massNet: integrated processing and classification of spatially resolved mass spectrometry data using deep learning for rapid tumor delineation.

MOTIVATION: Mass spectrometry imaging (MSI) provides rich biochemical information in a label-free manner and therefore holds promise to substantially impact current practice in disease diagnosis. However, the complex nature of MSI data poses computational challenges in its analysis. The complexity of the data arises from its large size, high-dimensionality and spectral nonlinearity. Preprocessing, including peak picking, has been used to reduce raw data complexity; however, peak picking is sensitive to parameter selection that, perhaps prematurely, shapes the downstream analysis for tissue classification and ensuing biological interpretation. RESULTS: We propose a deep learning model, massNet, that provides the desired qualities of scalability, nonlinearity and speed in MSI data analysis. This deep learning model was used, without prior preprocessing and peak picking, to classify MSI data from a mouse brain harboring a patient-derived tumor. The massNet architecture established automatically learning of predictive features, and automated methods were incorporated to identify peaks with potential for tumor delineation. The model's performance was assessed using cross-validation, and the results demonstrate higher accuracy and a substantial gain in speed compared to the established classical machine learning method, support vector machine. AVAILABILITY AND IMPLEMENTATION: https://github.com/wabdelmoula/massNet. The data underlying this article are available in the NIH Common Fund's National Metabolomics Data Repository (NMDR) Metabolomics Workbench under project id (PR001292) with http://dx.doi.org/10.21228/M8Q70T. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

High-Throughput Analysis of Tissue-Embedded Single Cells by Mass Spectrometry with Bimodal Imaging and Object Recognition.

In biological tissues, cell-to-cell variations stem from the stochastic and modulated expression of genes and the varying abundances of corresponding proteins. These variations are then propagated to downstream metabolite products and result in cellular heterogeneity. Mass spectrometry imaging (MSI) is a promising tool to simultaneously provide spatial distributions for hundreds of biomolecules without the need for labels or stains. Technological advances in MSI instrumentation for the direct analysis of tissue-embedded single cells are dominated by improvements in sensitivity, sample pretreatment, and increased spatial resolution but are limited by low throughput. Herein, we introduce a bimodal microscopy imaging system combined with fiber-based laser ablation electrospray ionization (f-LAESI) MSI with improved throughput ambient analysis of tissue-embedded single cells (n > 1000) to provide insight into cellular heterogeneity. Based on automated image analysis, accurate single-cell sampling is achieved by f-LAESI leading to the discovery of cellular phenotypes characterized by differing metabolite levels.

Quantitative MRI and laser ablation-inductively coupled plasma-mass spectrometry imaging of iron in the frontal cortex of healthy controls and Alzheimer's disease patients.

Accumulation of iron within the cortex of Alzheimer's disease (AD) patients has been reported by numerous MRI studies using iron-sensitive methods. Validation of iron-sensitive MRI is important for the interpretation of in vivo findings. In this study, the relation between the spatial iron distribution and T2∗-weighted MRI in the human brain was investigated using a direct comparison of spatial maps of iron as detected by T2∗-weighted MRI, iron histochemistry and laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS), in postmortem brain tissue of the medial frontal gyrus of three control subjects and six AD patients. In addition, iron levels measured by LA-ICP-MS and three quantitative MRI methods, namely R2∗ (=1/T2∗), image phase and quantitative susceptibility mapping (QSM), were compared between 19 AD and 11 controls. Histochemistry results we obtained with the modified Meguro staining were highly correlated with iron levels as detected by LA-ICP-MS (r2 â€‹= â€‹0.82, P â€‹< â€‹0.0001). Significant positive correlations were also found between LA-ICP-MS and the three quantitative MRI measurements: R2∗ (r2 â€‹= â€‹0.63), image phase (r2 â€‹= â€‹0.70) and QSM (r2 â€‹= â€‹0.74 (all p â€‹< â€‹0.0001)). R2∗ and QSM showed the strongest correlation with iron content; the correlation of phase with iron clearly showed increased variation, probably due to its high orientation dependence. Despite the obvious differences in iron distribution patterns within the cortex between AD patients and controls, no overall significant differences were found in iron as measured by LA-ICP-MS, nor in R2∗, phase or susceptibility. In conclusion, our results show that histochemistry as well as quantitative MRI methods such as R2∗ mapping and QSM provide reliable measures of iron distribution in the cortex. These results support the use of MRI studies focusing on iron distribution in both the healthy and the diseased brain.

