Considerations for MALDI-Based Quantitative Mass Spectrometry Imaging Studies

Abstract

Significant advances in mass spectrometry imaging (MSI) have pushed the boundaries in obtaining spatial information and quantification in biological samples. Quantitative MSI (qMSI) has typically been challenging to achieve because of matrix and tissue heterogeneity, inefficient analyte extraction, and ion suppression effects, but recent studies have demonstrated approaches to obtain highly robust methods and reproducible results. In this perspective, we share our insights into sample preparation, how the choice of matrix influences sensitivity, construction of calibration curves, signal normalization, and visualization of MSI data. We hope that by articulating these guidelines that qMSI can be routinely conducted while retaining the analytical merits of other mass spectrometry modalities.

Graphical Abstract

INTRODUCTION

Mass spectrometry imaging (MSI) provides a unique insight into the molecular distributions of a wide range of analytes for many scientific fields. It has emerged as a unique and powerful tool as an in situ, label-free imaging method, capable of mapping multiple molecular classes such as drugs,¹ lipids,² peptides,³ and proteins.⁴ As a technique, MSI has been driven at the forefront of research by matrix-assisted laser desorption/ionization (MALDI)-MS and has been the most widely used ionization technique for tissue imaging.⁵ This perspective will mostly focus on considerations regarding MALDI-based imaging. While desorption electrospray ionization (DESI) imaging is another mass spectrometry modality quickly gaining ground in absolute quantification, it will not be the primary focus of this article. Regardless of the ionization technique, data acquisition from a tissue sample is typically achieved by rastering the sample stage (or laser beam or solvent spray) relative to the sample in an x, y direction. The resulting data file is an array of mass spectra associated with a position in the sample, and subsequent data analysis can then be performed to generate heat maps of discrete mass-to-charge (m/z) values.

Improvements in bioinformatics and instrument design have increased the wealth of information that can be obtained in a single MSI analysis. While MSI experiments have typically provided the spatial distribution and relative abundance of analytes in a sample, they have also been used as a tool to obtain absolute quantification in recent years. The relative and absolute quantitative information on distinct molecules between different regions of the same tissue (e.g., normal versus tumor) or between different tissues (e.g., vehicle-treated versus drug-treated or control versus diseased) can give additional insight on the disease progression or pharmacological and toxicological action of a drug. Accurate quantification of target molecules between different regions of the tissue can give further insight into the degree of turnover or metabolism of target molecules. It can also accurately monitor the amount of therapeutic in a tissue, to better understand the pharmacokinetic-pharmacodynamic mechanisms of the molecule during development.

Over the past decade, several studies have explored methods to conduct quantitative MSI (qMSI) that are robust, reproducible, and accurate. These methods, which will be discussed more in detail below, seek to minimize several fundamental aspects of ion suppression effects, sample heterogeneity, and analyte recovery. Reviews by Ellis et al.⁶ and Rzagaliski et al.⁷ previously provided discussions on quantitative mass spectrometry imaging. Ellis et al. discussed ion suppression effects and different ionization sources used in MSI, while Rzagaliski et al. surveyed the quantitative determination of compounds on tissue surfaces by MALDI and compared different normalization strategies at that time. Buchberger et al.⁸ offers a broader review on the MSI technique as a whole, and a recent review by Spraker et al. discussed MSI in the scope of natural products discovery and the use of various ionization sources currently available.⁹ In this perspective, we will focus on and discuss various considerations in the experimental design and sample preparation, including methods for generating a calibration curve. We will expand on the discussion started by Ellis et al. and Rzagaliski et al. with current literature examples and also discuss aspects of the data analysis process to further streamline qMSI experiments and how each step would improve the analytical figures of merit as illustrated in Scheme 1.

Scheme 1. Considerations toward Conducting Quantitative Mass Spectrometry Imaging^a.

^aThe central yellow line is connected to other “metro lines” that represent a topic discussed in this perspective. The pink line represents experimental design and sources of variability, the purple line signifies the matrix selection process and application methods, the green line represents creating calibration curves for qMSI, the blue line symbolizes the image normalization aspect, while the red line represents the selection process for the proper color scheme of imaging data. The MS images depicted in this scheme are from the same murine heart section in a different color scheme and when the signal was normalized by different methods.

1. Establishing the Experimental Design: Minimizing the Error

Sample Procurement and Handling.

