Biology subjects

Weichert, W.

Publications and source records attributed to Weichert, W..

18 recordsLinked to original sources

bioRxiv2025

Latent plasticity of the human pancreas across development, health, and disease.

The pancreas plays a central role in major human diseases, yet our understanding of its cellular diversity and plasticity remains incomplete. Here, we present a single-cell multiomics atlas of the human pancreas, profiling over four million cells and nuclei from 57 donors across fetal development, adult homeostasis, and type 2 diabetes (T2D). Integrating sc/snRNA-seq, snATAC-seq, VASA-seq, spatial transcriptomics (Xenium), and multiplexed proteomics (CODEX), we resolve gene expression, chromatin accessibility, and spatial organization at high resolution. We identify transcriptionally plastic centroacinar-like cells (pCACs) in adults with fetal-like features, delineate endocrine and exocrine lineage trajectories during development, and uncover HNF1A-defined beta cell epigenetic states. In T2D, we observe shifts in beta cell subtypes and altered regulatory programs. Glucose perturbation of healthy islets reveals cell-type-specific adaptation and stress responses. This atlas provides a foundational framework to understand pancreas biology and the role of cellular plasticity in regeneration and disease.

genomics↗

bioRxiv2024

Cell binding, uptake and infection of influenza A virus using recombinant antibody-based receptors.

Human and avian influenza A viruses bind to sialic acid (Sia) receptors on cells as their primary receptors, and this results in endocytic uptake of the virus. While the role of Sia on glycoproteins and/or glycolipids for virus entry is crucial, the roles of the carrier proteins are still not well understood. Furthermore, it is still unclear how receptor binding leads to infection, including whether the receptor plays a structural or other roles beyond being a simple tether. To enable the investigation of the receptor binding and cell entry processes in a more controlled manner, we have designed a protein receptor for pandemic H1 influenza A viruses. The engineered receptor possesses the binding domains of an anti-HA antibody prepared as a single chain variable fragment (scFv) fused with the stalk, transmembrane and cytoplasmic sequences of the feline transferrin receptor type-1 (fTfR). When expressed in cells that lack efficient display of Sia due to a knockout of the Slc35A1gene which encodes for the Solute Carrier Family 35 transporter (SLC35A1), the anti-H1 receptor was displayed on the cell surface, bound virus or hemagglutinin proteins, and the virus was efficiently endocytosed into the cells. Infection occurred at similar levels to those seen after Sia reconstitution, and treatment with clathrin-mediated endocytosis (CME) inhibitors significantly reduced viral entry. IMPORTANCE.Influenza A viruses mostly circulate among avian reservoirs, and also can jump hosts to cause epidemics in mammals, including among humans. A key interaction of the viruses is with host cell Sia, which vary in chemical form, in their linkages within the oligosaccharide, and in the attachment to surface glycoproteins or glycolipids with different properties. Here we report a new method for examining the processes of receptor binding and uptake into cells during influenza A virus infection, by use of an engineered HA-binding membrane glycoprotein, where an antibody is used as the binding domain and the transferrin receptor uptake structures mediate efficient entry, which should allow us to test and manipulate the processes of cell binding, entry, and infection.

microbiology↗

bioRxiv2024

Towards routine proteome profiling of FFPE tissue: Insights from a 1,200 case pan-cancer study

Proteome profiling of formalin-fixed paraffin-embedded (FFPE) specimens has gained traction for the analysis of cancer tissue for the discovery of molecular biomarkers. However, reports so far focused on single cancer entities, comprised relatively few cases and did not assess the long-term performance of experimental workflows. Here, we did so by analyzing 1,220 tumors from six cancer entities processed over the course of three years. Key findings include the need for a new normalization method ensuring equal and reproducible sample loading for LC-MS/MS analysis across cohorts, showing that tumors can, on average, be profiled to a depth of >4,000 proteins and discovering that current software fails to process such large data sets. We report the first comprehensive pan-cancer proteome expression resource for FFPE material comprising 11,000 proteins which is of immediate utility to the scientific community by way of a web resource. It enables a range of analysis including quantitative comparisons of proteins between patients or cohorts or the discovery of protein fingerprints representing the tissue of origin, or proteins enriched in certain cancer entities.

