Biological classification of childhood arthritis: roadmap to a molecular nomenclature

Abstract

Chronic inflammatory arthritis in childhood is heterogeneous in presentation and course. Most forms exhibit clinical and genetic similarity to arthritis of adult onset, although at least one phenotype may be restricted to children. Nevertheless, pediatric and adult rheumatologists have historically addressed disease classification separately, yielding a juvenile idiopathic arthritis (JIA) nomenclature that exhibits no terminological overlap with adult-onset arthritis. Accumulating clinical, genetic and mechanistic data reveal critical limitations in this strategy, necessitating a new approach to defining biological categories within JIA. Here we review current evidence for biological subgroups of arthritis in children, delineate forms that appear continuous with adult-onset arthritis, and consider integrative genetic and bioinformatic strategies to identify discrete entities within inflammatory arthritis across the lifespan.

INTRODUCTION

Chronic inflammatory arthritis is most commonly a disease of adults. Rheumatoid arthritis (RA) affects approximately 0.6% of the US population, affecting females twice as often as males and with peak incidence in the seventh decade^1,2. Diseases in the spondyloarthritis family impact an estimated 0.9-1.4% of adults, often beginning at an earlier age and affecting males more frequently than females³. More rarely, adults may develop a febrile arthritis termed adult onset Still’s disease (AOSD).

Children also develop arthritis. Prevalence worldwide ranges between 15 to 400 per 100,000, with considerable geographical variability in phenotype^4,5. Childhood-onset arthritis peaks between the ages of 1 and 4 years, during which girls outnumber boys by approximately 3:1; the first year of life is relatively spared, and in adolescence the female gender skew is less pronounced (Figure 1). These epidemiological shifts are accompanied by changes in clinical presentation, strongly suggesting that childhood arthritis encompasses more than one pathophysiological entity.

Figure 1. Epidemiology of juvenile idiopathic arthritis.

Age of onset in 1081 children with arthritis evaluated in the Pediatric Rheumatology Clinic at The Hospital for Sick Children, Toronto, Canada. Note the paucity of JIA onset in the first year of life; the incidence peak in girls between age 1 and 4; and the relatively balanced gender ratio of arthritis that begins in adolescence. Modified from ref ¹³⁹.

The first extended description of inflammatory arthritis in children was provided by Still in 1897, who highlighted features including presentation early in life, a predilection for knee involvement, and in many cases a relentless and disabling course⁶. Characterizing this population further, in 1959 Ansell and Bywaters introduced the now-traditional definition of juvenile arthritis as beginning before the 16^th birthday, although they considered this threshold arbitrary^7-9. Formal classification of childhood arthritis developed contemporaneously in North America and Europe. In 1972, a commission of the American Rheumatism Association (now the American College of Rheumatology) endorsed the term “juvenile rheumatoid arthritis” (JRA), later divided into pauciarthritis (≤4 joints in the first 6 months of disease), polyarthritis (≥5 joints in the same period), or systemic arthritis (with fever)^10-12. In Europe, the European League Against Rheumatism adopted the term “juvenile chronic arthritis”, using overlapping but distinct terms and definitions, while including categories for juvenile ankylosing spondylitis and juvenile psoriatic arthritis^13,14. Spondyloarthritis in younger children was recognized in 1982 as “seronegative enthesopathy and arthropathy”¹⁵. An effort to unify these disparate criteria under the umbrella of “juvenile idiopathic arthritis” (JIA) was launched in 1994 through the International League Against Rheumatism (ILAR; later the International League of Associations for Rheumatology), culminating in the current categories: systemic, oligoarthritis (persistent or extended), polyarthritis rheumatoid factor (RF) negative, polyarthritis RF positive, psoriatic, and enthesitis related arthritis (ERA) (Figure 2)^14,16.

Figure 2. Evolution of classification criteria for childhood-onset arthritis.

Since 1959, juvenile and adult arthritis have been distinguished by an age cutoff at the 16^th birthday. Nomenclature has evolved through the American Rheumatism Association’s juvenile rheumatoid arthritis (JRA) and EULAR’s juvenile chronic arthritis (JCA) to ILAR’s current juvenile idiopathic arthritis (JIA) system, for which a provisional revision has been proposed by PRINTO.

Always regarded as provisional, the JIA nomenclature nevertheless codifies certain assumptions about how arthritides should be distinguished from one another. No form of JIA has the same name as any adult-onset arthritis, tacitly implying that childhood and adult arthritides are fundamentally distinct (Figure 3)¹⁷. Within JIA, the number of joints affected at or near presentation is used as a marker of disease type rather than severity, while psoriatic arthritis is distinguished categorically from enthesitis-related disease and recognized even in the youngest children^17-19. These features of ILAR shape how clinicians and investigators conceptualize childhood arthritis, and frame the ways in which pediatric and adult rheumatologists are and are not able to work together. In this review, we consider limitations of the current JIA nomenclature and new strategies to identify a classification of childhood arthritis that is more securely grounded in disease biology.

