Abstract

Objective

   Understanding the heterogeneity and pathophysiology of osteoarthritis
   (OA) is critical to support the development of tailored
   disease-modifying treatments. To this aim, transcriptomics tools are
   highly relevant to delineate dysregulated molecular pathways and
   identify new therapeutic targets.

Methods

   We review the methodology and outcomes of transcriptomics studies
   conducted in OA, based on a comprehensive literature search of the
   PubMed and Google Scholar databases using the terms “osteoarthritis”,
   “OA”, “knee OA”, “hip OA”, “genes”, “RNA-seq”, “microarray”,
   “transcriptomic” and “PCR” as key words. Beyond target-focused RT-qPCR,
   more comprehensive techniques include microarrays, RNA sequencing
   (RNA-seq) and single cell RNA-seq analyses.

Results

   The standardization of those methods to ensure the quality of both RNA
   extraction and sequencing is critical to get meaningful insights.
   Transcriptomics studies have been conducted in various tissues involved
   in the pathogenesis of OA, including articular cartilage, subchondral
   bone and synovium, as well as in the blood of patients. Molecular
   pathways dysregulated in OA relate to cartilage degradation, matrix and
   bone remodeling, neurogenic pain, inflammation, apoptosis and
   angiogenesis. This knowledge has direct application to patient
   stratification and further, to the identification of candidate
   therapeutic targets and biomarkers intended to monitor OA progression.

Conclusion

   In light of its high-throughput capabilities and ability to provide
   comprehensive information on major biological processes,
   transcriptomics represents a powerful method to support the development
   of new disease-modifying drugs in OA.

   Keywords: DMOADs, Molecular profiling, Osteoarthritis, Transcriptomics

1. Introduction

   Osteoarthritis (OA) is a highly prevalent chronic musculoskeletal
   condition, affecting up to 16% of the population worldwide and 23% of
   individuals aged 40 and over [[35]1]. As one of the most disabling and
   painful disease, OA has been ranked as the 11th highest priority by the
   World Health Organization. Furthermore, it causes a huge economic
   burden estimated to up to 2.5% of Growth National Product in Western
   countries, as a consequence of significant healthcare expenses and
   indirect costs associated with comorbidities [[36]2]. As of today,
   there is no curative treatment for OA, with current interventions
   solely based on symptomatic non-pharmacological and pharmacological
   treatments having a limited effect on function recovery and pain. Thus,
   a high unmet medical need for disease-modifying osteoarthritis drugs
   (DMOADs) persists to slow down progression of the disease. Ideally,
   future treatments should provide benefits on both function and pain,
   while delaying or avoiding the need for surgical joint replacement.

   The development of various DMOAD candidates has been unsuccessful to
   this date, mainly because OA is a heterogeneous disease with various
   pathogenic factors eventually leading to common clinical and
   radiographic manifestations. The complexity of OA stems from the
   possible involvement of several periarticular tissues beyond cartilage,
   including the subchondral bone, fat pad, tendons, ligaments, synovial
   tissue and skeletal muscle. Consequently, the cross-talk between
   cartilage and its surrounding tissues should be recognized as an
   important contributor to OA, broadly regulating cartilage remodeling
   and overall disease processes. Previous attempts to establish patient
   classifications were based on key underlying pathophysiological
   mechanisms, clinically relevant patient characteristics, disease stage,
   nature of joints involved or degree of associated inflammation
   [[37][3], [38][4], [39][5]]. However, no consensus could be established
   as these features often overlap in patients. Whereas a stratified
   representation of OA as endotypes does not exist as of today, it could
   be facilitated by the recent availability of profiling approaches
   relying upon omics technologies, well suited to better understand
   molecular changes associated with the disease [[40]6]. Next generation
   sequencing (NGS) technologies have made remarkable progress to allow
   high throughput DNA and RNA sequencing analyses in large cohorts of
   patients. In particular, RNA-seq allows to measure quantitatively the
   level of gene expression in the blood or in target tissues at the whole
   transcriptome level, thus providing for each patient comprehensive
   information regarding his/her inflammatory, immune and metabolic
   status, all being relevant to design better tailored treatments.

   In this context, we review herein the methodology and quality controls
   used in transcriptomics studies. We further report on the outcomes of
   such studies conducted in OA patients, both in affected tissues
   (cartilage, subchondral bone and synovium) and in the blood. We discuss
   how those transcriptomics studies are shedding light on dysregulated
   molecular pathways associated with OA as well as on potential new
   therapeutic targets or biomarkers useful to inform patient
   stratification and treatment strategies.

