying disease states and diagnosing patients.

Due to differences in MSI techniques compared to traditional chromatographic-based MS approaches, specific considerations are necessary in establishing a standardized QC/SST protocol in the field. First, MSI uniquely presents the spatial distribution of all ions detected in a sample. While most components of existing MS-based QC protocols may be integrated into a model for MSI, metrics to measure data quality with regard to spatial information must also be included. While tissue sections or homogenates are an option for QC procedures, this approach is costly, requires animal sacrifice, and presents challenges in creating a reproducible quality control matrix and workflow. Alternatively, an analyte mixture is an attractive option to satisfy the requirements of an SST protocol, as this would omit the necessity of sacrificing a biological organism and is more cost-effective. However, to be inclusive of multiple omics and ionization sources, analyte selection must be strategic to cover a broad m/z range, include diverse chemical structures, and be adaptable to different MSI sources (e.g., MALDI, MALDI-2, DESI, IR-MALDESI). Additionally, QC/SST methods require the ability to accurately interpret deviations in the data to rapidly determine whether analyses may continue or if cleaning and/or troubleshooting is necessary. To date, the only precedent of QC in MSI is a report from Condina et al., where variations of egg whites were utilized as a QC standard for matrix-assisted laser desorption/ionization (MALDI) MSI of peptides and N-glycans¹⁵; however, this approach was not necessarily designed as a generalized QC/SST for a variety of omics for multiple platforms.

Herein, we utilize the infrared matrix-assisted laser desorption electrospray ionization (IR-MALDESI) platform to develop a QC/SST mixture and protocol that is the first of its kind in the MSI community. A QC/SST mixture comprised of five analytes was sprayed onto glass slides and regions of interest (ROIs) were imaged at various conditions of instrument suitability to develop a statistical model and describe the state of the instrument. The constructed model utilizes a scaled principal components analysis (PCA) which is a multivariate analysis algorithm that has been widely studied as a method for data dimensionality reduction and as a classification tool^16–18. This PCA implements characteristics tailored to imaging data, such as monoisotopic and isotopic peak detection frequency, and general characteristics of mass spectrometry data such as mass measurement accuracy, resolution of isotopic distributions, and peak abundance. Ultimately, the procedure and scoring protocol were developed to use any number of these compounds (i.e., 1–5) that are reliably detected by a given ionization method and can be applied to any MSI platform. The scoring algorithm will report a final SST score that will recommend the researcher to continue analysis on a suitable system or to pursue troubleshooting measures on a deficient system.

2. EXPERIMENTAL

2.1. QC/SST Mixture Composition and Sample Preparation

The QC/SST mixture is comprised of five analytes at equimolar concentration (15 μM) dissolved in 50% MeOH (Fisher Scientific, Hampton, NH, USA) in LC-MS-grade water (Fisher Scientific, Hampton, NH, USA). The compounds included in the mixture are as follows: caffeine (Sigma Aldrich, Carlsbad, CA, USA), emtricitabine (Fisher Scientific, Hampton, NH, USA), propranolol (Fisher Scientific, Hampton, NH, USA), fluconazole (Sigma Aldrich, Carlsbad, CA, USA), and fluoxetine (Sigma Aldrich, Carlsbad, CA, USA). All species selected are exogenous compounds that have been previously detected by a variety of MSI ionization sources^19–23.

An HTX TM-sprayer (HTX Technologies, LLC, Chapel Hill, NC, USA) was used to uniformly coat plain glass slides (1 mm, Fisher Scientific, Hampton, NH, USA) with the QC mixture. All slides were sprayed with four passes in a crisscross (CC) pattern. The nozzle velocity was 500 mm/min with 3 mm track spacing, and the solvent flow rate was 10 μL/min with nitrogen gas flowing at 10 psi. Fresh slides were sprayed the day before each round of analysis and were stored overnight at −20°C. Relevant information about the analytes included in this study as well as the experimental workflow are included in Figure 1.

Figure 1.

A) Schematic of the experimental and data processing workflow. B) Structures of the five QC/SST analytes and a table depicting relevant information. The highlighted atom in the associated chemical formula indicates whether the isotopic peak utilized in the model is ¹³C or ³⁴S.

