Foundations of Osteoarthritis Section 3: Clinical Monitoring in OA Chapter 15: Biomarkers

Abstract

Objective:

The purpose of this overview of osteoarthritis (OA) biomarkers is to provide the non-specialist with a toolbox, based on experience acquired by biomarker researchers over many years, to understand biomarkers in general and their use in the OA field.

Methods:

We provide an update on this subject since the OARSI Primer on osteoarthritis (OA) nearly a decade ago.

Results:

Since the last update, the importance of molecular biomarkers has been increasingly recognized in the field, but no OA-related biomarkers have been adopted for routine use in clinical practice. The current lack of chondroprotective treatments for OA impairs the assessment, validation and qualification of the potential role of biomarkers as tools for monitoring disease status and patient responses to treatment of OA. Yet there is no lack of an evolving compendium of OA-related biomarkers, ever more fit-for-purpose, that could currently facilitate drug development for OA. We provide an abbreviated update and overview of specific soluble OA-related biomarkers for this new OARSI Primer on OA with OA-relevant examples encompassing the concepts of biomarker nomenclature, qualification, interpretation, measurement, reporting requirements, application to research, drug discovery and clinical care, and future needs for biomarker advancement.

Conclusion:

Appropriate biomarkers should play a role at all stages of OA diagnosis, prognosis, drug development, and treatment. The future of OA biomarker research and development holds great promise as its foundation is increasingly robust.

Introduction

Although we provide an abbreviated update and overview of specific Osteoarthritis (OA)-related molecular biomarkers, the purpose of this overview of OA biomarkers for this new OARSI Primer on OA is not meant to be an exhaustive review of all specific biomarkers that have been evaluated in OA; there are many fine reviews that fulfill this purpose¹, including the prior biochemical marker chapter of the OARSI Primer on OA of a decade ago focused on soluble markers measured in biological fluids (serum, urine or synovial fluid) by an immunoassay or by Immuno-affinity, liquid chromatography, and/or mass spectrometry assays (IA/LC/MS-MS)². Rather, this primer on OA biomarkers is intended to provide the non-specialist with a toolbox, based on experience acquired by biomarker researchers over many years, to understand biomarkers in general with examples relevant to the OA field encompassing the concepts of biomarker nomenclature, qualification, interpretation, measurement, reporting requirements, application to research, drug discovery and clinical care, and future needs for biomarker advancement.

The Case for Biomarkers in OA

It is said that a disease starts when detected by the best marker we develop to define it. The detection of OA has traditionally relied upon consideration of a patient’s medical history, physical examination, and radiographic (x-ray) images of the affected joints. Interestingly, upon the discovery of x-rays in 1895 by Wilhelm Conrad Roentgen, one of his first applications, producing an image of the hand of his wife, Bertha, was musculoskeletal³. Dependence on radiographic evidence for the diagnosis of OA ever since has resulted in a focus on anatomic derangements of the disease rather than its molecular pathogenesis.

A focus on radiographic diagnostic criteria leads to failure to confirm a diagnosis of OA in 24-85% of painful knees⁴. Conversely, knee pain has only a 23% sensitivity, and 88% specificity for a diagnosis of radiographic OA⁵. An inevitable consequence of this is that a large proportion of symptomatic but pre-radiographic OA is likely unable to be classified due to lack of gold standard criteria. Moreover, reliance on anatomic derangement (radiographic, MRI, etc.) for a definitive OA diagnosis precludes the opportunity to intervene early in the disease process, as practiced for the treatment of silent early phases of osteoporosis and heart disease⁶. Although efforts to develop classification criteria for identifying early OA have been underway⁷, they need validation based on ability to predict the subsequent development of anatomic OA (radiographic or MRI abnormality).

Molecular biomarkers can potentially fill this clinical diagnostic need by providing a molecular context for joint symptoms or providing a harbinger of joint tissue abnormalities in the early silent stages of the disease process. In addition to identifying early OA, molecular biomarkers also hold the promise of providing a means to objectively discern and monitor the distinct molecular causes, pathways and endotypes that are currently classified as one disease, despite the fact that OA has at times been referred to as a multitude of orphan diseases due to its heterogeneity of etiologies with a common end stage. Biomarker discovery could be informed by idiopathic, joint injury-related (from trauma⁸, hemophilia⁹ or other etiology) and even rare forms (such as alkaptonuria/ochronosis and Kashin-Beck Disease¹⁰) of OA, as well as from other arthritides, such as rheumatoid arthritis. The extent to which a common core set of OA biomarkers may inform the totality of OA is yet to be determined. The revelation that cartilages throughout the body share some constituents but have their own distinct differences¹¹ and different levels of innate turnover¹² suggests that a core set of biomarkers may be achievable as well as biomarkers to monitor specific joint sites or OA types from among all affected joints.

Biomarker Nomenclature

It is often believed that only symptomatic OA is clinically relevant. However, a disease classification system that takes into account joint tissue metabolism biomarkers makes it possible to go beyond diagnoses based primarily on signs and symptoms and offers the hope of identifying the stages of disease characterized by molecular derangement, prior to anatomic or clinical abnormalities¹³ when, like other diseases, OA is likely more amenable to reversal, or at least slowing, to limit the pathologic sequelae¹⁴. To use biomarkers for disease classification and drug development, a shared understanding of biomarker concepts and nomenclature, described below, is needed.

Biomarkers come in many forms; they can be broadly separated into variant (soluble peptides or proteins, genomic and imaging markers, epigenetic markers such as DNA methylation and microRNA), and invariant (DNA sequence) markers. Variant molecular biochemical markers can be highly dynamic, reflecting disease activity, and able to return to baseline during disease quiescence (Figure 1); whereas anatomic biomarkers, quantified by radiographic imaging, tend to reflect the cumulative change in state of the joint (Figure 1). Anatomic biomarkers have not yet been shown to reverse or return to a baseline state, no doubt due to the current lack of an effective disease modifying therapy required to test this potentiality.