Figure 3. Current arthritis classification in children and adults.

A. Major diagnostic categories of idiopathic inflammatory arthritis beginning in childhood (before the 16^th birthday) according to the ILAR nomenclature. B. Major diagnostic categories of idiopathic inflammatory arthritis beginning in adulthood (age 16 and onward). Adapted from refs ^3,16.

Note to illustrator – for PRINTO, the early-onset ANA-positive category includes some seroneg polys, some of each oligo category, and some psoriatics. Other/unclassified includes others in each of these categories plus some psoriatics.

BIOLOGICAL DIFFERENTIATION OF ARTHRITIS SUBTYPES

RA was first distinguished from gout in the 19^th century^20-22. Discovery of RF in 1939 enabled further subdivision of RA into RF positive and RF negative^23,24. In 1956, RA was differentiated from a family of diseases subsequently termed spondyloarthritis, including psoriatic arthritis and ankylosing spondylitis^25,26. These divisions have since been sharpened by the identification of distinct genetic associations for RF-positive RA, RF-negative RA, and spondyloarthritis; by recognition of enthesitis as a prominent feature of spondyloarthritis; and by the discovery of citrullinated peptides as antigenic targets within RF-positive RA, such that the “seropositive RA” now encompasses disease accompanied by either RF or anti-citrullinated protein antibodies (ACPAs)^3,26-33 (Figure 3).

Studies in animal models show that arthritis can arise via several distinct mechanisms³³. While human arthritis may be more complex, compelling data extend key biological “fault lines” into human disease, providing a useful guide to the ways in which arthritides may diverge from one another (Figure 4).

Figure 4. Biological “Fault lines” in arthritis.

Studies in murine models and in humans identify three central dichotomies in arthritis biology: synovitis vs. enthesitis, antibody-related vs. antibody-independent, and autoimmune vs. autoinflammatory. Distinct forms of arthritis may be characterized by understanding where they reside on each spectrum.

Synovitis vs. enthesitis.

The synovium is composed of fibroblasts, macrophages, and other cells residing within a loose connective tissue substructure. In RA, synovitis is the primary driver of musculoskeletal pathology. In spondyloarthritis, arthritis often involves (or may even begin with) inflammation of the entheses, specialized sites where ligaments, tendons and joint capsules insert into bone^34,35. Mechanisms of enthesitis differ from those that cause synovitis. Animal models show that enthesis-resident T lymphocytes programmed to produce Th17 cytokines cause florid enthesitis when triggered by systemic IL-23, an effect that may be magnified by mechanical stress^36-38. Correspondingly, whereas RA targets mainly synovium-rich diarthrodial joints, spondyloarthritis affects the Achilles and patellar tendon insertions, as well as locations rich in ligamentous attachments such as the spine and sacroiliac joints. Enthesitis likely contributes to dactylitis (sausage-like digital swelling) through swelling of ligaments, pullies, and joint capsules within the fingers and toes, a reflected in the high specificity of this clinical feature for psoriatic arthritis, at least in adults^39-42. The difference between synovitis and enthesitis provides a useful conceptual framework to differentiate RA from spondyloarthritis, although the proximity of entheses and synovial tissues (including synovium-lined tendon sheaths) means that the distinction is rarely absolute (Figure 4a).

Antibody-related vs. antibody-independent synovitis.

In mice, many though not all models of experimental synovitis are mediated through pathogenic antibodies³³. These antibodies typically trigger arthritis as immune complexes, formed in blood against circulating antigen or within the joint against an antigen that is either intrinsically local (e.g. collagen) or deposited from the circulation (e.g. glucose 6 phosphate isomerase). Joints may be uniquely susceptible to immune complex disease because the cartilage surface is acellular, lacking either intrinsic clearance mechanisms or membrane-bound complement inhibitors³³. By contrast, murine arthritis can also arise in an antibody-independent manner, through pathogenic T cells or other mechanisms^43,44. Human arthritis likely exhibits a parallel divergence. In RA, seropositive patients exhibit immune complexes embedded in cartilage and synovium, together with complement breakdown products^33,45,46. Patients with seronegative RA and spondyloarthritis (e.g. psoriatic arthritis) typically lack both immune complexes and complement fixation, suggesting that immune complexes play little or no pathogenic role. Correspondingly, tissue-resident B cell helper T cells, called T peripheral helper (Tph) cells, are abundant in joints from seropositive RA (and potentially in ANA-positive early-onset JIA) but are rare in seronegative RA^47,48. The distinction between seropositive and seronegative RA thus reflects at least in part a divide between arthritis with joint-depositing antibodies and arthritis without (Figure 4b). Importantly, RA develops through several phases, including an asymptomatic preclinical period characterized by high-titer autoantibodies without synovitis⁴⁹. We do not know yet what distinguishes pathogenic from “bystander” autoantibodies, or whether the role of antibody varies over the course of the disease⁵⁰.

