Abstract

   Complexities in degenerative disorders, such as osteoarthritis, arise
   from multiscale biological, environmental, and temporal perturbations.
   Animal models serve to provide controlled representations of the
   natural history of degenerative disorders, but in themselves represent
   an additional layer of complexity. Comparing transcriptomic networks
   arising from gene co-expression data across species can facilitate an
   understanding of the preservation of functional gene modules and
   establish associations with disease phenotypes. This study demonstrates
   the preservation of osteoarthritis-associated gene modules, described
   by immune system and system development processes, across human and rat
   studies. Class prediction analysis establishes a minimal gene
   signature, including the expression of the Rho GDP dissociation
   inhibitor ARHGDIB, which consistently defined healthy human cartilage
   from osteoarthritic cartilage in an independent data set. The age of
   human clinical samples remains a strong confounder in defining the
   underlying gene regulatory mechanisms in osteoarthritis; however,
   defining preserved gene models across species may facilitate
   standardization of animal models of osteoarthritis to better represent
   human disease and control for ageing phenomena.

Joint degeneration: linking animal and human disease

   Cartilage, and other tissues in our joints, begins to degenerate with
   age resulting in pain and reduced mobility; this is termed
   osteoarthritis (OA). To understand OA better researchers have often
   used animal models to represent this disease; however, these models
   have never been fully-evaluated against human cartilage. This study
   considered the messages produced by cartilage cells in both humans and
   rats. Using a method that creates a network of messages the study was
   able to define “blocks” of cell messages that were associated with
   diseased cartilage in both the rat and human. As part of this study the
   authors also defined a set of messages that could be used to
   distinguish healthy and disease cartilage. In this way it may be
   possible to define patients with early OA that may benefit from
   therapeutic interventions. (135)

Introduction

   Disorders of cartilage and joints account for a high incidence of
   disability^[30]1 and are prevalent co-morbidities of the ageing
   population.^[31]2 Debilitating in their own right musculoskeletal
   disorders contributes significantly to the global burden of disease,
   being the fourth most prevalent disorder^[32]3 and have a wider impact
   on the rehabilitation of co-occurring pathologies (obesity, stroke, and
   cardiovascular disease), thereby representing a major health policy
   issue.

   Osteoarthritis (OA), considered a chronic, degenerative condition of
   multiple tissues that comprise a joint,^[33]4 results in the
   destruction of cartilage, the friction-free interface, leading to
   considerable functional impairment. The major cell population of
   cartilage, chondrocytes, account for the unique extracellular matrix
   (ECM), which confers compression resistance and gliding characteristics
   of normal cartilage.^[34]5 Despite considerable efforts to characterize
   the nature of degenerate cartilage the pathophysiological process is
   not fully understood, disease-associated genetic variants are limited,
   and there are no disease-modifying therapeutics available.^[35]6

   Disease complexity, arising from multiscale perturbations, makes a
   mechanistic understanding of OA difficult. OA is a complex disease
   because it involves multiple tissues, environmental factors, behaviors,
   signaling pathways and genes. For example, numerous genetic risk loci,
   epigenetic effects, inflammation associated with ageing^[36]7 and
   obesity^[37]8 and biomechanical factors contribute to joint
   degeneration. Although heritable factors account for 50% of an
   individual’s risk of developing OA, only 16 disease risk loci have been
   consistently identified^[38]9 with candidate genes such as GDF5 and
   SMAD3 harboring the most promising risk alleles;^[39]10 overall,
   multiple risk alleles are likely to contribute to OA susceptibility.
   Additionally, OA is dynamic, being progressive and chronic, and so is
   likely to involve the dysregulation of several biological systems over
   multiple timescales. As with other multifactorial diseases (e.g.,
   neurological disorders), analysis of individual components cannot
   adequately explain the properties of the whole system (the contributing
   tissues) as novel properties emerge with increasing complexity of the
   system.^[40]11

   Animal models of multifactorial disorders are used to provide a
   controlled representation of subsets of human disease and aim to
   reproduce the natural history and progression. Rodent models of
   cartilage pathophysiology are frequently employed and include
   surgical-induced (destabilization of the medial meniscus) and
   chemical-induced (monoiodoacetate joint injection) OA. The rat is
   frequently used in the study of OA; however, there is no single
   standardized in vivo model.^[41]12 Gene expression studies arising from
   these models are often poorly controlled, underpowered, combine joint
   tissues, and use multiple different gene expression analysis platforms,
   making comparison across studies problematic. Overall, animal models
   that better represent human OA are required.^[42]13