Figure 1: Heterogeneity in disease progression, disease activity and disease status.

The level of disease activity biomarkers may be independent of the status and level of disease based on clinical or anatomic data. Based on several real life examples, elevations of soluble molecular biomarkers can predate changes in anatomic (radiographic) progression as described by Sharif et al. for cartilage oligomeric matrix protein (COMP)⁷⁹. Consequently, as OA progression is usually phasic, levels of prognostic biomarkers may be elevated during periods of high disease activity (that are coincident with or precede disease progression) but low in periods of low disease activity (that are coincident with or precede periods of no or slow disease progression). Reproduced with permission from Bihlet AR, Karsdal MA, Bay-Jensen AC, Read S, Kristensen JH, Sand JM, Leeming DJ, Andersen JR, Lange P, Vestbo J. Clinical Drug Development Using Dynamic Biomarkers to Enable Personalized Health Care in COPD. Chest 2015;148(1):16-23⁹⁹. Copyright © 2015 Elsevier Ltd. Figure recreated with BioRender.com (paid license)

In PubMed, the term “biomarker” is first found in 1977 in an article reporting serum RNase level as an indicator of renal function and “not a biomarker either for the presence or extent” of a plasma cell tumor¹⁵. In 1996, Van Gestel and Van Brummelen in ecotoxicology distinguished a ‘biomarker’ (any biological response below the individual level, measured inside an organism or in its products restricted to biochemical, physiological, histological and morphological measurements of ‘health’ that exclude behavioral effects) from a ‘bioindicator’ (a behavioral effect of an organism)¹⁶. In 1998, the National Institutes of Health Biomarkers Definitions Working Group defined a biomarker as “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention.” The term surrogate endpoint was reserved for a subset of biomarkers intended to substitute for a clinical endpoint¹⁷. These definitions have been broadly adopted for biomedical applications. To date, two biomarker classification schemes have been utilized in OA research, BIPEDS^18,19 and BEST²⁰, with 6 or 7 defined categories, respectively (see Supplementary Table 1 with practical examples). A glossary of common terms related to biomarkers is provided in Table 1.

Table 1.

Summary of useful biomarker terms.

Term	Definition	References
Accuracy	assessed by measuring recovery of the spiked analyte tested in triplicate	21
Anabolic	Synthesis
Catabolic	Degradative
Clinically meaningful limits	the maximum amount of variability in laboratory test results that will not affect patient care	22
Direct Biomarker	Mechanistic, in pathway of disease
Duplicate	quantification of a biomarker from two independent volumes of the same sample for the purposes of assessing precision and variability
ELISA	Enzyme Linked Immunosorbent Assay, an immunological assay technique used primarily for measuring an antigenic region of a protein and making use of an enzyme bonded to an antibody to detect the antigen
Endotype	a subtype of a condition, which is defined by a distinct functional or pathobiological mechanism
Good biomarker practices	Use of best practices in biobanking (including standardized specimen acquisition, storage and preparation, and acquisition of high-quality associated phenotypic data), translational research (standardized and reproducible analytical methods to ensure the results are reproducible and usable) and data stewardship (data are findable, accessible, interoperable and reusable—FAIR)	23
Harmonization	the process of ensuring that the results of different laboratories using different clinical laboratory tests at different times to measure the same substance are equivalent within clinically meaningful limits	22
Indirect Biomarker	An indirect sign of a pathway that is not fundamental to the key disease processes
Invariant	Genetic markers (polymorphisms) in DNA
LOD (limit of detection)	The minimum concentration that can be detected reliably based on a signal-to-noise approach; a signal-to-noise ratio between 2:1 or 3:1 is generally considered acceptable for estimating the detection limit; it is determined as the concentration corresponding to 2-3 standard deviations above the background (zero calibrator). Some assays use Minimal detectable dose (MDD) which is determined by adding two SDs to the mean optical density (OD) value of twenty zero standard replicates and calculating the corresponding concentration. At LOD and LOQ calculations, signal of analyte (S) should be 3 or 10 times of blank signal (N), respectively.	DOI:10.1016/j.jfda.2015.04.009
LOQ (limit of quantification)	The quantification limit of an individual analytical procedure is the lowest amount of analyte in a sample which can be quantitatively determined with suitable precision and accuracy.
Multiplex	quantification of multiple biomarkers in a single sample volume
Precision	measured using a minimum of five determinations per concentration. The precision determined at each concentration level should not exceed 15% of the coefficient of variation (CV) except for the LLOQ, where it should not exceed 20% of the CV.	21
Singleplex	quantification of a single biomarker
Qualification	providing evidence that biomarker is linked with a certain biological process and clinical endpoint; process applied to a particular biomarker to support its use as a surrogate endpoint in drug discovery, development or post-approval and, where appropriate, in regulatory decision-making	24
Quantification	the act of counting and measuring
Reliability	Reliability tells you how consistently a method measures something. The same method applied to the same sample under the same conditions should yield the same results. If not, the method of measurement may be unreliable.
Reproducibility	assessed by replicate measurements using the assay	21
Singlicate	quantification of a biomarker from a given volume of sample
Stability	Minimal variation under expected sample handling and storage conditions, including, among others, effects of freeze-thaw. For a stable biomarker, a mean value of a minimum of three concentrations in the range of expected study sample concentrations, should be within 15% of the nominal value except at LLOQ, where it should not deviate by more than 20%.	21
Standard curve	Assay standards or calibrators representing serial dilutions of one known concentration of the analyte across a range of concentrations near the expected (but unknown) concentration The concentration of the unknown may be calculated from the mass in the assay
Standardization	the achievement of equivalent results by different clinical laboratory tests conducted by different laboratories using reference samples that can be traced to a reference measurement procedure	22
Surrogate biomarker	A “surrogate marker” can be defined as “a laboratory measurement or physical sign that is used in therapeutic trials as a substitute for a clinically meaningful endpoint that is a direct measure of how a patient feels, functions, or survives and is expected to predict the effect of the therapy.”5 The primary difference between a biomarker and a surrogate marker is that a biomarker is a “candidate” surrogate marker, whereas a surrogate marker is a test used, and taken, as a measure of the effects of a specific treatment. Russell Katz, Biomarkers and Surrogate Markers: An FDA Perspective, NeuroRx 1(2): 189-195, April 2004	http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=534924
Theragnostic	A molecular diagnostic that guides treatment
Validation	The action of proving good technical performance characteristics of an assay, such as, for example, precision, accuracy, detection limit and robustness.	21, 25
Validity	How accurately a method measures something	26
Validity—face validity	The content of the test appears to be suitable to its aims	26
Validity—construct validity	The test measures the concept that it s intended to measure?	26
Validity—content validity	The test is fully representative of what it aims to measure	26
Validity—criterion validity	The results correspond to a different test of the same thing	26
Variant	Dynamic markers such as protein, RNA, and miRNA
‘wet’ and ‘dry’ biomarkers	soluble and insoluble biomarkers

