Quantitative Mass Spectrometry Imaging Reveals Mutation Status-independent Lack of Imatinib Penetration into Liver Metastases of Gastrointestinal Stromal Tumors

Mass spectrometry imaging (MSI) is an enabling technology for label-free drug disposition studies at high spatial resolution in life science- and pharmaceutical research. We present the first extensive clinical matrix-assisted laser desorption/ionization (MALDI) quantitative mass spectrometry imaging (qMSI) study of drug uptake and distribution in clinical specimen, analyzing 56 specimens of tumor and corresponding non-tumor tissues from 28 imatinib-treated patients with biopsy-proven gastrointestinal stromal tumors (GIST). For validation, we compared MALDI-TOF-qMSI with conventional UPLC-ESI-QTOF-MS-based quantification from tissue extracts and with ultra-high resolution MALDI-FTICR-qMSI. We introduced a novel generalized nonlinear calibration model of drug quantities based on focused computational evaluation of drug-containing areas that enabled better data fitting and assessment of the inherent method nonlinearities. Imatinib tissue spatial maps revealed striking inefficiency in drug penetration into GIST liver metastases even though the corresponding healthy liver tissues in the vicinity showed abundant imatinib levels beyond the limit of quantification (LOQ), thus providing evidence for secondary drug resistance independent of mutation status. Taken together, these findings underline the important application of MALDI-qMSI for studying the spatial distribution of molecularly targeted therapeutics in oncology.


Introduction
Matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) has emerged as a powerful label-free technology for studying spatial distributions of analytes in tissues. Owing to its high chemical specificity and ability to track hundreds of analytes simultaneously, MALDI-MSI has gained traction in various areas of biomolecular research such as quantitative profiling of metabolites, lipids, peptides, proteins and drugs (1)(2)(3)(4). In the latter case, MALDI-MSI has found its way into pharmaceutical research and development, where disposition of drugs and their carriers can be effectively monitored alongside their pharmacodynamics and toxic effects (4)(5)(6)(7)(8).
Traditionally, quantitative assessment of drug distribution in patient tissues employs bioanalytical techniques such as UPLC-ESI-MS, which assume homogenous distribution of analytes and require tissue homogenization resulting in a complete loss of spatial context (5). In contrast, MALDI-MSI provides spatial information, and much effort has recently been devoted to establish quantitative MSI (qMSI) techniques (1)(2)(3)(4)(9)(10)(11)(12). Linear calibration based on tissue mimetic models or compound dilution series spotted onto tissue, as well as signal normalization against stable isotope-labeled internal standards (IS) (11) and the calculation of tissue extinction coefficients (TEC) (10) were introduced to compensate for the inherently high technical variability of MALDI-MSI (3,4,12). Furthermore, advanced rational testing criteria for evaluating linearity of response, variability, reproducibility and limits of detection of qMSI have been suggested (13). Despite these technical advances, MALDI-qMSI has not yet been widely adapted to clinical pharmacology.
In clinical oncology, the ability to monitor drug penetration into tumors constitutes a key medical need (14). However, tumor heterogeneity and the poorly understood spatial organization of the tumor microenvironment present major challenges for drug uptake and, hence, effective cancer treatment (15,16). This challenge has prompted qMSI studies of drug disposition and of pharmacological/toxic effects in tumor tissues and their surroundings in mice (17,18). Most mouse studies and pioneering qualitative MSI studies of the tyrosine kinase inhibitor (TKI) erlotinib in patient tissue report high degrees of variability and intratumor heterogeneity as well as highly heterogeneous drug distribution (19)(20)(21). However, clinical qMSI proof-of-concept studies of drug distribution are still lacking.
Here, we present the first systematic MALDI-qMSI drug disposition study of a TKI, imatinib, in 56 resection specimens of tumor and surrounding non-tumor tissues from 28 patients with biopsy-proven gastrointestinal stromal tumor (GIST). Imatinib is the standard first line treatment for GIST, effectively stopping autophosphorylation and tumor proliferation particularly in exon 11 mutated GIST (22)(23)(24). However, tumors are prone to imatinib resistance, which is mainly attributed to secondary somatic mutations in the KIT or the platelet-derived growth factor receptor alpha (PDGFRA) tyrosine kinases (25,26). In this study, we introduce generalized nonlinear regression as a superior calibration method based on imatinib-containing pixels, report MALDI-qMSI of three full technical replicates, and compare these results [fast time of flight MALDI-TOF-qMSI and ultra-high-resolution Fourier-Transform Ion Cyclotron Resonance (FTICR-qMSI)] with conventional UPLC-ESI-QTOF-MS quantification. Taken together, our data suggests that MALDI-qMSI compares well with UPLC-ESI-QTOF-MS quantification, and that spatially resolved MS has utility in clinical pharmacology. We can demonstrate that independent of mutation status of the tumor, imatinib failed to penetrate or to be retained in tumor tissue in all GIST liver metastasis cases tested.