Autoimmunity vs. autoinflammation.

Autoimmune diseases arise from misrecognition of self antigens due to a break in immune tolerance. By contrast, autoinflammatory diseases reflect antigen-independent hyperactivation of immune pathways, most commonly from defects within innate immunity⁵¹. RA appears to be primarily autoimmune in etiology, as reflected in HLA associations for both seropositive and seronegative forms. By contrast, AOSD and systemic JIA (sJIA) exhibit features suggestive of autoinflammation, including fever, rash, and often brisk response to IL-1 blockade, although an HLA class II linkage complicates straightforward categorization as autoinflammatory^52,53. Diseases in the spondyloarthritis family likely also exhibit a prominent autoinflammatory component, reflecting the tendency of HLA-B27 to fold and traffic aberrantly, provoking an unfolded protein response exacerbated by mechanical stress at the entheses⁵⁴. Autoimmunity and autoinflammation are not mutually exclusive, and polygenic inflammatory diseases commonly reflect elements of each⁵¹ (Figure 4c).

WHY DOES CLASSIFICATION MATTER?

A nomenclature establishes which cases belong together and which do not. Clinical manifestations and pathogenesis both contribute, but the latter is usually prioritized where known. Within pathogenesis, shared etiology early in the chain of events, e.g. a common genetic cause, takes precedence over shared downstream pathways. Extending these principles to childhood arthritis, the goal is to identify groups of patients whose disease reflects similar biological mechanisms.

Classification risks two kinds of error: grouping together patients with different diseases (over-lumping), and dividing patients with the same disease (over-splitting). Over-splitting may be especially pernicious because it can render underlying similarities difficult to appreciate later, as exemplified by the current chasm between pediatric and adult arthritis. Over-splitting also has important practical consequences. Approval of new medications for JIA has been slowed by a requirement for randomized controlled trials showing efficacy in this population. Despite progress, this requirement poses a substantial barrier to drug access because eligible patients can be difficult to find and because the financial incentive for companies to conduct pediatric studies is modest⁵⁵. If certain forms of arthritis were recognized to extend across the age spectrum, then pediatric studies could restrict their focus to pharmacokinetics, pharmacodynamics, and safety, considerably accelerating drug approval for children⁵⁶.

THE NEED FOR NEW CLASSIFICATION CRITERIA

The ILAR criteria for JIA represented a major landmark in pediatric rheumatology, providing for the first time a worldwide language of communication about childhood arthritis and serving the community well for more than 25 years. More generally, segregation of juvenile from adult patients helped pediatric rheumatology to define itself as a distinct subspecialty, while focusing clinical and scientific attention on children with arthritis^57,58. Nevertheless, limitations in this nomenclature have become evident. From a practical point of view, the criteria are difficult to apply because of their complex network of inclusion and exclusion criteria, some with counter-intuitive implications. For example, no patient with sJIA may have a first-degree relative with psoriasis¹⁶. The age 16 cutoff does not correspond to the legal divide between childhood and adulthood, now typically age 18⁵⁹. The definition of seropositive JIA has not been updated to encompass ACPAs, measured clinically as antibodies against cyclic citrullinated peptide (CCP). Many patients with features consistent with sJIA are excluded from this category at disease onset because they lack overt arthritis, and with early treatment some escape joint inflammation altogether^60,61. These “bookkeeping” issues are all amenable to straightforward textual corrections^59,62-64.

Other difficulties with the ILAR system are more fundamental. By including within JIA all arthritis that begins before the 16^th birthday, the nomenclature provides no mechanism to recognize phenotypes that exist in both children and adults. The JIA criteria distinguish oligoarticular and polyarticular JIA by the number of joints affected within 6 months of disease onset. Neither 5-joint threshold nor the 6-month cutoff is well supported by data, while being further complicated by the limited inter-examiner reproducibility of the joint exam, the capacity of imaging to disclose inflamed joints not evident clinically, and the impact of early treatment^65-68. The oligoarticular/polyarticular divide introduces a further untested assumption that the number of joints and/or the speed of joint accrual reflects disease type rather than severity^18,19, an assumption that has become increasingly uncertain in light of genetic data that identify broad similarity between oligoarticular and seronegative polyarticular JIA^17,69-71. Joint count undervalues the importance of joint distribution in the identification of distinct forms of arthritis^72-75. More broadly, the ILAR distinction between ERA and psoriatic JIA obscures the central role of enthesitis in psoriatic arthritis⁴². Further, although there is no “gold standard” to identify children with psoriatic arthritis, some evidence suggest that current criteria distribute nearly 60% of children with this phenotype into other JIA categories^74,76.