Molecular Biomarkers in OA

Although we have many biomarkers under investigation for OA (see Table 4), we have no biomarkers that are formally accepted as surrogates that could substitute for a clinical endpoint. Nevertheless, subsets of imaging and soluble biomarkers are associated with clinically relevant outcomes, such as pain, functional status and total joint replacement; these include magnetic resonance imaging (MRI) assessed cartilage thickness loss over two years, change in shape of the knee femur and tibia, high urinary CTXII and low serum PIIANP, reflecting high degradation and low synthesis of type II collagen, respectively²⁷. Although there are well known racial and ethnic disparities in joint replacement surgery²⁸, this outcome is particularly compelling as a clinically relevant endpoint; the ability of several select OA-related biomarkers to predict this outcome underscores the emerging level of surrogacy for a small subset of OA progression-related biomarkers.

Table 4.

Representative Osteoarthritis Biomarkers.

Biomarker	Process	Description	Examples of Clinical Associations
Bone
C1M	Matrix destruction	MMP mediated tissue destruction. MMP-cleaved fragment of the interhelical region of Col1 and destroyed by cathepsin K 42	C1M are associated with pain outcomes 43 and pharmacodynamic responses to diet/exercise 44 and anti-inflammatory interventions 45
CTX-I	Bone resorption	A C-terminal crosslinked and isomerized fragment of collagen type I generated by cathepsin K cleavage during osteoclast-mediated bone resorption of mature collagen	Reflects Cathepsin K and MMP mediated bone resorption 46; in FNIH OA consortium, CTX-I was associated with disease progression 47; also modified with anti-resorptive treatment in OA 48,49
αCTX-I	Bone resorption	A C-terminal crosslinked fragment of Col1 generated by cathepsin K cleavage during osteoclast-mediated bone resorption of newly synthesized collagen. A measurement of young bone resorption such as woven bone in cancer 50 and subchondral remodeling on OA	Associated with subchondral bone turnover, JSN, osteophyte progression 51 and prognostic for knee OA progression in the important FNIH study 47
ICTP	Matrix destruction	MMP-cleaved type I collagen generated by many cell types including osteoclasts	Associated with JSN in hip OA 52
NTX-1	Bone resorption	An N-terminal crosslinked fragment of type I collagen generated by cathepsin K during osteoclastic bone resorption	OAI-FNIH, NTX was associated with disease progression 47
OPG	Osteoclast function	OPG is the natural ligand to RANK-L and is important for bone turnover in combination with RANK-L, SOST and DKK-1	Osteoclast activity is often investigated in osteoporosis studies 53
Osteocalcin	Bone formation or resorption	More bone specific than type I collagen; synthesized by osteoblasts, odontoblasts, and hypertrophic chondrocytes 54; intact protein indicative of bone formation; fragments indicative of bone resorption 46	Associated with progression of knee OA and severity of hand OA (reviewed in 46; used in most bone pharmacology studies 55
PICP	Bone formation	The C-terminal pro-peptide is enzymatically released from newly synthesized pre-pro-collagen prior to incorporation of the collagen molecule into the extracellular bone matrix. A measurement of bone and soft tissue formation	Association with progression of knee OA 56; Induction by pro-anabolic bone treatment and inhibited by anti-resorptives
PINP	Bone formation	The N- and C- terminal pro-peptide are trimeric, globular peptides enzymatically released from newly synthesized pre-pro-collagen prior to incorporation of the collagen molecule into the extracellular bone matrix. A measurement of bone and soft tissue formation	Prognostic and diagnostic for knee OA, especially for progressive osteophytosis 56
RANK-L	Osteclast formation	RANK-L is essential for osteoclast formation	Osteoclast activity is often investigated in osteoporosis studies 53
SOST/DKK-1	Osteoblast activity	SOST and DKK1 are Wnt signaling inhibitors important for osteoblasts and bone formation	SOST and DKK-1 are often used as serological biomarkers in bone research 57
TRACP	Osteoclast number	A specific enzyme for osteoclasts	Osteoclast activity is often investigated in osteoporosis studies 53, 58
Cartilage
CTX-II	Cartilage degradation	MMP degraded type II collagen	CTX-II was associated with progression of OA 59, and diagnosis, and responded to therapy 60–62. CTX-II was prognostic for progression in the important FNIH study 47
C1,2C	Connective tissue degradation	Collagenase mediated helical breakdown of Col2 andCcol1	Increase in serum C1,2C was associated with cartilage thinning (5 year follow-up) 63
C2C-HUSA	Cartilage degradation	MMP-mediated degradation of type II collagen	C2C concentrations were correlated with CTX-II, ARGS, osteocalcin, osteopontin and IL-8, but not structural joint injury by MRI 64. C2C-HUSA was prognostic for progression in the important FNIH study47
C2M	Cartilage degradation	MMP-mediated inter-helical degradation of Col2 quantified by serum/plasma ELISA	C2M was associated with KL-2 score and levels of chronic inflammation 65
Coll2-1	Cartilage degradation	MMP-mediated inter-helical degradation of Col2	Curcumin treatment reduced Coll2-1 serum levels, suggesting that the marker may act as a pharmacodynamic marker 66, 67; Coll2-1 is both a biomarker of OA and activates innate immunity as a disease associated molecular pattern31
Coll2-NO2	Cartilage degradation	MMP-mediated inter-helical degradation of nitrosylated Col2	Baseline levels were negatively associated with incidence of knee OA 67, 68
PIIANP	Cartilage formation	Type IIA-collagen N-terminal pro-peptide	PIIANP was associated with 2-year radiographic progression in OAI-FNIH and severity of disease 47, 69. PIIANP was prognostic for progression in the important FNIH study 47; a positive balance of collagen formation (by serum PIIANP) exceeded degradation (by urinary CTXII) from 6-24 months following knee joint distraction 70
PIICP/ CPII	Cartilage formation	Type II-collagen C-terminal pro-peptide	Levels in multi-site OA patients were lower than in hip OA only 71
PRO-C2	Cartilage formation	Type IIB-collagen N-terminal pro-peptide	Serum PRO-C2 is significantly higher in controls (KL0/1) compared to OA groups (KL2/3/4 and low PRO-C2 is associated with 2-year radiographic progression in the oral calcitonin trials 72–74
C10C	Hypertrophic chondrocytes	Collagen type X is expressed by hypertrophic chondrocytes	A urinary Type X collagen epitope was diagnostic for OA, associated with more severe OA and produced by hypertrophic chondrocytes 75
TIINP	Cartilage degradation	Collagenase mediated helical breakdown of Col2	uTIINE was able to distinguish between OA patients and healthy controls, and between symptomatic and asymptomatic radiographic OA patients 76
COMP	Cartilage turnover	COMP (cartilage oligomeric protein) is important in the process of collagen formation	COMP is associated with incidence and progression of OA 77, 78, serum elevations are phasic and predate radiographic progression and occur with joint replacement 79
ARGS	Cartilage destruction	When aggrecan is destroyed by the protease ADAMTS, this unique fragment is generated	ARGS have demonstrated pharmacodynamic potential 80, 81 and is a well-used biomarker for joint destruction in OA
FFGV	Cartilage destruction	When aggrecan is destroyed by MMP activity, this unique fragment is generated	FFGV in particular together with ARGS activity been used in vitro for assessment of drug mechanism of action 82, 83
CS846	Chondroitin sulfate	A side chondroitin sulfate chain of aggrecan that may reflect pathological increase in turnover of newly formed aggrecan, the most abundant proteoglycan in the cartilage	CS846 has been shown to be associated WOMAC stiffness 84
HA	Hyaluronan	HA is the backbone by which aggrecan binds to in cartilage and contributes to the viscoelastic properties of synovial fluid	HA has been shown to be a predictor of disease progression 85, and prognostic for progression in the important FNIH study 47
Synovium
MMP-3	Synovial inflammation	MMP-3 is expressed by synoviocytes	MMP-3 response to inflammatory treatment, is released from synovial explants in culture 86, is elevated in joint fluid after injury 87 and predicts joint space narrowing 88
C3M	Synovial inflammation	Type III collagen degraded by MMP	Respond to anti-inflammatory treatment 89 and predicts response to treatment 90
C1M	Soft tissue inflammation	MMP mediated soft tissue inflammation and destruction	C1M is predictive for progression of joint destruction in RA and respond to treatment for anti-inflammatory diseases 89–91
CRPM	Soft tissue inflammation	A MMP derived fragment of CRP, which is released during tissue remodeling	CRPM respond to anti-inflammatory treatment, predicts response to treatment and is associated with progression of OA 65, 77, 89