2.2. Collection of QC/SST Datasets by IR-MALDESI-MSI

The workflow for imaging QC/SST regions-of-interest (ROIs) is intended to mimic the workflow for tissue imaging by IR-MALDESI, therefore the procedure is consistent to those described in previous reports²⁴. On the Next Generation (NextGen) IR-MALDESI system²⁵ an electrospray emitter is directed coaxial to the mass spectrometer and orthogonal to the Peltier-cooled sample stage. The electrospray plume consists of 50% acetonitrile (Fisher Scientific, Hampton, NH, USA) in LC-MS-grade water with 0.2% formic acid modifier (Sigma Aldrich, Carlsbad, CA, USA) and is stabilized by applying high voltage (3.0–4.0 kV) and optimizing the solvent flow rate (1–2 μL/min). To form an ice matrix for analysis, the glass slide was immediately transferred from −20°C to the sample stage and cooled to −8°C. While equilibrating, the enclosure was purged with nitrogen gas (Arc3 Gases, Raleigh, NC, USA) to reduce the relative humidity (RH) of the enclosure below 12%. The stage temperature and nitrogen gas flow were maintained for 15 minutes. Afterwards, humidity was re-introduced by opening the enclosure and incubating beakers of warm water in the enclosure. Once ice was formed and stabilized for 10 minutes, the source enclosure was closed and purged again with nitrogen gas to achieve <12% RH. The desired QC/SST ROI was indicated in the RastirZ software, triggering a 10 kHz²⁶ 2970 nm laser (JGM Associates, Burlington, MA, USA) to be fired at discrete coordinate locations and desorb neutral sample material for subsequent ionization in the electrospray plume. The optical path of the laser includes beam expansion with a −75 mm plano-concave beam expander (Thorlabs, Newton, NJ, USA) before collimation with a +250 plano-convex lens positioned 175 mm away (Thorlabs, Newton, NJ, USA). The laser beam was re-shaped into a square, or top-hat, shape with a custom ZnSe diffractive optical element^27,28 (DOE) (HOLO-OR, FO-1933 #2) prior to being focused on the sample with a +25 mm aspheric lens (Thorlabs, Newton, NJ, USA). The laser spot was focused and aligned using burn paper (Edmund Optics, Barrington, NJ, USA) prior to data acquisition. The laser was set at the appropriate pulses-per-burst (ppb) for each round of analysis to provide 2.2–2.4 mJ of energy for each laser shot.

All ROIs collected utilized a step size of 200 μm and consisted of 400 scans (i.e., 20 scan lines × 20 scans per line). The NextGen IR-MALDESI source is currently coupled to an Orbitrap Exploris 240 (Thermo Fisher Scientific, Bremen, Germany), where analyses were conducted in centroid mode²⁹ with a resolving power (RP) of 240,000_FWHM at m/z 200 in positive mode only. An m/z range from 100–500 was utilized and the EASY-IC internal standard (fluoranthene, [M^●+] m/z 202.0777) was enabled to ensure high mass measurement accuracy (MMA) within ± 2.5 ppm. To synchronize timing events between the source and instrument to optimize ion accumulation, automatic gain control (AGC) was disabled and the injection time (IT) was fixed at 15 ms³⁰. Routine mass calibration was performed on the mass spectrometer on the morning of each analysis with FlexMix (Pierce Biotechnology, Rockford, IL, USA).

Three different groups of data were collected based on the state of the instrument to construct, refine, and test the QC/SST model (i.e., “clean”, “compromised”, “newly clean” datasets). For the initial “clean” dataset, the front-end ion optics and the quadrupoles of the instrument were cleaned prior to collecting 118 ROIs (47,200 scans) across three days. Concurrent studies were conducted on the platform until another cleaning was required. At this point, 68 “compromised” ROIs (27,200 scans) were collected across three days to evaluate the QC/SST model and define the SST score cutoff for our platform. To evaluate recovery of the instrument, the front-end optics only were cleaned after collecting the compromised data and 10 “newly clean” ROIs were collected on the platform within a single day of analysis.

