AI-Based Digital Pathology-Enabled Spatial-Omics Data Analyses of the Human Kidney.

Role of Kidney Biopsies for Biomarker Discovery in Diabetic Kidney Disease.

Introduction Clinical research into human intraocular disease UM FFPE tissue sections is described. UM remains the most common intraocular malignancy in adults, with poor prognosis for UM within the choroid region and distant sites UM metastasis. Two imaging MS techniques Matrix Assisted Laser Desorption/Ionisation (MALDI) and Desorption Electrospray Ionisation (DESI) MSI were applied. Molecular profiles were obtained and analysed by multivariate statistical approaches, providing insight into the biochemical and biological differences/similarities within the patient sample cohort (n = 15). Methods Enucleations were collected over 6 years and subjected to the standard fixation and paraffin embedding protocols. In preparation for MS, 5 μm sections were produced with removal of the paraffin followed by heat induced antigen retrieval. MALDI MSI tissue sections were prepared by applying a matrix solution onto the tissue sections to help ionization of molecules directly from the tissue. All MSI experiments (MALDI and DESI) were carried out using a SYNAPT mass spectrometer (Waters Corporation, Manchester, UK). Multivariate analyses were performed using MATLAB (MathWorks, Inc., Natick, MA, US) and the Eigenvector PLS_Toolbox. Results Initial MALDI MSI images acquired with a spatial resolution of 100 μm x 100 μm in positive ion mode showed distinct spatial distribution of many molecular species throughout the choroid, cornea, retina, lens and UM tumor regions. Further MALDI MSI experiments using consecutive tissue sections at 50 μm x 50 μm spatial resolution show substantial variation in the spatial distribution of species within the low molecular mass range. DESI MSI experiments were carried out at 200 μm x 200 μm in negative ionization mode. Deprotonated molecular ions were detected from a variety of lipid related species, localized to specific regions within the eye including the tumor region. Multivariate analysis classified UM samples (good vs. poor prognosis). Using unsupervised PCA, an unbiased representation of the data was generated, with clear sample grouping and differentiation observed based upon tumor status. Use of the supervised PLS-DA technique provided even clearer separation between the selected samples, with tumor profiles displaying dominant discriminatory peaks. These peaks could be identified as sphingolipids and Lyso-phosphocholine, a phosphatidylcholine degradation product. Conclusions It was possible to analyze clinical research FFPE tissue sections by MALDI and DESI MSI, illustrating that specific small molecular species remained localized to certain tissue types including the UM tumor. Initial correlation with tumor status was determined from the statistical analysis of the MALDI MSI datasets. Citation Format: Laura M. Cole, Hardeep S. Mudhar, Karen Sisley, Andrew Peck, Mike Batey, Emmanuelle Claude, Malcolm Clench. Analysis of FFPE treated clinical tissue sections obtained from human intraocular malignancy, uveal melanoma by mass spectrometry imaging (MSI). [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 4231.