Information derived from the references cited in the table and from Bay-Jensen et al. 2019 ⁹². The table describes biomarkers used in osteoporosis and osteoarthritis research, with selected examples from rheumatoid arthritis with inflammation driven biomarkers that have proven valuable in different clinical settings

Molecular biomarkers can be monitored in blood (serum or plasma), urine, and/or synovial fluid (for methods related to processing and storing of these samples, see Supplementary Methods). Biomarkers measured in biological fluids proximal to a tissue (such as synovial fluid or cerebrospinal fluid) have a greater likelihood of being directly associated with the causal pathway of a disease. A biomarker can be a direct (mechanistic) or indirect (an imperfect but much more common) type of biomarker²⁹ (Figure 2). Direct biomarkers rooted in the pathogenesis of disease, can be considered ‘superior’¹⁴. The OA field is in the enviable position of being able to obtain a biological fluid proximal to disease (synovial fluid), and having biomarkers that reflect disease activity (direct and indirect biomarkers) and that mediate disease (direct biomarkers); this is the case because synovial fluid contains inflammatory mediators associated with disease progression, and fragments released by joint tissue catabolism (such as fibronectin fragments, collagen fragments such as the biomarker Coll2-1 and other disease associated molecular proteins or DAMPs) that reflect disease activity and stimulate the innate immune system^30,31.

Figure 2. Diagram of types of biomarkers.

Direct biomarkers are in the disease pathway and are therefore mechanistic biomarkers that reflect disease activity and contribute to disease pathogenesis. Changes in direct biomarkers would have the potential to reflect changes in disease state over time or in response to a drug. Indirect biomarkers report on some aspect of the disease but are not directly in the disease pathway. Created with BioRender.com (paid license)

In contrast to synovial fluid biomarkers, systemic (blood serum or plasma and/or urine) biomarkers in skeletally mature adults can reflect whole body burden of disease as demonstrated by the correlation of urinary CTXII, serum cartilage oligomeric matrix protein (sCOMP) and serum hyaluronan (sHA) with total body burden of radiographic joint space loss and osteophyte³² These results suggest that biochemical biomarkers of the systemic circulation or urine provide holistic measures of disease activity at the whole person level that in the future can inform treatment effects on multi-joint OA. However, systemic biomarkers are typically used to pick “the needle out of the haystack”, i.e. diagnose or prognose local disease of a single joint despite the fact that this local OA usually exists in a “sea of OA” – the additional affected joints at contralateral or other sites. Moreover, systemic biomarkers from articular cartilage may also originate from the cartilaginous growth plates throughout the body in skeletally immature individuals and animals. As long as any growth plates are open (closure occurs at different times/ages according to sex and bone location in the body), they can confound interpretation of biomarker results for any joint site. This challenge can be overcome with age-matched controls as a reference population.