The experiment starts as soon as the sample has been procured for subsequent sample preparation. Sample procurement should be carefully planned to maintain the integrity of the shape of the tissue, as mishandling or improperly storing tissue samples can lead to premature molecular delocalization. For tissue samples such as the murine brain, it is essential to surgically remove them carefully and to snap freeze in liquid nitrogen or over dry ice before storing in a −80 °C environment. For storage, we recommend using 24-well cell culture plates for murine organs, as it is important to avoid the use of conical tubes whenever possible so as to not damage the spatial integrity of the tissue. For gelatin-embedded samples such as spheroids,¹⁰ or organoids,¹¹ the use of 12 or 24-well cell culture plates are typically used to form the gelatin blocks prior to storage and cryosectioning. For formalin-fixed paraffin embedded (FFPE) samples, Buchberger et al. has previously discussed sample handling and preparation in more detail.⁸

Replicates.

Conducting an qMSI study will most certainly require multiple measurements for the same biological or chemical condition (e.g., vehicle-treated versus drug-treated or control versus diseased), or when generating calibration curves for an analyte. Multiple tissue sections are typically mounted on the same ITO slide or MALDI target plate during cryosectioning to conduct different imaging runs in positive or negative ion modes. If possible, biological replicates, which are tissue sections originating from another biologically distinct source of the same condition, should be prepared and analyzed on the same slide or plate to minimize error from MALDI matrix deposition. It is worth considering the use of technical replicate sections, which are tissue sections originating from the same biologically distinct source, to construct calibration curves as they would exhibit a similar chemical microenvironment. However, this approach does not consider biological variation, and thus researchers might consider constructing calibration curves from multiple biological replicates as well. Finally, at a minimum, consider procuring three biological replicates for each condition in the study to verify that the changes is due to the condition and not by random variation.

Sources of Variability and How to Minimize Their Effects.

Addressing experimental variability has always been an important consideration in MALDI imaging experiments, which can reveal substantial variation in analytical sensitivity within the tissue or between tissues. Several studies have discussed potential pitfalls and solutions in conventional, dried-drop MALDI and imaging analyzes.^12,13 Shot-to-shot variability can occur in a single spot/pixel. Typically, a mass spectrum is generated from several laser shots (10s–1000s) on a sample and averaged together. Therefore, this type of variability can mainly be observed in instruments that have lasers of varying beam profiles for each pulse, which can provide inconsistent laser fluences. This variation can be mitigated by increasing the number of shots to average for every pixel. However, this modification will inherently increase the acquisition time for each imaging run.

Spot-to-spot variability is a commonly observed issue in MALDI. This can occur from several points: (1) when heterogeneous matrix crystals form because of inconsistent deposition, (2) or due to tissue inhomogeneity as a result of a diverse chemical microenvironment. The use of automated sprayers and sublimation chambers for matrix application has helped in providing uniform matrix deposition. Also, spectral normalization is often conducted as a preprocessing step to address this type of variability such as pixel-specific normalization, which will be discussed further below.

Another source of variability in MALDI imaging experiments occurs between sample-to-sample, which can be further separated between technical (serial tissue sections from the same biological source) or biological (tissue sections different biological source but from the same biological condition). As previously mentioned, there is a need for replication in imaging studies to provide reproducible research and strengthen biological conclusions. Frequent mass calibration, especially for time-of-flight instruments, which are subject to mass deviations over time can aid in minimizing sample-to-sample variability. If possible, the matrix application process should be performed at the same time for all samples to minimize the variability. Finally, borrowing from the -omics field of mass spectrometry, the use of a quality control (QC) sample can mitigate sample-to-sample variation as a way to monitor system stability. Condina et al. recently demonstrated the use of boiled egg white as a QC sample for a large scale FFPE MALDI-MSI study using a TOF/TOF instrument.¹⁴ The mean signal intensities from peptides arising from the egg white section were used to monitor the instrument performance throughout the experiment. Meanwhile, Barry et al. utilized a mimetic tissue model, spiked with an internal standard to monitor the performance of their multicenter study to conduct qMSI.¹⁵ Overall, the use of QC samples is certainly warranted in any MSI experiment to monitor system performance and the quality of the resulting data.

2. Sample Preparation: Improving the Basics

Pretreatment of Samples (Washing).

Proper experimental design, especially the sample preparation, is critical in qMSI studies to obtain the most representative and quantitative data. Biological tissue sections are composed of diverse microenvironments that contain the target analyte, endogenous salts, and other molecules. Sometimes, sample pretreatment may be necessary to simplify the biological matrix before further steps can be taken. Sample pretreatment such as washing the sample in a solvent or buffer can aid in suppressing more abundant molecules or to increase the ionization efficiency of the target analyte. Sample clean up processes have found use in mass spectrometry imaging analysis prior to matrix application without compromising the spatial localizations of molecules. For example, to image a protein and its peptides, an effective but involved sample preparation consists of a series of fixation wash steps before subjecting the tissue to an on-tissue proteolytic digestion.^16–19 Another type of sample pretreatment involves aqueous washing using ammonium formate to increase the ionization of lipids such as gangliosides.^20,21 However, the delocalization will affect some drugs, depending on their solubility, so it is always prudent to test with practice samples.