pathology↗

bioRxiv2023

Viral capsid, antibody, and receptor interactions: experimental analysis of the antibody escape evolution of canine parvovirus

Canine parvovirus (CPV) is a small non-enveloped single-stranded DNA virus that causes serious diseases in dogs worldwide. The original strain of the virus (CPV-2) emerged in dogs during the late-1970s due to a host range switch of a virus similar to the feline panleukopenia virus (FPV) that infected another host. The virus that emerged in dogs had altered capsid receptor- and antibody-binding sites, with some changes affecting both functions. Further receptor and antibody binding changes arose when the virus became better adapted to dogs or to other hosts. Here, we use in vitro selection and deep sequencing to reveal how two antibodies with known interactions select for escape mutations in CPV. The antibodies bind two distinct epitopes, and one largely overlaps the host receptor binding site. We also engineered antibody variants with altered binding structures. Viruses were passaged with the wild type or mutated antibodies, and their genomes deep sequenced during the selective process. A small number of mutations were detected only within the capsid protein gene during the first few passages of selection, and most sites remained polymorphic or were slow to go to fixation. Mutations arose both within and outside the antibody binding footprints on the capsids, and all avoided the TfR-binding footprint. Many selected mutations matched those that have arisen in the natural evolution of the virus. The patterns observed reveal the mechanisms by which these variants have been selected in nature and provide a better understanding of the interactions between antibody and receptor selections. IMPORTANCEAntibodies protect animals against infection by many different viruses and other pathogens, and we are gaining new information about the epitopes that induce antibody responses against viruses and the structures of the bound antibodies. However, less is known about the processes of antibody selection and antigenic escape and the constraints that apply in this system. Here, we use an in vitro model system and deep genome sequencing to reveal the mutations that arise in the virus genome during selection by each of two monoclonal antibodies or their engineered variants. High-resolution structures of each of the Fab: capsid complexes revealed their binding interactions. The engineered forms of the wild-type antibodies or mutant forms allowed us to examine how changes in antibody structure influence the mutational selection patterns seen in the virus. The results shed light on the processes of antibody binding, neutralization escape, and receptor binding, and likely have parallels for many other viruses.

evolutionary biology↗

bioRxiv2022

Transcriptionally defined morphological subtypes of pancreatic ductal adenocarcinoma

Tumour heterogeneity remains a major obstacle to effective and precise therapy for pancreatic ductal adenocarcinoma (PDAC), the most common pancreatic cancer. Several transcriptional subtypes of PDAC with differential prognosis have been described, but they co-occur within tumours and are difficult to distinguish in routine clinical workflows. To investigate the relationship between transcriptional PDAC subtypes, local tissue morphology and the tumour microenvironment, we employed in situ sequencing to profile single cells in their spatial tissue context. We identify five transcriptional subtypes of PDAC cells occurring in three distinct morphological patterns, including secretory tumour cell monolayers, invasive tumour cells with high expression of cell adhesion molecules CEACAM5 and CEACAM6, and spatially distributed tumour cells associated with inflammatory-type fibroblasts. Analysis of bulk RNA-sequencing datasets of the TCGA-PAAD and PACA-AU cohorts according to these spatio-transcriptional subtypes confirmed their prognostic significance. Our results thus indicate an automatable substratification based on spatially-resolved transcriptomics of PDAC and identify distinct subtypes of classical PDAC, representing most cases of this devastating malignancy.

cancer biology↗

bioRxiv2022

Proteogenomic analysis reveals RNA as an important source for tumor-agnostic neoantigen identification correlating with T-cell infiltration

