(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Topological data analysis reveals a core gene expression backbone that defines form and function across flowering plants [1]

['Sourabh Palande', 'Department Of Computational Mathematics', 'Science', 'Engineering', 'Michigan State University', 'East Lansing', 'Michigan', 'United States Of America', 'Joshua A. M. Kaste', 'Department Of Biochemistry']

Date: 2023-12

Since they emerged approximately 125 million years ago, flowering plants have evolved to dominate the terrestrial landscape and survive in the most inhospitable environments on earth. At their core, these adaptations have been shaped by changes in numerous, interconnected pathways and genes that collectively give rise to emergent biological phenomena. Linking gene expression to morphological outcomes remains a grand challenge in biology, and new approaches are needed to begin to address this gap. Here, we implemented topological data analysis (TDA) to summarize the high dimensionality and noisiness of gene expression data using lens functions that delineate plant tissue and stress responses. Using this framework, we created a topological representation of the shape of gene expression across plant evolution, development, and environment for the phylogenetically diverse flowering plants. The TDA-based Mapper graphs form a well-defined gradient of tissues from leaves to seeds, or from healthy to stressed samples, depending on the lens function. This suggests that there are distinct and conserved expression patterns across angiosperms that delineate different tissue types or responses to biotic and abiotic stresses. Genes that correlate with the tissue lens function are enriched in central processes such as photosynthetic, growth and development, housekeeping, or stress responses. Together, our results highlight the power of TDA for analyzing complex biological data and reveal a core expression backbone that defines plant form and function.

Funding: This work was funded primarily by National Science Foundation Research Traineeship training grant (NSF 1828149 to ATM, DHC, and RV) which established the Integrated training Model in Plant And Compu-Tational Sciences (IMPACTS) program at Michigan State University. This grant funded fellows within this program (JAMK, MDR, KSA, CC, JD, RD, TBJ, HRJ, AM, EMR, AMS, JY) as well as the project-based curriculum for the Plants and Python Course that formed the backbone of this manuscript. This work is also supported by NSF Plant Genome Research Program awards IOS-2310355 to EM, DHC, and RV, IOS-2310356 to AH, and IOS-2310357 to AK, NSF Developmental Mechanisms award IOS-2039489 to AH, and NSF Biological Integration Institute award (DBI-2213983 to RV). Several students (JAMK, MDR, KSA, HMP, JP) were supported by predoctoral training award (T32-GM110523 to RV) from the National Institute of General Medical Sciences of the NIH. This project was supported by the USDA National Institute of Food and Agriculture, and by Michigan State University AgBioResearch to AMT, DHC, and RV. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Copyright: © 2023 Palande et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Surveys of gene expression capture tens of thousands of data points per sample, and this high dimensionality can be represented by a unique shape that underlies emergent biological features. This shape explains gene expression along evolutionary, developmental, and environmental trajectories, leading to innovations that have marked the successful adaptation and proliferation of plant species. To visualize this shape is to better understand what transcriptional profiles are possible and to know the boundaries or constraints that permit or limit gene expression. Here, we analyzed publicly available gene expression profiles across diverse flowering plant families and visualized the underlying structure of gene expression in plants as a graph using the Mapper algorithm. We identified unique topological shapes of plant gene expression when viewed through lenses that delineate different tissue or stress responses. These complex, emergent patterns were largely hidden by biological complexity and sample heterogeneity. Our results demonstrate the ability of Mapper to uncover these patterns in high-dimensional plant gene expression datasets and its potential as a powerful tool for biological hypothesis generation.

Data visualization lies at the heart of exploratory data analysis and provides us with a powerful tool for generating hypotheses that can later be examined using standard statistical techniques. In the era of Big Data, the development of new data visualization pipelines has become increasingly important due to the high dimensionality of the datasets generated and the need to identify patterns and structures that can then become targets for more focused studies. Just as we can look upon the shape of a leaf and derive insights into how it functions from multiple perspectives (developmental, physiological, and evolutionary), we can visualize the shape of any type of data using a Mapper graph [ 4 ]. The Mapper algorithm takes as input a filter function that describes a biological aspect of the data and uses mathematical ideas of shape to return a graph that reveals the underlying structure of the data. Even abstract data types like gene expression datasets, therefore, have a shape that we can visualize and derive insights from. For example, Nicolau and colleagues visualized the structure of breast cancer gene expression, identifying 2 distinct branches with differing underlying genotypes and prognostic outcomes that traditional statistical and bioinformatic approaches fail to resolve [ 5 ]. This structure was revealed using a pairwise correlation distance matrix as input and modeling of the residuals of each sample from a vector of healthy gene expression as a measure of disease severity. In a second example, using a lens of developmental stage on single-cell RNASeq data, Rizvi and colleagues visualized the underlying structure of gene expression during murine embryonic stem cell differentiation, revealing transient states as well as asynchronous and continuous transitions between cell types [ 6 ]. In both examples, Mapper allowed the shape of data, through a selected lens, to be visualized. The resulting topology of the graph—in the form of loops, branch points, or flares—allowed previously hidden structures to be seen and novel insights to be derived. Loops, branch points, and flares in topological data analysis (TDA)-based Mapper graphs are visual representations of patterns, transitions, and outliers in the data. They provide insights into the topological structure and organization of the data, helping to identify clusters, subgroups, and potential anomalies. Loops represent recurring patterns or relationships in the data, branch points occur when different subsets of data points exhibit distinct topological characteristics, and flares typically indicate outliers or subgroups within a larger cluster and can help identify regions of interest or anomalous behavior in the data.