2. Methodology and quality control in transcriptomics

   The technologies currently available to assess gene expression are
   presented in [41]Table 1. Among those, real-time reverse transcription
   polymerase chain reaction (RT-PCR) has been broadly used as a highly
   sensitive and reproducible method to measure the expression of specific
   sets of genes [[42]7]. Microarrays rely upon synthetic oligonucleotides
   spotted on a microchip to assess the expression of thousands of
   different genes simultaneously [[43]8]. Whereas those methods can
   analyze multiple genes at once, RT-PCR does not differentiate between
   differentially spliced RNAs (high sensitivity, low specificity) and
   microarray lacks sensitivity and specificity. They are being
   complemented by new RNA-Seq technologies which do not require species
   nor transcript-specific probes and can help identifying novel
   transcripts [[44]9]. RNA-Seq thus encompasses a large range of gene
   expression quantification methods, allowing to detect differentially
   expressed genes (DEG) with a higher specificity and sensitivity than
   microarrays [[45]10]. More recently, single cell RNA-Seq (scRNA-Seq)
   has been broadly adopted to relate gene expression to a specific cell
   within a tissue [[46]11]. Single cell RNA-Seq can also be combined with
   focused proteomics, and further, to spatial representation within a
   tissue ([47]Table 1). One limitation is that sufficient numbers of
   cells need to be analyzed to overcome the inter-cell heterogeneity in
   gene expression profiles.

Table 1.

   Characteristics and quality controls of transcriptomics analyses.
   Characteristics
   Technologies Research purposes Advantages Limitations Quality Controls
   RT-PCR[48]^a
     * •
       Qualitative and quantitative assessment of the expression of known
       genes

     * •
       Wide dynamic range of quantification
     * •
       High sensitivity

     * •
       Focused on selected genes (hypothesis driven)
     * •
       Costly
     * •
       Potential bias increased by cycles of PCR amplification

     * •
       RIN[49]^b

                Microarrays
     * •
       Application to the expression of thousands of genes

     * •
       Species and transcript-specific detection of gene expression

     * •
       Limited dynamic range of detection

     * •
       RIN

   RNA-seq
     * •
       Qualitative and quantitative analysis of the whole transcriptome,
       including novel genes

     * •
       Comprehensive analysis of the whole transcriptome (non
       hypothesis-driven)
     * •
       High throughput analysis applicable at patient cohort level

     * •
       Costly and time consuming
     * •
       Complex data integration and interpretation

     * •
       RIN
     * •
       DV200[50]^d

   Single cell
   RNA Seq
     * •
       Analysis of gene expression at the transcriptome level in
       individual cells

     * •
       Comprehensive analyses of biological processes at individual cell
       level
     * •
       Possibility to combine with single cell proteomics and spatial
       representation within tissues

     * •
       Need to analyze sufficient numbers of cells to overcome sequencing
       coverage bias
     * •
       Low capture efficiency
     * •
       Complex data integration and interpretation

     * •
       Detected genes per cell
     * •
       UMI[51]^c counts per cell
     * •
       Percentage of detected mitochondrial genes per cell

   [52]Open in a new tab
   ^a

   RT-PCR: real-time reverse transcription polymerase chain reaction.
   ^b

   RIN: RNA Integrity Number.
   ^c

   UMI: unique bar code.
   ^d

   DV200: percentage of RNA fragments with a length over 200 nucleotides.

   Irrespective of the technology, both the process of RNA extraction
   [[53]12] and the quality control of RNA are critical to ensure the
   reliability of outcomes [[54]13] ([55]Fig. 1 A, B, C). Although RNA is
   a thermodynamically stable molecule, it is rapidly digested by
   ubiquitous RNase enzymes present in biological fluids and tissues. A
   gradual degradation of RNAs leads to a continuous shift towards shorter
   fragment sizes affecting the results of quantitative analyses. In order
   to evaluate RNA integrity, bioanalyzer electrophoresis is commonly used
   to confirm that 28 ​S and 18 ​S RNAs are properly detected. RNA
   Integrity Number (RIN) is a software algorithm routinely used to assess
   RNA quality [[56]14], with a RIN value of at least 6, commonly set as a
   cutoff for an acceptable RNA integrity [[57]15]. When using
   formalin-fixed, paraffin-embedded (FFPE) tissue samples, RNA signals
   are often more difficult to detect, resulting in low RIN values. RNA
   signals are considered as detectable when RIN is above 2 in FFPE tissue
   [[58]16,[59]17]. Quality control of such FFPE-derived RNA is then
   better performed by determining percentages of RNA fragments with a
   length over 200 nucleotides (DV200) [[60]13] ([61]Fig. 1 A, B). A
   commonly accepted threshold for DV200 to ensure a sufficient RNA
   quality is over 70 when using an Illumina platform.

Fig. 1.

