Chemical QuantArray: A Quantitative Tool for Mass Spectrometry Imaging

Abstract

Matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) is a powerful analytical technique that provides spatially preserved detection and quantification of analytes in tissue specimens. However, clinical translation still requires improved throughput, precision, and accuracy. To accomplish this, we created “Chemical QuantArray”, a gelatin tissue microarray (TMA) mold filled with serial dilutions of isotopically labeled endogenous metabolite standards. The mold is then cryo-sectioned onto a tissue homogenate to produce calibration curves. To improve precision and accuracy, we automatically remove pixels outside of each TMA well and investigated several intensity normalizations, including the utilization of a second stable isotope internal standard (IS). Chemical QuantArray enables the quantification of several endogenous metabolites over a wide dynamic range and significantly improve over current approaches. The technique reduces the space needed on the MALDI slides for calibration standards by approximately 80%. Furthermore, removal of empty pixels and normalization to an internal standard or matrix peak provided precision (<20% RSD) and accuracy (<20% DEV). Finally, we demonstrate the applicability of Chemical QuantArray by quantifying multiple purine metabolites in 14 clinical tumor specimens using a single MALDI slide. Chemical QuantArray improves the analytical characteristics and practical feasibility of MALDI-MSI metabolite quantification in clinical and translational applications.

Graphical Abstract

INTRODUCTION

Altered cellular metabolism is a hallmark of many different cancers and has proven to be a promising source of therapeutic targets.¹ Exploiting these vulnerabilities, however, requires accurate metabolic characterization in heterogeneous specimens, which are often composed of neoplastic, stromal, and immune cells in a complex tumor microenvironment. Although liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) remains the gold standard for many analytical applications, most methods involve liquid matrices, such as serum, plasma, or urine. Tissue-based quantification of small molecules using LC-MS/MS requires extensive sample preparation, making it challenging to implement in clinical laboratories. Furthermore, since these approaches involve homogenization or other tissue-destructive processes, knowledge of spatial variability is lost, an important characteristic of endogenous metabolites in heterogeneous tumors.^2,3

Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) provides a powerful platform to map the spatial distribution of large biopolymers, proteins, peptides, lipids, small molecules, and drugs directly in tissue sections.^4–7 Accurate and reliable quantification using MALDI-MSI, however, has faced several technical challenges.^8–10 One challenge, common to most MS-based techniques, is variable ionization efficiency and ion suppression due to matrix effects.¹¹ This has been addressed in LC-MS/MS methods by creating calibration curves spiked into comparable biological matrices, as well as through normalization to spiked stable isotope internal standards (IS). However, matrix effect challenges are considerably more challenging when mapping metabolites in tissue specimens due to the wide pixel-to-pixel variability in tissue composition and the lack of chromatographic separation.

Several approaches to account for matrix effects and to allow for better accuracy and precision have been applied to MALDI-MSI. To generate calibration curves, the use of tissue mimetics for the absolute quantitation for MALDI-MSI has been presented by several groups.^12–17 The first mimetic approach presented by Groseclose and Castellino consisted of spiking a range of different drug concentrations into a set of tissue homogenates that were preweighed into microcentrifuge tubes and then transferred into a home-built mold.¹⁸ This was further refined by creating cylindrical molds consisting of layers of serially frozen spiked-tissue homogenates in increasing concentrations to create calibration curves.¹⁹ Despite these improvements, both methods are still arguably cumbersome to generate and require considerable space on a relatively small MALDI slide, leaving minimal room for experimental samples. In addition to the use of calibration curves, stable isotope internal standards have also been applied, though most often to analyze drugs in tissue and less commonly for endogenous metabolites. In past studies, we have used this approach in several drug quantification studies to map the ion distribution through the blood–brain barrier and assess the concentration of drug reaching the tumor target.^13,20–22

Herein, we present a further developed approach named Chemical QuantArray to use tissue microarray (TMA) molds filled with different analytes and normalize measurements to isotopically labeled internal standards to improve the precision, accuracy, and quantitative strength of MALDI-MSI experiments. This workflow is more efficient than more time-consuming mimetic techniques, uses fewer materials, and has better reproducibility. Moreover, by using different stable isotope internal standards, one for normalization and another for calibration curves, we create a more robust platform to quantify endogenous metabolites that are naturally present in all or most tissues.