Systemic pan-tumor analyses may reveal the significance of common features implicated in cancer immunogenicity and patient survival. Here, we provide a comprehensive multi-omics data set for 32 patients across 25 tumor types by combining proteogenomics with phenotypic and functional analyses. By using an optimized computational approach, we discovered a large number of novel tumor-specific and tumor-associated antigens including shared common target candidates. To create a pipeline for the identification of neoantigens in our cohort, we combined deep DNA and RNA sequencing with MS- based immunopeptidomics of tumor specimens, followed by the assessment of their immunogenicity. In fact, we could detect a broad variety of non-wild type HLA-binding peptides in the majority of patients and confirmed the immunogenicity of 24 neoantigens. Most interestingly, the majority of total and immunogenic neoantigens originated from variants identified in the RNA dataset, illustrating the importance of RNA as a still understudied source of cancer antigens. Moreover, the amount of these mainly RNA-based immunogenic neoantigens correlated positively with overall CD8+ tumor-infiltrating T cells. This study therefore underlines the importance of RNA-centered variant detection for the identification of shared biomarkers and potentially relevant neoantigen candidates. Statement of significanceThe significance of this study lies not only in the potential of our optimized proteogenomic workflow for the discovery of neoantigens (in particular RNA-derived neoantigens) for clinical application, but sheds light on the entity-agnostic prevalence of HLA class I peptide presentation of RNA processing events to be used for tumor targeting.

cancer biology↗

bioRxiv2022

The Limits of Molecular Signatures for Pancreatic Ductal Adenocarcinoma Subtyping

Molecular signatures have been suggested as biomarkers to classify pancreatic ductal adenocarcinoma (PDAC) into two, three or four subtypes. Since the robustness of existing signatures is controversial, we performed a systematic evaluation of three established signatures for PDAC stratification across eight publicly available datasets. Clustering revealed inconsistency of subtypes across independent datasets and in some cases a different number of PDAC subgroups than in the original study, casting doubt on the actual number of existing subtypes. Next, we built nine classification models to investigate the ability of the signatures for tumor subtype prediction. The overall classification performance ranged from [~]35% to [~]90% accuracy, suggesting instability of the signatures. Notably, permuted subtypes and random gene sets achieved very similar performance. Cellular decomposition and functional pathway enrichment analysis revealed strong tissue-specificity of the predicted classes. Our study highlights severe limitations and inconsistencies that can be attributed to technical biases in sample preparation and tumor purity, suggesting that PDAC molecular signatures do not generalize across datasets. How stromal heterogeneity and immune compartment interplay in the diverging development of PDAC is still unclear. Therefore, a more mechanistic or at least multi-omic approach seems necessary to extract more robust and clinically exploitable insights.

bioinformatics↗

bioRxiv2021

GATA4 and GATA6 loss-of-expression is associated with extinction of the classical programme and poor outcome in pancreatic ductal adenocarcinoma