Taken together, these accumulated limitations highlight the emerging need to replace the ILAR criteria with a nomenclature that is based on disease biology. This nomenclature should be informed by the growing understanding of arthritis mechanisms as well as by a “big picture” view of joint inflammation as it appears across the lifespan.

FORMS OF ARTHRITIS THAT AFFECT BOTH CHILDREN AND ADULTS

Many diseases exhibit phenotypic variance as a function of age at onset, reflecting factors such as genetic load, physiological maturation and/or senescence, and environmental exposures (Box 1)^77-79. In arthritis, accumulating data indicate that most forms affect both adults and children.

Box 1: Age of onset effects in arthritis.

Diseases that share a common pathophysiology may still manifest differently because of factors that vary with age. A familiar example is parvovirus B19 infection, which presents in children as the “fifth disease” slapped-cheek exanthem, in adult women as joint inflammation or miscarriage, and in patients with sickle cell disease as aplastic crisis. Phenotypic variation with age is similarly evident in arthritis. For example, spondyloarthritis beginning in childhood is associated with a higher risk of joint replacement than adult-onset disease, while sJIA presenting in very early childhood exhibits more macrophage activation syndrome and a worse prognosis^98,127. Age-dependent variation arises in several ways.

• Genetic loading. Within a polygenic disease, earlier onset often reflects a stronger genetic predisposition. For example, in systemic lupus erythematosus, patients with more risk variants present earlier and more frequently develop nephritis^78,128,129. In adult RA, risk HLA alleles predict lower age at onset, early adult presentation confers higher RA risk for family members^77,130.

• Substrate differences. Children and adults differ anatomically and physiologically in ways that are relevant to arthritis. Immune function changes with age, including under the influence of sex hormones^131,132. Developing tissues, such as joints or eyes, could theoretically expose antigens lacking in adult joints or otherwise exhibit differential vulnerability to disease. For example, early-onset JIA is often associated with CAU, but a similar condition affects young children without arthritis¹⁰⁷. Conversely, aging cartilage may be less resistant to complement fixation and therefore favor immune complex arthritis^33,133.

• Environment. The environment to which an individual is exposed varies with age. RA risk increases with smoking, occupational silica inhalation, and obesity, all more prevalent in adults than children^134,135. Children experience an evolving gut microbiome, recurrent viral infections, and in some cases frequent antibiotic treatments, exposures with immunological consequences^136-138.

These mechanisms likely result in continuous rather than dichotomized variation, illustrating the difficulty in assigning a specific age cutoff to any particular form of arthritis, even as such a cutoff may sometimes be necessary for classification purposes.

Seropositive arthritis

Childhood-onset seropositive arthritis typically appears first in late childhood and early adolescence. As in adults, the disease usually affects many joints, is often both RF and CCP positive, may be accompanied by rheumatoid nodules, requires sustained and often aggressive disease-modifying therapy, and is associated with HLA class II alleles that share specific citrulline-binding residues in the antigen binding pocket, as well as shared non-HLA risk loci^70,80,81. This phenotype, termed RF-positive polyarticular JIA, exhibits no more phenotypic variability from adult-onset disease than is otherwise observed within adult seropositive RA, and is to all intents and purposes the same disease^17,59,70,81.

Spondyloarthritis

Adults with ankylosing spondylitis often report onset of symptoms as teenagers⁸². Typical sacroiliitis is well documented in older children⁸³. In younger children, spondyloarthritis commonly presents in a less differentiated form, characterized by enthesitis, arthritis, and prominent arthralgia, often without sacroiliitis¹⁵. Gathered together under the ILAR term ERA, associated HLA alleles overlap with those of ankylosing spondylitis, supporting the continuity of spondyloarthritis across the age spectrum^70,84.

Psoriatic arthritis in children is more controversial. In young children, definitive identification is challenging because of skin disease can lag for many years. Some of these children closely resemble the proposed early-onset ANA-positive subset of JIA (see below), rendering a unique psoriatic identity uncertain^18,85. However, classic adult-type psoriatic arthritis is readily observed in adolescents with overt psoriasis vulgaris, and associated enthesitis of the dactylitic digit has been identified by imaging and histology⁴⁰. Further, the frequency of psoriatic JIA – 5-20% of JIA, varying with population and criteria employed – greatly exceeds the expected chance association of these two diseases (prevalence of psoriasis among children 1-2%)^73,76,85,86.

Seronegative arthritis

Seronegative arthritis is likely a heterogeneous mixture of conditions in both adults and children. Evidence for this suggestion includes substantial patient-to-patient clinical variation, lower heritability than seropositive RA in adults, and the abundance of pathways to antibody-independent arthritis in animal models^19,33,77. Studies of childhood seronegative arthritis remain insufficient to determine the role of autoantibodies and immune complexes. Clinical similarities have been noted between early-onset oligoarticular JIA and a subset of seronegative polyarticular JlA^18,19,87,88. HLA associations are shared among oligoarticular JIA, seronegative polyarticular JIA, and adult seronegative RA⁷⁰. Less is known about genetic associations beyond the HLA, but available data similarly suggest extension across the age spectrum, further supported by familial aggregation of seronegative RA with JIA^89,90. Together, these considerations favor continuity between children and adults in at least some forms of seronegative arthritis.