   [62]Fig. 1
   [63]Open in a new tab

   Quality controls for RNA extraction and sequencing. A,B: Principal
   Component Analysis (PCA) of RNA sequencing data obtained from the blood
   of 286 patients with knee OA (APPROACH cohort). Sample clustering is
   displayed based on DV200(A) or RIN (B) values, respectively. Each dot
   represents an RNA sample from an individual patient with the following
   color code: green means high values; red color means a low value; black
   dots correspond to samples without any RIN value. No significant
   correlations are observed between RIN and DV200 values. Samples
   corresponding to red and black dots in both analyses are considered as
   outliers and removed from subsequent data analyses focusing exclusively
   on samples corresponding to green dots. C: Kernel plots of FPKM
   obtained from each of the 286 OA samples. Each line represents a Kernel
   plot displaying for individual samples the log2(FPKM) value on the X
   axis and its density on the Y axis (1C). Samples with good quality RNA
   sequencing data follow a similar pattern on the graph (blue lines). In
   contrast, poor quality samples (red lines in the bottom or upper part
   of the Kernel plot) depart from this pattern and are eliminated from
   the data analysis. D: RIN and DV200 correlation plot of 286 OA samples.
   Each dot represents a sample with RIN value on X axis and DV200 on Y
   axis. The regression line obtained by Pearson analysis is displayed in
   blue.

   QC data combining RIN and DV200 values can be represented by principal
   component analysis (PCA). A PCA analysis of RNA samples obtained from
   286 OA patients included in the APPROACH cohort illustrates that RIN
   and DV200 values are independent (with a Pearson correlation of 0.56)
   and thus provide complementary information on RNA quality
   ([64]Fig. 1D). Samples exhibiting either low RIN or low DV200 values
   appear as outliers ([65]Fig. 1A and B) and can be excluded from
   subsequent data analyses. In addition, the measurement of FPKM
   (fragments per kilobase of exon model per million reads) is a sample
   normalization method frequently used to overcome the gene-length and
   library-size effects [[66]14]. Collectively, biases in gene expression
   may reflect RNA sample or sequencing quality issues, distinguishable in
   profiles obtained with a Kernel plot representation ([67]Fig. 1C).
   Specific quality controls for scRNA-Seq are commonly based on three
   main criteria, including (i) number of genes per cell [[68]15]; (ii)
   unique bar code (UMI) counts per cell; (iii) percentage of
   mitochondrial genes per cell. The number of genes detected varies
   depending on the quality of the library and cell types used, while
   further reflecting the heterogeneity of single cell datasets [[69]18].
   The two first metrics are often evaluated together to document RNA
   quantity. Single cell samples exhibiting few (ie below 2000) expressed
   genes or low UMI counts usually contain small RNA quantities. The third
   metric reflects samples with damaged/dying cells whose cytoplasmic mRNA
   leaked out through a broken membrane, thus releasing mitochondrial mRNA
   [[70]18,[71]19].

3. Overview of transcriptomics studies in OA

   [72]Table 2 provides a comprehensive overview of transcriptomic studies
   conducted in joint tissues and in the blood of patients with OA since
   the availability of reliable quantitative transcriptomics methods (ie
   circa 2010). We reviewed the methodology and outcomes of
   transcriptomics studies conducted in human OA based on a comprehensive
   literature search of the PubMed and Google scholar databases using the
   terms “osteoarthritis”, “OA”, “knee OA”, “hip OA”, “genes”, “RNA-seq”,
   “microarray”, “transcriptomic” and “PCR” as key words. The inclusion
   criteria were as follows: (1) original articles published in
   peer-reviewed journals, (2) publications focused on primary knee or hip
   OA, (3) studies reporting transcriptomic changes at mRNA levels.
   Publications describing effects on circulating, mitochondrial or long
   coding RNA only were excluded from our review. When two publications
   used the same dataset, only the first one was mentioned.

Table 2.