The Challenges of Biomarker Qualification in OA

Molecular biomarkers identify abnormalities of the structure and function of body organs and systems that constitute disease³³. Efforts to qualify biomarkers are typically based on contexts of illness (symptoms) or anatomic (imaging-based) outcomes. Molecular biomarkers have the potential to differentiate different subtypes of disease, termed endotypes, that are rooted in different pathobiologies (Figure 3A). There is a need to identify molecular endotypes for the purposes of precision medicine to select the patients most likely to benefit from specific treatments. However, the ability of a biomarker to be qualified, that is reflect these outcomes or context, is hampered by what has been described as the ‘cracked mirror’ problem (Figure 3B). The cracked mirror represents the traditional symptomatic or anatomic outcome measures for OA³⁴ or eminence-based disease phenotypes of OA; these measures and/or phenotypes may not reflect the same biological pathways or time courses of disease alterations as molecular biological markers and may not adequately represent true biological subsets of disease. Moreover, unlike systemic molecular biomarkers, imaging biomarkers typically inform on the state of local and not whole body burden of disease. Thus, the ultimate extent to which a molecular biomarker can be qualified is limited by the best gold standard outcome/context (mirror) used for the qualification process. Successful qualification also requires reliable biomarker measurement as described below.

Figure 3. Qualification challenge for OA biomarkers.

A) An endotype is a subtype of a condition, which is defined by a distinct pathobiological mechanism. Thus, an endotype could potentially be identified by a specific molecular biomarker. Greater confidence in assigning endotypes could be achieved by multiple biomarkers specific to particular pathobiological mechanisms, such as inflammation in the lower example. B) However, the ability to qualify a molecular biomarker for a particular endotype is limited by the degree to which the context (mirror) for qualification--traditionally symptoms and/or imaging abnormalities of OA—are also representative of the pathobiological process of the particular endotype. Image in (B) reproduced with permission from Kraus VB. Do biochemical markers have a role in osteoarthritis diagnosis and treatment? Best Pract Res Clin Rheumatol 2006;20(1):69-80³⁴, Copyright © 2006 Elsevier Ltd.

important Issues Related to Biomarker Measurement

Numerous methods are available for measuring variant, soluble biomarkers; the most common methods are enzyme-linked immunosorbent assays (ELISA), mass spectrometry proteomics, and chemical assays. In general, ELISA assays are the gold standard for protein-based biomarker quantification. Regardless of assay type, technical and analytical validation under ‘Good Laboratory Practice’ (GLP) or ‘Good Manufacturing Practice’ (GMP) conditions are necessary for the establishment of a clinically useful assay. In general, good biomarker practices require best practices in biobanking (including standardized specimen acquisition, storage and preparation, and acquisition of high-quality associated phenotypic data), translational research (standardized and reproducible analytical methods to ensure the results are reproducible and usable) and data stewardship (data are findable, accessible, interoperable and reusable--FAIR)²³. Guidance on bioanalytical validation, as well as recommended best practices in OA biomarkers research, are provided in the OARSI Clinical Trials Recommendations document on soluble biomarkers²¹. A list of common parameters influencing the results of biomarker analyses is provided in Supplementary Table 2.

It is important to recognize that the units of measure attached to a biomarker concentration, such as ng/ml or pg/ml etc., are arbitrary as they are based on the standards/calibrators used as the basis for the unit calculation. A recent real-life example, testing three different commercially available assays for neurofilament light chain protein (NfL), yielded highly correlated results across the three assays but units of either pg/ml or ng/ml for the same sample. In this case, a read out of pg/ml did not indicate an assay with higher analytical sensitivity as these units were simply a result of a protein standard assigned a concentration three orders of magnitude lower than a similar standard of another manufacturer. The analytical sensitivity, the lowest quantity of the given analyte that an assay can detect is termed the Limit of Detection (LOD) or Lower Limit of Detection (LLOD). This is in contrast to diagnostic sensitivity, which is related to the ability of an assay to correctly identify individuals with the disease. One of the many challenges of biomarker research is that biomarkers of interest are often present in very low concentrations and therefore small increases that may be biologically relevant and associated with diseased states are unable to be identified due to lack of assay sensitivity.

Reliability is about the consistency of a measure, and validity is about the accuracy of a measure. Results from different laboratories and/or assay methods are both more reliable and valid if they are harmonized. Laboratories tend to harmonize laboratory values when there is an effective treatment that has measurable clinical outcomes that benefit from standardized monitoring. There are several ways of achieving laboratory test harmonization (standardization) of results across a field or laboratories²². For one, harmonization to an internationally agreed upon standard, that can be shared across assay providers, provides inter-laboratory assay agreement by assigning an absolute concentration to lots or batches of different standards. In the OA biomarker field, a precedent was set by Dr. Eugene Thonar for use of a universal standard (highly purified keratan sulfate-2) for the keratan sulfate (5D4) ELISA assay³⁵; unfortunately, to our knowledge, this precedent has not been followed for other OA-related assays. For another, a shared set of samples (ideally sustainable reference materials) can be used to determine relative equivalencies of assay concentrations across laboratories. There are other methods of comparing biomarker results that rely on statistical approaches, such as z scores (see below) and meta-analyses. Moreover, comparison across platforms is facilitated by using the same unit of measure. For instance, blood sample analyses by quantitative mass spectrometry were typically reported on a unit mass basis; however, recently, analyses are more often reported on a unit volume basis and thereby better able to be related and compared to traditional ELISA assays and clinical tests, that are also usually reported on a unit volume basis.