Erlotinib has been used to treat patients with EGFR-mutated non-small-cell lung cancer (NSCLC) for almost two decades; however, acquired resistance sooner or later develops against its blockade, thus, low efficacy is inevitable in some patients.1, 2 Many studies have aimed to discern the cause of this resistance by exploring the underlying molecular mechanisms of erlotinib. To investigate an underlying mechanism of erlotinib's resistance, its distribution in tumor, liver, and kidney tissues were analyzed with matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) in drug-resistant and drug-sensitive NSCLC mouse xenograft models. The low in vivo distribution of erlotinib in tumor tissues in a drug-resistant NSCLC mouse xenograft model suggests the existence of a new resistance mechanism in NSCLC. To investigate the resistance of NSCLC cell lines to erlotinib treatment, the drug effects on proliferation were studied in two NSCLC cell lines, H1299 (EGFR WT; erlotinib-resistant) and PC9 (Exon19 del; erlotinib-sensitive).3, 4 Erlotinib weakly inhibited cell proliferation in the H1299 cells (IC50 = 65 μM, Figure 1A). In contrast, robust inhibition was observed in the PC9 cells (IC50 = 0.7 μM) (Figure 1B). Additionally, we investigated whether erlotinib inhibited EGFR kinase activity in H1299 cells to confirm previously reported results.5 Erlotinib (30 μM) was applied to EGF-induced H1299 cells to examine the drug effects on the EGFR signaling pathway. As presented in Figure 1C, EGF activated the EGFR, whereafter the activation was suppressed by erlotinib treatment, which led to the suppression of both AKT and ERK phosphorylation. We also used the DARTS assay to investigate the binding of erlotinib to EGFR (Figure 1D).6, 7 Pronase treatment, that is, digestion, significantly reduced EGFR level. However, this digestion was suppressed by pre-treatment with erlotinib (30 μM) due to a conformational change induced by erlotinib binding to EGFR. In contrast, the amount of VDAC1, a non-erlotinib binding protein used as a control, was significantly decreased even when incubated with erlotinib prior to pronase treatment. This implies that erlotinib is directly binding to EGFR to inhibit EGF-induced EGFR kinase activity. Based on these results, H1299 was selected to explore the erlotinib resistance mechanisms in this study. Next, the in vivo responses of erlotinib in H1299 (erlotinib-resistant) and HCC827 (erlotinib sensitive) cells were investigated in xenograft mice models. The PC9 cell line was not used in this study because PC9 derived tumors were associated with severe ulceration, which would also be difficult to analyze for MALDI-MSI.8 Additionally, HCC827 cell line harbors exon19 del of EGFR such as PC9 cell line. Notably, erlotinib did not reduce the volume of H1299 tumors (Figure 2A) in contrast to HCC827 tumors (Figure 3A), which confirmed the cells were erlotinib-sensitive (IC50 = 0.2 μM) (Figure S2), and there was no apparent toxic indication of the erlotinib treatment (Figures 2B and 3B). H1299 tumor tissue, liver, and kidney were isolated from the xenograft mice previously treated with either vehicle (N = 4) or erlotinib (N = 5) (10 mg/kg) for analyzing erlotinib distribution. The distribution of erlotinib was then examined in the isolated tissue sections with MALDI-MSI. The erlotinib target protein, EGFR, was visualized with immunofluorescence. The images generated from the vehicle and erlotinib-treated groups were subsequently compared and are presented in Figures 2C-2R and 3C-3R. We confirmed that erlotinib co-localized with EGFR in the erlotinib-treated tumor tissue samples. Erlotinib-resistant mouse xenograft model with H1299 cells and erlotinib distribution in tumor, liver, and kidney tissue. (A and B) Mean tumor volumes ± S.D. and body weight over time in response to treatment with erlotinib are shown. (C-R) Distribution of erlotinib and its target protein, EGFR, in various tissues from vehicle and erlotinib-treated mice visualized with MALDI-MSI and IF, respectively. The nucleus is visualized with Hoechst 33342 (blue) (Figures S12A-S12D), and EGFR is visualized with IF (green) in the merged images. The gray and the red lines in the MSI images represent boundaries of the analyzed tissue sample and EGFR distributions, respectively. MALDI-MSI images of metabolites M13/M14 (m/z 380.160) and erlotinib (m/z 394.178) are presented. These specific m/z values of erlotinib were examined by MALDI-MS using erlotinib standard solution (Figure S1). The signal intensity of specific m/z values is presented as an RGB color gradient from blue (low) to red (high). The xenograft mice were treated with either vehicle (N = 4) or erlotinib (N = 5) (10 mg/kg) for analyzing erlotinib distribution. Abbreviation: NS, not significant. Erlotinib-sensitive mouse xenograft model with HCC827 cells and erlotinib distribution in tumor, liver, and kidney tissue. (A and B) Mean tumor volumes ± S.D. and body weight over time in response to treatment with erlotinib are shown. ***p < 0.001. (C-R) Distribution of erlotinib and its target protein, EGFR, in various tissues from vehicle