Molecular and Machine Learning Based Characterization of Human Skeletal Muscle to Decipher Complex Biological Processes Governing Muscle Wasting in Surgical Patients with Cancer

• Author / Creator

  Bhatt, Bhumi J

• Cancer cachexia is a multifactorial syndrome characterized by progressive loss of weight (WL), muscle, and fat tissues. Skeletal muscle wasting, in particular, is strongly associated with morbidity and mortality. Understanding the pathophysiological mechanisms underlying muscle wasting in humans is urgently needed. While clinical variables, e.g., Body Mass Index (BMI), WL, and Skeletal Muscle Index (SMI), are helpful in prognostication, these variables provide no insight into the underlying molecular mechanisms of cachexia. Molecular mechanisms are described in animal models of cachexia with unknown translational relevance; clinical studies of human skeletal muscle biopsies are sparse and did not yield consistent findings. Patient classification heterogeneity, limited sample size, aggregate cancer types, and sex, and focus only on coding transcripts also limited the value of molecular findings in humans.
  An emerging consensus is that non-coding RNAs are key players in gene regulation and may play a role in muscle homeostasis. Therefore, comprehensive human studies are needed to address these gaps. My thesis addresses these gaps, and all subsequent work describes the findings in a sex-specific context.
  I hypothesize that expression profiling of coding and non-coding RNAs from skeletal muscle of patients with cancer, when subjected to unsupervised machine learning clustering approaches, would facilitate the identification of the underlying molecular architecture (subtypes) of skeletal muscle.
  Rectus abdominis skeletal muscle biopsies (n=84; males, n=48; and females, n=36) were obtained from a surgical cohort of patients with cancer. Clinical data included age, sex, cancer diagnosis, WL, BMI, plasma C-reactive protein, and SMI determined by computed tomography.
  mRNAs, long non-coding RNAs/ lncRNAs, and small RNAs (miRNA, piRNA, snoRNA, tRNA) were profiled using high-throughput Next Generation Sequencing. Transcriptome profiles were subjected to unsupervised clustering using the integrative Non-negative Matrix Factorization (intNMF) algorithm to address the clinical classification heterogeneity. K=2 clusters (or subtypes) were identified in patients of the male and female sex. Differential Expression (DE) analysis of the subtypes identified dysregulated RNAs. However, the intricate interplay through which RNAs co-regulate each other via post-transcriptional competing endogenous RNA (ceRNA) mechanism is not studied in clinical cachexia. ceRNAs communicate through common miRNA binding sites or response elements (MREs). I identified lncRNAs and mRNAs acting as ceRNAs. Amongst the top six identified hub lncRNAs, three were common.
  Although 2 muscle subtypes were determined, there is no standard method to discern the intrinsic characteristics of subtypes. I adopted two independent benchmark paradigms and referred to them as (i) clinical and (ii) molecular and functional benchmarking.
  
  I performed clinical benchmarking based on WL grading (severity Grade 0-4) and age- and sex-specific SMI z-score. These analyses revealed that Subtype 1 in both male and female patients was proportionally associated with high-grade WL and low SMI z-score and regarded as a cachexia group; Subtype 2 had minimal WL and high SMI.
  There are no available datasets in the literature to guide functional benchmarking. I employed two experimental model systems to approach molecular benchmarking. mRNA profiles were generated from (i) human Rhabdomyosarcoma cells, a model of myoblast proliferation and differentiation. and (ii) from gastrocnemius muscle of rats bearing a colon adenocarcinoma. Comparison of pathways from DE mRNAs in these models with that in human Subtype 1 vs. Subtype 2 showed overlap. These findings led me to propose a schematic model explaining the disruption of homeostatic response from component pathways necessary for skeletal muscle integrity and function. These include Extracellular Matrix regulation, nNOS signaling, Matrix Metalloproteases, calcium signaling, and transcriptional regulatory network in embryonic stem cells.
  Overall, my work has identified two de novo clinically applicable molecular subtypes. This is the first large-scale investigational study that provides a transcriptional landscape of human skeletal muscle from patients with cancer along with multilayered RNA crosstalk. The study results in a paradigm shift from applying heterogeneous patient classification criteria to the pragmatic utility of unsupervised machine learning algorithms for pursuing molecular studies in the cachexia domain. Subtype 1 is demonstrated to be a group affected by cachexia by WL and muscle depletion.
  Data curated for the entire transcriptional atlas of mRNAs, lncRNAs, and small ncRNAs will be accessible publicly to enable cachexia researchers to examine their genes or pathways of interest.

• Subjects / Keywords

  • Cancer Cachexia
  • Machine Learning
  • Skeletal Muscle Wasting
  • Next Generation Sequencing
  • RNA sequencing
  • Human skeletal muscle biopsies
  • Small RNA sequencing
  • Molecular subtypes
  • Transcriptomics

• Graduation date

  Spring 2023

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-vjmx-g214

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.