ABSTRACT
Optical microscopy has long been the gold standard to analyse tissue samples for the diagnostics of various diseases, such as cancer. The current diagnostic workflow is time-consuming and labour-intensive, and manual annotation by a qualified pathologist is needed. With the ever-increasing number of tissue blocks and the complexity of molecular diagnostics, new approaches have been developed as complimentary or alternative solutions for the current workflow, such as digital pathology and mass spectrometry imaging (MSI). This study compares the performance of a digital pathology workflow using deep learning for tissue recognition and an MSI approach utilising shallow learning to annotate formalin-fixed and paraffin-embedded (FFPE) breast cancer tissue microarrays (TMAs). Results show that both deep learning algorithms based on conventional optical images and MSI-based shallow learning can provide automated diagnostics with F1-scores higher than 90%, with the latter intrinsically built on biochemical information that can be used for further analysis.
ABSTRACT
Background: Metabolites, especially lipids, have been shown to be promising therapeutic targets. In conjugation with genes and proteins they can be used to identify phenotypes of disease and support the development of targeted treatments. The majority of clinically collected tissue samples are stored in formalin-fixed and paraffin embedded (FFPE) blocks due to their tissue conservation ability and indefinite storage capacity. For metabolic analysis, however, fresh frozen (FF) samples are currently preferred over FFPE samples due to concerns of metabolic information being lost when preparing the samples. With little or no sample preparation, desorption electrospray ionisation mass spectrometry imaging (DESI-MSI) allows for the study of spatial as well as spectral information. Methods: DESI-MSI analysis was performed on FFPE breast cancer tissue microarray samples from 213 patients collected between the years 1935-2013. Logistic regression (LR) models were built to classify samples based on age and FF samples were used for feature validation. Results: LR models developed on the FFPE samples achieved an average classification accuracy of 96% when predicting their age with a 10-year grouping. Closer examination of the metabolic change over time revealed that the mean signal intensities for the lower mass range (100 - 500 m/z) linearly decrease over time, while the mean intensities for the higher mass range (500 - 900 m/z), remained relatively constant. Conclusions: In our samples, which span over 70 years, sample age has a weak yet quantifiable impact on metabolite content in FFPE samples, while the higher mass range is seemingly unaffected. FFPE samples thus provide an alternative avenue for metabolic analysis of lipids.
ABSTRACT
An amendment to this paper has been published and can be accessed via a link at the top of the paper.
ABSTRACT
Human breast cancer is believed to arise in luminal progenitors within the normal breast. A subset of these are double positive (DP) for basal and luminal keratins and localizes to a putative stem cell zone within ducts. We here present a new protocol based on a combination of CD146 with CD117 and CD326 which provides an up to thirty fold enrichment of the DP cells. We show by expression profiling, colony formation, and morphogenesis that CD146high/CD117high/CD326high DP cells belong to a luminal progenitor compartment. While these DP cells are located quite uniformly in ducts, with age a variant type of DP (vDP) cells, which is mainly CD146-negative, accumulates in lobules. Intriguingly, in specimens with BRCA1 mutations known to predispose for cancer, higher frequencies of lobular vDP cells are observed. We propose that vDP cells are strong candidates for tracing the cellular origin of breast cancer.