ObjectiveGATA6 is a master regulator of pancreatic differentiation and a key regulator of the classical phenotype in pancreatic ductal adenocarcinoma (PDAC). Low GATA6 expression is associated with poor patient outcome. GATA4 is the second most expressed GATA factor in the pancreas. The aim was to assess whether, and how, GATA4 contributes to PDAC phenotype and to analyze the association of expression with clinical outcome. DesignWe analyzed PDAC transcriptomic data, stratifying cases according to GATA4 and GATA6 expression, and identified differentially expressed genes and pathways. A multicenter TMA study to assess GATA4 and GATA6 expression in PDAC samples (n=745) from patients undergoing tumour resection was performed using immunohistochemistry with antibodies of validated specificity. GATA4 and GATA6 levels were dichotomized into high/low categorical variables; association with outcome was assessed using univariable and multivariable Cox regression models. ResultsSubtype classification using transcriptomic data revealed that GATA4 mRNA is enriched in classical, compared to basal-like tumours. We classified samples in 4 groups as high/low for GATA4 and GATA6. Reduced expression of GATA4 did not have a major transcriptional impact. However, concomitant low expression of GATA4 enhanced the transcriptomic effects of GATA6 low expression. Reduced expression of both proteins in tumours was associated with the worst patient survival. GATA4 and GATA6 expression significantly decreased in metastases and negatively correlated with basal markers. ConclusionsOur analyses uncover a cooperative interaction between GATA4 and GATA6 to maintain the classical PDAC phenotype and provide compelling clinical rationale for assessing their expression as biomarkers of poor prognosis. SUMMARY BOXO_ST_ABSWhat is already known about this subject?C_ST_ABS- Patients with classical-type PDAC have a better outcome - Retrospective analyses suggest that classical-type PDAC is more sensitive to 5-FU-based chemotherapy - GATA6 is a surrogate biomarker of classical tumours and its expression is associated with better survival - GATA4 and GATA6 have overlapping and unique functions during pancreatic and gastrointestinal development What are the new findings?- Tumours displaying only low GATA4 expression have a transcriptomic profile similar to those with preserved expression of both transcription factors - Combined low expression of GATA4 and GATA6 has the highest transcriptomic impact - In a large multicenter tissue microarray study, patients with tumours showing low expression of both GATA4 and GATA6 have the worst overall survival - Low expression of GATA4 and GATA6 is an independent predictor of survival in patients with resectable PDAC - GATA4 levels are down-regulated in liver metastases and are negatively correlated with basal markers such as KRT5/6, KRT14, and TP63 How might it impact on clinical practice in the foreseeable future?- The combined assessment of GATA4 and GATA6 expression may improve prognostic stratification of patients with PDAC - Prospective studies should confirm whether GATA4 and GATA6 expression is also predictive of response to chemotherapy in PDAC patients

cancer biology↗

bioRxiv2021

Spatial and temporal plasticity of neoantigen-specific T-cell responses bases on characteristics associated to antigen and TCR

Neoantigens derived from somatic mutations have been demonstrated to correlate with therapeutic responses mediated by treatment with immune checkpoint inhibitors. Neoantigens are therefore highly attractive targets for the development of personalized medicine approaches although their quality and associated immune responses is not yet well understood. In a case study of metastatic malignant melanoma, we performed an in-depth characterization of neoantigens and respective T-cell responses in the context of immunotherapy with Ipilimumab. Three neoantigens identified either by immunopeptidomics or in silico prediction were investigated using binding affinity analyses and structural simulations. We isolated seven T-cell receptors (TCRs) from the patient immune repertoire recognizing these antigens. TCRs were compared in-vitro and in-vivo with multi-parametric analyses. Identified immunogenic peptides showed similar binding affinities to the human leukocyte antigen (HLA) complex and comparable differences to their wildtype counterparts in molecular dynamic simulations. Nevertheless, isolated TCRs differed substantially in functionality and frequency. In fact, TCRs with comparably lower functional avidity and higher potential for cross-reactivity provided at least equal anti-tumor immune responses in vivo. Of note, these TCRs showed a reduced activation pattern upon primary in vitro stimulation. Exploration of the TCR-{beta} repertoire in blood and in different tumor-related tissues over three years, offered insights on the high frequency and particular long-term persistence of low-avidity TCRs. These data indicate that qualitative differences of neoantigen-specific TCRs and their impact on function and longevity need to be considered for neoantigen targeting by adoptive T-cell therapy using TCR-transgenic T cells. Statement of translational relevanceImmunotherapy has demonstrated high efficacy in diverse malignancies. Neoantigens derived from mutations provide promising targets for safe and highly tumor-specific therapeutic approaches. Yet, single determinants of an effective and enduring T-cell mediated tumor rejection are still not well understood. We analyzed in detail seven neoantigen-specific T-cell receptors (TCRs) derived from a melanoma patient targeting three different altered peptide ligands identified by mass spectrometry and prediction analyses. Functional characterization of these TCRs revealed potent anti-tumor reactivity of all TCRs. Of special interest, TCRs with comparably lower affinity demonstrated effective in vivo activity as well as dominant spatial and temporal distribution in blood and tissue. Functional differences of TCR may require further T-cell and/or TCR engineering and should be considered for future clinical trial designs.