Beyond a common currency that links the subdisciplines of biology, gene expression links its emergent levels. Below gene expression, the genome gives rise to transcriptional networks and protein interactions that are directly responsible for the complexity of gene expression. Above it, gene expression orchestrates cell-specific expression and the development of the organism itself, impacting phenotypes ranging from physiology to plasticity that propagate further to the population, community, and ecological levels. These features, from molecular (DNA, promoter sequences, -omics datasets) to the organismal, population, and ecological levels (life history traits, climatic data from species distributions, etc.) have been used in the past as labels and predicted outputs of machine learning models [ 2 , 3 ]. The structure—the shape—of gene expression in flowering plants is therefore a constraint that is formed by and impacts biological phenomena below and above it, respectively.

Over 300,000 gene expression datasets have been collected for thousands of diverse plant species spanning over 900 million years of divergence [ 1 ]. This wealth of publicly available datasets spans ecological niches, species, developmental stages, tissues, stresses, and even single cells, providing a largely untapped reservoir of biological information. These diverse datasets provide an opportunity to link insights from various biological disciplines, including ecology, development, physiology, genetics, evolution, biochemistry, and cell biology through a common computational and mathematical framework. These gene expression datasets have been analyzed individually for specific experiments and hypotheses, but large-scale meta-analyses across the publicly available expression datasets are largely nonexistent for plants.

Results

A representative catalog of flowering plant gene expression The vast number of gene expression datasets in plants provides a unique opportunity to search for patterns of conservation and divergence throughout angiosperm evolution, across developmental time, tissues, and stress response axes. Previous studies have tried to find common signatures that define different plant tissues or responses to abiotic/biotic stresses, but these have been limited in species breadth [7], depth [8], or had limited downstream analyses [9]. Here, we reanalyzed public expression data on the NCBI sequence read archive (SRA) and applied a topological data analysis method to map the shape of gene expression in plants. We included 54 species that captured the broadest phylogenetic diversity within angiosperms while maximizing the breadth of expression at the tissue and stress levels (Fig 1A). This includes 44 eudicots across 13 families and 9 monocot species across 2 families, as well as Amborella trichocarpa, which is sister to the rest of angiosperms. Raw reads were downloaded, cleaned, and reprocessed through a common RNAseq pipeline to remove artifacts related to the different algorithms and downstream analyses used by each group. After filtering datasets with low read mapping, our final set of expression data includes 2,671 samples across 7 distinct developmental tissues and 9 stress classifications for 54 species. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Dimensional space of plant gene expression across evolution, development, and stress. (A) Representative phylogeny of the 54 plant species included in this study. Nodes (species) are colored by plant family as denoted in Fig 1C. Dimensionality reduction of all samples by principal components (left) and t-SNE (right) are shown for tissue type (B), plant family (C), and abiotic/biotic stress (D). Individual samples are quantified and colored by tissue, family, and stress as shown in the respective bar plots. (E) Hierarchical clustering of samples with various biological features highlighted (stress, family, and tissue). Raw expression data underlying the graphs in this figure can be found in S7 Dataset, and code to regenerate analyses can be found in https://zenodo.org/records/8428609 [65]. https://doi.org/10.1371/journal.pbio.3002397.g001 To facilitate comparisons of gene expression across species, we limited our analysis to a set of 6,328 orthologous low-copy genes that were conserved across all 54 plant species using Orthofinder [10]. These sets of orthologous genes or orthogroups are mostly single copy in our diploid species and scale with ploidy in polyploid species. The orthogroups are conserved across a diverse selection of Angiosperm lineages and correspond to well-conserved biological processes. Gene ontology (GO) term enrichment analysis on the Arabidopsis thaliana loci associated with these orthogroups show enrichment for basic biological functions like “DNA replication initiation” and “tRNA methylation” at the top of the list of enriched GO terms, as well as functions specific to photosynthetic organisms like “photosystem II assembly,” and “tetraterpenoid metabolic process.” Although the remaining orthogroups contain significant biological information, they were excluded from analysis as multigene families typically have diverse functions with divergent expression profiles that would conflate downstream comparative analyses. The transcript per million (TPM) counts were summed for all genes within an orthogroup for a given species and merged into a single dataframe to create a final matrix of 6,335 orthologs by 2,671 samples. Principal component analysis (PCA) [11] and t-distributed stochastic neighbor embedding (t-SNE) [12] based dimensionality reduction show some separation of samples by different biological factors (Fig 1). The sample space is most clearly delineated by tissue, where both PC1 (explaining 25.4% variation) and t-SNE1 separate the samples into a gradient from root to leaf tissues with other plant tissues sandwiched in between (Fig 1B and 1D). This distribution largely correlates with tissue function, as the sink tissues of flowers, seeds, and fruits resolve closer to the root samples along t-SNE1 and PC1. No tissue type is separated fully by either dimensionality reduction approach. Samples from the 16 plant families are distributed throughout the dimensional space, suggesting that family- or species-level traits are not masking emergent features of distinct tissues (Fig 1C). Interestingly, abiotic and biotic stresses are similarly distributed throughout the dimensional space, with no clear grouping of the same stress across species or individual experiments. This could be due to intrinsic differences in how individual species respond to stress or to differences in the way stress experiments are carried out by different research groups. To account for batch effects and the influence of unmodeled factors, we applied surrogate variable analysis (SVA) to generate estimates of surrogate variables and their effects on our expression matrices. We identified 24 surrogate variables within the dataset, but these latent variables were intrinsically linked to the primary factors in our study (e.g., stress, tissue, and family). Removing surrogate variables would have masked much of the biology we were attempting to quantify, so we chose not to use these “data cleaning” approaches (see Text A in S1 Text for more details).

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002397

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.