   Transcriptomics studies Focusing on patients with OA.
   Studies Study objectives Sample types and patient numbers Methods Main
   Results
   Cartilage and chondrocytes
   Xu J et al., 2012 [[73]22] Transcriptomic analyses of normal and
   end-stage OA hip cartilage. Hip cartilage
   N ​= ​9 OA/N ​= ​10 CTLR Microarray (validation by RT-qPCR) 998 DEG
   between healthy femoral neck and OA cartilage associated with 71
   canonical pathways; potential role of IL-17 signaling in OA. Low
   overlap between knee and hip OA gene profiles. MMP1, ADAMTS-1 and
   ADAMTS-5 increased in knee OA but decreased in hip OA cartilage.
   Chou C.-H et al., 2013 [[74]27] Identification of OA related genes in
   site-matched tissues. Cartilage and subchondral bone from knee
   N ​= ​50 OA qPCR Up to 27 among 61 genes dysregulated in the same
   direction in both tissues. 19 genes related to OA severity. GDF10
   downregulated in both articular cartilage and subchondral bone and
   associated with severe OA.
   Soul J et al., 2018 [[75]23] Study of OA heterogeneity and patient
   subgroups for stratification. Knee cartilage
   N ​= ​60 OA/N ​= ​10 CTLR RNA-seq Unsupervised clustering identifies 2
   subgroups of patients. A 10 gene-subgroup classifier was validated by
   RT-qPCR in 16 additional OA patients. Intact OA cartilage sampled in
   both groups exhibited very low expression and no differential
   expression of pro-inflammatory cytokines.
   Steinberg J et al., 2018 [[76]24] Identification of molecular
   differences between chondrocytes from osteophytic, low- and high-grade
   articular cartilages from OA patients. Primary chondrocytes from hip
   N ​= ​9 OA RNA-seq in combination with DNA methylation and proteomics
   Higher differences in gene expression and protein abundance in
   osteophytic versus low-grade articular cartilage compared to
   high-grade. Fewer significant differences between chondrocytes from
   high- and low-grade articular cartilage. Identification of 56 genes
   differentially expressed between osteophytic and low-grade articular
   cartilages including IL6, ALDH1A2, MMP13, CLIP. Gene ontology
   enrichment analysis shows that differences are at the level of
   extracellular matrix organization, skeletal system development,
   platelet aggregation and regulation of ERK1 and ERK2 cascade.
   Ji Q et al., 2018 [[77]18] Understanding molecular mechanisms
   underlying human cartilage degeneration by analyzing the profile of
   human chondrocytes at different stages of OA. Isolated chondrocytes
   from knee
   N ​= ​10 OA sc-RNA seq and histological assays Identification of 7
   different chondrocyte phenotypes including 3 novels populations/gene
   signatures associated with human OA progression.
   He A et al., 2018 [[78]20] Study the genes and functional pathways
   deregulated in OA patients. Knee cartilage
   N ​= ​10 OAN ​= ​5 CTLR Microarray Identified 145 common aberrant genes
   in both DNA methylation and mRNA expression levels, and 26 associated
   biological pathways in osteoarthritic cartilage.
   The pathway enrichment analysis showed that the MAPK signaling is
   associated with OA development.
   Rai M ​− ​F et al., 2018 [[79]25] Assessment of changes in the medial
   and lateral tibial plateau cartilage in tandem with obesity in OA
   patients. Knee cartilage medial and lateral compartments
   N ​= ​23 Microarray
   RT-PCR BMI has a distinct and differential impact on medial and lateral
   compartments. Transcripts elevated with obesity in the medial
   compartment were enriched for metabolic processes including blood
   pressure.
   Mimpen J.Y et al., 2021 [[80]26] Comparison of the transcriptome of
   chondrocytes and synovial fibroblasts derived from cartilage and
   synovium of patients with end-stage knee OA. after treatment with
   IL-17 ​A, IL17F or IL-17AF. Knee cartilage and synovium
   N ​= ​6 OA RNA-seq
   RT-qPCR Both genes (IL17RA and IL17RC) were expressed by chondrocytes
   and synovial fibroblasts. Treatment with these cytokines induced
   activation of pathways related to inflammatory responses, complement,
   hypoxia, angiogenesis, and glycolysis and modulated genes associated
   with experimental arthritis, knee osteoarthritis, and musculoskeletal
   disease, which further highlights the potential importance of this
   cytokine in OA. Treatment of these samples with these cytokines and
   increasing dose of secukinumab (anti-IL17A) significantly inhibited
   IL17-A induced genes such as MMP1 in chondrocytes but also IL-6, PDPN,
   NKFBIZ in synovial fibroblasts.
   De Almeida R Coutinho al., 2021 [[81]21] Identification of OA subtypes
   based on pathophysiological processes and/or clinically relevant
   characteristics. Knee and hip cartilage
   N ​= ​66 OA RNA-seq Analysis of unaffected cartilage samples led to the
   identification of two OA clusters. Cluster A is enriched in
   cartilage-related pathways while Cluster B is associated with an
   upregulation of chemokine genes as well as higher joint space narrowing
   (JSN) and low osteophyte (OP) scores.
   Wang X et al., 2021 [[82]37] Understanding the changes occurring in the
   chondrocytes from OA patients, Kashin-Beck and controls. Knee cartilage
   N ​= ​1 OA
   N ​= ​1 CTLR Singe-cell RNA-seq
   qRT-PCR Identification of a new population of chondrocytes expanded in
   OA (RegC) while CPCs populations were markedly expanded in healthy
   controls.
   Subchondral bone and bone marrow lesions
   Kuttapitiya A et al., 2017 [[83]29] Assessment of cellular and
   molecular expression of bone marrow lesions (BML) in OA. Bone marrow
   lesions from knee