immunology↗

bioRxiv2020

Maintenance of epithelial traits and resistance to mesenchymal reprogramming promote proliferation in metastatic breast cancer

Despite important advances in the treatment of breast cancer, the 5-year survival rate for patients with distant metastasis remains less than 30%. Metastasis is a complex, multi-step process beginning with local invasion and ending with the outgrowth of systemically disseminated cells into actively proliferating metastases that ultimately cause the destruction of vital organs. It is this last step that limits patient survival and, at the same time, remains the least understood mechanistically. Here, we focus on understanding determinants of metastatic outgrowth using metastatic effusion biopsies from stage IV breast cancer patients. By modelling metastatic outgrowth through xenograft transplantation, we show that tumour initiation potential of patient-derived metastatic breast cancer cells across breast cancer subtypes is strongly linked to high levels of EPCAM expression. Breast cancer cells with high EPCAM levels are highly plastic and, upon induction of epithelial-mesenchymal transition (EMT), readily adopt mesenchymal traits while maintaining epithelial identity. In contrast, low EPCAM levels are caused by the irreversible reprogramming to a mesenchymal state with concomitant suppression of metastatic outgrowth. The ability of breast cancer cells to retain epithelial traits is tied to a global epigenetic program that limits the actions of EMT-transcription factor ZEB1, a suppressor of epithelial genes. Our results provide direct evidence that maintenance of epithelial identity is required for metastatic outgrowth while concomitant expression of mesenchymal markers enables plasticity. In contrast, loss of epithelial traits is characteristic of an irreversible mesenchymal reprogramming associated to a deficiency for metastatic outgrowth. Collectively, our data provide a framework for the intricate intercalation of mesenchymal and epithelial traits in metastatic growth.

cancer biology↗

bioRxiv2020

Influenza A viruses serially passaged in different MDCK cell lines exhibit limited sequence variation across their genomes, with the exception of the hemagglutinin gene

New methods for deep sequence analysis provide an opportunity to follow the emergence and dynamics of virus mutations in real time. Although viruses are commonly grown in cell culture for research and for vaccine development, the cells used to grow the virus are often not derived from the same tissue or even the same host that the virus naturally replicates in. The selective pressures of culturing virus in vitro are still only partially understood. MDCK cells are the standard cell for growing influenza viruses yet are derived from the epithelium of the canine kidney and are also heterogenous. We passaged human H3N2, H1N1 pandemic, and canine H3N2 influenza A viruses (IAV) in different lineages of MDCK cells, as well as lines engineered to express variant Sia receptors, including 2,3- and 2,6-linkages or N-glycolylneuraminic acid (Neu5Gc) or N-acetylneuraminic acid (Neu5Ac) forms. MDCK-Type II cells had lower infection efficiency and virus production, and infection appeared more dependent on protease activation of the virus. When viruses were passaged in the different cells, they exhibited only small numbers of consensus-level mutations, and most were within the HA gene. Both human IAVs showed selection for single nucleotide minority variants in the HA stem across cell types, as well as low frequency variants in the HA receptor binding site of virus passaged in cells expressing Neu5Gc. Canine H3N2 also showed minority variants near the receptor-binding site in cells expressing Neu5Gc and also in those expressing 2,6-linkages. IMPORTANCEThe genetic variation and adaptability of viruses are fundamental properties that allow their evolutionary success in the face of differing host environments and immune responses. The growth of viruses in cell culture is widely used for their study and for preparing vaccines. However, the selection pressures that cell passaging imposes on viruses are often poorly understood. We used deep sequence analysis to define, in detail, how three different influenza A viruses respond to passaging in different lineages of canine MDCK cells that are commonly used for their growth, as well as in variant cells engineered to express different forms of their cell surface receptor, sialic acid. This analysis revealed that most mutations occur in the HA gene and few sequence changes in the virus population reached high proportions. This is relevant for understanding the selective pressures of virus growth in cell culture and how it shapes evolutionary patterns.