Choosing a Matrix to Maximize Analytical Sensitivity.

The choice of matrix can vary the range of molecular weights and molecular species that can be ionized by the mass spectrometer. Several matrices have found popularity for their widespread applicability including α-cyano-4-hydroxycinnamic acid (CHCA) and 2,5-dihydroxybenzoic acid (DHB) for peptides and metabolites. Recent studies have routinely used 1,5-diaminonaphthalene (DAN) or 9-aminoacridine (9-AA) to image lipids in mouse brain.^22,23 It is important to consider that these commonly used matrices can ionize target molecules in both positive and negative ion modes, but their analytical sensitivity will vary depending on the polarity. Their volatility is also of consideration especially for high resolution imaging, where it would typically take several hours to complete an analysis. Specifically, sublimation in vacuum conditions inside the mass spectrometer is a concern when using 2,5-dihydroxyacetophenone (2,5-DHA) for protein imaging. Recently, Yang et al. developed a new novel vacuum stable matrix based on 2,5-DHA, which has improved vacuum stability while maintaining the benefits of 2,5-DHA.²⁴ The analytical sensitivity for lipid imaging can also vary when employing different types of matrices. An extensive study on the matrix choice and the number of lipids annotated was recently reported by Perry et al.²⁵ They have shown increased ionization efficiency of lipids with DAN and 9-AA in negative ion mode compared to the other matrices in the study. They also reported that different adducts (sodium or potassium) in positive mode are more prevalent when using DHB, 2,5-DHA, or 5-chloro-2-mercaptobenzothiazole (CMBT). Testing the most appropriate matrix for samples can be easily accomplished as researchers often obtain serial sections; it is highly recommended preparing multiple sample sections for the primary purpose of matrix selection in which you apply different matrices and assess the signal of the target molecules. At a minimum, test out 2–3 matrices to be applied on test samples to determine the best instrumental signal response for the experimental workflow. A list of common MALDI matrices and selected manuscripts using each matrix is shown in Table 1.

Table 1.

Popular MALDI Matrices^a

matrix (abbreviation)	target molecule
α-cyano-4-hydroxycinnamic acid (CHCA)	proteins,19 peptides,26 N-glycans,27 lipids25,28
2,5-dihydroxybenzoic acid (DHB)	peptides,16,29 drugs,30,31 neuropeptides,3 lipids25,32
sinapinic Acid (SA)	proteins33
1,5-diaminonaphthalene (DAN)	lipids,23,34,35 metabolites36
9-aminoacridine (9-AA)	lipids25,37
2,5-dihydroxyacetophenone (2,5-DHA)	proteins35,38
norharmane	bile acids,39 lipids40
Nl,N4-dibenzylidenebenzene-1,4-diamine (DBDA)	fatty acids41

^aSelected manuscripts are also listed with the target molecule analyzed by the authors.

Reactive Matrices.

Recently, matrices designed to enhance the analytical sensitivity of specific analytes have received a lot of attention to elucidate their spatial distributions. Selective in situ chemical derivatization makes it possible to conduct a targeted mass spectrometry imaging for these specific molecules that would not normally be detected by conventional MSI workflows. They enhance the detection of target molecules that are either are low-abundant or contain certain chemical moieties such amines or double-bonds in fatty acyls of lipids, while also serving as the chemical matrix for laser desorption/ionization.^42,43 For example, neurotransmitters in the nervous system are important chemical messengers between neurons and have been the focus of MSI studies. Shariatgorji et al. presented an MSI approach for mapping neurotransmitters that contained phenolic hydroxyl and/or primary or secondary amine groups by using a reactive matrix called FMP-10.⁴³ They were successfully visualized major neurotransmitters of the dopaminergic and serotonergic systems as derivatized molecules in rat and primate models of Parkinson’s disease and a Parkinsonian human brain tissue section. As compared to using CHCA or DHB, they found that FMP-10 gave higher detection limits of these neurotransmitters such as dopamine and γ-aminobutyric acid (GABA) when derivatized.