Systemic JIA / adult onset Still’s disease.

The febrile arthritis designated sJIA is highly distinctive within the JIA family. Current definitions segregate sJIA from AOSD by age of onset, a requirement for overt arthritis, and several other minor differences^16,91. However, Bywaters’ original description of AOSD in 1971 explicitly considered patients to have the pediatric disease⁹². Similarities between sJIA and AOSD include the ubiquity of fever, a characteristic rash, a generally even male-female ratio, elevation in ferritin and D-dimer, high circulating IL-18, and a characteristically brisk response to IL-1 or IL-6 blockade^17,93-96. Rash is more common in children whereas sore throat is reported less frequently; progression to chronic arthritis may occur somewhat less commonly in adults⁹⁶. Importantly, heterogeneity is present even within sJIA, as a function of age at onset and in response to therapy, where it has been suggested that a brisk and complete response to IL-1 inhibitors may identify a subset of sJIA with predominantly autoinflammatory features^97,98.

A FORM OF ARTHRITIS POTENTIALLY RESTRICTED TO CHILDREN

The most remarkable feature of the epidemiology of arthritis in children is a peak in incidence during early childhood, typically between the ages of 1 and 4 years (Figure 1). Most patients in this early peak present with an oligoarticular phenotype, many with only a single swollen knee. These patients lack RF and CCP but are commonly positive for antinuclear antibody (ANA), albeit typically at a modest titer (≤1:320). This population is at highest risk for chronic anterior uveitis (CAU), an indolent but destructive disease distinct from the acute anterior uveitis observed in both adult and pediatric spondyloarthritis. CAU has no counterpart in adult-onset arthritis. Up to 50% of patients with early-onset arthritis enter long-term drug-free remission, an outcome rarely observed in adult arthritis^99,100. Corroborating studies suggest an inflection point around age 6 with respect to HLA associations and peripheral blood transcriptomic signatures^71,101,102.

The optimal way to delimit this population remains undefined. The Paediatric Rheumatology International Trials Organisation (PRINTO) has proposed a form of arthritis termed “early-onset ANA-positive JIA” that encompasses arthritis beginning ≤ age 6 years and with at least 2 ANA titers ≥ 1:160, without other recognizable forms of JIA⁵⁹. This proposal is based on a literature review¹⁸ and two studies that found children with JIA who met the specified ANA threshold to exhibit a younger age of onset (80% under 6 years), fewer affected joints, and more uveitis than JIA patients in seronegative oligoarticular, polyarticular and psoriatic JIA categories who had never tested positive for ANA^18,87,88. One study found that synovial biopsies from JIA patients with a positive ANA (at any titer) exhibited an excess of lymphoid aggregates, although it is less clear that untreated JIA synovial tissues echo this finding^103,104. Patients within the ANA+ early-onset group may exhibit an abundance of synovial fluid T cells that could be T peripheral helper (Tph) cells, potentially implicating pathogenic autoantibodies⁴⁸. Unresolved questions include the sensitivity and specificity of ANA and whether ANA predicts clinical features or is independent of age, although interestingly the abundance of Tph-like cells was not explained by early onset alone^{17,48,102,105,106}.

While data supporting a unique form of arthritis in young children are compelling, they are not yet conclusive. Differences in anatomy, immunology, and exposures – collectively the “substrate” in which a disease occurs – could translate into phenotypic differences between younger and older patients with the same disease (Box 2). Fine-mapping failed to identify a unique set of HLA associations for persistent oligoarticular JIA, the closest fit for this phenotype among existing JIA categories⁷⁰. The CAU so characteristic of these patients can occur in young children without arthritis, raising the possibility that this hallmark feature reflect the pediatric substrate rather than a unique arthritis biology¹⁰⁷. Systemic JIA, genetically unrelated to oligoarticular and polyarticular JIA, also peaks in younger children, as do immune-mediated conditions such as Kawasaki disease, type I diabetes, and dermatomyositis^96,108. Further investigation will be required to determine whether early-onset (ANA-positive) arthritis represents a distinct disease of childhood.

Box 2. Queries for any proposed subtype of chronic inflammatory arthritis.

1. Does the subtype comport with, or convincingly overturn, established understanding of arthritis biology?

2. Does the subtype display substantial genetic coherence, as reflected in internal homogeneity and differences from other types of arthritis?

3. Does the subtype include all patients with sufficiently similar disease biology, or are many closely related cases excluded?