EXPERIMENTAL SECTION

Chemical and Materials.

1,5-Diaminonaphthalene (DAN), L-glutamic acid-¹⁵N₅, D-glutamate-d₅, adenosine-¹⁵N₅ 5′-triphosphate (ATP) disodium salt solution, adenosine-¹⁵N₅ 5′-diphosphate disodium salt (ADP), adenosine-¹³C₁₀¹⁵N₅ 5′-monophosphate (AMP) disodium salt, and adenosine-¹³C₁₀¹⁵N₅ were purchased from Sigma-Aldrich (St. Louis, MO). HPLC-grade methanol, ethanol, trifluoroacetic acid, hydrochloric acid, water, and Rat Tail Collagen Type I, Rate Tail (A10483-01, Gibco, Grand Island, NY) were purchased from Fisher Chemical (Pittsburgh, PA). Indium tin oxide (ITO)-coated IntelliSlides glass sides were obtained from Bruker Daltonics (Billerica, MA).

Chemical QuantArray Preparation.

A 40/60 (w/v) gelatin solution was made, autoclaved, and poured onto a tissue microarray mold (TMA) containing 120 wells, each with a 1.5 mm core and a 10 μL depth. A 50 mM concentration stock solution was prepared from isotopically ¹⁵N labeled glutamate in water. A dilution series was made in collagen to obtain final concentrations ranging from 0.005 to 20 mM. Similarly, isotopically labeled ATP, ADP, AMP, and adenosine were diluted from a 100 mM stock solution to obtain final concentrations ranging from 0.005 to 50 mM solutions. 10 μL of each dilution mixture was dispensed into the TMA wells at −20 °C in a Microm HM550 cryostat (Thermo Scientific). The final mimetic was stored at −80 °C until MALDI tissue preparation.

Tissue Homogenates.

Approximately 10 g of control human or mouse brain was placed in a 15 mL cryovial. Tissue homogenization was performed at room temperature using a Tissue Master 125 homogenizer for 60 s. Once the tissue was completely homogenized, 20% (w/v) water was added to become fluent enough to transfer the peptide using a syringe. The addition of water minimized the holes in the tissue that are generated while sectioning. The tissue homogenate was blended for an additional 60 s and then transferred to a rectangular plastic mold (25 mm × 25 mm × 10 mm), frozen, and stored at −80 °C.

LC-MS/MS.

The brain homogenate samples were extracted by a previously published method using acetonitrile followed by sonication.^{23 15}N glutamate was quantified by a Sciex (Framingham, MA) 5500 ESI (electrospray ionization) triple-quadruple mass spectrometer coupled with a Waters Acquity ultrapressure liquid chromatograph (UPLC, Waters Corp., Milford, MA) as previously described.²⁴ The instrument was operated in negative-ion mode, and the analyte of interest was detected using an Acquity HSS T3 1.8 μM, 2.1 mm × 150 mm UPLC column (Milford, MA). Solvent A (water with 0.1% FA) and solvent B (acetonitrile with 0.1% FA) were combined in a 5-min-long gradient: 0–0.45 min, 10% B; 3.00–3.75 min, 95% B; 3.8–5.0 min, 10% B. Analyst software 1.7.3 (Framingham, MA) was used to control the instrument and data analysis. Extracted samples were diluted accordingly, and glutamate-d₅ was used as an internal standard before the LC-MS/MS analysis. All of the samples were injected in duplicate and quantified against a six-point standard curve with a 200-fold dynamic range (0.01 μm to 2 μM). The ESI was operated in multiple reaction monitoring (MRM) mode with the following MRM transitions as a precursor to production ions: ¹⁵N glutamate, m/z 146.9 → 103.0; glutamate-d₅, m/z 150.9 → 107.0 along with a declustering potential (DP) of −10 V, collision energy (CE) −18 V, curtain gas (CUR) 20, ion gas source 40, ion spray voltage −4500, and temperature 500 °C.

Clinical Tissue Samples Characterization.

