Literature Review: February 2023

Tissue agnostic driven neoadjuvant treatment, molecular signatures of brain aging, serine/threonine kinome, AI prediction in gastric cancers, Kaposi sarcoma diagnosis, AI lung cancer risk prediction

DALL.E 2 representation of keywords in this post

Neoadjuvant Pembrolizumab in Localized Microsatellite Instability High/Deficient Mismatch Repair Solid Tumors

JOURNAL OF CLINICAL ONCOLOGY

A phase II, open-label, single-center trial, in which 35 patients were enrolled, the majority of whom had colorectal adenocarcinoma (77%) and clinical stage III disease (74%) and are treated with Pembrolizumab. The primary endpoints were safety and rate of pathologic complete response (pCR) in patients with surgically resected specimens receiving at least three cycles of Pembrolizumab. The secondary endpoints included best overall response, rate of organ-sparing at one year for non-surgical patients and the rate of pCR for all patients. Exploratory endpoints include endoscopic response rates and ctDNA kinetics as a surrogate for clinical efficacy.

Of the 17 (49%) patients who underwent surgery, pCR rate was 65%. Of the 35 patients, 33 were evaluable for radiographic response, with an overall response rate of 82% (n = 27): 10 (30%) complete responses (CRs) and 17 (52%) partial responses (PRs). Grade 3 events were reported in two patients (6%), while there were no grade 4 events. At the time of data cutoff, two patients (6%) had expired.

Importance: This is one of the first studies to investigate the use of immunotherapy as neoadjuvant treatment in patients with localized solid tumors that are microsatellite instability high (MSI-H) or deficient mismatch repair (dMMR). Appropriate patient selection in the neoadjuvant setting is of utmost importance to prevent overtreatment. It would be interesting to see if a tissue agnostic approach in this setting is safe and effective in larger cohorts and patients with noncolorectal disease. A similar approach using tumor mutational burden in non-small cell lung cancer has shown promise.

Molecular and spatial signatures of mouse brain aging at single-cell resolution

CELL

Interesting study that used spatially resolved single-cell transcriptomics to create a high-resolution map of the changes in cells in the frontal cortex and striatum of the brain in aging female mice to understand the mechanisms underlying age-related functional decline. They found that changes in non-neuronal cells, such as glial cells, were more pronounced than in neurons and identified molecular and spatial changes in glial cells that may be activated during aging, particularly in the subcortical white matter. This study in essence aims to provide a molecular and spatial signature of glial and immune cell activation patterns associated with aging. It suggests that changes in the oligodendrocytes and myelinated axons and their associated microglial and astrocytic reactivity in the white matter may be an important factor in age-associated cognitive deficits and that there may be specific molecular mechanisms underlying inflammatory activation.

Importance: Spatial analysis is important in this setting because it allows for understanding the location of changes to gain insights into how those changes may impact specific brain functions. For example, changes that occur during aging in one part of the brain may have different implications for brain function than changes that occur in another part of the brain. Additionally, by understanding the spatial relationships between different cell types, one can begin to identify patterns and mechanisms that may be driving the changes. Furthermore, understanding the spatial distribution of changes across brain regions can give insight into which regions are more prone to aging effects or in which regions aging effects are more severe. Application of machine learning methods for spatial pattern analysis in this study may have provided additional clues.

An atlas of substrate specificities for the human serine/threonine kinome

NATURE

This well-conducted study outlines a spectrum of substrate motifs of the human serine/threonine kinome, which can potentially be used as a framework to explore their cellular functions. The study found that the motifs were more diverse than expected, suggesting a broader substrate repertoire of the kinome. The Ser/Thr kinases were also found to be strongly discriminatory against specific motif features, which may suggest that fidelity in kinase signaling pathways is achieved through selective pressure on substrates to avoid phosphorylation by irrelevant kinases. This is achieved by tuning the amino acid sequences surrounding the phosphorylation sites to be disfavored by non-cognate kinases. Using the kinome-wide dataset, the study was able to predict the specific kinases responsible for substrate phosphorylation based on the amino acid sequence surrounding the phosphorylation site. The study suggests that the motif-based approach will provide a valuable resource for researchers who study signaling pathways in human biology and disease, and that it can be used to understand complex signaling in human disease progressions, mechanisms of cancer drug resistance, dietary interventions, and other physiological processes.