Reactive matrices have also proven useful in the mapping of lipids. Mass spectrometry has greatly expanded the ability to identify lipids and the study of lipid structures by ion fragmentation techniques such as collision-induced fragmentation (CID). In addition, other instrument advances such as ultraviolet photodissociation, ozonolysis and UV light-induced Paternò-Büchi (PB) reactions allows for the assignment of double bond positions. A recent report detailed the use of reactive matrices in the determination of double bond positions in fatty acyl chains of lipids. Wäldchen et al. reported on the use of benzophenone as a reactive matrix, which functionalized olefins via a PB reaction upon irradiation of the ultraviolet laser in the MALDI instrument.⁴² The structural determination of the double bond was then determined by MS1 accurate mass comparisons and by CID fragmentation that yielded ions 182.07 Da higher in mass than the unfunctionalized counterparts and yielded fragmentation spectra assigned to the retro-PB reaction, respectively.

There is great promise in the use of reactive matrices as they can provide added insight into the fundamental metabolic processes in many biological systems. However, more work is warranted to determine the robustness and reproducibility of reactive matrices for qMSI studies. In the previously mentioned study, Shariatgorji et al. also revealed a linear response between the amount of compound present on a MALDI plate and the signal intensity of the derivatized counterpart, as shown from their calibration standard curves for 3-methoxytyramine, homovanillic acid and dopamine. Other potential uses for reactive matrices could be their application in a time-course analysis during derivatization process to monitor the appearance of new compounds over the course of the reaction. This approach can potentially narrow down the list of m/z peaks for candidate for reaction products from the derivatization process. Finally, as several new derivatized compounds are produced during these on-tissue reactions, there will be a need to automatically assign the m/z peaks as the spectral complexity would certainly increase.

Matrix Application: Sieving, Manual Spraying, Sublimation, Robotic Sprayer.

Matrix application needs to be uniform, produce small crystal sizes, and appropriately extract analytes without introducing artifacts such as spatial delocalization. A low cost and robust method to apply matrix uses stainless steel sieves. The use of sieves is popular in the imaging of agar-based microbial colonies and has been previously described in more detail by Yang et al.⁴⁴ Manual spraying with airbrushes was the initial approach for applying matrix on tissue samples as it is economical and easy to implement. Meanwhile, using sublimation devices or robotic sprayers have become standard practice to obtain the most uniform matrix application. The first sublimation device for matrix deposition was reported by Hankin et al.⁴⁵ Most homemade sublimation systems consist of a flat-bottom condenser to which indium–tin-oxide (ITO)-coated slides or small MALDI target plates can be attached by a double-side copper tape and affixed on top of a bed of matrix crystals. In comparison, homemade and commercial robotic sprayers mainly utilize a nozzle that is affixed on top of a stage where ITO slides or target plates can be placed. The nozzle, which is coupled to a HPLC pump, rasters along the stage while spraying the matrix solution. Additionally, Gemperline et al. observed by optical microscopy that airbrushes produce larger matrix crystals when compared to sublimation and a robotic sprayer.⁴⁶ This study compared the reproducibility and analyte detection between the three different matrix application methods. They found that both sublimation and robotic sprayers can achieve uniform applications compared to the airbrush, which also showed increased analyte diffusion. Finally, given the robust functionality of matrix sprayers, robotic sprayers can also be used to apply internal standards on samples to conduct absolute quantification studies.^11,31,47 Therefore, it is strongly suggested to use robotic sprayers or sublimation devices to streamline the matrix application process and minimize irregularities between users and the periods of application. Robotic sprayers can also be utilized to uniformly apply internal standards, which is a critical element in absolute quantification.

3. Absolute Quantification Using a Calibration Curve

Obtaining the Analytical Figures of Merit in qMSI.

Considerations to conduct absolute quantification in MSI experiments can be accomplished using the same guidelines for conventional mass spectrometry techniques such as liquid chromatography–mass spectrometry (LC-MS) and gas chromatography–mass spectrometry (GC-MS). Zabell et al. previously proposed several guidelines for constructing calibration curves of analytes by LC-MS.⁴⁸ They suggested that calibration curves should contain a minimum of seven calibrants, including a blank solution that only contains the suspension solvent and to evenly distribute the calibration points in the linear working range of the curve. They claim this arrangement will maintain a 95% confidence of the curve’s accuracy, even if one calibration point is excluded from the curve. The signal response for each calibrant is typically generated from an averaged intensity of the standard over an area or region-of-interest (ROI). The monoisotopic mass intensities of the [M + H]⁺, [M + Na]⁺, or [M + K]⁺ ion species of the standard can be plotted with respect to the amount deposited on the tissue (or surface), which should be expressed in units of moles or moles per unit area. Accounting for the size of the ROI is crucial as it contains n number of mass spectra or pixels that can be averaged together. Porta et al. suggested to average at least 4–5 pixels to generate an averaged mass spectrum to compensate for instrumental variability and potential pixel-to-pixel matrix heterogeneity.⁴⁹ They determined that four pixels is the minimum amount required to achieve a relative standard deviation below 15% in signal response. Signal from user-defined ROIs can be extracted in vendor-neutral MSI software packages such as BioMap, MSiReader, msIQuant, SCiLS Lab, or LipostarMSI for each calibration standard as well as within the tissue samples or a specific microenvironment.^50–52