2.3. Data Analysis

Raw mass spectra were viewed in QualBrowser in XCalibur 4.1.50 (Thermo Fisher Scientific, Bremen, Germany). MSConvert³¹ was used to convert .RAW data to .mzML files, which were subsequently converted to .imzML files using imzMLConverter³² for visualization and processing in MSiReader Pro v2.60^33,34 (MSI Software Solutions, LLC, Raleigh, NC, USA). All datasets required extraction of each ion’s experimental m/z, MMA, ion flux (ions/sec), spectral accuracy ion count³⁵, and detection frequency (DF) of the monoisotopic and isotopic peaks. These data were extracted on a scan-by-scan (observation) basis utilizing the following tools: Auto MSI QC, MMA data export, and spectral accuracy data export. Further, the data were categorized based on experiment day and ROI number.

Following extraction, all data were compiled and formatted into CSV files. These data were analyzed in R (version 4.3.1) for construction of the QC/SST statistical models. The code for developing the model, clean code for generating SST scores, and the CSV files for all data can be found in the Supplemental Material.

2.4. Construction of the System Suitability Testing Model with Clean MSI Data

The QC/SST model utilizes a preliminary filter before an ROI is tested in the model. If the minimum criteria are not met in the pre-filter, the QC immediately fails, and troubleshooting is advised. Otherwise, the ROI will pass through to the model and an SST score will be calculated to characterize the current state of the instrument, then compared to a statistically selected threshold. Before evaluating the scores of the compromised datasets, these data metrics and thresholds require characterization. The procedure described herein was developed using a random selection of 89 clean ROIs, hereafter termed training ROIs, where the remaining 29 were held-out and used for testing the procedure.

Detection frequency (DF) of the monoisotopic peak was selected as the preliminary filtering metric to determine if testing data will be subjected to the QC/SST model, and is described in Equation 1.

$$\mathit{\text{detection frequency}}\left(DF\right)=\frac{\mathit{\text{number of scans analyte detected}}}{\mathit{\text{total number of scans}}}$$ (1)

The use of this filter was justified because the lack of signal across an ROI would supersede the importance of any other data metrics. Of the clean data, the minimum DF reported for each analyte was 99.5% for caffeine, 97% for emtricitabine, 98.25% for propranolol, 100% for fluconazole, and 98% for fluoxetine; since emtricitabine reported the lowest DF, this threshold was utilized for all analytes in the QC panel. Further, the criteria would require that two or more analytes have a detection frequency less than or equal to 97% for the QC to fail. This condition was added to make the pre-filter more lenient, since the final SST score will also be reflective of the platform’s working condition.

The rest of the model considers the following data metrics: mass measurement accuracy (MMA) (Equation 2) of the monoisotopic peak, monoisotopic abundance, isotopic peak detection frequency (isoDF), and spectral accuracy (SA).

$$MMA\left(ppm\right)=\frac{m/z_{exp}-m/z_{\mathit{\text{theo}}}}{m/z_{\mathit{\text{theo}}}}\times {10}^6$$ (2)

The last two metrics, isoDF and SA, consider the ¹³C peak for all analytes except emtricitabine, where the ³⁴S peak is used instead. Specifically, SA involves estimating the number of carbon or sulfur atoms and calculating the difference between the expected and observed quantity³⁵, which is described in Equation 3.

$$\mathit{\text{known}}-\mathit{\text{observed }}\#of\mathit{\text{ atoms}}=\frac{\left(\mathit{\text{known }}\#\mathit{\text{ of atoms}}\right)\times \left(\mathit{\text{isotope }}\%\right)-\left(\mathit{\text{observed isotope RA}}\right)}{\mathit{\text{isotope }}\%}$$ (3)

These data metrics were chosen to validate the quality of MS data while also including features indicative of image quality for MSI experiments. Further, the included metrics are consistent across MSI platforms regardless of ionization source or the associated mass analyzer.

Next, all score metrics within the model were scaled by analyte to bring all values to a similar scale, as described in Equation 4, where x is the experimental value for each voxel, j, and respective metric, i (i.e., MMA, abundance, SA, isoDF). The mean of the values across the training ROIs is represented as $\mathit{\text{cente}}{r}_i$, and $\mathit{\text{scal}}{e}_i$ is the average standard deviation of the values in each training ROI.

$$\mathit{\text{normalized valu}}{e}_{ij}= \frac{x_{ij}-\mathit{\text{cente}}{r}_i}{\mathit{\text{scal}}{e}_i}$$ (4)

After scaling, a principal component analysis (PCA) was conducted, reducing the data dimensionality to four principal components (PCs) that explain the variability of the data in the training ROIs for each analyte. The metrics are combined and reduced to PCs using a covariance matrix of eigenvectors³⁶, which was all completed in R. Each metric contributes to each PC in some capacity; therefore, each PC does not directly correspond to an individual data metric. The variability explained by each PC was visualized by Scree plots in Figure 2.