Human tissue was obtained in compliance with all relevant ethical regulations and was reviewed and approved by the Institutional Review Boards (IRB) at Brigham and Women’s Hospital (BWH). The principal investigator is responsible for ensuring that this project was conducted in compliance with all applicable federal, state, and local laws and regulations, institutional policies, and requirements of the IRB.

Frozen and formalin-fixed paraffin-embedded (FFPE) tissue specimens were retrieved from the archives of Brigham and Women’s Hospital (BWH) under excess discarded tissue protocol 2018P001627 (reviewed and managed by the Mass General Brigham Institutional Review Board), which waives the requirement for patient consent. Meningioma specimens subjected to genomic analysis were used after written informed patient consent had been obtained under Dana-Farber Cancer Institute IRB protocol #10-417.

All tissue specimens were classified according to the revised W.H.O. 2016 Classification of Tumours of the Central Nervous System. All cases were reviewed, and the diagnoses were confirmed by two board-certified pathologists (S.M.C., S.S.). The cases analyzed included six W.H.O. grade I meningiomas, including 4 tumors with no special histologic subtype, one angiomatous/microcystic meningioma, and one case with some atypical features, and six W.H.O. grade II atypical meningiomas. The cohort additionally included one metastatic melanoma and one solitary fibrous tumor (SFT). All specimens were derived from independent patients (Table S1).

Immunohistochemistry.

FFPE sections were deparaffinized and dehydrated, and endogenous peroxidase activity was blocked. Antigen retrieval was performed in Dako citrate buffer 123 °C ± 2, 45 s, at 15 ± 2 PSI. Slides were incubated with anti-CD73 antibody, (Abcam, EPR6115, rabbit monoclonal clonal) at 1:5000 for 45 min, washed, and then incubated with Labeled Polymer-HRP antirabbit secondary antibody (Dako-Cytomation, K4011) for 30 min. Slides were then incubated with DAKO DAB+ solution for 3–5 min and counterstained with hematoxylin. We considered the membranous expression of CD73 to represent the physiologically relevant (active) expression of the protein.

Immunohistochemical expression of CD73 was evaluated using a semiquantitative scoring system.²⁵ In brief, cases were scored according to the following system: Absent (0): No expression of CD73 by tumor cells, or weak membranous staining in <10% of tumor cells; Weak (1): Weak membranous staining in >10% of tumor cells; Moderate (2): Moderate membranous staining in >10% of tumor cells, or strong/circumferential membranous staining in <10% of tumor cells; Strong (3): Strong/circumferential membranous staining in >10% of tumor cells. The overall percentage of tumor cells expressing CD73 in each specimen was also assessed (Table 1).

Table 1.

Clinical Tissue Sample Characterization and Tissue Quantitation of Purines Using MALDI-MSI

sample no.	diagnosis	W.H.O. grade	CD73 [index]	estimated % of tumor with membranous CD73 expression	AMP (μM)	ADP (μM)	ATP (μM)	adenosine (μM)
7	meningioma	1	3	95	163.9	359.3	109.5	229.8
2	meningioma	1	3	80	666.6	261.4	17.0	885.2
3	meningioma	1	3	95	606.2	908.4	169.5	622.5
11	meningioma	1	3	95	167.1	553.7	273.5	122.1
9	meningioma, with angiomatous/microcystic features	1	3	95	330.0	315.4	63.6	96.3
8	meningioma, with some atypical features	1	3	80	63.1	261.4	185.0	112.0
4	atypical meningioma	2	2	20	512.3	584.0	88.4	481.9
6	atypical meningioma	2	3	95	318.7	711.0	263.5	199.7
5	atypical meningioma	2	2	10	293.5	860.7	380.8	222.2
14	atypical meningioma	2	3	80	361.8	905.6	391.0	82.5
13	atypical meningioma	2	3	70	372.8	605.0	193.5	134.2
1	atypical meningioma	2	3	95	295.8	1017.4	443.9	172.1
10	solitary fibrous tumor	-	3	50	180.3	414.7	107.0	60.6
12	metastatic melanoma	-	2	10	128.6	147.4	23.2	1671.5

Tissue Preparation for MALDI-MSI and Microscopy.