   N ​= ​84 OA/N ​= ​12 CTLR Microarray Comparison between OA BML and
   normal bone revealed 218 DEGs. In BML, high expression of genes
   involved in neuronal development (STMN2, THBS4, PSIP1, NYAP2), pain,
   cartilage/bone formation, ECM turnover (MMP-13, COL16A1) and
   angiogenesis was documented
   Tuerlings Met al., 2020 [[84]34] Identification of pathophysiological
   processes in subchondral bone and cartilage. Subchondral bone from knee
   and hip
   N ​= ​24 OA RNA-seq (Validation by RT-qPCR) 1569 DEG were identified
   between lesioned and preserved OA subchondral bone tissue. Of those,
   750 were upregulated and 819 were downregulated. Some neural markers
   were upregulated in lesioned compared to preserved OA, such as NGF and
   THBS3, suggesting that OA related pain originates from bone.
   Hip and knee subchondral bone showed different gene expression profiles
   with only a small DEG overlap (9%, 305 DEG). The latter include IL11
   and CHADL previously related to genetic susceptibility to OA.
   Yuan C et al., 2020 [[85]39] Classification and determination of the OA
   subtypes. Synovial and subchondral bone from knee OA
   N ​= ​227 OA
   N ​= ​5 CTLR RNA-seq Unsupervised clustering of OA cartilage samples
   identifying four distinct OA subtypes:
   - the C1-glycosaminoglycan metabolic may present a relatively earlier
   stage with no specific symptoms
   - The C2 collagen metabolic disorder represented osteophytes
   - in the clinic
   - The C3-activated sensory neuron OA subtype with high expression of
   neural markers
   - The C4-inflammation OA subtype presented high inflammation and a
   narrow joint space in the clinic.
   Synovial tissues
   Wang Q et al., 2011 [[86]30] Determination of the role for the
   inflammatory complement system in the pathogenesis of osteoarthritis.
   [87]GSE32317
   [88]GSE12021
   N ​= ​10 OA
   N ​= ​13 CTLR Microarray qPCR Gene expression of complement components
   (C1s, C4A, factor B, C9, C3, C5) are markedly higher expressed while
   the complement inhibitors (clusterin, factor H, C4b, C1 inhibitor) are
   lower expressed in synovial fluids from OA patients compared to healthy
   synovial membranes. The excessive complement activation is part of the
   OA synovitis.
   Ritter S·Y et al., 2013 [[89]31] Gene expression analysis from
   cartilage and synovium in parallel with proteomic profiling of in knee
   SF from OA patients or controls. Cartilage & synovial tissue and fluid
   from knee
   Synovial tissue N ​= ​19
   Cartilage
   N ​= ​12 OA/N ​= ​13 CTLR Microarray and comparison with 2-DIGE and
   mass spectrometry The study identified 66 proteins differentially
   expressed in OA SF compared to healthy SF. Among those, 3 dominant
   pathways were identified, related to acute-phase response signaling
   complement activation, and coagulation. Many complement proteins have a
   concordant expression in both synovium and SF suggesting that the
   synovium may be the primary source of these proteins in the SF.
   Comparison with cartilage microarray data were less clear.
   Lambert C et al., 2013 (32] Comparison of the gene expression in
   synovial cells from inflamed or normal synovial membrane in the same OA
   patient. Synovial membrane from knee
   N ​= ​12 OA Microarray Among the 896 DEG identified between normal and
   inflamed areas, a significant number of genes were categorized as
   belonging to inflammation (upregulation of IL6, IL8, CXCL5, CXCL6,
   CXCL16, CXCL2, CXCL1, TREM-1 and S100A9), cartilage extracellular
   matrix metabolism (upregulation of MMP3, MMP9) showing an imbalance
   between anabolic and catabolic genes in favor of catabolic genes. DEG
   also impact the Wnt signaling with an upregulation of Wnt-5a involved
   in chondrocyte differentiation, and angiogenesis pathway with the
   upregulation of STC1.
   Zhang X et al., 2018 [[90]9] Identification of DEGs in joint synovial
   tissues from OA vs normal individuals. Synovial tissue
   N ​= ​54 OA/N ​= ​43 CTLR Microarray (Validation by RT-qPCR)
   Identification of USP46, CPVL, FKBP5, FOSL2, GADD45B, PTGS1, ZNF423,
   ADAMTS1, and TFAM genes as being dysregulated, confirming their
   important role in the pathophysiology of OA.
   Blood studies
   Attur M et al., 2011 [[91]40] Comparative analysis of gene expression
   in PBMCs from OA patients compared to non-OA controls. PBMC from knee
   OA patients
   N ​= ​167 OA/N ​= ​37 CTLR Microarray
   Validation by qPCR Two clusters were identified from symptomatic OA
   patients based on inflammatory gene expression. Patients with
   inflammatory gene signature had higher pain scores and were at higher
   risk of radiographic progression at 2 years. at 2 years.
   Attur M et al., 2015 [[92]41] Determination whether inflammation,
   evidenced in plasma and peripheral blood leukocytes (PBLs), reflects
   the presence, progression, or specific symptoms of symptomatic knee OA.
   PBMC from knee OA patients
   N ​= ​227 OA
   N ​= ​88 CTLR Microarray qPCR Elevated expression of PGE synthase mRNA
   in PBLs was observed from patients with symptomatic knee OA vs non-OA
   controls and is associated with an increased PGE2 production. A
   dichotomization of patients based on the presence of a “PBL
   inflammatory phenotype” (above median expression levels of IL1b, TNFa
   and COX-2 determined by qPCR) revealed that patients expressing these
   inflammatory genes had more rapid radiographic progression than
   patients with expressions levels at or below the median. Finally,
   patients with high levels of IL-1b mRNA had significantly higher WOMAC
   and VAS pain scores.
   [93]Open in a new tab