Methods (see also Supporting Information) Tissue Samples
Human GIST and corresponding non-tumor control tissues (total of 56 specimens) had been surgically removed from 28 patients (Supporting Information; Table 1 Figure 1b). To achieve comparable conditions after preparation, all slides were stored in a desiccator at room temperature for one hour.
MALDI-TOF-qMSI data acquisition was performed on an ultrafleXtreme MS using

DATA IMPORT, CONVERSION AND PRE-PROCESSING
A total of 48 MALDI-TOF-and three MALDI-FTICR-qMSI-datasets were acquired and exported into imzML format (28) using flexImaging 4.1 (Bruker Daltonics) with a binning rate of 120,000. The imzML datasets were subsequently imported into R 3.3.1 (R Foundation for Statistical Computing, Vienna, Austria)(29) using MALDIquantForeign and processed using MALDIquant packages (30). Mass spectra were normalized to the maximum peak intensity over the mass range m/z 502.32 ± 100 ppm (MALDI-TOF-qMSI) and ± 10 ppm (MALDI-FTICR-qMSI) of the sprayed IS imatinib-D8. Neither baseline correction nor further pre-processing of spectra were performed. All subsequent analysis and visualization (mainly with ggplot2 and fmsb packages) were performed using R.

REGRESSION MODELS AND QUANTIFICATION
Linear regression analysis for calibration of the spotted imatinib dilution series was performed as described (9,10).

Generalized nonlinear regression for calibration in MALDI-ToF-and -FTICR-qMSI
For clinical MALDI-qMSI proof-of-concept, we systematically examined GIST tumorand corresponding non-tumor samples from 28 patients who received the TKI imatinib as first line treatment but presented with refractory disease. Some patients had developed metastatic lesions, mostly in the liver (Supporting Information; Table 1).
Similar nonlinear behavior of calibration curves has been reported by others (11). We therefore wondered if nonlinear regression fitting might be a better option for calibration than a linear regression fit. As illustrated for exemplary dilution series, the generalized nonlinear calibration provided a much better fit indicated by the 6-and 3-fold better RSE for MALDI-TOF-qMSI and MALDI-FTICR-qMSI (Figure 1a,b), respectively.  Figure   2d). MALDI-FTICR-qMSI matched the UPLC-ESI-QTOF-MS results even more closely (Figure 2e).