4. Does the subtype distinguish disease type from disease severity?

5. Does the subtype distinguish disease type from variation due to age of onset (Box 1)?

6. Does the subtype validate using approaches distinct from those used for its derivation?

MOVING TOWARDS A NEW ARTHRITIS CLASSIFICATION

Segregating childhood-onset arthritis from adult disease has had several benefits. Not only did it help to distinguish pediatric rheumatology as a field, but it also drew the attention of regulators, funding agencies, and pharmaceutical companies to arthritis affecting children^57,58,64. In this sense, the JIA terminology served the field better than JRA, with its tacit implication that childhood arthritis might simply be “baby rheumatoid”. A distinct nomenclature supported development of research organizations with specialized expertise in pediatric rheumatology, including the Pediatric Rheumatology Collaborative Study Group (PRCSG), PRINTO, and CARRA, as well as organizations of parents and other advocates for children with arthritis. Collectively, these organizations have drawn significant funding into pediatric rheumatology research, revolutionizing the understanding of disease phenotype, course, and treatment and leading to regulatory approval of a broad range of medications for JIA.

A further benefit has been to underscore the special needs of children. Developing joints are highly vulnerable to arthritis-mediated injury and deformity, including the temporomandibular joint that is (for unknown reasons) rarely affected in adults. Linear growth of the skeleton is easily impaired by systemic inflammation or corticosteroid therapy. Children with arthritis require screening for CAU and drug dosing adjusted by age and weight. Children and their families must be managed in a manner that is developmentally appropriate, with attention to issues such as school performance, body image, pregnancy risk, vocational aspirations, and transition to adulthood. The use of a unique terminology has helped to ensure that children with arthritis are not managed simply as little adults.

These advantages are real but are irrelevant to the question of whether childhood arthritis is biologically unique enough to merit its own terminology. As noted above, the age 16 demarcation was never intended to reflect (or worse, to create) a fundamental difference between pediatrics and adult arthritis^8,109. No other domain in medicine has found it useful to segregate a whole category of disease by hard age cutoff; even within rheumatology, only arthritis is treated in this way, whereas systemic lupus erythematosus, vasculitis, myositis, scleroderma, and other conditions are largely recognized as existing on a pediatric-adult continuum.

Four interim conclusions

The authors of this Review agree upon four conclusions derived from the considerations summarized above. First, the categorical divide between childhood-onset arthritis and adult-onset arthritis was an unintended consequence of the historical processes used to develop the current nomenclature and is at odds with emerging data. Second, the traditional division between oligoarticular and polyarticular JIA is unlikely to represent an important distinguishing feature of types of arthritis. Third, most arthritis phenotypes extend across the age spectrum, including seropositive RA, seronegative RA, spondyloarthritis, and sJIA/AOSD. Fourth, early-childhood (ANA-positive) arthritis may represent a distinct disease, although its pathophysiological uniqueness and boundaries remain to be established. These conclusions represent the starting point for ongoing studies with respect to disease classification.

New classification models

Two models incorporating these biological insights have been proposed. These models bear on primary idiopathic arthritis. Arthritis that arises as part of a distinct process, for example familial Mediterranean fever or Blau syndrome, could be termed secondary and is not considered further here.

PRINTO model.

In 2015, PRINTO embarked on a multi-step effort to define disease entities within childhood arthritis (defined for this purpose as onset <18 years) that are homogeneous from a clinical and laboratory point of view. A web Delphi process was employed to revise current ILAR JIA categories, followed by the use of nominal group technique (NGT) at a consensus conference to achieve provisional criteria for new arthritis categories. Four principal forms were identified, entitled Systemic JIA, RF-positive JIA, Enthesitis/spondylitis-related JIA, and Early-onset ANA-positive JIA (Figure 5a)⁵⁹. Psoriatic arthritis was not included because consensus was not reached on its definition. Childhood arthritis outside of these categories is considered either Other JIA (fits criteria for no definition) or Unclassified JIA (fits criteria for more than one definition). All definitions are considered provisional. An effort is underway to collect over 1000 children with new-onset arthritis, together with clinical descriptors, routine laboratory tests, and where possible biosamples. These data will be analyzed to see if clustering of clinical and laboratory descriptors allows identification of homogeneous entities (including psoriatic arthritis) in the group of patients provisionally included in the “other” category, and a further process using NGT will be organized to discuss the data and to provide evidence-based validation of the provisional criteria. Strengths of the PRINTO approach are its rigorous methodology and focus on the use of clinical measures readily available to clinicians at the point of care. Limitations include the restriction of patients analyzed to children and the possibility that not all features relevant for biological categorization may be evident in the data available.

Figure 5. Proposed subdivisions within arthritis: PRINTO and the four-cluster model.