   Studies have been classified according to sample types and further, by
   chronological order within each category.

   Abbreviations: ADAMTS: A disintegrin and metalloprotease with
   thrombospondin; BML: Bone marrow lesion; CTRL: Control; DEG:
   Differentially expressed genes; ECM: Extracellular matrix; FJOA: Facet
   joint osteoarthritis; FLS: Fibroblasts like synoviocytes; HSF: Healthy
   synovial fibroblast; IHC: Immunohistochemistry; JSN: Joint space
   narrowing; KL: Kellgren and Lawrence; MMP: Metalloproteinase; OA:
   Osteoarthritis; OASF: Osteoarthritis synovial fibroblast; OP:
   Osteophytes; PBMC: Peripheral blood mononuclear cells; RA: Rheumatoid
   arthritis; RASF: Rheumatoid arthritis synovial fibroblast; SF: Synovial
   fluid; ST: Synovial tissue; VAS: Visual analog scale.

   Collectively, gene expression profiling studies in OA patients were
   performed using a broad range of transcriptomics methods, ranging from
   target-focused qRT-PCR, and microarrays to next-generation RNA-Seq and
   scRNA-Seq, as described above, thus providing a genome-wide coverage
   and broad dynamic range at both research and cohort-patient levels.
   Eighteen studies focused on gene expression profiles in target tissues
   highlighting molecular changes related to pathophysiological processes
   occurring within the joint during the course of OA ([94]Table 2). Such
   analyses at a tissue level, however, require invasive biopsies
   difficult to obtain in clinical practice. Two additional studies
   reported gene expression profiles in peripheral blood leukocytes
   obtained from OA patients, providing insights on molecular
   dysregulations at a systemic level ([95]Table 2). Such studies on blood
   samples are also useful to identify circulating biomarkers of interest
   to monitor disease status and/or treatment efficacy. Analyses of genes
   differentially expressed between damaged and intact cartilage as well
   as synovial areas or bone marrow lesions within the same joint of OA
   patients provided deeper insights on the relative molecular changes
   occurring in periarticular tissues. Such approaches can help relating
   apparently independent events, such as structural changes, pain, or
   inflammation, into an integrated view to identify interlinked molecular
   pathways at the whole joint level.

   From these analyses, chondrocytes were found to play a critical role in
   cartilage degradation, mainly resulting from an activation of MAPK
   signaling [[96]20] leading to an increased production of cytokines and
   enzymes involved in the cleavage of extracellular matrix components,
   such as collagens and proteoglycans [[97][21], [98][22], [99][23],
   [100][24], [101][25]] Phenotypical alterations of chondrocytes in OA
   appear as a consequence of major changes in gene expression profiles
   when compared to normal cartilage. Of note, one hurdle in such analyses
   is that chondrocytes within articular cartilage are scarce
   (representing 1–5% of the tissue volume), even more so for OA
   cartilage. Thus, adapted methods have been set up in order to collect
   sufficient amounts of high-quality RNA from articular cartilage tissue
   and perform transcriptomics analyses [[102]24]. Other approaches
   consisted in prior isolation and expansion of chondrocytes in primary
   cultures before RNA extraction. These transcriptomic analyses provided
   converging evidence for an upregulation of genes associated with
   cartilage degradation as well as matrix and bone remodeling during OA
   [[103]21,[104]22,[105]26]. Differences in the expression of genes
   primarily associated to chondrocyte function were linked to biological
   processes leading to both the formation of osteophytes and OA
   progression. In addition, chondrocytes were found to express multiple
   proinflammatory proteins likely contributing to the pathogenesis and
   evolution of OA. However, it is still unclear whether the association
   between OA and inflammatory markers is causative or subsequent to
   cartilage degradation or mechanical dysfunction [[106]27].