Limited imatinib uptake in metastatic GIST independent of mutation status
To be effective, cancer-targeting drugs must adequately penetrate into tissue. Hence, we sought to investigate imatinib's tumor penetration capability in human GIST samples. Figure 3 illustrates an overview of the measured tissue sections for all 28 patient samples, done in triplicates, with green and red pixels indicating detected (S/N ≥ 3) and undetectable drug signal, respectively (also see Supporting Information; Table 1 for clinical and histopathological information and Supporting Information;  Surprisingly though, the orally administered imatinib had limited uptake or retention in metastatic GIST in liver tissue, leading to amounts below LOQ despite the high abundance of the drug (well above LOQ) within surrounding normal liver tissue.
Sample A appears to be a notable exception, as comparable amounts of imatinib were present in both normal and tumor tissue (also observed with UPLC-ESI-QTOF-MS in Supporting Information; Figure 7). However, upon histopathological reinvestigation, all three replicates of that sample were found to contain both normal hepatic and metastatic GIST tissues (Figure 4b). As revealed by spatially-resolved-qMSI, also this sample contained imatinib only in its non-tumor part (Figure 4b).
Apparently, the drug was unable to penetrate into the metastatic tumor despite high concentrations within the surrounding tissue. To rule out the possibility that observed differences in imatinib signal were simply a consequence of different desorption/ionization characteristics of the respective tissues, we calculated TEC for all normal tissue-tumor tissue pairs (Supporting Information; Figure 5): Importantly, For proper quantification, the systematic nonlinearity observed in all 48 and 3 dilution sets (Supporting Information; Figure 4) of MALDI-TOF-and -FTICR qMSI, respectively, is particularly relevant. The same behavior has already been reported by Pirman and co-workers (11), and was attributed to the matrix-to-analyte ratio.
Moreover, the complex MALDI process, non-uniform tissue-ion suppression, and interferences from matrix background signals are all factors that could contribute to nonlinear responses (33). Even though it is common for MALDI-TOF-and -FTICR-qMSI studies to use linear calibration for drug dilution series, this linearity cannot always be guaranteed for rare heterogeneous tissue samples even with optimized matrix-to-analyte ratios (see (10)). Furthermore, especially in a clinical screening setting where samples are obtained from different individuals and contain substantially different amounts of target analyte, it is unlikely that any method can guarantee optimal matrix-to-analyte ratios in all scenarios and for all samples. Therefore, a linear behavior of calibration curves for all samples cannot be guaranteed. Hence, nonlinear calibration seems to be the method of choice. On another cautionary note, we suggests that for a reliable reporting of a drug's uptake and distribution, the level of the drug should always be reported relative to its LOQ; drawing conclusions based on low intensity MS ion images of the drug can be potentially misleading and unreliable.
For GIST metastases in liver, the discrepancy between imatinib amounts in normal hepatic versus metastatic tissue is striking (e.g. replicates of sample A (Figure 4b)).
Further work is required to elucidate if lack of imatinib in liver metastases reflects effective export/metabolism or inefficient uptake. For instance, higher expressions of P-glycoprotein and multiresistance protein 1 (MRP1), which are involved in active efflux of a wide spectrum of drugs including imatinib, have been reported in gastric and nongastric GISTs, respectively (34). Other, rather passive, reasons for such drugresistance that is independent of the KIT and PDGFRA mutation types, a poorly organized vasculature, increased interstitial fluid caused by the lack of functional lymphatics and/or inflammation and an abnormal structures of the extracellular matrix (14,15). Additionally, it is well-known that tumor vascularity decreases in GIST metastases after imatinib treatment, which may lay the basis for subsequent tumor progression due to inadequate local drug concentrations (35,36). The observed apparent lack of drug also confirms the clinical practice for GIST patients with liver metastases that the best treatment option is a multidisciplinary one with continued TKI drug therapy and possibly surgical intervention (37). This result based on a small cohort of patient-derived tissues is in line with results from several MALDI-qMSI mouse studies on various chemotherapeutic drug agents in different cell line-based and patient-derived xenograft models (18,19,21,(38)(39)(40).

Conclusion
While providing additional spatial information on imatinib distribution, MALDI-qMSI based on generalized nonlinear calibration and focused computational analysis of drug-bearing pixels provided quantitative measures of imatinib that deviated from reference analysis by UPLC-ESI-QTOF-MS by less than two-fold in 78% of cases and provided a generalized method for modeling analytes dilu-tion series as well as assessing the degree of the apparent nonlinearity of the system. While further refinement of the method and testing in larger patient cohorts will be crucial, our data provides proof-of-concept for clinical utility of MALDI-qMSI. Furthermore, spatial mapping of imatinib distribution within GIST patient tissues revealed striking inefficiency in its penetration into liver metastases irrespec-tive of their mutation status.
Our study therefore strongly suggests that also mechanisms other than driver mutations in receptor tyrosine kinases such as alterations in drug up-take, efflux or intratumor metabolism may play an im-portant role in imatinib resistance in GIST. This finding underlines the important application of MALDI-qMSI for studying the spatial distribution of molecularly targeted therapeutics in oncology.