Researchers need to be aware of interferents—substances in a specimen that cause interference with the analysis of another substance. In this case, interferents are substances that cause a significant difference in the measured biomarker result due to the effect of another component or property of the sample. An appreciation of this concept can aid proper performance of biomarker assays and interpretation of results. An interferent can be either endogenous (originating in the sample such as hemoglobin or bilirubin) or exogenous (such as a drug or its metabolites in the sample). Because effects of interferents can be concentration-dependent, a change in concentration of interferent in a sample could be misinterpreted as a change in the status of a patient or study participant. It is important to recognize that an interferent could increase or decrease the apparent biomarker result depending on the nature of the interfered. To test for interference, a sample from a single donor or pool of donors is spiked with a range of concentrations of potential interferents (a list of routine interferents is provided in Table 2). Screening can be performed with both low and high analyte pools, as interference may be dependent upon analyte concentration. A concentration variance of >20% is considered significant. For most experiments, 4-6 mL of a sample or sample pool should be sufficient for testing, using half for the control (no added interferents) and half for the test pool (sample with interferents added). In the absence of such definitive information, a sample can be analyzed undiluted and at multiple dilutions; an apparent increase in concentration with dilution is evidence for the presence of an interferent inhibitor. Due to the presence of assay inhibitors in most biospecimens, samples such as serum or plasma are generally diluted at least two-fold for analysis. To learn about other methodologic issues, such as the importance of standard operating procedures, reagent characterization and storage, matrix effects, calibration curves, selectivity, and accuracy etc., see Kraus et al. 2015²¹. Suggested standard operating procedures for obtaining and processing biospecimens (serum, plasma, urine, synovial fluid, small volume joint lavage) are provided in Supplemental Methods.

Table 2.

Routine interferents that can impact biomarker assays^*.

Common endogenous interferents	Interferent concentrations to test	Examples of causes of high interferent concentrations	Examples of assays impacted by interferent
Hemolysate (Hemoglobin)	0 - 500 mg/dL	hemolysis	Potassium, aspartate transaminase (AST), amylase, calcium, phosphorus, magnesium, total protein, and total and direct bilirubin
Triglycerides	0 - 1000 mg/dL (CLSI recommendation 3000 mg/dL)	hyperlipemia	phosphorus, creatinine, total protein, calcium, hemoglobin
Total Protein (from Albumin and gamma-globulins)	0 - 12 g/dL	dehydration, multiple myeloma	carbon dioxide, apolipoprotein AI, and apolipoprotein B
Bilirubin, conjugated	0 - 20 mg/dL	hemolysis, gall bladder disease	Creatinine, total protein and assays that use a peroxidase as the detection system
Bilirubin unconjugated	0 - 20 mg/dL	liver disease	assays that use a peroxidase as the detection system

^*modified from information available online from Sun Diagnostics (https://sundiagnostics.us/assurance/); CLSI, Clinical and Laboratory Standards Institute

Minimal Reporting Requirements

Researchers routinely examine the impact of covariates, such as age, sex, and body mass index (BMI) on biomarker associations with outcomes. However, as aptly demonstrated by recent evaluation of C-reactive protein in nutrition studies, reporting of assay technical performance in publications is not standardized³⁶; assessment of 20 publications showed that 80% failed to report the LLOD or LLOQ, 65% failed to report assay coefficients of variation, 20% failed to report the assay/kit manufacturer and 90% failed to report whether duplicate or singlicate analyses were performed. Standardization of laboratory practices and their reporting would result in reduced error and variation arising from inconsistencies in specimen handling, assay selection, and assay performance and thereby enhance the reproducibility of results within and between laboratories. To facilitate such standardizations, it is important to provide sufficient information on assay methodology in manuscripts reporting biomarker results. Suggested minimal reporting requirements for biomarker results are listed in Table 3.

Table 3.

Suggested minimal reporting requirements for biomarker publications.

Consideration	Assay characteristic	Comments
Biospecimen protocols	Specimen collection, handling and storage conditions and duration	Could include details on number of freeze-thaw cycles; fasting/non-fasting status of individual
Analytical Considerations	LOD	Concentration that is reliably distinguished from “analytical noise”
	LLOQ	May be higher than the LOD and is the lowest concentration that is acceptably quantified by a particular assay; typically refers to the concentration of lowest standard on the calibration curve; alternatively, the lowest concentration at which the CV of the calculated concentration is <20% and the value within 80-120% of the known value
	Data handling method below LOD/LLOQ	For instance imputation, if so what type of imputation; or exclusion of sample, patient or biomarker from analysis
	ULOQ	This is the highest calibrator concentration; alternatively, the highest concentration at which the CV of the calculated concentration is <20% and the value is within 80-120% of the known value.
	Data handling method above ULOQ	For instance, re-run sample at higher dilution or impute or exclude
	Inter-assay and/or intra-assay CV	Inter-assay (between plates) and/or intra-assay (within plate) CV
	Specific analyzer and/or assay manufacturer	Include manufacturer product number
	Duplicate measurements performed for each sample	If duplicate analyses cannot be performed due to limited availability of sample volume, then either a pooled sample control should be run in duplicate on all plates and/or a small number of samples with sufficient volume should be run in duplicate on all plates
Statistical Considerations	Understanding the assumptions needed for valid statistical inference and assessing validity of those assumptions for the data at hand	For instance, assessing normality of biomarker data before performing a t-test
	Adjustment for multiple testing	Avoid p-value hacking
	Assessing stability of results using cross-validation and/or bootstrapping	For instance, cross validation of areas under the curve [AUC] from receiver operator characteristic curves
	Accounting for paired/longitudinal study design	For instance, use of generalized estimating equations (GEEs) to control for inter-individual correlation when examining synovial fluid biomarker results from both knees of the same subject
	Sensitivity analysis	For instance, assessing how results change when outliers are included versus excluded

LLOQ, lower limit of quantification; LOD, limit of detection; ULOQ, upper limit of quantification; CV, coefficient of variation