A. Provisional PRINTO JIA categories for childhood-onset arthritis⁵⁹. B. Four-cluster model defining mechanistic subgroups within arthritis across the age spectrum (modified from ¹⁷). These two models represent hypotheses with differing methodology and purposes. The PRINTO model reflects an ongoing effort based on consensus methodology to provide preliminary criteria defining clinically homogeneous entities to enable structured validation studies as well as biological research. The four-cluster model integrates clinical and biological data to define groups that exhibit pathophysiologic similarity, irrespective of age of onset. These models exhibit more resemblance than difference, recognizing seropositive arthritis, spondyloarthritis, and systemic/Still’s disease as distinct entities. Early-onset ANA-positive children may or may not form a distinct subset within seronegative arthritis.

Four-cluster model.

A four-cluster model has been proposed for arthritis in both adults and children based on the arthritis mechanistic “fault lines” of seropositive (antibody-related)/seronegative (antibody-independent), synovitis/enthesitis, and autoimmune/autoinflammatory (Figure 5b)¹⁷. These categories are not operationalized as inclusion and exclusion criteria, but instead are drawn as Venn diagrams to highlight mechanistic overlap, recognizing for example that many genetic risk loci are shared between seropositive and seronegative arthritis, and that a family history of either confers genetic risk for the other^17,77,89,90. This model emphasizes lumping over splitting and allows that each category remains internally heterogeneous. Pending the development of stronger evidence, early-onset arthritis remains within the seronegative category, although some evidence hint at a potential role for B cells and therefore potentially antibodies (and their helper Tph cells) despite lack of RF^48,102. Strengths of the four-cluster model are its pathophysiologic foundation and its embrace of both children and adults. However, unlike the PRINTO model, it is not directly applicable to the clinic, serving as a guide for concept generation and research rather than practice.

Strategies to identify biological categories within arthritis

An important goal of childhood arthritis research in the next decade will be to identify biologically homogeneous subgroups, in the expectation that discovery of distinct “molecular signatures” will enable partition of patients into groups amenable to mechanism-based intervention. It is not sufficient to identify clinically homogeneous entities, because a similar phenotype can emerge from distinct etiologies. On the other hand, clinical similarity within a category will markedly simplify downstream mechanistic studies. These efforts are critical to the integration of pediatric-onset arthritis into disease prevention research ongoing in RA and psoriatic arthritis^49,110.

Genetic approaches to disease clustering.

Genetic studies have the major advantage that an individual’s primary DNA sequence can be assessed without regard to disease activity or treatment. Further, genes reside at the origin of the etiopathogenic sequence, eliminating the possibility that a genetic association reflects a disease effect or epiphenomenon rather than a cause.

In rare cases, a single-gene defect can give rise to arthritis. Examples include mutations in LACC1 and MYD88 ^111,112. However, most arthritis reflects both genes and environment, as exemplified by imperfect concordance between siblings and even identical twins^113-115. Genome wide association studies (GWAS) have identified approximately 30 loci with strong evidence of genetic association with oligoarticular and RF-negative polyarticular JIA^69,116. The impact of each variant is small, with odds ratios ranging from 6 for the HLA region to 1.1-1.6 for other loci; these effect sizes reflect the impact of the common variants amenable to study by GWAS, and should not be interpreted as reflecting of the importance of particular genes^17,69. Genetic associations provide critical information at the population level. Such studies helped to establish the fundamental differences between seropositive RA, seronegative RA, and spondyloarthritis in adults¹⁷. HLA associations support the identity of seropositive RA with seropositive polyarticular JIA, the continuity between ankylosing spondylitis and ERA, and the similarity among seronegative forms of arthritis across the age spectrum^70,81. Variants outside the HLA are also highly informative. Systemic JIA exhibits no non-HLA overlap with non-systemic forms of JIA, and thus is likely biologically quite distinct^52,108. By contrast, non-HLA variants are shared across seropositive arthritis, irrespective of age of onset, helping to emphasize the continuity of this condition across the age spectrum⁸¹. Too little is known about seronegative RA, but available data regarding associations beyond the HLA region suggest overlap with oligoarticular and seronegative polyarticular JIA⁸⁹.

Genetics alone will not allow definitive arthritis subclassification. Fundamental immunoregulatory mechanisms, and the genetic variants that affect them, are often shared across autoimmune diseases¹¹⁷. However, as illustrated above, similarity of HLA and non-HLA associations is strong evidence of identity whereas dissimilarity is strong evidence to the contrary, allowing genetics to serve as a “touchstone” of correct subset identification. Of first importance will be closer analysis of the early-onset (ANA-positive) subset of children. Broadly, oligoarticular JIA exhibits similar HLA associations to seronegative polyarthritis in children and adults⁷⁰. However, earlier studies found certain HLA alleles to carry risk or protection only within age-specific windows – for early-onset JIA, HLA-A2, DRB1*03, DRB1*05 (later refined to DRB1*11), DRB1*06 (later refined to DRB1*13) and DRB1*08 conferring susceptibility and DRB1*04 and HLA-B27 conferring protection¹⁰¹. It will be important to confirm these results and test whether they vary with ANA status. Advanced methodologies to define the causal non-coding variants that represent at least 85% of GWAS hits will provide an opportunity to genetically “fingerprint” arthritis across the age spectrum¹¹⁸. Genetics and epidemiology will further inform understanding of mechanism. For example, synergy between HLA alleles and smoking led to the hypothesis that seropositive RA could begin in the lung, and recent studies defining associations between HLA alleles and specific autoantibodies suggest new ways in which the seropositive/seronegative division within RA may be further refined^119,120.