and erlotinib-treated mice visualized with MALDI-MSI and IF, respectively. The nucleus is visualized with Hoechst 33342 (blue) (Figures S12A-S12D) and EGFR is visualized with IF (green) in the merged images. The gray and the red lines in the MSI images represent boundaries of the analyzed tissue sample and EGFR distributions, respectively. MALDI-MSI images of metabolites M13/M14 (m/z 380.160 ) and erlotinib (m/z 394.178 ) are presented. These specific m/z values of erlotinib were examined by MALDI-MS (Figure S1) using erlotinib standard solution. The signal intensity of specific m/z values are presented as an RGB color gradient from blue (low) to red (high). The xenograft mice were treated with either vehicle (N = 6) or erlotinib (N = 6) (10 mg/kg) for analyzing erlotinib distribution. Abbreviation: NS, not significant. In the tumor tissues from H1299 xenograft mouse model, the Total Ion Current (TIC) normalized average signal intensities of erlotinib, and M13/M14 precursor ions (at m/z 394.178 and 380.160, respectively) were analyzed in vehicle- and erlotinib-treated mice (Figures 2E, 2F, 2I, and 2J). M13/M14, which are the biologically active metabolites of erlotinib were also detected in erlotinib treated mice tissues.9 Our measurement also confirmed that erlotinib signal is detectable in liver and kidneys of drug-treated mice, showing relatively high signal intensities (Figures 2M, 2N, 2Q, and 2R). In the tumor tissues from HCC827 xenograft model, the TIC normalized average signal intensities of erlotinib and M13/M14 precursor ions (at m/z 394.178 and 380.160, respectively) were analyzed in both vehicle and erlotinib-treated mice (Figures 3E, 3F, 3I, and 3J). We also confirmed the intensity of the precursor ions in liver and kidney from erlotinib-treated mice (Figures 3M, 3N, 3Q, and 3R). These localization data in both xenograft mouse models clearly demonstrate that erlotinib has affected the tumor by binding to EGFR in vivo. The average drug signal intensities per tissue unit were calculated for each tissue sample from the HCC827 and H1299 xenograft models to compare the drug distributions in the two in vivo mouse groups. We observed that the vehicle-treated groups exhibited different basic tumor tissue intensities in each model (Figure S3). The average intensity was 4.50E-8 for HCC827 tumors, 1.36E-7 for H1299 tumors (Figures S4 and S7), 4.03E-7 for the liver (Figures S5 and S8), and 1.32E-8 for the kidney (Figures S6 and S9) in the vehicle-treated mice. Erlotinib was detected in several tissues but with different intensities in the HCC827 xenograft model group (tumor tissue (13.89), liver (12.32), and kidney (46.42)) similar to the H1299 xenograft model group, where erlotinib was also detected in all three tissue types (tumor tissue (4.55), liver (37.58), and kidney (300.25)) (Figure 4A). Interestingly, as predicted by the high affinity of mutated EGFR in HCC827 cells,10 erlotinib content was 2.95 times higher in HCC827 tumors when compared to H1299 tumors. Furthermore, erlotinib showed stronger localization in the kidney of the drug-treated H1299 xenografts. It is noteworthy that erlotinib was highly localized in normal organ tissues (liver, kidney) in the erlotinib-resistant H1299 mouse xenograft model. In summary, these results demonstrate that in erlotinib-resistant H1299 xenografts, erlotinib preferentially distributed in the liver and kidney rather than in the tumor tissues (Figure 4B). Although erlotinib still has a binding affinity to the target protein, EGFR, even in the H1299 tumor cells, the reduced distribution within in tumor tissues suggests a new mechanism of erlotinib resistance in vivo. We are grateful to Dr. Ho-young Lee (Seoul National University, Seoul, Korea) for providing the NSCLC cell lines (HCC827, PC9, and H1299). The authors declare that there is no conflict of interest. This work was partly supported by grants from the National Research Foundation of Korea and was funded by the government of the Republic of Korea (MSIP; 2015K1A1A2028365, 2016K2A9A1A03904900), Brain Korea 21 Plus Project, and ICONS (Institute of Convergence Science), Yonsei University, Republic of Korea as well as the Berta Kamprad Foundation, Lund, Sweden and the KNN121510 grant by the National Research, Development and Innovation Office of Hungary. Tae Young Kim and Ho Jeong Kwon participated in project conception and experimental design. Tae Young Kim performed cell and molecular biology assays, DARTS assays, in vivo assays (mouse xenograft models injecting H1299 and HCC827), and analyzed the data. Tae Young Kim, Boram Lee, Melinda Rezeli, Yonghyo Kim, and Yutaka Sugihara analyzed the MALDI-MSI data. Tae Young Kim, A. Marcell Szasz, Balazs Dome, Melinda Rezeli, Gyorgy Marko-Varga, and Ho Jeong Kwon wrote the paper. All authors edited and approved the final manuscript. Materials are available upon a reasonable request from the corresponding author. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Nomenclature for kidney function and disease: Executive summary and glossary from a Kidney Disease: Improving Global Outcomes (KDIGO) consensus conference