Automatic 3D Nonlinear Registration of Mass Spectrometry Imaging and Magnetic Resonance Imaging Data.

Multimodal integration between mass spectrometry imaging (MSI) and radiology-established modalities such as magnetic resonance imaging (MRI) would allow the investigations of key questions in complex biological systems such as the central nervous system. Such integration would provide complementary multiscale data to bridge the gap between molecular and anatomical phenotypes, potentially revealing new insights into molecular mechanisms underlying anatomical pathologies presented on MRI. Automatic coregistration between 3D MSI/MRI is a computationally challenging process due to dimensional complexity, MSI data sparsity, lack of direct spatial-correspondences, and nonlinear tissue deformation. Here, we present a new computational approach based on stochastic neighbor embedding to nonlinearly align 3D MSI to MRI data, identify and reconstruct biologically relevant molecular patterns in 3D, and fuse the MSI datacube to the MRI space. We demonstrate our method using multimodal high-spectral resolution matrix-assisted laser desorption ionization (MALDI) 9.4 T MSI and 7 T in vivo MRI data, acquired from a patient-derived, xenograft mouse brain model of glioblastoma following administration of the EGFR inhibitor drug of Erlotinib. Results show the distribution of some identified molecular ions of the EGFR inhibitor erlotinib, a phosphatidylcholine lipid, and cholesterol, which were reconstructed in 3D and mapped to the MRI space. The registration quality was evaluated on two normal mouse brains using the Dice coefficient for the regions of brainstem, hippocampus, and cortex. The method is generic and can therefore be applied to hyperspectral images from different mass spectrometers and integrated with other established in vivo imaging modalities such as computed tomography (CT) and positron emission tomography (PET).

Data-driven identification of prognostic tumor subpopulations using spatially mapped t-SNE of mass spectrometry imaging data.

The identification of tumor subpopulations that adversely affect patient outcomes is essential for a more targeted investigation into how tumors develop detrimental phenotypes, as well as for personalized therapy. Mass spectrometry imaging has demonstrated the ability to uncover molecular intratumor heterogeneity. The challenge has been to conduct an objective analysis of the resulting data to identify those tumor subpopulations that affect patient outcome. Here we introduce spatially mapped t-distributed stochastic neighbor embedding (t-SNE), a nonlinear visualization of the data that is able to better resolve the biomolecular intratumor heterogeneity. In an unbiased manner, t-SNE can uncover tumor subpopulations that are statistically linked to patient survival in gastric cancer and metastasis status in primary tumors of breast cancer.

Interactive Visual Exploration of 3D Mass Spectrometry Imaging Data Using Hierarchical Stochastic Neighbor Embedding Reveals Spatiomolecular Structures at Full Data Resolution.

Technological advances in mass spectrometry imaging (MSI) have contributed to growing interest in 3D MSI. However, the large size of 3D MSI data sets has made their efficient analysis and visualization and the identification of informative molecular patterns computationally challenging. Hierarchical stochastic neighbor embedding (HSNE), a nonlinear dimensionality reduction technique that aims at finding hierarchical and multiscale representations of large data sets, is a recent development that enables the analysis of millions of data points, with manageable time and memory complexities. We demonstrate that HSNE can be used to analyze large 3D MSI data sets at full mass spectral and spatial resolution. To benchmark the technique as well as demonstrate its broad applicability, we have analyzed a number of publicly available 3D MSI data sets, recorded from various biological systems and spanning different mass-spectrometry ionization techniques. We demonstrate that HSNE is able to rapidly identify regions of interest within these large high-dimensionality data sets as well as aid the identification of molecular ions that characterize these regions of interest; furthermore, through clearly separating measurement artifacts, the HSNE analysis exhibits a degree of robustness to measurement batch effects, spatially correlated noise, and mass spectral misalignment.