   Weighted gene co-expression network analysis^[43]14 is a systems
   biology methodology that considers the connectivity between genes based
   upon their co-expression. This method facilitates investigation of the
   global network properties of a transcriptome and provides functional
   insights into the organization of a co-expression network by utilizing
   the concept of scale-free networks.^[44]15 As the co-expression of
   genes encodes the downstream protein interactions, the study of
   transcriptional co-expression patterns can reveal emergent functional
   properties of the cellular system under investigation. Weighted gene
   co-expression network analysis (WGCNA) has been used widely to define
   candidate genes for human disorders including prognostic signatures for
   cancers, and has demonstrated preservation of functional gene modules
   between human and mouse brains.^[45]16 In this context network, nodes
   represent genes that are expressed in a sample. Edges connect nodes
   based upon their weighted co-expression across samples. WGCNA assumes
   that all nodes are connected and the connections have different
   strengths; highly connected genes within a network can be gathered as
   modules, with “hubs” being the most highly connected genes within a
   module. The modularity of networks is inherent to cell biology,^[46]17
   and biological phenomena arise from molecular interactions organized
   into functional modules. The network topology (or architecture of these
   module structures) can be compared across networks to assess
   conservation of modules in different conditions or between species.

   The system under consideration in this study was the chondrocyte,
   either as whole cartilage or isolated cells. Transcriptomic profiling
   from different environments and conditions provided information on
   perturbations to that system. The study sought to establish, from
   publically available gene expression data, a comprehensive analysis of
   the gene–gene co-expression networks from transcriptomic profiles of
   different chondrocyte phenotypes in human and rat. By performing this
   analysis on human and rat data, an understanding of the preservation of
   network module topology would inform the validity of rodent in vivo
   models of OA. Additionally, by establishing a subnetwork of genes
   associated with the phenotype of interest, osteoarthritic cartilage,
   rational therapeutic and diagnostic targets may be proposed for future
   study. This study demonstrates high preservation of modules across
   species associated with physiological functions in addition to modules
   associated with inflammatory mediators and system development that are
   characteristic of a subset of human osteoarthritic cartilage samples.
   Importantly, genes with class discrimination potential were established
   that may serve to define early cartilage degeneration.

Results

Construction of rat and human co-expression networks

   Global co-expression networks were constructed from rat (115 arrays)
   and human (129 arrays) gene expression data using 5982 genes with
   common annotation by WGCNA. Overall, 12 modules were defined for the
   human network (Fig. [47]1a, b) and 20 modules for the rat (Fig. [48]2a,
   b), inclusive of a module of unassigned genes for each species. All
   further characterizations were undertaken on these modules. An
   alphanumeric code for each module (H-human, R-rat) is provided for
   reference, Supplementary Figs. [49]4a and [50]5a. The genes assigned to
   each module are listed Supplementary Data [51]SD1 and [52]SD2.

Fig. 1.

   Fig. 1
   [53]Open in a new tab

   Definition of co-expression modules in the human (a). Hierarchical
   cluster dendrogram derived from merged human gene expression data
   (derived from n = 129 arrays and 5982 genes) defines 12 modules.
   Branches of the dendrogram represent groups of genes. Dynamic
   tree-cutting was used to define modules; where these had significant
   overlap, they were assigned the same label (arbitrary module color).
   The co-expression distance (1-topological overlap (TO)) between the
   genes is defined by the y-axis; the genes are plotted along the x-axis;
   b top band—gene modules (clusters of highly co-expressed genes) coded
   by color; unassigned genes are colored “gray”. Key modules of interest
   (H2 and H4) are annotated; the associated consensus modules (C1 to C5),
   modules found in both rat and human networks, are also defined above
   the module bar. Alphanumeric module codes are provided in Supplementary
   Fig. [54]4a; bands 2–4—selected samples showing positive or negative
   correlations with genes enriched in each module (see figure key—red
   corresponds to positive correlation). Clinical samples represent
   whole-cartilage gene expression derived from human donors; gene
   expression profiles from samples were allocated to one of three
   clinical sample groups (“Clinical Groups 1–3”); a fourth (“Articular
   cartilage”) represents ostensibly normal cartilage. The overlap between
   human and consensus modules is provided in Supplementary Fig. [55]4b.
   “Clinical Group 2” shows a positive correlation with H2 and H4 modules,
   which overlap with the C4 and C5 consensus modules; c module eigengene
   expression (y-axis) is defined for two consensus modules (C4 and C5)
   across selected samples (x-axis) relative to all samples contributing
   to the human network (“Other”). For both consensus modules there is a
   significant difference in the expression of the corresponding module
   eigengenes across samples using a non-parametric Kruskal–Wallis one-way
   analysis of variance (C4, brown—p = 1.6e−09; C5, yellow—p = 2.7e−11)
   with high expression found in “Clinical Group 2” relative to normal
   articular cartilage. Overall, whole-cartilage samples demonstrate
   heterogenous gene expression and differ in their association with
   network modules

Fig. 2.

   [56]Fig. 2
   [57]Open in a new tab

   Definition of co-expression modules in the rat. a Hierarchical cluster
   dendrograms in the rat, derived from gene expression data (n = 115
   arrays and 5982 genes) show 20 modules. An alphanumeric code is
   provided to clarify references to specific modules (Supplementary