In addition to adequate reporting of analytical or technical performance of the assay, it is important to provide details regarding methods related to statistical analyses of the biomarker results. Although analyses intended to be discovery or hypothesis generating do not necessarily require adjustment for multiple testing or comparisons, it is helpful to disclose how this issue was handled. Valid statistical inference relies on specific assumptions about the data being true, or at least reasonable/approximately true. Which exact assumptions are needed depend on the particular statistical analysis performed. Many parametric tests, such as the Welch 2-sample t-test comparing means, make a normality assumption about the data. Biomarker data, particularly when generated from a small number of samples, often make this normality assumption untenable; in this case, the data can be transformed (e.g. with logarithms, see also Tukey’s ladder of powers, etc.) such that the parametric assumption is valid so parametric analyses can be applied. Alternatively, for non-normal data, non-parametric statistical analyses can be used. Often predictions are codified as AUCs; these can be cross-validated to provide a more conservative estimate of predictive capability. There are no set rules regarding missing values, only different approaches to dealing with this very common challenge. Some ways of dealing with missing biomarker values include imputation of missing values (putting in plausible guesses for the missing values) or exclusion of the biomarker from analysis (for instance when more than 20-25% of values are missing). If a value is <LLOD then it is low, missing not at random; however, if a value is >LLOD and <LLOQ, this is a measurable value with questionable reliability. When a value is ‘missing not at random’ it can be assigned a value such as ½ LLOD, or a random value between zero and LLOD, or dichotomized (whereby values below LLOD are assigned value=0 while values above LLOD are assigned value=1). This latter approach can be especially useful for discovery research. However, when possible, it is always preferable to analyze a biomarker as a continuous or an “ordinal” variable (multiple levels) rather than a dichotomized variable, as dichotomization reduces statistical power³⁷; in fact, dichotomizing a variable at the median reduces power by the same amount as would discarding a third of the data³⁷. If there is a concern that a linear regression would not truly represent the relation between the outcome and predictor variable, biomarker transformation (such as a log transformation) can be explored. A good rule of thumb regarding reporting of methodology of biomarker studies, ‘when in doubt, disclose’; such disclosure will better inform the field, facilitate assay interpretation and harmonization, and thereby accelerate advances in the field of OA biomarker research.

Concepts to Inform Biomarker Interpretation

Biomarkers are dynamic indicators that change continuously. Concentrations over time reflect net production and clearance. The bathtub analogy has been used to aid interpretation of measured biomarker concentrations in a particular matrix, e.g. blood, urine, tissue, etc. For instance, the measurement represents how much water (biomarker) is currently present in the bathtub (the matrix); this biomarker concentration however, represents the net effect of how much water (biomarker) is running into the bathtub (that could be estimated by a tissue/molecular formation biomarker) versus how much water is running out of the bathtub (that could be estimated by a tissue/molecular degradation biomarker). Therefore, an ideal panel of biomarkers, to be indicative of tissue homeostasis, would include indicators of both anabolic and catabolic processes; such indicators could come in the form of different molecular entities or variations or fragments of the same molecular entity.

It is now apparent that OA is a disease of joint tissue degradation as well as inadequate joint tissue reparative response³⁸. The ability to measure both degradation and synthesis of a joint tissue component is therefore expected to provide a more accurate indication of the trajectory of disease in an individual than either type of indicator alone. As singular molecular entities, fibrillar collagens are particularly well suited as indicators of tissue homeostasis and joint remodeling because biomarkers of both catabolism and anabolism exist for these collagens³⁸. For instance, such biomarkers are available for type II collagen (PIIANP, PIIBNP, CTXII) used in OA research and type I collagen (PICP, CTXI and NTXI) used for osteoporosis research and clinical surveillance³⁸. In addition, ratios of a catabolic to an anabolic epitope have been evaluated in hopes of providing a better predictor of a patient’s likelihood of disease progression³⁹.

Degradation of joint tissues plays a key role in OA. For this reason, biomarkers quantifying neoepitopes generated in response to molecular degradation of joint tissue components are especially useful in OA as indicators of joint disease and joint tissue catabolism. Neo-epitopes are regions of antigens generated by modification of the original antigen, often by pathologic processes. To date, neoepitope biomarkers of aggrecan, collagen⁴⁰ and cartilage oligomeric matrix protein (COMP)⁴¹ have been developed in the OA field as well as others (see Table 4). Because neoepitope biomarkers reflect degraded substrates, they provide information about the activity of an enzyme or enzyme family and thereby the activity state of pathological catabolic processes in OA. Two neoepitopes, one N-terminal and one C-terminal, are generated for every proteolytic cleavage event, thus doubling the opportunities for monitoring or quantifying proteolysis⁴⁰.

Knowing the half-lives of biomarkers aids interpretation of tissue turnover rate. Half-lives of only a few OA related biomarkers are known with any precision; for instance, the half-life of serum hyaluronan is estimated to be minutes⁹³. A recent methodology based on half-life determination by proteomic measurement of native and deamidated forms of peptides has yielded cartilage half-life estimates for 7 major cartilage proteins including aggrecan (G1 and G3 domains), cartilage oligomeric matrix protein, fibronectin, prolargin, clusterin, collagen type III alpha 1 chain and collagen type II alpha 1 chain¹²; biomarker half-lives ranged from 2 days (for aggrecan G3 domain) to 2420 days (for guanidine-soluble collagen type II). This method relies on knowledge of protein epitope structure to estimate deamidation rates; with the recent advances in automated protein structure modeling⁹⁴, this method is in theory applicable to any protein in any tissue containing either Asn or Gin residues.

It is often desirable to compare biomarkers for their ability to predict an outcome. Because the units of measure and range of different biomarkers can be very different, it is necessary to standardize the biomarker measurements in some way in order to compare them. Standardization is often done by transforming biomarker values to z-scores [(biomarker value – group mean)/group SD], Z-scores of biomarkers put into a logistic regression can provide odds ratios based on units of standard deviation and can be very useful as a means of comparing strengths of association among various biomarkers. In addition, ‘standardized betas’ generated by statistical analysis software in the performance of linear regression analyses standardize the biomarkers in this same way so the biomarkers can be directly compared to one another.