microbiology↗

bioRxiv2019

Single nucleus RNA sequencing maps acinar cell states in a human pancreas cell atlas

Molecular evidence of cellular heterogeneity in the human exocrine pancreas has not been established, due to the local concentration of hydrolytic enzymes that can rapidly degrade cells and RNA upon resection. Here we innovated single-nucleus RNA sequencing protocols, and profiled more than 120,000 cells from adult and neonatal human donors to create the first comprehensive atlas of human pancreas cells, including epithelial and non-epithelial constituents. Adult and neonatal pancreata shared common features, including the presence of previously undetected acinar subtypes, but also showed marked differences in the composition of the endocrine, endothelial, and immune compartments. Spatial cartography, including cell proximity mapping through in situ sequencing, revealed dynamic developmental cell topographies in the endocrine and exocrine pancreas. Our human pancreas cell atlas can be interrogated to understand pancreatic cell biology, and provides a crucial reference set for future comparisons with diseased tissue samples to map the cellular foundations of pancreatic diseases.

genomics↗

bioRxiv2019

Influenza Viruses in Mice: Deep Sequencing Analysis of Serial Passage and Effects of Sialic Acid Structural Variation

Influenza A viruses have regularly jumped to new hosts to cause epidemics or pandemics, an evolutionary process that involves variation in the viral traits necessary to overcome host barriers and facilitate transmission. Mice are not a natural host for influenza virus, but are frequently used as models in studies of pathogenesis, often after multiple passages to achieve higher viral titers that result in clinical disease such as weight loss or death. Here we examine the processes of influenza A virus infection and evolution in mice by comparing deep sequence variation of a human H1N1 pandemic virus, a seasonal H3N2 virus, and a H3N2 canine influenza virus during experimental passage. We also compared replication and sequence variation in wild-type mice expressing N-glycolylneuraminic acid (Neu5Gc) with that seen in mice expressing only N-acetylneuraminic acid (Neu5Ac). Viruses derived from plasmids were propagated in MDCK cells and then passaged in mice up to four times. Full genome deep sequencing of the plasmids, cultured viruses, and viruses from mice at various passages revealed only small numbers of mutational changes. The H3N2 canine influenza virus showed increases in frequency of sporadic mutations in the PB2, PA, and NA segments. The H1N1 pandemic virus grew well in mice, and while it exhibited the maintenance of some minority mutations, there was no clear adaptive evolution. The H3N2 seasonal virus did not establish in the mice. Finally, there were no clear sequence differences associated with the presence or absence of Neu5Gc.\n\nSIGNIFICANCEMice are commonly used as a model to study the growth and virulence of influenza A viruses in mammals, but are not a natural host and have distinct sialic acid receptor profiles compared to humans. Using experimental infections with different subtypes of influenza A virus derived from different hosts we found that evolution of influenza A virus in mice did not necessarily proceed through the linear accumulation of host-adaptive mutations, that there was variation in the patterns of mutations detected in each repetition, and the mutation dynamics depended on the virus examined. In addition, variation in the viral receptor, sialic acid, did not effect influenza evolution in this model. Overall this shows that mice provide a useful animal model for influenza, but that host passage evolution will vary depending on the virus being tested.

microbiology↗

bioRxiv2019

A machine learning algorithm predicts molecular subtypes in pancreatic ductal adenocarcinoma with differential response to gemcitabine-based versus FOLFIRINOX chemotherapy