Figure 2.

Scree plots for each analyte describing the amount of variability explained by each PC, where data points are accompanied by the respective percentages.

After estimating the variance explained by each PC for each analyte, the 2.5% and 97.5% quantiles of the PC values were defined accordingly to characterize the middle 95% of the PC values for each of the four PCs. These bounds and the variance explained by each PC will be used in later calculations to report an SST score for the ROI. The 2.5% and 97.5% quantiles for each PC have been reported in the Supplemental Material (Table S1).

This procedure was evaluated on the 29 held-out ROIs to evaluate data overlap between the two groups. Since these ROIs are randomly selected and were all collected on a clean instrument, these data should agree. Since the PC distributions are very similar between the two groups of clean data (Figure 3), we can confirm acceptable model replication. The SST scores for all clean datasets are provided in the Supplemental Material (Table S2).

Figure 3.

Density histograms demonstrating agreement between PC values of the held-out clean datasets (testing, red, 29 ROIs) and those that were used to construct the SST model (training, blue, 89 ROIs).

3. RESULTS AND DISCUSSION

3.1. Employing the QC/SST Protocol on Compromised Datasets

Compromised datasets were converted, processed, and formatted in the same manner as the original clean datasets and the relevant data was provided to the model in R to undergo preliminary testing. Representative spectra of a clean and compromised ROI are shown in Figure 4A, accompanied by ion images of all ROIs collected for each dataset discussed (Figure 4B).

Figure 4.

A) Representative mass spectra comparing a clean and a compromised ROI (400 scans averaged). B) Ion images for all ROIs collected for the clean, compromised, and newly clean data sets. Both the clean and compromised datasets were collected across three days of analysis, while the newly clean ROIs were analyzed in the same day.

Any ROIs where two or more analytes did not achieve a DF of >97% failed immediately and were not scored in the model. Of the 68 compromised ROIs that were collected, 58 continued through to the model. For each compromised ROI that passes the pre-filter, the algorithm produces new PC values for each analyte and were calculated as follows (Equation 5):

$$P{C}_{ijq}=w_{ijq}\frac{y_{ij}-\mathit{\text{cente}}{r}_i}{\mathit{\text{scal}}{e}_i}, j=1,\dots ,400, q=1,\dots ,4$$ (5)

As described previously, each metric, i(i.e., MMA, abundance, isoDF, SA), was scaled by its mean ($\mathit{\text{cente}}{r}_i$) and standard deviation ($\mathit{\text{scal}}{e}_i$). To convert the data to a PC value, the scaled value was weighted by the eigenvector of the covariance matrix ($w_{ijq}$). Finally, $P{C}_{ijq}$ is the value determined at the $j^{th}$ voxel for the metric of interest, i, for the $q^{th}$ PC. This calculation was performed for all 400 scans in the ROI.

Subsequently, the calculated values were compared to the previously determined bounds of the respective PC for each analyte. The ratio of values out of 400 that lie within the 2.5% and 97.5% quantile bounds defined with the clean data defined our proportion, $d_q$ (Equation 6).

$$\begin{gathered} d_q=\frac{\mathit{\text{number of the }}{q}^{th}\mathit{\text{ PC values within }}95\%\mathit{\text{ quantiles of }}{q}^{th}\mathit{\text{ PC training data values}}}{\mathit{\text{number of scans }}\left(j=400\right)} \\ q=1,\dots ,4 \end{gathered}$$ (6)

These proportions were calculated for each PC per analyte used the in model. Further, this proportion was weighted by the appropriate variance explained ($v_q$) by the respective PC (Figure 2) to render an individual SST score for each analyte (Equation 7).

$$SS{T}_{\mathit{\text{analyte}}}=\left(v_1*d_1\right)+\left(v_2*d_2\right)+\left(v_3*d_3\right)+\left(v_4*d_4\right)$$ (7)

This scoring was repeated for each analyte and the average score yielded the final SST score that describes the working state of the platform (Equation 8).

$$SS{T}_{\mathit{\text{final}}}= \frac{SS{T}_{\mathit{\text{caffeine}}}+SS{T}_{\mathit{\text{emtricitabine}}}+SS{T}_{\mathit{\text{propranolol}}}+SS{T}_{\mathit{\text{fluconazole}}}+SS{T}_{\mathit{\text{fluoxetine}}}}{5}$$ (8)

The workflow for scoring a QC/SST ROI is visualized stepwise in Figure 5.