Prior to sectioning, frozen tissue samples were placed at −20 °C in a Microm HM550 cryostat (Thermo Scientific) for 15 min to allow thermal equilibration. Tissue samples and a tissue homogenate were cryo-sectioned at 10 μm thickness and thaw-mounted onto indium tin oxide (ITO) coated slides. A serial section was collected for hematoxylin and eosin (H&E) staining. The Chemical QuantArray was then cryo-sectioned at the same thickness as the tissue samples and thaw-mounted over either the human or mouse tissue homogenate on the same ITO-coated slide for MALDI-MSI analysis. The slides were dried in a vacuum desiccator. Optical microscopy images were acquired using a bright-field microscope (Zeiss Observer Z.1) with a 20× objective.

Matrix Deposition.

A 1,5-diaminonaphthalene (DAN)-HCl matrix solution was prepared for metabolite quantification. A stock solution of D-glutamate-d₅ was prepared at 100 mM in 1 M HCl to be used as an IS. The stock solution (5 μL) was added to 5 mL of 1,5-DAN (4.3 mg/mL) dissolved in 4.5/5.0/0.5 HPLC-grade water/ethanol/1 M HCl. Matrix was deposited using a TM-sprayer (HTX imaging, Carrboro) with a two-pass cycle with a flow rate of (0.09 mL/min), spray nozzle velocity (1200 mm/min), spray nozzle temperature (75 °C), nitrogen gas pressure (10 psi), and track spacing (2 mm).

MALDI Fourier Transform-Ion Cyclotron Resonance (FT-ICR) Mass Spectrometry Imaging.

The mass spectrometry imaging experiments were performed by using a 15 T solariX FT-ICR mass spectrometer with a dual ESI/MALDI source (Bruker Daltonics, Billerica, MA) operating in negative-ion mode. The MSI methods were optimized and calibrated by using the electrospray source. Tune mix solution (Agilent Technologies, Santa Clara, CA) was used to calibrate the mass range. The instrumental parameters were optimized to a defined pixel step size of (50 μm) covering the m/z range of 92.15–3000. Each pixel consisted of 250 laser shots with a frequency of 1000 Hz.

MALDI-MSI Visualization, Data Processing, and Statistical Analyses.

The ion images and mass spectra were analyzed and visualized using SCiLS Lab software version (2019c premium, Bruker Daltonics, Billerica, MA). Further data processing and statistical analyses were conducted using an in-house R package that expands on the rMSI²⁶ and rMSIproc²⁷ framework.

An MS feature unique to the embedding collagen solution (790.5381 m/z) was used as a quality control (QC) to exclude pixels outside the TMA wells or in tissue homogenate cracks. Four different intensity normalizations were used: no normalization, total ion current (TIC), matrix peak (312.1379 m/z), and stable-isotope-labeled glutamate internal standard (IS) (151.0773 m/z).

A weighted linear regression $\left(w=\frac{1}{y^2}\right)$ was fitted to 10 replicates of each concentration. Points with regression residuals greater than the (regression standard error) × (t-value at 95% confidence interval) were considered outliers and thus removed.

In compliance with May 2018 FDA guidelines, accuracy was defined as the mean relative error of recalled to nominal concentrations (%) and precision as the relative standard deviation of the recalled concentrations (%RSD). Due to the lack of MSI-specific quantification guidelines, we abide by chromatographic assay guidelines (minimum ±20% accuracy and precision requirement).²⁸

The influence of the number of replicates (from 1 to 6) on accuracy was estimated by using a Monte Carlo simulation.

RESULTS AND DISCUSSION

Optimization of Chemical QuantArray.

The steps involved in building a Chemical QuantArray for MALDI-MSI in situ metabolite quantification are displayed in Figure 1. A gelatin imprint mold is created from a tissue microarray (TMA) mold. Serial concentrations of isotopically labeled metabolites are dispensed into the channel. The array is then frozen at −20 °C and sectioned at 10 μm thickness over a tissue homogenate at the same thickness mounted on a single ITO slide. The whole slide is then sprayed with the matrix of choice and analyzed by using MALDI-MSI.

Figure 1.