PurposeDevelopment of a supervised machine-learning model capable of predicting clinically relevant molecular subtypes of pancreatic ductal adenocarcinoma (PDAC) from diffusion-weighted-imaging-derived radiomic features.\n\nMethodsThe retrospective observational study assessed 55 surgical PDAC patients. Molecular subtypes were defined by immunohistochemical staining of KRT81. Tumors were manually segmented and 1606 radiomic features were extracted with PyRadiomics. A gradient-boosted-tree algorithm (XGBoost) was trained on 70% of the patients (N=28) and tested on 30% (N=17) to predict KRT81+ vs. KRT81-tumor subtypes. The average sensitivity, specificity and ROC-AUC value were calculated. Chemotherapy response was assessed stratified by subtype. Radiomic feature importance was ranked.\n\nResultsThe mean{+/-}STDEV sensitivity, specificity and ROC-AUC were 0.90{+/-}0.07, 0.92{+/-}0.11, and 0.93{+/-}0.07, respectively. Patients with a KRT81+ subtype experienced significantly diminished median overall survival compared to KRT81-patients (7.0 vs. 22.6 months, HR 1.44, log-rank-test P=<0.001) and a significantly improved response to gemcitabine-based chemotherapy over FOLFIRINOX (10.14 vs. 3.8 months median overall survival, HR 0.85, P=0.037) compared to KRT81-patients, who responded significantly better to FOLFIRINOX over gemcitabine-based treatment (30.8 vs. 13.4 months median overall survival, HR 0.88, P=0.027).\n\nConclusionsThe machine-learning based analysis of radiomic features enables the prediction of subtypes of PDAC, which are highly relevant for overall patient survival and response to chemotherapy.

bioinformatics↗

bioRxiv2019

Multi-task learning of a deep k-nearest neighbour network for histopathological image classification and retrieval

Deep neural networks have achieved tremendous success in image recognition, classification and object detection. However, deep learning is often criticised for its lack of transparency and general inability to rationalize its predictions. The issue of poor model interpretability becomes critical in medical applications, as a model that is not understood and trusted by physicians is unlikely to be used in daily clinical practice. In this work, we develop a novel multi-task deep learning framework for simultaneous histopathology image classification and retrieval, leveraging on the classic concept of k-nearest neighbors to improve model interpretability. For a test image, we retrieve the most similar images from our training databases. These retrieved nearest neighbours can be used to classify the test image with a confidence score, and provide a human-interpretable explanation of our classification. Our original framework can be built on top of any existing classification network (and therefore benefit from pretrained models), by (i) adding a triplet loss function with a novel triplet sampling strategy to compare distances between samples and (ii) a Cauchy hashing loss function to accelerate neighbour searching. We evaluate our method on colorectal cancer histology slides, and show that the confidence estimates are strongly correlated with model performance. The explanations provided by nearest neighbors are intuitive and useful for expert evaluation by giving insights into understanding possible model failures, and can support clinical decision making by comparing archived images and patient records with the actual case.

pathology↗

bioRxiv2019

Expression of 9-O- and 7,9-O-acetyl modified sialic acids in cells and their effects on influenza viruses