   Bone marrow lesions, known to be an early event in joint damage, are
   associated with angiogenesis and fibrosis features, in addition to new
   cartilage formation and increased bone turnover. Finer-grained
   transcriptomics analyses of associated dysregulated pathways showed an
   important contribution of genes related to cartilage remodeling,
   neurogenesis, pain sensitization, as well as chemokine and cytokine
   signaling [[107]28]. Besides these findings, the differential gene
   expression analysis of joint synovial tissues from OA patients and
   healthy individuals supported the hypothesis of a low-grade
   inflammation in OA [[108]9,[109]29]. The molecular changes in
   acute-phase, complement and coagulation pathways suggested that chronic
   and progressive articular damage observed during OA occurs at least in
   part through a sustained inflammatory response associated with
   unresolved tissue injury [[110]30]. Dysregulation of angiogenesis,
   apoptosis and Wnt signaling pathways were also identified as prominent
   contributors to cartilage degradation and OA progression
   [[111]9,[112]20,[113]28,[114]31]. The co-expression of osteogenic and
   angiogenic genes and their association with tissue remodeling highlight
   that vascular proliferation and bone formation are coupled events in
   bone marrow lesions.

   Blood vessels being formed within neurovascular bundles, the
   upregulation of genes consecutive to NGF signaling and neurotrophin
   pathway activation in subchondral bone tissue likely accounts for new
   nerve formation, neoangiogenesis and pain [[115]32,[116]33]. It was
   also hypothesized that damages to innervated joint structures and OA
   synovial inflammation may cause peripheral pain as inflammation
   decreases the activation threshold of local afferent nerve fibers
   [[117]34]. As mentioned above, the activation of genes encoding for
   inflammatory mediators such as chemokines or cytokines was also
   frequently reported in target tissues, either as a consequence of
   cartilage degradation but also as a possible local cause of OA, raising
   the hypothesis of inflammation as one of the phenotypic manifestations
   of this heterogeneous disease [[118]35].

   Transcriptomics analyses also revealed differences between hip and knee
   OA, with evidence for an enrichment in epigenetic modifications such as
   DNA methylation in the knee. Although there was a poor overlap in
   differentially expressed genes between both sites, IL11 and CHADL were
   found as common dysregulated genes. Both were previously linked to
   genetic susceptibilities [[119]33]. Interestingly, new chondrocyte
   subtypes, markers of chondrocyte subpopulations and signaling pathways
   contributing to OA have been identified through the use of scRNA-seq
   [[120]36,[121]37]. Differential transcriptomics analyses were also
   applied to better understand the various phenotypes of OA, allowing to
   define subgroups of patients likely to benefit from personalized
   treatment strategies. Network analysis of RNA-Seq data allowed to
   distinguish two subgroups of patients with knee OA. Group A was
   characterized by changes in chemokine signaling, inflammasome
   activation, glycosaminoglycan synthesis, Toll receptor activation and
   innate immune responses. Group B rather exhibited dysregulations in
   complement regulation, Wnt signaling, eicosanoid receptor signaling and
   syndecan interactome. Further analysis revealed little evidence for
   inflammatory pathway activation within intact OA cartilage nor for
   activation by factors from synovial fluid [[122]23]. Nonetheless,
   common changes in circadian clock regulators were observed in both OA
   groups highlighting a possible general mechanism affecting tissue
   integrity. Furthermore, similar biological processes (including
   inflammation) exist despite the molecular heterogeneity in OA cartilage
   and synovium. However, molecularly defined patient clusters differ
   between tissues, potentially reflecting differences in tissue-specific
   disease processes [[123]24]. In this regard, two patient subgroups were
   identified with low-grade OA cartilage associated with gender and
   proton pump inhibitor prescription, while in synovium two additional
   patient subgroups were associated with inflammation, extracellular
   matrix and cell adhesion. Using an independent dataset, similar
   clusters were defined using unsupervised hierarchical clustering
   [[124]21] with considerable overlap (45%) of differentially expressed
   genes with the ones identified by Soul et al. [[125]23]. In the latter
   study, the intact cartilage in cluster A patients was enriched in genes
   encoding for cartilage and extracellular matrix-related pathways,
   whereas genes involved in chemokine signaling pathways were rather
   upregulated in cluster B. Cluster B was also characterized by high
   joint space narrowing (JSN) and low osteophyte scores.