Histology-Guided High-Resolution Matrix-Assisted Laser Desorption Ionization Mass Spectrometry Imaging.

Mass spectrometry imaging (MSI) is widely used for clinical research because when combined with histopathological analysis the molecular signatures of specific cells/regions can be extracted from the often-complex histologies of pathological tissues. The ability of MSI to stratify patients according to disease, prognosis, and response is directly attributable to this cellular specificity. MSI developments are increasingly focused on further improving specificity, through higher spatial resolution to better localize the signals or higher mass resolution to better resolve molecular ions. Higher spatial/mass resolution leads to increased data size and longer data acquisition times. For clinical applications, which analyze large series of patient tissues, this poses a challenge to keep data load and acquisition time manageable. Here we report a new tool to perform histology guided MSI; instead of analyzing large parts of each tissue section the histology from adjacent tissue sections is used to focus the analysis on the areas of interest, e.g., comparable cell types in different patient tissues, thereby minimizing data acquisition time and data load. The histology tissue section is annotated and then automatically registered to the MSI-prepared tissue section; the registration transformation is then applied to the annotations, enabling them to be used to define the MSI measurement regions. Using a series of formalin-fixed, paraffin-embedded human myxoid liposarcoma tissues, we demonstrate an 80% reduction of data load and acquisition time, thereby enabling high resolution (mass or spatial) to be more readily applied to clinical research. The software is freely available for download.

Automatic generic registration of mass spectrometry imaging data to histology using nonlinear stochastic embedding.

The combination of mass spectrometry imaging and histology has proven a powerful approach for obtaining molecular signatures from specific cells/tissues of interest, whether to identify biomolecular changes associated with specific histopathological entities or to determine the amount of a drug in specific organs/compartments. Currently there is no software that is able to explicitly register mass spectrometry imaging data spanning different ionization techniques or mass analyzers. Accordingly, the full capabilities of mass spectrometry imaging are at present underexploited. Here we present a fully automated generic approach for registering mass spectrometry imaging data to histology and demonstrate its capabilities for multiple mass analyzers, multiple ionization sources, and multiple tissue types.

Automatic registration of mass spectrometry imaging data sets to the Allen brain atlas.

Mass spectrometry imaging holds great potential for understanding the molecular basis of neurological disease. Several key studies have demonstrated its ability to uncover disease-related biomolecular changes in rodent models of disease, even if highly localized or invisible to established histological methods. The high analytical reproducibility necessary for the biomedical application of mass spectrometry imaging means it is widely developed in mass spectrometry laboratories. However, many lack the expertise to correctly annotate the complex anatomy of brain tissue, or have the capacity to analyze the number of animals required in preclinical studies, especially considering the significant variability in sizes of brain regions. To address this issue, we have developed a pipeline to automatically map mass spectrometry imaging data sets of mouse brains to the Allen Brain Reference Atlas, which contains publically available data combining gene expression with brain anatomical locations. Our pipeline enables facile and rapid interanimal comparisons by first testing if each animal's tissue section was sampled at a similar location and enabling the extraction of the biomolecular signatures from specific brain regions.

Spatially resolved characterization of tissue metabolic compartments in fasted and high-fat diet livers.

Cells adapt their metabolism to physiological stimuli, and metabolic heterogeneity exists between cell types, within tissues, and subcellular compartments. The liver plays an essential role in maintaining whole-body metabolic homeostasis and is structurally defined by metabolic zones. These zones are well-understood on the transcriptomic level, but have not been comprehensively characterized on the metabolomic level. Mass spectrometry imaging (MSI) can be used to map hundreds of metabolites directly from a tissue section, offering an important advance to investigate metabolic heterogeneity in tissues compared to extraction-based metabolomics methods that analyze tissue metabolite profiles in bulk. We established a workflow for the preparation of tissue specimens for matrix-assisted laser desorption/ionization (MALDI) MSI that can be implemented to achieve broad coverage of central carbon, nucleotide, and lipid metabolism pathways. Herein, we used this approach to visualize the effect of nutrient stress and excess on liver metabolism. Our data revealed a highly organized metabolic tissue compartmentalization in livers, which becomes disrupted under high fat diet. Fasting caused changes in the abundance of several metabolites, including increased levels of fatty acids and TCA intermediates while fatty livers had higher levels of purine and pentose phosphate-related metabolites, which generate reducing equivalents to counteract oxidative stress. This spatially conserved approach allowed the visualization of liver metabolic compartmentalization at 30 µm pixel resolution and can be applied more broadly to yield new insights into metabolic heterogeneity in vivo.