Sialic acids (Sia) are widely displayed on the surfaces of cells and tissues. Sia come in a variety of chemically modified forms, including those with acetyl modifications at the C7, C8, and C9 positions. Here, we analyzed the distribution and amounts of these acetyl modifications in different human and canine cells. As Sia or their variant forms are receptors for influenza A and influenza C viruses, we examined the effects of these modifications on virus infections. We confirmed that 9-O-acetyl and 7,9-O-acetyl modified Sia are widely but variably expressed across cell lines from both humans and canines. While they were expressed on the cell surface of canine MDCK cell lines, they were located primarily within the Golgi compartment of human HEK-293 and A549 cells. The O-acetyl modified Sia were expressed at low levels of 1-2% of total Sia in these cell lines. We knocked out and over-expressed the sialate O-acetyltransferase gene (CasD1), and knocked out the sialate O-acetylesterase gene (SIAE) using CRISPR/Cas9 editing. Knocking out CasD1 removed 7,9-O- and 9-O-acetyl Sia expression, confirming previous reports. However, over-expression of CasD1 and knockout of SIAE gave only modest increases in 9-O-acetyl levels in cells and no change in 7,9-O-acetyl levels, indicating that there are complex regulations of these modifications. These modifications were essential for influenza C infection, but had no obvious effect on influenza A infection.\n\nIMPORTANCESialic acids are key glycans that are involved in many different normal cellular functions, as well as being receptors for many pathogens. However, Sia come in diverse chemically modified forms. Here we examined and manipulated the expression of 7,9-O- and 9-O-acetyl modified Sia on cells commonly used in influenza virus and other research by engineering the enzymes that produce or remove the acetyl groups.

cell biology↗

bioRxiv2019

A prospectively validated machine learning model for the prediction of survival and tumor subtype in pancreatic ductal adenocarcinoma

PurposeTo develop a supervised machine learning algorithm capable of predicting above vs. below-median overall survival from medical imaging-derived radiomic features in a cohort of patients with pancreatic ductal adenocarcinoma (PDAC).\n\nMaterials and Methods102 patients with histopathologically proven PDAC were retrospectively assessed as the training cohort and 30 prospectively enrolled patients served as the external validation cohort. Tumors were segmented in pre-operative diffusion weighted-(DW)-MRI derived ADC maps and radiomic features were extracted. A Random Forest machine learning algorithm was fit to the training cohort and tested in the external validation cohort. The histopathological subtype of the tumor samples was assessed by immunohistochemistry in 21/30 patients of the external validation cohort. Individual radiomic feature importance was evaluated.\n\nResultsThe machine learning algorithm achieved a sensitivity of 87% and a specificity of 80% (ROC-AUC 90%) for the prediction of above- vs. below-median survival on the unseen data of the external validation cohort. Heterogeneity-related features were highly ranked by the model. Of the 21 patients for whom the histopathological subtype was determined, 8/9 patients predicted by the model to experience below-median overall survival exhibited the quasi-mesenchymal subtype, while 11/12 patients predicted to experience above-median survival exhibited a non-quasi-mesenchymal subtype (Fishers exact test P<0.001).\n\nConclusionThe application of machine-learning to the radiomic analysis of DW-MRI-derived ADC maps allowed the prediction of overall survival with high diagnostic accuracy in a prospectively collected cohort. The high overlap of clinically relevant histopathological subtypes with model predictions underlines the potential of quantitative imaging workflows in pre-operative subtyping and risk assessment in PDAC.

bioinformatics↗

bioRxiv2019

Distinct immune evasion in APOBEC-enriched, HPV-negative HNSCC

Immune checkpoint inhibition leads to response in some patients with head and neck squamous cell carcinoma (HNSCC). Robust biomarkers are lacking to date.\n\nWe analyzed viral status, gene expression signatures, mutational load and mutational signatures in whole exome and RNA-sequencing data of the HNSCC TCGA dataset (N = 496) and a validation set (DKTK MASTER cohort, N = 10). Public single-cell gene expression data from 17 HPV-negative HNSCC were separately re-analyzed.\n\nAmong HPV-negative HNSCC, APOBEC3-associated TCW motif mutations but not total single nucleotide variant burden were significantly associated with inflammation. APOBEC3-enriched HPV-negative HNSCC showed higher T-cell inflammation and immune checkpoint expression. Mutations in immune-evasion pathways were enriched in these tumors. APOBEC3B and 3C expression was identified in tumor cells and correlated with tumor inflammation.\n\nWe identified an APOBEC-enriched subgroup of HPV-negative HNSCC with a distinct immunogenic phenotype, potentially mediating response to immunotherapy.

cancer biology↗