Figure 5.

The workflow for testing a QC/SST ROI in the model. The QC must first pass a preliminary filter before continuing through for scoring. The individual scores for each utilized analyte (a) are averaged and compared against the statistically determined threshold to characterize the state of the instrument.

From the 58 final SST scores derived from the compromised data, the 95% quantile was 0.93. This value is utilized as the final SST score cutoff and provides a Type-I error rate of 5%, meaning that a QC/SST would determine the system to be in proper working condition when cleaning/troubleshooting was in fact necessary only 5% of the time. With this cut-off, four of the 58 compromised ROIs would pass and indicate proper working condition of the instrument. Further, if we apply this threshold to the original clean data that was held-out, 20 of the 29 ROIs would pass this criterion, providing a statistical power of 69%, or Type-II error rate of 31%. Ultimately, we can expect the SST protocol to indicate troubleshooting to be necessary when the instrument is in proper working condition, on average, 31% of the time.

The selection of the SST threshold requires compromise between the statistical power and the Type-I error rate. Therefore, while the power of this cut-off could be improved by increasing the minimum score required to pass the SST protocol, there would be a complimentary increase in the Type-I error rate. As a result, the SST threshold was preferentially optimized to minimize the Type-I error rate, as analyzing precious samples on a compromised platform is more costly than cleaning or troubleshooting the system when it is not required. Even at the cost of some statistical power, minimizing the Type-I error rate is more advantageous than the alternative for this application. All final SST scores for compromised ROIs can be found in the Supplementary Material (Table S3).

3.2. Illustration on Newly Clean Data

To evaluate instrument recovery after cleaning, 10 ROIs were collected after a routine cleaning of the front-end ion optics only; ion images of the “newly clean” ROIs are shown in Figure 4B. These ROIs were tested through the model, and eight out of the 10 passed the DF preliminary filter. Of the eight ROIs that continued through to the model, none of the final scores achieved the SST threshold of 0.93. All of these scores are reported in the Supplementary Material (Table S4).

This highlights the necessity to clean the entirety of an instrument platform for full recovery and optimal performance. This is particularly true for imaging platforms, as intact tissue samples are often the samples of interest and may compromise an instrument platform at a faster rate compared to other MS techniques. Further, these results agree with data from preliminary studies (data not shown) and emphasize that implementation of a QC/SST protocol would be advantageous in characterizing the state of the platform and not rely on the instinct and experience of the researcher.

3.3. Future Work and Considerations

This protocol was developed to implement a universal QC/SST protocol and collect a significant amount of data to evaluate the viability of the approach. While this method has shown promising results, the procedure will require optimization to accommodate compatibility with the platform of interest. For MALDI imaging, this may involve co-crystalizing the QC/SST mixture with the appropriate matrix, as an example. Additionally, given the nature of this protocol, other labs will need to replicate this protocol to understand and characterize the performance of the data metrics specific to their instrument and platform. Finally, a graphical user interface (GUI) would be created to score the QC/SST efficiently and minimize the time added to an experimental workflow. Such a development is essential to the practicality of the method and to encourage use of QCs in MSI.

In future work, the sampling procedure will be optimized to improve the efficacy and ease of implementation of a QC/SST protocol into MSI workflows. Additionally, other statistical approaches will be investigated for data analysis. The QC/SST panel will also be expanded to incorporate more structurally diverse analytes, increase the m/z range, and make the mix more applicable to multiomics studies.

CONCLUSIONS

In this work, we developed a QC/SST mix of analytes that are applicable to the most common MSI platforms. We utilized this mix to construct and test an SST scoring method employing a scaled PCA model and established a cutoff score for accurate determination of the instrument condition. Details of the model developed and used for this protocol are described along with considerations that are necessary to implement this procedure within other labs of various platform designs. Ultimately, this work addresses an important gap in the field of MSI and proposes an amenable protocol that can be utilized to improve data quality and enhance the ability to compare data across time and between different platforms.