Important analytical merits such as the limit of the blank (LOB), limits of detection (LOD), and the limit of quantification (LOQ) can then be determined for any qMSI assay, especially for clinical and pharmaceutical studies that require stricter guidelines for reporting quantified analytes. These key figures of merit can also determine what impact each iterative modification will make during method development. Since each calibration spot in the calibration curve will contain multiple laser ablation spots that correspond to a single mass spectrum, we can obtain the mean analyte signal and the standard deviation of the mean to calculate the following figures of merit. Regulatory authorities such as the United States Food and Drug Administration (FDA), United States Pharmacopoeia (USP), International Union of Pure and Applied Chemistry (IUPAC), and Association of Analytica Communities (AOAC) have different guidelines for each analytical figures of merit with some not having a specified definition and procedure. Barry et al. utilized the following equations to calculate their analytical figures of merit, which were recommended by Clinical and Laboratory Standards Institute (CLSI).^53,54 The LOB is described as the highest apparent analyte concentration that can be distinguished from a blank sample.⁵⁵ Therefore, the LOB can be calculated from the average signal of the analyte and the standard deviation of the signal using the following equation:

$${\text{LOB=mean}}_{\text{blank}}\text{+1}.645\left({\text{standard deviation}}_{\text{blank}}\right)$$

The LOD is lowest analyte concentration that can be confidently distinguished from the LOB and is greater in value than the LOB. The LOD can be calculated considering the LOB and the standard deviation of the signal of the lowest concentration in the calibration curve using the following equation:

$$\text{LOD=LOB+1}.645\left({\text{standard deviation}}_{\text{lowest concentration}}\right)$$

Taken together, reporting these values would provide a suitable metric on how well the experimental set up can quantify the analyte and provide a basis for future validation studies (Figure 1).

Figure 1.

A graphical representation of the limit of detection (LOD) and the limit of quantitation (LOQ), relative to the linear range of a calibration curve.

Critical to the accuracy and validity of any quantification assay is preparing the calibration curve standards in the same biological matrix as the samples being analyzed. Three experimental designs have emerged in recent years to conduct qMSI studies: (1) an in-solution method, (2) an on-tissue method, and (3) an in-tissue method, which are pictorially depicted in Figure 2.

Figure 2.

Three common approaches to generate calibration curves. (A) The in-solution method is conducted by spotting calibration standards directly on the target plate or ITO slide. (B) The on-tissue method utilizes a control sample which is sectioned and mounted next to the sample section. The control section is spotted with calibration standards prior to the imaging analysis. (C) The in-tissue method utilizes a tissue mimetic model which has been serially molded and frozen to contain different concentrations of the calibration standard. This tissue mimetic model is then sectioned and mounted next to the sample section. (D) A summary table of the characteristics of each method, in which the star indicates the performance (low = 1-star, high = 5-star). Adapted with permission from Porta et al.⁴⁹ Copyright 2015, Springer.

Creating the Calibration Curve.

The in-solution method relies on a set of standard solutions spotted on the MALDI target plate, preferably near the tissue sample. Ideally, we recommend analyzing the calibration points and the sample tissue at the same time to minimize variability. This relatively simple method has been used extensively in the literature to quantify various analytes.^56,57 However, utilizing this method will fail to consider the biological matrix effects when quantifying the analyte of interest in the sample tissue, which may impact desorption of the analyte.

