Genetics and the classification of arthritis in adults and children

Abstract

Current classification of primary inflammatory arthritis begins from the assumption that adults and children are different. No form of juvenile idiopathic arthritis bears the same name as an adult arthritis, a nomenclature gap with implications for both clinical care and research. Recent genetic data question this adult/pediatric divide, revealing instead broad patterns that span the age spectrum. Combining these genetic patterns with demographic and clinical data, we propose that inflammatory arthritis segregates into four main clusters, largely irrespective of pediatric or adult onset: seropositive, seronegative (likely including a distinct group that usually begins in early childhood), spondyloarthritis and systemic. Each of these broad clusters is internally heterogeneous, highlighting the need for further study to resolve etiologically discrete entities. Eliminating divisions based on arbitrary age cutoffs will enhance opportunities for collaboration between adult and pediatric rheumatologists, thereby helping to promote the understanding and treatment of arthritis.

Introduction to a family feud

In The Canterville Ghost, English author Oscar Wilde wrote that “we have really everything in common with America nowadays, except, of course, language.” An analogous relationship holds between adult and pediatric rheumatology. We examine patients in the same way, order the same tests, and prescribe the same medications. Curiously, however, we don’t treat the same diseases.

For adult rheumatologists, most inflammatory arthritis falls under the umbrella of rheumatoid arthritis (RA), divided in turn into seropositive and seronegative. Beyond RA, most arthritis falls into a second family, spondyloarthritis, characterized by a tendency to enthesitis and spine involvement and whose members include ankylosing spondylitis and psoriatic arthritis (PsA).

Pediatric rheumatologists live in a different world. Primary inflammatory arthritides are grouped together as juvenile idiopathic arthritis (JIA), and divided into six mutually-exclusive International League of Associations for Rheumatology (ILAR) subcategories, with a seventh – undifferentiated JIA – as a catchall for cases that fit no category or more than one (1). No type of JIA shares the exact name of an adult arthritis (Table 1). Kids are not little adults, after all.

Table 1. Current classification of the idiopathic arthritides in adults and children.

Modified from Taurog et al. (54) and Petty et al. (1). RA, rheumatoid arthritis; IBD, inflammatory bowel disease; JIA, juvenile idiopathic arthritis; RF, rheumatoid factor.

Adult arthritis	Pediatric arthritis
Rheumatoid arthritis seropositive RA seronegative RA Spondyloarthritis axial spondyloarthritis peripheral spondyloarthritis psoriatic IBD-associated infection-associated other spondyloarthritis Adult Onset Still’s Disease	Oligoarticular JIA persistent oligoarthritis extended oligoarthritis Polyarticular JIA, RF negative Polyarticular JIA, RF positive Psoriatic JIA Enthesitis related arthritis Systemic JIA Undifferentiated JIA

Or are they? Nomenclature has consequences. Different names imply different diseases. Nomenclature determines who can enroll in studies, who can be managed per algorithms, and who can receive insurance coverage for medications. Nomenclature underlies our ability to track diseases longitudinally, including when JIA patients grow up. If adult and pediatric rheumatologists use different names for what turn out to be the same diseases, then nomenclature represents a wedge between our communities that complicates efforts to work together, obscuring not only areas of similarity but also of genuine divergence.

Fortunately, recent studies, principally in the field of genetics, have begun to point the way toward new ways of identifying patterns across the age spectrum. The present review seeks to integrate these observations in a “big picture” view of arthritis classification in adults and children, suggesting ultimately that the nomenclature gap between pediatric and adult arthritides reflects more about how physicians cluster than how diseases do.

The limits of eminence-based classification

Historically, arthritis has been classified by clinical phenotype, as adjudicated by experts. Reliance on expert opinion is sometimes termed eminence-based medicine, a tongue-in-cheek contrast with evidence-based medicine. Sometimes clinical observation works well. For example, RA, gout and rheumatic fever were distinguished from one another on the basis of patient demographics, joint distribution, chronicity, and blood uric acid levels (2, 3). Yet pattern recognition has its limits, and RA itself is an excellent example. The identification of rheumatoid factor (RF) and then anti-citrullinated peptide antibodies (ACPAs) uncovered a fault line that would have been difficult to identify clinically, since seropositive and seronegative RA exhibit extensive phenotypic overlap. Yet these conditions differ in genetic associations, in immune complexes and complement fixation within the joint, and in synovial T cell populations (4–8). RA thus exemplifies how shared phenotype is an imperfect guide to shared pathophysiology.

Another problem with eminence-based classification is that physicians know best what they see. In general, pediatric rheumatologists treat children and adult rheumatologists treat adults. Indeed, the dividing line between JIA and adult arthritis – a sharp cutoff at the 16^th birthday – has always reflected medical custom more than disease biology, despite its decisive role in nomenclature, regulatory and insurance approvals, and arthritis research.