   Recently, a new classification model was established using an
   unsupervised clustering of transcriptomic data from OA cartilage
   samples, allowing the identification of four OA subtypes associated
   with distinct functional signatures [[126]38]. The C1-glycosaminoglycan
   metabolic disorder OA subtype presented classical clinical symptoms,
   while the C2-collagen metabolic disorder OA subtype was associated with
   clinical symptoms linked to osteophytes. The C3-activated sensory
   neuron OA subtype, enriched in the youngest patients, displayed a high
   expression of neural molecules such as LRRC4C, CTNNA2, and GRID2,
   suggesting a strong susceptibility to pain in this subtype. Lastly, the
   C4-inflammation OA subtype presented high inflammation and clinical
   symptomatology of a narrow joint space.

   Early diagnosis and treatment are crucially needed to decrease the
   burden of OA. For this purpose, biomarkers detected in the blood
   possibly in the form of specific and easily monitorable gene signatures
   could facilitate the monitoring of OA patients with different clinical
   phenotypes. The search for such a signature has been the focus of some
   transcriptomics studies ([127]Table 2). Blood cell analyses suggested
   that symptomatic knee OA patients could be classified into two distinct
   clusters based on levels of inflammatory gene expression. OA patients
   with an inflammatory gene signature (including IL-1β, IL-8, COX-2)
   exhibited higher pain scores and were at a higher risk of disease
   progression in a 2-year follow-up [[128]39]. Elevated baseline
   expression of IL-1β, TNFα and COX-2 mRNA further identified a subset of
   SKOA patients at higher risk for radiographic progression. These
   molecular changes likely reflecting the low-grade inflammation of OA
   have been proposed as potential diagnostic and prognostic biomarkers
   for disease-modifying OA drugs [[129]40]. In the absence of clinically
   validated biomarkers to predict OA progression, these findings
   nonetheless suggest that transcriptomics tools and gene expression
   profiling in PBMCs from OA patients can help identifying prognostic and
   follow-up biomarkers for the disease.

4. Discussion

   OA remains a devastating, slowly progressive disease affecting millions
   of patients worldwide, still in a need for efficient disease-modifying
   therapies. Relating clinical phenotypes to well-defined patient
   subpopulations in terms of underlying pathophysiological mechanisms
   remains a key challenge to design better targeted therapies and
   envision a personalized medicine approach [[130]30]. Also, a better
   understanding of OA pathophysiology could lead to the identification of
   new therapeutic targets to address the unmet medical needs of millions
   of patients [[131]41].

   As a first step, the transcriptomics studies described in the present
   review have contributed to better understand at a molecular level some
   of the biological processes occurring during OA in articular cartilage,
   synovium tissue and subchondral bone in association with alterations in
   joint structure and pain features ([132]Table 2). Gene expression
   studies conducted on target organs also resulted in a better
   understanding of patient heterogeneity, with a first proposed
   classification in the form of two distinct molecularly defined
   clusters. Given the importance of a patient stratification to envision
   a differentiated precision-medicine strategy, further replication of
   these findings in independent datasets with clinical data integration
   are required. Transcriptomics analyses in the blood of OA patients also
   appear very interesting, most particularly if they can be in the future
   combined with gene expression profiling performed in target organs from
   the same patients. Identification of blood-borne biomarkers reflecting
   pathophysiological mechanisms occurring in the joints would represent
   an outstanding achievement with multiple applications to early
   diagnosis, patient stratification and therapeutic monitoring in OA.

   In light of the complexity of OA, transcriptomics should in the future
   be combined with proteomics, lipidomics, metabolomics, epigenetics,
   imaging technologies as well as artificial intelligence and robust
   computational methods to integrate massive and multimodal data coming
   from those analyses. Such a strategy is currently applied in the
   context of the APPROACH IMI consortium on a cohort of 300 patients with
   OA, with a clinical evaluation conducted at baseline and after 1 and 2
   years of follow-up [[133]31]. This APPROACH study aims to relate the
   outcome of multi-omics and imaging analyses with disease progression.
   Such an integrated analysis, in which blood transcriptomics will
   undoubtedly play an important role, is well suited to pave the way for
   precision medicine-based therapies better designed to treat OA patients
   on the basis of a comprehensive understanding of their biological and
   clinical specificities.

Author contributions

   HK, FDC, LL and PM have conducted the literature search to mine and
   analyze published transcriptomics studies in OA. YL, AL and SC have
   generated the data and provided the overview on quality controls for
   RNA samples. All authors have contributed to writing and critically
   reviewing the manuscript.

Role of the funding source

   This study was funded by Servier.

Declaration of competing interests

   All authors are employees at Servier. Servier is a member of the
   APPROACH IMI consortium. The authors have no conflict of interest to
   declare in relationship with this study.

Acknowledgements