The second strategy to generate calibration curves is by spotting calibration standards on a reference tissue, which is usually a control or untreated tissue section. Ideally, the analytes used to make the calibration standards should not be endogenously present on the sample tissue being analyzed. For example, quantifying endogenous analytes such as lipids or metabolites would be accomplished by using the stable isotope-labeled (SIL) analogue of the analyte (or a close chemical analogue) as calibration standards and spotting them on the control tissue section. By using a SIL compound of the target molecule, it retains a similar chemical structure and can mimic the ionization efficiency while having a shifted m/z ratio to enable differentiation by MS. Meanwhile, for exogenous analytes such as pharmaceutical compounds, the unlabeled variant can be used to generate the calibration curve and spotted on a control or untreated tissue section. It is worth noting that during this part of the sample preparation, an internal standard solution (typically another SIL analogue) can also be applied to all samples to correct for signal variability and normalization which will be further discussed below. The on-tissue method attempts to correct any matrix ionization effects, ion suppression or analyte extraction during data acquisition that could hinder accurate quantification. For example, Buck et al. created a calibration curve for irinotecan (CPT-11) and its drug metabolite SN-38 on an untreated murine liver tissue section. The calibration curve generated was then used to interpolate drug concentrations from mouse tissues which were previously dosed with CPT-11.⁵⁸ Another study by Chumbley et al. explored various routes to quantify rifampicin, an antibiotic for tuberculosis. They tested four methods of applying an isotopically labeled internal standard and the MALDI matrix on rat liver tissue using an acoustic robotic spotter and compared the qMSI measurements by LC-MS of the tissue extracts.⁴⁷ The MALDI matrix and internal standard were applied as 250 μm microspots onto tissue sections where they were able to quantify the amount of drug in different regions of the tissue with a relatively high agreement of 90.4% similarity with their LC-MS validation run. As an adaptation to the on-tissue method, Pirman et al. demonstrated an under-tissue method in which calibration standards of cocaine were pipetted on a slide and used an inkjet printer to apply a deuterated cocaine analogue as an internal standard before thaw-mounting human nucleus accumbens on top of the calibration spots.⁵⁹ However, it is unclear if on-tissue/under-tissue methods truly mimic the analyte extraction upon laser desorption. The deposition of an internal standard can disrupt the native microenvironment, which can increase the signal variability between each pixel and uncertainties in the degree of penetration of the standard solution can vary. In addition, the use of isotopically labeled standards are the gold standard to quantify endogenous analytes, but they can be expensive or lacking in availability. Therefore, further validation and assessment for ion suppression is warranted if one takes this route to conduct a qMSI study. Nonetheless, the on-tissue method provides a relatively simple approach for generating a calibration curve where a series of standards are spotted on a control tissue section and analyzed with the same instrumental parameters as the sample tissue.

The third method used for qMSI studies is using a mimetic tissue model (in-tissue) as first described by Castellino and others.^30,60,61 The most recent method of preparing the mimetic tissue model is using tissue homogenates that have been serially frozen into a cylindrical mold and contain serial concentrations of spiked internal standard(s).⁵⁴ This method aims to incorporate the internal standard within the tissue mimetic and should resemble the matrix effects and analyte extraction during a MALDI imaging experiment. However, creating the tissue model requires an involved sample preparation and a large amount of tissue material. Barry et al. recently compared three case studies involving the use of the tissue mimetic model by quantifying three different ion distributions and validating with an LC-MS run of the sample homogenates. They suggest that the ion distribution should be the main driver of how LC-MS validation should be conducted as whole tissue homogenate samples can provide misleading numbers, especially if the target molecule is spatially localized in a small region of the tissue. A key caveat for using tissue mimetic models is that the tissue homogenate will only be a matrix-match model for that type of tissue or microenvironment. Therefore, qMSI experiments involving multiple types of tissue might require multiple tissue mimetic models to truly match each region. Hansen and Janfelt investigated differences in ion suppression between homogenates from rabbit tissue relative to a spiked internal standard. They found that even if an internal standard was used, it did not fully correct for all the differences between the tissue homogenates.⁶²

Overall, a single method cannot be fully established as the gold standard for generating calibration curves for MSI. The method choice will be a case-to-case basis and the sample amount (by area and thickness), and overall sample complexity needs to be considered. Also, a series of trial and error experiments is suggested to obtain the most appropriate concentration dynamic range and for each calibration point be equidistant from one another. If quantitative validation by LC-MS is needed, laser capture microdissection (LCM) might be necessary to validate the amount, especially for spatially localized analytes. Although LC-MS validation might not be needed; as a recent multicenter validation study indicated that qMSI can achieve quantitation with relative standard deviation in the low teens and accuracies around 80% between multiple analysts.¹⁵ The study compared the reproducibility and accuracy among three analysts from three different sites, which followed a common protocol to conduct the sample preparation and mass spectrometry analysis. Each analyst quantified the amount of clozapine that was perfused in rat liver using the on-tissue and in-tissue method as well as testing the impact of SIL normalization on the precision and accuracy of both methods when compared to quantification by liquid-chromatography-MS (LC-MS). They found that each analyst can provide a high degree of accuracy when compared to the LC-MS analysis. When taken together as a pooled result, the accuracy using the on-tissue and in-tissue method was 71% and 78%, respectively. When they further normalized using the SIL, the accuracy further improved to 76% and 84%, respectively. Furthermore, the relative standard deviation (RSD) of the pooled result from three analysts was at 29% at 14% for the on-tissue and in-tissue methods, respectively. While after SIL normalization, these results further improved to 15% and 13%, respectively, where the in-tissue method greatly benefitted to the SIL normalization.