Mimetic schematic of the workflow for large-scale metabolite and drug quantification. Using a tissue array mold, gelatin is poured to create an imprint mold. Varied concentrations of isotopically labeled metabolites or drugs are dispensed into the wells. The mimetic is then frozen at −20 °C and sectioned at 10 μm thickness over a human tissue homogenate at the same thickness next to a tissue sample. The tissues are then sprayed with the desired matrix and analyzed using MALDI-MSI.

We evaluated different casting media and dilution solvents to ensure mold integrity, cryosectioning ability, mass spectrometry compatibility, and minimal analyte diffusion (Figure 2). We investigated molds with low (40%) and high (60%) concentrations of gelatin—a well-established embedding medium compatible with MSI workflows.²⁹ While the 60% gelatin solution was more uniformly distributed and contained fewer air bubbles, the mold fractured during cryosectioning. Therefore, a 40% gelatin solution was selected as the optimal casting medium. We later studied the interwell analyte diffusion when using water, agarose, gelatin, or collagen as solvent mediums to dilute the analyte standards. The solutions were dispensed into the array wells in a cryostat at >−15 °C to prevent the analyte mixture from freezing during deposition. Water led to high analyte diffusion into the gelatin mold (Figure 2A). Regardless of concentration (evaluated from 1 to 6%), the water-based agarose gel resulted in highly viscous solutions that did not dispense uniformly into the wells. Additionally, agarose- and gelatin-based gels require continuous heating incompatible with many heat-sensitive labeled metabolites. The collagen type 1 solution demonstrated optimal viscosity and minimal interwell analyte diffusion (Figure 2B) and can be prepared at room temperature. We therefore concluded that the Chemical QuantArray should use a 40% gelatin mold and collagen solvent to ensure robust and reproducible sectioning and analysis.

Figure 2.

Method comparison of three quantitation mimetic techniques. (A) Testing of media for Chemical QuantArray showed a large amount of diffusion between wells (left: ion image; right: optical image) from using water as the medium in comparison to collagen (B) that showed limited interwall contamination. (C) Chemical QuantArray of ¹⁵N glutamate using both water and collagen as the medium. Six arrays of each were sectioned as well as spiked brain tissue homogenate with 2 and 7 mM concentrations of ¹⁵N glutamate. (D) Calibration curves of both the collagen and water-based arrays and the calculated spiked concentrations plotted onto each curve indicating with a star. (E) Comparison of calculated glutamate concentrations using LC-MS/MS to collagen- and water-based Chemical QuantArray at 2 mM (left columns) and 7 mM (right columns).

Chemical QuantArray Improves over Current Quantification Methods.

The main goal of the Chemical QuantArray is to minimize the slide area needed for calibration and quality control (QC) and to maximize the space available for experimental specimens. The existing method proposed by Castellino et al. uses 7 wells per analyte and employs ~16% of a microscope slide. Using this approach, triplicate QC mimetics would cover more than half of the slide and leave little space for experimental sections. Conversely, a Chemical QuantArray containing two analyte series with nine concentration points each takes up ~5%. Six replicate mimetics for QC measurements take up ~33% of the slide, leaving ample space for the experimental tissue sections. The effective 8-fold space reduction allows Chemical QuantArray to multiplex analytes; for example, a mold containing 150 wells (15 rows × 10 columns) can quantify 15 analytes with 10 varied concentrations in a single acquisition, allowing quantification of multiple intermediate metabolites within a metabolic pathway. Additionally, this reduction in space ensures that all quantification and QC can be run along with the experimental samples, blocking batch effects and ensuring accurate and precise quantification.

Additionally, previously described methods rely on analyte-spiked tissue homogenates, locking each preparation to a particular species or tissue type. By decoupling the analyte dilutions from the tissue homogenate, Chemical QuantArray can be generalized to any species or tissue type. The same analyte dilution mold can be sectioned on demand and placed over the required tissue homogenate. This facilitates bulk analyses by reducing preparation time and technical variability.