The use of mass spectrometry imaging (MSI) for the analysis of 3D tissue models of human skin has been shown to provide an elegant label-free methodology for the study of both drug absorption and drug biotransformation. The main aim of the work presented in this thesis was to develop methodology for quantitative assessment of percutaneous absorption using matrix assisted laser desorption ionisation mass spectrometry imaging (MALDI-MSI). Quantitative assessment of the absorption of an antifungal agent, terbinafine hydrochloride, into the epidermal region of a commercial full thickness living skin equivalent model (Labskin) was used as a model system. Different approaches to generate robust and sensitive quantitative mass spectrometry imaging (QMSI) data were developed and compared. The combination of microspotting of analytical and internal standards, matrix sublimation, and recently developed software for quantitative mass spectrometry imaging provided a high-resolution method for the determination of terbinafine hydrochloride in Labskin. A quantitative assessment of the effect of adding a penetration enhancer (dimethyl isosorbide (DMI)) to the delivery vehicle was also performed, and data was compared to LC–MS/MS measurements of isolated epidermal tissue extracts. Comparison of means and standard deviations indicated no significant difference between the values obtained by the two methods. In this thesis the localisation of hydrocortisone hydrochloride in ex-vivo skin was also investigated. Hydrocortisone exhibits a low ionisation efficiency that makes its detection challenging with mass spectrometry techniques. An in-solution and on-tissue chemical derivatisation reaction using the Girard reagent T, a hydrazine based reagent, significantly increased the sensitivity and detection of the respective hydrocortisone-derivative using MALDI-MSI. In an additional study, MALDI-MSI was used to assess the metabolic activity in Labskin by employing the approach of "substrate-based mass spectrometry imaging" (SBMSI). Preliminary MALDI-MSI data detected the activity of the carboxylesterase 1 enzyme in the epidermal layer of skin. The MALDI-MSI data was supported by preliminary LC-MS/MS analysis. To investigate the reproducibility of the results future investigations are required.