Finally, it is worth noting the expectations for the analytical figures of merit in an qMSI study. Table 2 highlights some previously published studies that utilized MSI to quantify molecules using calibrations curves. The ionization efficiency will differ across molecules and will be heavily influenced by the chemical environment. Relative standard deviations in signal between pixels should be at least 20% or lower to expect appropriate quantification by qMSI, as previously demonstrated by Barry et al.¹⁵

Table 2.

qMSI Studies and the Target Molecules Quantified Using Calibration Curves^a

molecule	calibration curve type	reported concentration range	instrument	reported LOD/LOQ/LOB	reference, (year)
erlotinib	in-solution	0.2–2 pmol	Q-TOF	not reported	Signor et al.,56 2007
propranolol	in-solution	0.02–10 μM	TOF	LOD: 0.006 pmol/mm2 LOQ: 0.018 pmol/mm2	Hamm et al.,63 2012
olanzapine	in-solution	1–60 μM	TOF	LOD: 0.3 pmol/mm2 LOQ: 0.9 pmol/mm2	Hamm et al.,63 2012
epertinib	in-solution	0.01–0.71 pmol/mm2	LTQ XL	LOD: 0.02 pmol/mm2 LOQ: 0.05 pmol/mm2	Tanaka et al.,64 2018
irinotecan	on-tissue	0.4–40 pmol	7T FT-ICR	not reported	Buck et al.,58 2015
paclitaxel	on-tissue	0.5–10 pmol	TOF/TOF	not reported	Giordano et al.,65 2016
rifampicin	on-tissue	1.0–10 μM	LTQ XL	not reported	Chumbley et al.,47 2016
rifampicin	on-tissue	3.00–100 μM	TOF/TOF	LOQ: 3.00 μM	Prentice et al.,66 2017
lactate	on-tissue	10–500 ng	TOF/TOF	not reported	Swales et al.,67 2018
glutamate	on-tissue	0.025–2 mM	TOF/TOF	not reported	Swales et al.,67 2018
imatinib	on-tissue	0.78–25 pmol	TOF/TOF	LOD: 0.73 pmol/section LOQ: 1.82 pmol/section	Abu Sammour et al.,68 2019
irinotecan	on-tissue	0.05–50 μM	15T FT-ICR	not reported	Tobias et al.,69 2019
dabrafenib	in-tissue	750–37 500 ng/g	7T FT-ICR	not reported	Groseclose et al.,70 2015
amitriptyline	in-tissue	0.5–20 μg/g	LTQ XL	not reported	Hansen et al.,62 2016
clozapine	in-tissue	0.5–50 μg/g	7T FT-ICR	LOD: 1.05 μg/g (liver), 1.30 μg/g (kidney), 0.97 μg/g (brain) LOB: 0.16 μg/g (liver), 0.47 μg/g (kidney), 0.48 μg/g (brain)	Barry et al.,54 2019

^aEach study has their reported concentration range as in their original units, the mass spectrometer used, and any analytical figures of merit reported. The selected publications are listed in the order of calibration curve typed used.

4. Normalizing MS Images for Visualization

Normalizing the MS signal in MSI studies is a common step to appropriately visualize the target analyte and achieve relative quantification within the tissue or between samples. This step is often necessary to minimize the effects of artifacts such as biological heterogeneity of the tissue surface (interfering molecules and different salt contents) and nonuniform distribution of matrix crystals. It is normally achieved at the software-level during initial data analysis. Current software platforms and packages have several pixel-specific normalization options such as total ion count (TIC), root-meansquare (RMS), median normalization or normalization to a specific m/z signal. The latter is particularly useful if the intensity of an internal standard, matrix peak, or an endogenous analyte is used as the normalization factor. TIC and RMS normalization applies a global-wide scaling for all the mass spectra corresponding to each pixel in the MSI experiment with the assumption that all mass spectra do not contain peaks that facilitate ion suppression. TIC have been used successfully to conduct relative quantification between control and diseased tissue sections, time-course drug-treated spheroids, and have been used to correct the signal for absolute quantification of endogenous molecules in rodent brain.^23,71,72 However, as mentioned before, the addition of an internal standard and other pixel-specific normalization methods may not completely normalize the signal between different tissues and microenvironments as there could be additional factors such as protein binding of target molecules or overall different cell and tissue densities.⁶² To minimize misleading interpretation, the use of TIC and RMS normalization should be more appropriately used if the sampled area is a relatively similar chemical environment or to omit the lower m/z ranges where matrix peaks are present and which can contribute to incorrect normalization.⁷ Normalizing to an internal standard would be more appropriate if the sampled area is more heterogeneous and contains multiple different microenvironments such as mouse brain or whole-body animal sections.