Multimodal platform for assessing drug distribution and response in clinical trials.

BACKGROUND: Response to targeted therapy varies between patients for largely unknown reasons. Here, we developed and applied an integrative platform using mass spectrometry imaging (MSI), phosphoproteomics, and multiplexed tissue imaging for mapping drug distribution, target engagement, and adaptive response to gain insights into heterogeneous response to therapy. METHODS: Patient-derived xenograft (PDX) lines of glioblastoma were treated with adavosertib, a Wee1 inhibitor, and tissue drug distribution was measured with MALDI-MSI. Phosphoproteomics was measured in the same tumors to identify biomarkers of drug target engagement and cellular adaptive response. Multiplexed tissue imaging was performed on sister sections to evaluate spatial co-localization of drug and cellular response. The integrated platform was then applied on clinical specimens from glioblastoma patients enrolled in the phase 1 clinical trial. RESULTS: PDX tumors exposed to different doses of adavosertib revealed intra- and inter-tumoral heterogeneity of drug distribution and integration of the heterogeneous drug distribution with phosphoproteomics and multiplexed tissue imaging revealed new markers of molecular response to adavosertib. Analysis of paired clinical specimens from patients enrolled in the phase 1 clinical trial informed the translational potential of the identified biomarkers in studying patient's response to adavosertib. CONCLUSIONS: The multimodal platform identified a signature of drug efficacy and patient-specific adaptive responses applicable to preclinical and clinical drug development. The information generated by the approach may inform mechanisms of success and failure in future early phase clinical trials, providing information for optimizing clinical trial design and guiding future application into clinical practice.

Peak learning of mass spectrometry imaging data using artificial neural networks.

Mass spectrometry imaging (MSI) is an emerging technology that holds potential for improving, biomarker discovery, metabolomics research, pharmaceutical applications and clinical diagnosis. Despite many solutions being developed, the large data size and high dimensional nature of MSI, especially 3D datasets, still pose computational and memory complexities that hinder accurate identification of biologically relevant molecular patterns. Moreover, the subjectivity in the selection of parameters for conventional pre-processing approaches can lead to bias. Therefore, we assess if a probabilistic generative model based on a fully connected variational autoencoder can be used for unsupervised analysis and peak learning of MSI data to uncover hidden structures. The resulting msiPL method learns and visualizes the underlying non-linear spectral manifold, revealing biologically relevant clusters of tissue anatomy in a mouse kidney and tumor heterogeneity in human prostatectomy tissue, colorectal carcinoma, and glioblastoma mouse model, with identification of underlying m/z peaks. The method is applied for the analysis of MSI datasets ranging from 3.3 to 78.9 GB, without prior pre-processing and peak picking, and acquired using different mass spectrometers at different centers.

Interim clinical trial analysis of intraoperative mass spectrometry for breast cancer surgery.

Optimal resection of breast tumors requires removing cancer with a rim of normal tissue while preserving uninvolved regions of the breast. Surgical and pathological techniques that permit rapid molecular characterization of tissue could facilitate such resections. Mass spectrometry (MS) is increasingly used in the research setting to detect and classify tumors and has the potential to detect cancer at surgical margins. Here, we describe the ex vivo intraoperative clinical application of MS using a liquid micro-junction surface sample probe (LMJ-SSP) to assess breast cancer margins. In a midpoint analysis of a registered clinical trial, surgical specimens from 21 women with treatment naïve invasive breast cancer were prospectively collected and analyzed at the time of surgery with subsequent histopathological determination. Normal and tumor breast specimens from the lumpectomy resected by the surgeon were smeared onto glass slides for rapid analysis. Lipidomic profiles were acquired from these specimens using LMJ-SSP MS in negative ionization mode within the operating suite and post-surgery analysis of the data revealed five candidate ions separating tumor from healthy tissue in this limited dataset. More data is required before considering the ions as candidate markers. Here, we present an application of ambient MS within the operating room to analyze breast cancer tissue and surgical margins. Lessons learned from these initial promising studies are being used to further evaluate the five candidate biomarkers and to further refine and optimize intraoperative MS as a tool for surgical guidance in breast cancer.