Although technically challenging, precise knowledge of the identity and source of the measured biomarker epitope can greatly aid in the interpretation of biomarker results and in the confidence of target engagement by a therapy impacting the biomarker. The process of experimentally identifying the binding site, or “epitope”, of an antibody on its target antigen (usually, on a protein) is referred to as epitope mapping. This is typically performed using the antibodies that constitute the assay in combination with methods such as mass spectrometry. Biomarkers for which the epitope has been mapped provide the strongest disease understanding.

Application of Biomarkers to OA Research, Drug Discovery and Clinical Care

Biomarkers may be productively used at different stages of research, drug discovery and clinical care of OA (Figure 4). In fact, biomarkers reduce the attrition rate of drugs at each phase transition during the process of drug development⁹⁵. A list of Clinical and Laboratory Standards Institute (CLSI) Guidelines for seeking qualification of biomarkers for drug development from regulatory bodies are provided in Supplementary Table 3. Although OA affords multiple opportunities for surveillance by soluble biomarkers, due to the availability of synovial fluid, blood and/or urine in which to conduct measurements, the current lack of chondroprotective treatments for OA impairs the assessment, validation and qualification of the potential role of biomarkers as tools for monitoring disease status and patient responses to treatment of OA. Nevertheless, to date, a rich set of analytes have been assessed for many stages and types of OA, serving as a foundation for the hope and expectation that results of biomarker testing will someday be used to guide clinical management of disease.

Figure 4. Application of biomarkers to research, drug discovery, and clinical care.

A summary is provided of the specific uses of biomarkers for research, different phases of drug development (pre-clinical and phases I-V), and clinical care. Information adapted from^19,21,25. NDA, New Drug Application; BLA, Biologic License Application; MAA, Marketing Authorization Application; Created with BioRender.com (paid license)

It is anticipated that a combination of biomarkers likely improves the prediction of disease and progression due to the fact that OA is a disease of the joint organ involving multiple tissues simultaneously. As aptly stated by Williams⁹⁶, “no single biomarker will be a perfect descriptor of OA any more than one single pathological feature encapsulates the coordinated changes in the three tissues along with the cytokines released”. Therefore, there is a need for combinatorial biomarkers and algorithms for improving disease detection and risk of progression. A masterful and comprehensive review of the OA candidate biomarker literature was provided by van Spil et al. in 2010¹. New approaches based on single cell genetics, epigenetics and proteomics⁹⁷, that makes it possible to quantify >1000 proteins from individual mammalian cells, will undoubtedly transform the OA field and lead to the next generation of biomarkers. To provide a brief overview of the most important biomarkers that currently reflect quantitative and dynamic variations in joint remodeling and the more recent developments in the field, a summary is provided in Table 4 with biomarkers separated on the basis of bone, cartilage and/or synovial origin.

To be applicable to clinical care, the outcome of biomarker measurements must be actionable. The concept of an actionable biomarker is based on the expectation that results of biomarker testing could be used to guide clinical management of disease¹⁴. Although biomarkers do not need to be directly involved in disease pathogenesis to be useful, ‘mechanistic’ biomarkers (also known as ‘direct’ biomarkers, see Table 1 and Figure 2), are a subtype of actionable biomarker that is embedded in disease pathogenesis and, therefore, represent a potentially superior biomarker¹⁴. Mechanistic biomarkers could help to establish a molecular taxonomy of diseases¹⁴ that would be particularly valuable for OA for distinguishing different pathological pathways or endotypes, in contrast to classical diagnostic imaging criteria that represent a common end stage of a variety of pathogenic etiologies.

As mentioned above, the current lack of available disease modifying OA drugs (DMOADs) currently precludes use of biomarkers for actionable outcomes. Once such treatments are available, well-validated biomarkers will also likely need to cross the ‘translation gap’ in order to be employed in healthcare; namely, they must offer an advantage in terms of cost per quality adjusted life year over existing treatment and surveillance paradigms. Given the general cost-effectiveness of biomarkers and their widespread application in other fields, for instance high sensitivity C-reactive protein and brain natriuretic protein in cardiology, it is highly likely that for a disease as widespread and disabling as OA that a subset of biomarkers will in future, successfully cross the translation gap.

Needs for Biomarker Advancements in the OA Field

Critical needs and recommendations to advance the science of biomarkers were previously outlined by the OARSI FDA Osteoarthritis Biomarkers Working Group¹⁹. These recommendations included collection of biospecimens in all OA trials. Although speculative, in cancer research there appear to be broad economic benefits to standardized centralized biobanking. However, as aptly stated by Rogers et al., “the true cost of operations for research biobanking is little understood, making return on investment analyses extremely difficult”⁹⁸. They estimate that the initial year start-up costs for even a relatively small biorepository could approach $4 million⁹⁸. Rogers et al., also note that the scientific community has historically tended to underestimate the value of biospecimen research⁹⁸; this has hampered development of improved biomarker practices that are vital to robust and reproducible biomarker results. Taken together, these challenges might be surmounted by a collection of “virtual repositories” across the globe representing OA biomarker laboratories that prioritize and standardize sample storage and processing to optimize biomarker studies for the future.

Conclusions

Biomarkers have the potential to aid clinical diagnosis when symptoms are present or, in the absence of symptoms, to provide a means of detecting early signs of disease¹⁴. Consequently, appropriate biomarkers should play a role at all stages of OA diagnosis, prognosis and drug development. We are cautiously optimistic that the OA field can advance toward new treatments with the assistance of biomarkers due to an improved understanding of the pathogenesis of OA resulting in more direct biomarkers, deeper and more refined OA phenotyping (fixing the ‘cracked mirror’ problem), emergence of biomarkers reflecting the multi-tissue pathology of OA, and bioinformatics and systems biology capable of dealing with the complexity of such a disease.