Importance: Protein phosphorylation is a common biological process and as the study states “phosphorylation of proteins at serine, threonine, tyrosine and histidine residues controls nearly every aspect of eukaryotic cellular function.” Mass-spectrometry-based phosphoproteomics have identified 90,000 sites of serine and threonine phosphorylation, with several thousand being associated with human diseases and biological processes. However, for the majority of phosphorylation events, it is not known which of the more than 300 protein serine/threonine kinases in the human genome are responsible. This study used synthetic peptide libraries to profile the substrate sequence specificity of 303 of these kinases, which is reportedly 84% of the ones predicted to be active in humans.

This study provides a way to link phosphorylation events to specific biological pathways, which can help to better understand cellular signaling responses. Additionally, by identifying the kinases responsible for phosphorylating specific sites in the human Ser/Thr phosphoproteome, the study can be useful in the development of therapeutics for human diseases that are associated with phosphorylation. Specifically, by identifying the kinases that are responsible for phosphorylating specific sites that have been associated with diseases, one can potentially develop inhibitors of those kinases. Also the data can be potentially used for computational annotation and identification of relevant kinases.

Predicting gastric cancer outcome from resected lymph node histopathology images using deep learning

NATURE COMMUNICATIONS

The study developed a deep-learning framework for detecting lymph node (LN) metastases of gastric cancer from whole histopathology slides, with high performance (sensitivity 98.5%, specificity 96.1%) reported to be comparable to trained pathologists. The study also found that the framework corrected misdiagnoses made by pathologists (~6.8% of cases) and was able to predict outcomes for patients with gastric cancer at each N stage, especially at the N1 and N2 stages.

Importance: The American Joint Committee on Cancer (AJCC) guidelines for gastric cancer provide a standardized system for staging the disease based on the size and location of the tumor, as well as the presence or absence of lymph node and distant metastases. Determining lymph nodes status is critical in staging, a process which can take pathologists up to 15 minutes per patient, depending on the total number of resected lymph nodes and the difficulty of classifying each lymph node. The deep-learning framework in this study can potentially serve as an objective method to augment the accuracy of diagnosis while speeding up the process. The study claims if their framework is used, only 2-6 minutes are needed for an AI-assisted pathologist to diagnose a patient’s lymph node in whole-slide images.

LAMP-enabled diagnosis of Kaposi’s sarcoma for sub-Saharan Africa

SCIENCE ADVANCES

LAMP (Loop-mediated Isothermal Amplification) is a technique for isothermal amplification of nucleic acid application, meaning that it can amplify DNA or RNA at a constant temperature, unlike PCR (Polymerase Chain Reaction) which requires temperature cycling. LAMP has several advantages over PCR, including faster amplification time, higher sensitivity, and the ability to detect multiple targets at once. This study aimed to assess the performance of LAMP-based biopsy analysis as an approach to diagnose Kaposi's sarcoma (KS) at the point of care in sub-Saharan Africa. Biopsies collected from patients in Uganda were compared to gold standard US based histopathology and the results showed a sensitivity of 97%, and specificity of 92%, exceeding the accuracy of local pathologists.

Importance: KS is a caner of lymphatic endothelial origin caused by the KS-associated herpesvirus (also known as human herpesvirus 8). KS has a high disease burden in sub-Saharan Africa where there is limited infrastructure for traditional histopathology, making timely and accurate diagnosis difficult. By using a molecular approach using LAMP-based biopsy analysis, the researchers were able to diagnose KS at the point of care with minimal training and equipment and in a timely manner.

Sybil: A Validated Deep Learning Model to Predict Future Lung Cancer Risk From a Single Low-Dose Chest Computed Tomography

JCO

This study introduces a deep learning model named Sybil that can predict the future risk of lung cancer up to 6 years from a single Low-Dose Computed Tomography (LDCT) scan without the need for demographic or clinical data or radiologists' annotations. It was developed using LDCTs from the National Lung Screening Trial (NLST) and validated on three independent datasets (NLST, Massachusetts General Hospital, and Chang Gung Memorial Hospital). Sybil had good performance in predicting lung cancer at 1 year with good concordance among the validation datasets over 6 years. The model and annotations are publicly available for further study.

Importance: The model's performance was evaluated on diverse patient populations from the US and Taiwan and showed promising results in predicting both short-term and long-term lung cancer risk. This could potentially improve current screening practices by providing additional information about future lung cancer risk with minimal disruption to normal clinical workflow. However, further evaluation through a prospective study to assess the performance and clinical benefit is required.