A unique subset of glycolytic tumour-propagating cells drives squamous cell carcinoma.

Head and neck squamous cell carcinoma (SCC) remains among the most aggressive human cancers. Tumour progression and aggressiveness in SCC are largely driven by tumour-propagating cells (TPCs). Aerobic glycolysis, also known as the Warburg effect, is a characteristic of many cancers; however, whether this adaptation is functionally important in SCC, and at which stage, remains poorly understood. Here, we show that the NAD+-dependent histone deacetylase sirtuin 6 is a robust tumour suppressor in SCC, acting as a modulator of glycolysis in these tumours. Remarkably, rather than a late adaptation, we find enhanced glycolysis specifically in TPCs. More importantly, using single-cell RNA sequencing of TPCs, we identify a subset of TPCs with higher glycolysis and enhanced pentose phosphate pathway and glutathione metabolism, characteristics that are strongly associated with a better antioxidant response. Together, our studies uncover enhanced glycolysis as a main driver in SCC, and, more importantly, identify a subset of TPCs as the cell of origin for the Warburg effect, defining metabolism as a key feature of intra-tumour heterogeneity.

Dysregulation of very-long-chain fatty acid metabolism causes membrane saturation and induction of the unfolded protein response.

The unfolded protein response (UPR) senses defects in the endoplasmic reticulum (ER) and orchestrates a complex program of adaptive cellular remodeling. Increasing evidence suggests an important relationship between lipid homeostasis and the UPR. Defects in the ER membrane induce the UPR, and the UPR in turn controls the expression of some lipid metabolic genes. Among lipid species, the very-long-chain fatty acids (VLCFAs) are relatively rare and poorly understood. Here, we show that loss of the VLCFA-coenzyme A synthetase Fat1, which is essential for VLCFA utilization, results in ER stress with compensatory UPR induction. Comprehensive lipidomic analyses revealed a dramatic increase in membrane saturation in the fat1Δ mutant, likely accounting for UPR induction. In principle, this increased membrane saturation could reflect adaptive membrane remodeling or an adverse effect of VLCFA dysfunction. We provide evidence supporting the latter, as the fat1Δ mutant showed defects in the function of Ole1, the sole fatty acyl desaturase in yeast. These results indicate that VLCFAs play essential roles in protein quality control and membrane homeostasis and suggest an unexpected requirement for VLCFAs in Ole1 function.

Localized Metabolomic Gradients in Patient-Derived Xenograft Models of Glioblastoma.

Glioblastoma (GBM) is increasingly recognized as a disease involving dysfunctional cellular metabolism. GBMs are known to be complex heterogeneous systems containing multiple distinct cell populations and are supported by an aberrant network of blood vessels. A better understanding of GBM metabolism, its variation with respect to the tumor microenvironment, and resulting regional changes in chemical composition is required. This may shed light on the observed heterogeneous drug distribution, which cannot be fully described by limited or uneven disruption of the blood-brain barrier. In this work, we used mass spectrometry imaging (MSI) to map metabolites and lipids in patient-derived xenograft models of GBM. A data analysis workflow revealed that distinctive spectral signatures were detected from different regions of the intracranial tumor model. A series of long-chain acylcarnitines were identified and detected with increased intensity at the tumor edge. A 3D MSI dataset demonstrated that these molecules were observed throughout the entire tumor/normal interface and were not confined to a single plane. mRNA sequencing demonstrated that hallmark genes related to fatty acid metabolism were highly expressed in samples with higher acylcarnitine content. These data suggest that cells in the core and the edge of the tumor undergo different fatty acid metabolism, resulting in different chemical environments within the tumor. This may influence drug distribution through changes in tissue drug affinity or transport and constitute an important consideration for therapeutic strategies in the treatment of GBM. SIGNIFICANCE: GBM tumors exhibit a metabolic gradient that should be taken into consideration when designing therapeutic strategies for treatment.See related commentary by Tan and Weljie, p. 1231.