In order to compare the performance of the optimized collagen Chemical QuantArray with the water-based approach, two types of QuantArrays were prepared using either collagen or water with varying concentrations of ¹⁵N glutamate (Figure 2C). Six arrays of each type were sectioned and imaged. However, the water-based array yielded a lower signal compared to the collagen array due to the diffusion of water into the wells. To validate the suitability of the optimized Chemical QuantArray for quantifying analytes in tissue samples, mouse brain tissue homogenates were spiked with 2 and 7 mM ¹⁵N glutamate and dispensed into a QuantArray mold and further cross-validated using LC-MS/MS analysis. These samples underwent the same preparation process as the Chemical QuantArrays. A d₅-labeled glutamate standard was used as an internal standard (IS) for normalization. Using the calibration curve derived from the collagen samples, the calculated concentrations for the 2 and 7 mM tissue-spiked mixtures were found to be 2.3 ± 0.2 mM (with a 14.3% error) and 7.0 ± 0.5 mM (with a 0.5% error), respectively (Figure 2D). On the other hand, the water-based Chemical QuantArray yielded less accurate results due to interwell diffusion. For the 2 mM tissue-spiked mixture, the calculated concentration was 3.9 ± 0.5 mM (with a 94.5% error), while for the 7 mM mixture, it was 13.3 ± 1.0 mM (with a 90.2% error). Compared to the LC-MS/MS data, the two spiked samples of 2 and 7 mM concentrations were determined to be 2.0 ± 0.2 mM (with a 6.3% error) and 6.9 ± 0.03 mM (with a 1.1% error), respectively, showing comparable results to the collagen-based Chemical QuantArray (Figure 2E).

Finally, Figure S1 compares the performance of Chemical QuantArray to the method proposed by Castellino et al. Overall, QuantArray is shown to preserve comparable accuracy and precision within the linear range (1–15 mM) of glutamate quantification.

Postacquisition Empty Pixel Removal.

Typically, the analyst will manually define the ROIs for each TMA to be acquired by MALDI-MSI. The calibration curves are then obtained by averaging all of the acquired pixels within each ROI. While practical, this manual assignment induces two main sources of variability. First, the ROI includes pixels that fall out of the TMA and thus do not contain calibrant. Second, the tissue homogenate typically presents cracks and imperfections that lead to inconsistencies in ionization. These two phenomena lead to overall inconsistencies in the pixel-wise intensities within each ROI, which in turn affect the overall performance of quantification.

To address both challenges, we rely on an MS feature unique to the embedding collagen solution (790.5381 m/z) (Figure S2). Figure 3A shows that this collagen-unique peak is present only within the TMA spots and is highly suppressed in cracks in the tissue homogenate. By removing all pixels with low intensities of the embedding collagen solution (<3e7 a.u.), we effectively improve precision (μ = 3.4%, SD = 7.1%) while preserving accuracy (Figure 3B).

Figure 3.

Data processing optimization. (A) AMP QuantArray optical image, AMP fragment (361.0746 m/z, log-scaled), collagen solution (790.5381 m/z), and collagen solution mask (>3e7) showing selected concentrations. (B) Accuracy and performance comparison before and after removal of pixels not expressing the collagen solution on AMP (log-scaled concentration axis). (C) Model fit (D) accuracy and performance comparison of 4 normalizations on Glutamate.

We recommend the removal of empty pixels when using tissue samples. The method of choice is application-dependent and can rely on: (1) on-tissue marker, (2) TIC, and (3) microscopy segmentation.

Normalization Strategies.

Although MALDI-MSI can provide chemical and spatial information directly from biological tissue, there are challenges including finding proper internal calibrations and signal normalization techniques to find the appropriate conditions to mimic physical and chemical tissue properties. These issues stem from but are not limited to uniform MALDI matrix deposition, pixel-to-pixel variability, ion suppression, and data interpretation.³⁰ Since ion intensities are influenced by their chemical environment and specific region resulting in different extraction and ionization efficiencies, internal calibrants are needed for normalization.^31,32 While MALDI-MSI normalization of signal intensity has additional challenges, here we demonstrate how variability in extraction and ionization efficiencies may be diminished by the application of an IS to the matrix and, in turn, improve accuracy and precision. Common normalization strategies were investigated, such as the total ion current (TIC), matrix peak, no normalization, and the normalized intensity of the IS.