The genetics revolution

Fortunately, over the last decade, the shaky ground of clinical phenotype has found a robust ally in genetics. Genetic associations have one critical advantage over almost any other kind of clinical or laboratory observation: genes predate disease. A serologic or histopathological finding may be a cause, an effect, or an epiphenomenon. By contrast, genetics means causality, even if establishing which variant is at fault and how it works is often a daunting challenge (9).

The first genetic associations with arthritis were with the human leukocyte antigen (HLA) region, responsible for peptide antigen presentation to T cells. These associations included RA with HLA-DRB1*04 and related alleles, and ankylosing spondylitis with HLA-B*27 (10–12). Advancing genetic methodology, including sequencing and statistical imputation, have enhanced the understanding of these associations, showing for example that the HLA-DRB1 association reflects key amino acids that enhance presentation of citrullinated peptides (13–15).

Definition of the broader genetic landscape has been transformed by genome-wide association studies (GWAS). GWAS take advantage of the fact that germline DNA recombination is relatively rare. Genetic polymorphisms therefore tend to travel together in large blocks of DNA termed haplotypes. By characterizing one or a few variants on each haplotype, we can assess which polymorphisms an individual is likely to carry without sequencing every nucleotide. Comparing individuals with and without disease can then identify loci associated with disease risk. The largest GWAS in RA was a meta-analysis that employed almost 30,000 cases and 74,000 controls to identify 101 risk loci, while a study of 2,800 cases and 13,000 controls identified 28 regions associated with risk for JIA (16, 17), with more added recently (18).

Sometimes GWAS data ignite a conceptual revolution. For example, GWAS helped to define the connection between ankylosing spondylitis and the IL-17 pathway, fueling a successful therapeutic development program (19, 20). More commonly, however, genetic findings are ambiguous. Only 16% of RA risk haplotypes contain any variant that alters protein sequence (16). This implies that most hits are regulatory rather than coding, and these are much harder to identify, in particular since haplotypes often contain several genes (on average nearly 4 in RA) (16). Finally, the increment in disease risk associated with each GWAS “hit” is often small. In RA, the HLA carries an odds ratio (OR) of only ~2.8 (although some alleles carry higher risk), whereas most non-HLA loci are in the 1.1–1.4 range; the same holds for JIA (21, 22).

These challenges have sometimes led to skepticism. How much can an OR 1.1 locus really tell us about how RA works? Such concerns are unfounded. GWAS evaluate the effect of common genetic polymorphisms (minor allele frequency >1%). Due to evolutionary pressure, common variants usually have relatively minor functional impact, and therefore a correspondingly small effect on disease risk (23). The impact of this natural variation does not necessarily reflect the importance of the gene itself. For example, a non-coding variant near HMGCR exhibits a decidedly modest effect on cholesterol levels, yet tags the target of blockbuster statin therapy, HMG CoA-reductase (9, 23). Correspondingly, RA GWAS findings identify targets with established therapeutic efficacy, including IL-6 and CTLA4/CD28 (16). Of course, if no functionally-divergent polymorphisms happen to be common, then GWAS cannot evaluate a locus. GWAS will not find rare variants, even if their effect size is large. Mechanisms related to arthritis incidence may be less important in established arthritis. Finally, even when a genetic variant is known, understanding how it modulates disease risk requires extensive investigation using complementary approaches (24). Despite these limitations, genetic association means human relevance, free of translatability concerns that trouble in vitro studies, cell lines, or mice. GWAS thus provide a roadmap of potential “pressure points” for human diseases, including arthritis, and represent an important tool that can be combined with other methodologies to unravel disease pathogenesis (9).

Using genetics to help define categories of arthritis

If genetics reflects mechanism, then shared genetics is strong prima facie evidence for common pathophysiology. This principle can help define disease clusters. Figure 1 illustrates the association of specific genetic loci with disease risk in different types of arthritis, anchored on oligoarticular plus RF-negative polyarticular JIA (17, 25), illustrating how other forms of arthritis overlap or diverge. For adult onset Still’s disease (AOSD), little or no data are available, and even seronegative RA is characterized only superficially. Nevertheless, the comparison makes clear that some diseases resemble each other and others do not.

Figure 1. Overlap of genetic susceptibility between JIA and other forms of inflammatory arthritis.