Signal transduction and activator of transcription 3 (STAT3) is a key transcription factor implicated in the pathogenesis of kidney fibrosis. Although Stat3 deletion in tubular epithelial cells is known to protect mice from fibrosis, vFoxd1 cells remains unclear. Using Foxd1-mediated Stat3 knockout mice, CRISPR, and inhibitors of STAT3, we investigate its function. STAT3 is phosphorylated in tubular epithelial cells in acute kidney injury, whereas it is expanded to interstitial cells in fibrosis in mice and humans. Foxd1-mediated deletion of Stat3 protects mice from folic-acid- and aristolochic-acid-induced kidney fibrosis. Mechanistically, STAT3 upregulates the inflammation and differentiates pericytes into myofibroblasts. STAT3 activation increases migration and profibrotic signaling in genome-edited, pericyte-like cells. Conversely, blocking Stat3 inhibits detachment, migration, and profibrotic signaling. Furthermore, STAT3 binds to the Collagen1a1 promoter in mouse kidneys and cells. Together, our study identifies a previously unknown function of STAT3 that promotes kidney fibrosis and has therapeutic value in fibrosis.

Diabetes mellitus and kidney disease are two of the most feared long-term conditions with their impact affecting all ages and potentially diminishing quality of life, both in the short and long term. Either one on its own is a major factor in any persons’ life, so one can imagine the impact for the individual, family members and carers when someone develops diabetic kidney disease. This is so not only for the individual but also for the health system where diabetes has long been the single most common cause of end-stage renal failure in the world. The impact of diabetic kidney disease is felt in all populations, and in some parts of the world, the dual assault of diabetes and kidney disease reaches ‘epidemic proportions’. In addition, the person with diabetic kidney disease is at very high risk of all other diabetic microvascular and macrovascular complications making the need for integrated and comprehensive care even more crucial as many different healthcare professionals and specialities will be involved in the care. The breadth of the papers in this issue highlight the complexities involved for someone with diabetic kidney disease and the need for the diabetes, kidney and primary care tribes to work closely together. Care for those with diabetic kidney disease has progressed in the United Kingdom over several years, however, we are still missing opportunities to improve outcomes and much more can and should be done. Meaningful integrated care with the patient at the centre is an absolute priority. Much of the kidney disease in type 1 diabetes affects those who develop the disease in childhood and who have to manage the complexities of diabetes through puberty and adolescence with all the challenges that brings and again there is a parallel in kidney care where transition from paediatric to adult practice is associated with up to 30–40% unnecessary kidney transplant loss. Our services in general are unfriendly to young people between the ages of 14 and 25 years. There is an urgent need for the UK health system and the practitioners within it, to find better ways to support individuals and their families at this time. As obesity increases in young people, not only is the prevalence of type 2 diabetes increasing but also its age of onset is falling, especially in those from ethnic minorities. Over the last few years, some centres have seen a frightening increase in the number of young people with type 2 diabetes. This can only increase the risk of future diabetes complications, including kidney disease. Different ethnicities are affected to different degrees in diabetic kidney disease with the prevalence of type 2 diabetes risk and kidney disease risk being higher in those of African or Caribbean origin and also in those of Asian origin however, all ethnicities can be affected. Overall, therefore, as the number of young people with diabetes increases, particularly in those from ethnic minorities, we face the possibility of a devastating increase the prevalence of diabetic kidney disease and other complications. Advances in the treatment of diabetes with more intensive insulin regimens and greater focus on good glycaemic control and in the early detection of kidney disease with the description, in the 1980s, of the utility of microalbuminuria as a predictor for progressive kidney damage have contributed greatly to the prevention of advanced diabetic kidney disease. Equally, the increasing knowledge of the importance of tight blood pressure control, the advent of renin–angiotensin