To provide accurate quantitation using MALDI-MSI, the approach must mimic the interaction, ionization, and desorption of the analyte in the tissue. Several IS deposition methods have been investigated, such as the standard spotted onto, under, and premixed with the MALDI matrix. It has been shown that the application of an IS followed by matrix spraying over a tissue section gave a similar quantitation performance to that of LC-MS/MS analysis.⁴ In our study, a deuterium-labeled glutamate standard was spiked into the DAN-HCl matrix and sprayed over the ¹⁵N glutamate Chemical QuantArray and different normalization parameters were applied. The linear response covering a dynamic range of 0–20 mM ¹⁵N glutamate using TIC, IS, matrix (peak at 312.138 m/z), and no normalization was generated (Figure 3C). An improvement in the correlation coefficient was observed when normalizing to an IS (R²= 0.94, LOQ = 0.44 μM), compared with no normalization (R² = 0.83, LOQ = 0.71 mM). However, similar results were shown when looking at normalization to the intensity of TIC (R² = 0.93, LOQ = 0.46 μM) and the DAN matrix peak (R² = 0.93, LOQ = 0.48 μM).

The best-performing normalization was IS (accuracy μ = 11.07% SD = 11.11% | precision μ = 14.21% SD= 4.24%). The matrix peak normalization showed better accuracy (μ = 8.48% SD = 11.11%) while compromising precision (μ = 18.29% SD = 3.09%). TIC normalization on the other hand showed a good precision (μ = 13.67% SD = 4.58%) at the expense of accuracy (μ = 13.85% SD = 12.14%). Finally, no normalization showed the worst performance (accuracy μ = 12.67% SD = 11.05% | precision μ = 27.43% SD = 10.35%) (Figure 3D).

Although the addition of an IS led to improved precision and accuracy and accounts for tissue heterogeneity, isotopically labeled IS are not always available. Because the Chemical QuantArray requires an isotopically labeled IS, two isotopologues are required for each analyte, which may not always be commercially available. Thus, normalization to the matrix peak may be feasible in place of the matrix IS since the precision and accuracy were similar. Overall, utilizing an IS or normalizing to the matrix peak can correct for several factors and improve the precision of analysis (SD, RSD, and accuracy), which is important when translating to a clinical setting. However, it is well noted here that for clinical diagnostic testing using MALDI-MSI, further optimization needs to be addressed, including more comprehensive validation using LC-MS/MS.

Number of Calibration Replicates Needed.

Quantification accuracy is influenced by the number of mimetic replicates used to fit the regression; thus, we investigated different numbers of technical Chemical QuantArray (Figures S3 and S4). All explored analytes (adenosine, AMP, ADP, or glutamate) indicate that the variability in accuracy is reduced with increasing number of replicates. Whenever possible, six replicates should be used as the SD in accuracy is well below 10%. However, we suggest the use of three replicates in applications with multiple analytes as the performance closely matches six replicates (with maximum deviations of 2.5%). The use of only one replicate is highly discouraged as it presents an accuracy SD 10% higher than when using six replicates.

Purinergic Pathway Analysis Using Chemical QuantArray.

Using our Chemical QuantArray design, we next demonstrated the proof of concept to use this technology to study the activation and the spatial distribution of the purine biosynthesis metabolic pathways, which is altered in many tumors.³³ Among the central metabolites involved in purine biosynthesis, 24 of 26 metabolites were detectable by MALDI-MSI. Importantly, 20 labeled isotopologues of these metabolites are commercially available and thus could be quantified with the Chemical QuantArray (Figure 4A).

Figure 4.

Calibration curves of isotopically labeled metabolite Chemical QuantArray. (A) Purinergic pathway coverage by MALDI-MSI for quantitation. (B) Composite ion images of adenosine (green), AMP (teal), ADP (pink), and ATP (blue) from 6 tissue mimetic sections showing concentration gradient. (C) Calibration curves of corresponding metabolites demonstrating linearity from 7 calibrant points proving reproducibility.