Big data/machine learning for disease clustering.

Technical and computational advances provide new ways to characterize biological phenotypes and to identify patterns within the resulting datasets. The range of methodologies of potential relevance for arthritis is large, and include genomic strategies (whole genome sequencing; transcriptional profiling at the bulk, cell subpopulation, or single-cell level; epigenetic analysis), cell profiling (flow cytometry including related DNA-tagged antibodies for single-cell surface/transcriptomic studies, mass cytometry, phosphoprotein assessment, functional assays), autoantibody arrays, proteomics, lipidomics, and advanced histological methods including quantitative immunostaining and spatial transcriptomics. Analysis methods include any of a diverse range of clustering strategies and supervised or unsupervised machine learning.

Unsupervised machine learning aims to be ‘data-driven’ for the discovery of underlying patterns and clusters. Such studies are never fully hypothesis-independent, reflecting investigator choices with respect to input sample, biological assays, and analytical assumptions. Given the propensity to find patterns in almost any dataset, any strategy should be informed by understanding of disease pathogenesis. Putative disease clusters can be assessed for plausibility by a number of specific queries (Box 2).

Early efforts illustrate both the promise and the challenges of subgroup identification in JIA^75,121,122. In one study, children with recent-onset non-systemic JIA, untreated aside from NSAIDs, were divided into a derivation cohort (n=157) and a validation cohort (n=102). Using principal components analysis (PCA), patients in the derivation cohort were clustered on the basis of demographic features, clinical and laboratory data, and a panel of cytokines and chemokines measured in plasma. The five resulting groups were distinct from ILAR categories and replicated in the validation cohort, exhibiting relatively homogeneous joint trajectories¹²¹. A related study employed a wider range of biomarkers, analyzing patients both at baseline and after 6 months, arriving at a different set of clusters (3 at baseline, 5 at follow-up)¹²². Assessment of joint trajectories in 640 new-onset JIA patients identified 7 patterns of joints that tended to be involved together⁷⁵. These studies demonstrate the power of dimensionality reduction strategies to identify patterns. However, to the extent that the patterns identified did not align, they illustrate the susceptibility of clustering to the choice of input data, highlighting the need for orthogonal, longitudinal, and biologically-informed approaches to ensure that the clusters recognized correspond to distinct pathophysiologic groups.

The mission to achieve biological phenotyping of arthritis

Understanding childhood arthritis will require collaborative effort. In March 2016, an international group of clinicians, clinical trial experts, translational and basic researchers gathered in London, seeking to speed progress in personalized medicine for children with rheumatic disorders. Participants endorsed a statement of principles, termed the London Declaration, “to improve care and ultimately cure childhood rheumatic disorders through worldwide collaboration”¹²³. Signatories to this group included PRINTO, CARRA, and two other translational research networks. The Understanding Childhood Arthritis Network (UCAN) is a federation of research networks focused on translational research in childhood arthritis. UCAN represents more than 50 countries and 300 sites in a ‘hub and spoke’ model with cores in Utrecht, the Netherlands (UCAN-U), Toronto, Canada (UCAN-CAN), and Singapore in Asia (UCAN-A)¹²⁴. Using harmonized procedures for collection, processing, transfer, storage and access to clinical and biologic data, UCAN efforts seek to use computational biology and machine learning to discover genetic, biologic and phenotypic markers that provide diagnostic and prognostic information to caregivers at the bedside. The CLUSTER Consortium is a UK-based effort based to study approximately 5000 children and young people with JIA to enable biomarker-driven stratified medicine in childhood arthritis and associated uveitis. Parallel efforts are underway to understand adult arthritis through detailed analysis of joint tissues, including the NIH-sponsored Accelerating Medicines Partnership and the UK’s Pathobiology of Early Arthritis Cohort^125,126.

CONCLUSIONS

Defining biological subtypes within childhood arthritis remains a dauntingly complex task. This review outlines a roadmap forward, based on advances in arthritis biology, genetics, and clinical science since the ILAR JIA categories were first proposed over 20 years ago. Reclassification of disease has risks as well as advantages. Practical concerns notwithstanding, certain ingrained assumptions are no longer tenable, most fundamentally that of a categorical difference between childhood and adult arthritis, offering the prospect of fruitful collaboration between pediatric and adult rheumatology in coming years. Approaches that take advantage of new opportunities in clinical and biological phenotyping, interpreted cautiously and through the lens of pathogenesis, promise to accelerate progress toward the personalized management of children with arthritis.