Rapid MALDI mass spectrometry imaging for surgical pathology.

Matrix assisted laser desorption ionization mass spectrometry imaging (MALDI MSI) is an emerging analytical technique, which generates spatially resolved proteomic and metabolomic images from tissue specimens. Conventional MALDI MSI processing and data acquisition can take over 30 min, limiting its clinical utility for intraoperative diagnostics. We present a rapid MALDI MSI method, completed under 5 min, including sample preparation and analysis, providing a workflow compatible with the clinical frozen section procedure.

Postmortem MRI and histology demonstrate differential iron accumulation and cortical myelin organization in early- and late-onset Alzheimer's disease.

Previous MRI studies reported cortical iron accumulation in early-onset (EOAD) compared to late-onset (LOAD) Alzheimer disease patients. However, the pattern and origin of iron accumulation is poorly understood. This study investigated the histopathological correlates of MRI contrast in both EOAD and LOAD. T2*-weighted MRI was performed on postmortem frontal cortex of controls, EOAD, and LOAD. Images were ordinally scored using predefined criteria followed by histology. Nonlinear histology-MRI registration was used to calculate pixel-wise spatial correlations based on the signal intensity. EOAD and LOAD were distinguishable based on 7T MRI from controls and from each other. Histology-MRI correlation analysis of the pixel intensities showed that the MRI contrast is best explained by increased iron accumulation and changes in cortical myelin, whereas amyloid and tau showed less spatial correspondence with T2*-weighted MRI. Neuropathologically, subtypes of Alzheimer's disease showed different patterns of iron accumulation and cortical myelin changes independent of amyloid and tau that may be detected by high-field susceptibility-based MRI.

Integrated mapping of pharmacokinetics and pharmacodynamics in a patient-derived xenograft model of glioblastoma.

Therapeutic options for the treatment of glioblastoma remain inadequate despite concerted research efforts in drug development. Therapeutic failure can result from poor permeability of the blood-brain barrier, heterogeneous drug distribution, and development of resistance. Elucidation of relationships among such parameters could enable the development of predictive models of drug response in patients and inform drug development. Complementary analyses were applied to a glioblastoma patient-derived xenograft model in order to quantitatively map distribution and resulting cellular response to the EGFR inhibitor erlotinib. Mass spectrometry images of erlotinib were registered to histology and magnetic resonance images in order to correlate drug distribution with tumor characteristics. Phosphoproteomics and immunohistochemistry were used to assess protein signaling in response to drug, and integrated with transcriptional response using mRNA sequencing. This comprehensive dataset provides simultaneous insight into pharmacokinetics and pharmacodynamics and indicates that erlotinib delivery to intracranial tumors is insufficient to inhibit EGFR tyrosine kinase signaling.

Whole-Brain Microscopy Meets In Vivo Neuroimaging: Techniques, Benefits, and Limitations.

Magnetic resonance imaging, positron emission tomography, and optical imaging have emerged as key tools to understand brain function and neurological disorders in preclinical mouse models. They offer the unique advantage of monitoring individual structural and functional changes over time. What remained unsolved until recently was to generate whole-brain microscopy data which can be correlated to the 3D in vivo neuroimaging data. Conventional histological sections are inappropriate especially for neuronal tracing or the unbiased screening for molecular targets through the whole brain. As part of the European Society for Molecular Imaging (ESMI) meeting 2016 in Utrecht, the Netherlands, we addressed this issue in the Molecular Neuroimaging study group meeting. Presentations covered new brain clearing methods, light sheet microscopes for large samples, and automatic registration of microscopy to in vivo imaging data. In this article, we summarize the discussion; give an overview of the novel techniques; and discuss the practical needs, benefits, and limitations.