Four metabolites of biological relevance at the metabolic and cell signaling levels from this pathway were selected to be quantified: adenosine, AMP, ADP, and ATP. We first determined a general concentration range for each compound by spotting four coarsely different concentrations of stable isotopes onto tissue homogenates. Once an average of concentrations in the tissue homogenate was established, a multiplexed Chemical QuantArray was cast containing the four metabolites, here each with eight concentrations (from 0 to 1 mM) and acquired MSI of the array. As expected, the ion image of these Chemical QuantArrays shows a gradient of signal from low to high concentrations (Figure 4B,C).

Chemical QuantArray Application in Clinical Tissue Specimens.

To translate Chemical QuantArray to the clinical setting, we applied this approach to 14 clinical samples consisting of 12 meningiomas, one metastatic melanoma, and one solitary fibrous tumor. Serial tissue sections were stained with hematoxylin and eosin (Figure 5A) and analyzed by using MALDI-MSI (Figure 5B–F). Chemical QuantArray provided the simultaneous metabolite quantification of biological tissue with up to 15 possible metabolites. For this study, the slide setup consisted of four purine metabolites, ATP, ADP, AMP, and adenosine, in two rows for reproducibility next to the 14 clinical tissues. Importantly, the mimetic is placed on the same glass slide as the tissue section of interest to eliminate the risk of instrumental changes between cycles, and irregularities from sample/matrix preparation. After normalizing all data to the matrix peak, we generated 4 plots to aid in our quantitation (Figure 5C–F). All metabolites showed a good linearity of calibration: adenosine (R² = 0.9908), ATP (R² = 0.9876), ADP (R² = 0.9229), and AMP (R² = 0.9997) (Figure 5B). The quantitation demonstrated very similar LOD values: adenosine (56.3 μM), ATP (9.7 μM), AMP (70.1 μM), and ADP (169.1 μM). The same trend was found for LOQ values of adenosine (187.66 μM), ATP (32.5 μM), AMP (233.5 μM), and ADP (563.7 μM). Using these values, we quantified the absolute metabolite concentration with each tissue meningioma sample. The concentration of adenosine ranged from 0.06 to 1.67 mM, and that of AMP, ADP, and ATP ranged from 0.06 to 0.67 mM, 0.15 to 1.02 mM, and 0.02 to 0.44 mM, respectively (Figure S5 and Table 1).

Figure 5.

Simultaneous metabolite quantification of ATP, ADP, AMP, and adenosine. (A) Hematoxylin & eosin of 14 human meningioma specimens. (B) Calibration curve of purine isotopically labeled metabolites. Individual ion images of (C) ATP, (D) adenosine, (E) AMP, and (F) ADP, from 14 tissue sections shown on half of an ITO slide. (G) Composite ion image projected onto the optical image of the slide containing the samples and Chemical QuantArray.

The clinical MSI tissue quantitation can be further applied to find correlations between established biomarkers such as CD73 and the respective metabolites.^33–35 In other studies, it has been shown that high CD73 expression levels led to higher extracellular levels of adenosine.³⁶ Thus, this can be a promising application and follow-up study to determine the correlation of CD73 with adenosine using MALDI-MSI. Furthermore, this MSI approach can be used to monitor oncometabolite such as 2-hydroxyglutarate (2-HG) concentrations in glioblastomas (GBM) expressing isocitrate dehydrogenase 1 (IDH1) mutant.³⁷ Correlations between biomarkers and respective metabolites can provide additional information about the TME to aid in the diagnosis of certain cancers and the development of new immunotherapies.

CONCLUSIONS

Accurate and precise mapping of endogenous metabolites in complex tumor specimens is critical for clinical diagnostics and understanding of fundamental disease pathogenesis and tumor heterogeneity. Chemical QuantArray with normalization to stable isotope internal standards or matrix peaks demonstrated an improvement in both precision and feasibility of metabolite quantification using MALDI-MSI in clinical and translational applications. This strategy yielded accurate, spatially resolved quantitative data, with a mimetic that can be easily changed and allows for the quantification of multiple metabolites at once. Importantly, this approach allows analysis of a greater number of specimens with accompanying calibration curves on a single slide, thereby increasing throughput and decreasing inter-run variability. Future directions include the incorporation of quality control samples and assessing the potential of using MALDI-MSI to rapidly identify and characterize tumors.