system inhibitors and other therapeutic advances have all reduced the risk of an individual with diabetes developing diabetic kidney disease and of the kidney disease progressing if it should be detected at an early stage. Indeed, for most individuals, advanced kidney disease takes 10–20 years after the onset of diabetes to develop, therefore, it should be very amenable to preventative strategies and indeed, the incidence in cohorts of newly diagnosed diabetes has fallen. Late detection in type 2 diabetes, with an estimated 850,000 with the disease who are undiagnosed in the United Kingdom, and a failure to grasp the opportunities of modern diabetes management, however, mean many do not gain the benefits of this progress. On a population level, as the prevalence of diabetes increases the numbers afflicted by diabetic kidney disease have grown alarmingly. Unfortunately not all individuals benefit from the advances in diabetes and kidney disease detection and management. A greater focus is required on engaging patients in their own care with more support, education and information, early detection of both diabetes and kidney disease and increasing health care determination to offer intensive interventions especially to those at higher risk. We need particular strategies for those who are different or difficult to reach. New treatments will also be welcome as some individuals fail to respond to current strategies and new therapeutic approaches will be required for these. For many, however, early and appropriate application and uptake of current knowledge and treatments will be highly effective in preventing this potentially devastating complication. For those who have developed advanced kidney disease, there has also been much progress made over the last 20 years. Twenty years ago many individuals of all ages with diabetic kidney disease did not routinely get offered renal replacement therapy or transplantation. Now not only is dialysis and renal transplantation standard options for those with diabetes but for those with type 1 diabetes pancreas and kidney transplantation or maybe, in the future, increasingly islet and kidney transplantation. However, one must not underestimate the morbidity and mortality associated with diabetes and renal replacement therapy. This supplement is a timely overview with well-written comprehensive papers and will, we hope, contribute to a greater awareness of the complexities of care that individuals unfortunate enough to develop this long-term condition must contend. The guest editors have no potential conflicts to declare. Donal has been a Consultant Renal Physician at Salford Royal NHS Foundation Trust since 1992. He was appointed the first National Director for Kidney Care in England in 2007. After gaining degrees in Physiology and Medicine from Manchester University, Donal trained in Renal and Internal Medicine in Leicester, Nottingham, Manchester and Edinburgh. Research training was as a Medical Research Council fellow at Hopital Necker in Paris. Donal has published more than 80 peer review papers, book chapters and articles across the spectrum of nephrology, dialysis and transplantation. Donal was the inaugural president of the multi-professional British Renal Society and is a former Treasurer and President elect of the Renal Association. Donal chaired the National Service Framework for renal services and leads the policy team and implementation strategy for kidney services in England. This has included aligning kidney policy with public health and vascular risk reduction programmes, early detection schemes, integrated care and development of a chronic disease management model of care for advanced kidney disease. Current research interests include epidemiology of chronic kidney disease and acute kidney injury, the biology and management of progressive kidney disease and models of service delivery to optimise outcomes in advanced kidney disease including support during adolescence and transplantation. Stephen Thomas is Consultant in Diabetes and Endocrinology at Guy's & St Thomas’ part of the King's Healthcare partners Academic Health Sciences Centre. He is active in a broad range of diabetes clinics including diabetes kidney clinics. He has both hospital and community diabetes responsibilities. His main research interest is in diabetic complications particularly kidney disease. He teaches on a number of local and national diabetes courses and contributes as an abstract marker for the Diabetes UK Annual Professional Conference. He has been on the NCCDG advising NICE on the treatment of anaemia in chronic kidney disease representing the Royal College of Physicians.