A tool for evaluating the amount of ion suppression (or enhancement) effects is to determine the tissue extinction coefficient (TEC) of a region in the sample tissue. It can be calculated by dividing the mean intensity of an analyte in that ROI by the mean intensity of the same analyte on a reference area or reference tissue. This TEC value can give a sense of how much attenuation (or amplification) is occurring in that region relative to a reference. This practice was first proposed by Hamm et al. as a way to determine the amount of ion suppression occurring in a region when compared to a reference region such as the target plate or a tissue homogenate.⁶³ More recently, Taylor et al. expanded on the use of the TEC by using a spatial segmentation algorithm to determine the ROIs in brain tissue and calculating the TECs for each ROI which corresponded to specific anatomical brain regions.⁷³ They also went further to demonstrate that by creating spatial segments of the tissues, a TEC-normalized image can be produced of their drug target. In summary, there are multiple routes to normalize MS imaging data to quantitatively visualize the target molecules. Therefore, it is imperative to understand the limitations of each normalization method as to not mislead the reader on the quantity in the sample.

5. Visualization Using Proper Color Schemes

One of the goals of every MSI experiment is to elucidate the spatial distribution and relative concentration of the molecules of interest based on the intensity of the color in an image. Therefore, the use of a proper color scheme to illustrate MSI data is vital. As suggested by Kovesi and others, color schemes for scalar data such as MSI data should have the following in their design principles: (1) a linear response in brightness and (2) colors in the scheme should be perceptually uniform, or change at a constant rate.^74,75 The use of a rainbow or heat color scheme has been popular for their attractive vibrancy but the reversals in the lightness gradients can be detrimental to the viewer’s perception of the ordering of the colors and can mislead the understanding of the data.^76,77 Greyscale and the use of red, green, and blue (RGB) color schemes can satisfy the linear change in lightness. However, if these color schemes are used to create an overlapping color map, such as when showing colocalizing m/z images in the same figure, each color scheme will have a different apparent contrast and brightness between one another which can mislead the viewer of their relative quantities. This issue was previously illustrated by Race et al. which further discussed optimization of color schemes for MS imaging.⁷⁸

Color schemes such as viridis and parula have found increased use for satisfying the two design principles listed above. Color vision deficiency (CVD) or “color blindness” reduces the ability to distinguish certain colors and affects around 8% of men and 0.5% of women worldwide.^79,80 Recently, Nuñez et al. have optimized the viridis color scheme for those with CVD in mind into the cividis color scheme, which still satisfies the two design principles. Mass spectrometry data analysis software for MSI such as MSiReader, Cardinal, SCiLS Lab, and LipostarMSI have such color schemes in place when generating MS images.^50,52,81

CONCLUSION

In conclusion, we have described several considerations based on previous studies and personal experiences for conducting quantitative mass spectrometry imaging for various research areas. Establishing an appropriate experimental design to determine the size of a study and methods to minimize experimental variability should be at the forefront of considerations. Optimizing the appropriate MALDI matrix for the sample and target analytes is critical to increasing the analytical sensitivity. The use of robotic sprayers is valuable to achieve highly reproducible matrix application, which will minimize the effects from heterogeneous coating and signal suppression. The appropriate considerations should also be taken when generating calibration curves so the analytical figures of merit can be obtained and appropriately reported. The mode of normalizing MS images should consider the chemical heterogeneity of the sampled space when choosing the appropriate technique. The use of perceptually uniform and linear responsive color schemes will provide MS images a more quantitative visualization and should be routinely used for publication. Finally, previous discussions in various MSI workshops and discussion groups have called for the harmonization of reporting MSI methods and analyzes. The need for standardization and general guidelines has already been suggested in several reports in the literature.^82,83 More importantly, greater openness in data availability is warranted in the form of data repositories, where raw data, identified results and the experiment metadata can be kept. Public dissemination of imaging data should create increased transparency by creating a centralized hub to share data with reviewers and to share findings with the scientific community. The metabolomics and proteomics fields have already started this duty of data dissemination for some time with PRIDE, MassIVE, Panorama, and Metabolights. While it is difficult to fully replicate a previously published study due to the differences in sample preparation equipment and mass spectrometry instrumentations, we hope these considerations will pave the way to a more uniform data reporting and for qMSI to become an even more established quantitate methodology.