Understanding, and Reversing, Metabolic Memory Is Within Reach

Direct Molecular Tissue Analysis by MALDI Imaging Mass Spectrometry in the Field of Gastrointestinal Disease

Investigating MALDI MSI parameters (Part 2) – On the use of a mechanically shuttered trigger system for improved laser energy stability

Metabolomics: Where seeing is believing

Markers of and Risk Factors for the Development and Progression of Diabetic Kidney Disease

This dissertation presents my efforts to advance the application of matrix-assisted laser desorption/ionization - mass spectrometry imaging (MALDI-MSI) to the chemical analysis of latent fingerprints. The first chapter contains a general introduction to MALDI-MSI, with a focus on the application to fingerprint analysis. The final chapter summarizes the presented work and future directions for the research. The second chapter presents the feasibility of using carbon fingerprint development powder (CFP) as an existing MALDI matrix. This study compared the ionization efficiency of CFP and other commonly used MALDI matrices. The data revealed that CFP is comparable or better than the currently utilized MALDI matrices for latent fingerprint analysis. MALDI-MSI was performed on fingerprints dusted with CFP and lifted with forensic lifting tape, demonstrating that more realistic samples can also be analyzed using MALDI-MSI. Most importantly, it was shown that MALDI-MSI does not destroy the fingerprint during analysis and the fingerprint can be preserved as forensic evidence. The third chapter investigated the use of titanium oxide development powder (TiO2) as a MALDI matrix and elaborates on the impact of adding additional matrices to the signal-to-noise (S/N) ratio of fingerprint compounds. It was demonstrated that TiO2 worked efficiently as an existing MALDI matrix and did not require the use of a high-resolution mass spectrometer. Additional matrices on top of the TiO2 showed limited success and caused a decrease in intensity for some compounds. However, additional matrix did allow the analysis of TiO2 developed fingerprints in negative mode. Importantly this work emphasized the need for knowledge of traditional matrix applicability in fingerprint analysis. In the fourth chapter, the potential for using MALDI-MSI to develop lifestyle profiles of unknown individuals is presented. Prior work studying exogenous fingerprint compounds focused on illicit substances. In this work, compounds related to consumer products, foods, and beverages could be detected in fingerprint residue using MALDI-MSI. These specific compounds could be used for brand or subtype determination of a particular source, such as subtype of citrus fruit. Each set of compounds detected tells a portion of an individual's lifestyle. In the fifth chapter, the mechanism of degradation of unsaturated triacylglycerols (TGs) in fingerprints aged under ambient environment conditions was investigated. MALDI-MSI was used to explore TG profiles of fresh and aged latent fingerprints. With time, the unsaturated TGs underwent ambient ozonolysis resulting in a decrease in the abundance of unsaturated TGs that was relatively reproducible in an individual. In addition,

Simple SummaryFor the effective treatment of lung cancer patients, correct tumor subtyping is of utmost importance, but it is often challenging in clinical routine. Using artificial intelligence and combining information from digital microscopy and matrix-assisted laser desorption/ionization mass spectrometry imaging data have the potential to support the pathologist’s decision-making process. We present a classification algorithm to distinguish between adenocarcinoma and squamous cell carcinoma of the lung based on the automatic detection of tumor areas in whole tissue sections and the determination of the tumor subtype with high accuracy.Artificial intelligence (AI) has shown potential for facilitating the detection and classification of tumors. In patients with non-small cell lung cancer, distinguishing between the most common subtypes, adenocarcinoma (ADC) and squamous cell carcinoma (SqCC), is crucial for the development of an effective treatment plan. This task, however, may still present challenges in clinical routine. We propose a two-modality, AI-based classification algorithm to detect and subtype tumor areas, which combines information from matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) data and digital microscopy whole slide images (WSIs) of lung tissue sections. The method consists of first detecting areas with high tumor cell content by performing a segmentation of the hematoxylin and eosin-stained (H&E-stained) WSIs, and subsequently classifying the tumor areas based on the corresponding MALDI MSI data. We trained the algorithm on six tissue microarrays (TMAs) with tumor samples from N = 232 patients and used 14 additional whole sections for validation and model selection. Classification accuracy was evaluated on a test dataset with another 16 whole sections. The algorithm accurately detected and classified tumor areas, yielding a test accuracy of 94.7% on spectrum level, and correctly classified 15 of 16 test sections. When an additional quality control criterion was introduced, a 100% test accuracy was achieved on sections that passed the quality control (14 of 16). The presented method provides a step further towards the inclusion of AI and MALDI MSI data into clinical routine and has the potential to reduce the pathologist’s work load. A careful analysis of the results revealed specific challenges to be considered when training neural networks on data from lung cancer tissue.

Developing a novel strategy for fabricating matrix film to assess the distribution of potassium perfluorooctanic sulfonate by matrix-assisted laser desorption/ionization mass spectrometry imaging