Loci include HLA regions associated with each disease and non-HLA loci meeting genome-wide significance in oligoarticular and RF-negative polyarticular JIA (17, 25). HLA alleles or amino acids are shown as red up arrows when conferring risk or green down arrows when protective. Non-HLA loci are marked in red if the lead SNP is the same as, or in linkage disequilibrium (R²>0.5) with, the listed polygoJIA SNP. In every case where markers were correlated, the direction of association was the same. Pink circles indicate potential overlap, where both diseases achieve genome-wide significance but with SNPs that are not in close linkage disequilibrium (R²<0.5), potentially reflecting distinct variants (direction of effect not specified). Other associated autoimmune diseases are listed in the last column, per Hinks et al. (17). References: polygoJIA (17, 25, 38); seronegative RA (47, 86, 87); seropositive polyJIA (25, 32, 33); RA (majority seropositive) (16, 88); psoriatic JIA (25); PsA (89–94); ERA (25); ankylosing spondylitis (95–98); sJIA (64, 65). Abbreviations: Chr, chromosome; HLA, human leukocyte antigen; polygoJIA, oligoarticular and RF-negative polyarticular JIA; RA, rheumatoid arthritis; JIA, juvenile idiopathic arthritis; T1D, type 1 diabetes; Vit, vitiligo; T1Dab, type 1 diabetes antibodies; MS, multiple sclerosis; AITD, autoimmune thyroid disease; SLE, systemic lupus erythematosus; SSc, systemic sclerosis; PBC, primary biliary cirrhosis; IBD, inflammatory bowel disease; HT, hypothyroidism.

Of course, genes cannot be the whole story, not least because disease concordance in monozygotic twins is relatively modest, estimated at 16% in ACPA-positive RA and 25% in JIA (22, 26). An effort to group “like with like” will also need to incorporate epidemiology and clinical features. The features we decide to include, and to disregard, represent an educated guess about what matters. Genetics helps us see underlying patterns, but does not (yet) allow us to escape eminence-based classification altogether.

Importantly, since all types of inflammatory arthritis share the same immune system and a common joint target, we can expect that they will often share biological pathways. Clusters within arthritis are therefore likely to resemble overlapping Venn diagrams, with some patients on the borders. If the radiologist’s favorite plant is the hedge, then the rheumatologist’s favorite color may need to be shades of gray.

Even if we cannot assume that children and adults are different, age of onset still means something. Environmental exposures such as viral infections, smoking and periodontitis vary with age, as do anatomy and physiology, as for example endocrine function. In many cases, early onset may also reflect genetic loading. For example, children with lupus bear a greater number of GWAS-defined risk variants than lupus patients who present as adults, while RA that begins early carries a higher disease risk for family members than RA presenting at an older age (27, 28). Ankylosing spondylitis of early onset is often more severe than later-onset disease, at least with respect to hip disease (29–31). Thus, even within the “same” disease, patients who present as children will be different those who develop disease in their 40s, who in turn will be different from those who present in their 70s. This is not a difference between children and adults per se, but rather reflects age of onset as a disease phenotype that – like others – is function of genes and environment.

Taking these considerations into account, we will outline four general clusters of immune-mediated arthritis that transcend the adult/pediatric divide. If the world is divided into “lumpers” and “splitters,” we will err on the side of lumping unless there are compelling reasons to split, recognizing that genetics, epigenetics, and other features will almost certainly permit these broad categories to be subdivided into more discrete etiological entities going forward.

Seropositive arthritis

Perhaps the clearest example of a disease that crosses age categories is the chronic arthritis known in adults as seropositive RA and in children as RF-positive polyarticular JIA (or sometimes childhood-onset rheumatoid arthritis, CORA). Both adult and pediatric variants are strongly associated with HLA alleles that favor presentation of citrullinated peptides, and share other genetic associations as determined by GWAS (Figure 1) (32, 33). Seropositive (RF+) arthritis essentially never occurs in very young children, and the chronic anterior uveitis characteristic of JIA is very rare in this population if it occurs at all. Indeed, below the age of 6 years, RA-associated HLA-DRB1 alleles are protective against arthritis (34). In both children and adults, seropositive patients exhibit a shared pattern of joint involvement and a predilection for nodule formation, while sustained drug-free remission almost never occurs. Thus, both genetically and clinically, seropositive RA and RF+ polyarticular JIA represent a convincing disease cluster.

Seronegative RA plus seronegative oligoarticular and polyarticular JIA

In adult arthritis, non-seropositive “leftovers” without other distinguishing features are grouped together as seronegative RA. In children, seronegative arthritis is divided by the number of joints affected in the first 6 months of disease: oligoarticular JIA (<5 joints) vs. polyarticular seronegative JIA (5 or more joints); oligoarticular JIA is further divided into arthritis that remains limited to few joints vs. arthritis that becomes polyarticular over time (persistent vs. extended oligoarticular JIA). Thus, current nomenclature divides these patients into 4 different groups.

This division is problematic. In JIA, early treatment may obscure the natural history of oligo-onset JIA. Pediatric rheumatologists know that children who start out with oligoarticular disease are unlikely to stay that way if the wrist or ankle is involved or if inflammatory markers are elevated (35). Such patients are usually treated with systemic disease modifiers, potentially forestalling the accumulation of additional inflamed joints. Some patients with oligoarthritis by exam have polyarthritis by imaging (36). Peripheral blood gene expression signatures and HLA associations cluster not along the oligo/poly divide but by age of onset, <6 vs. >6 years (37, 38). JIA investigators therefore often employ the etymologically-peculiar yet useful term “polygoJIA” to indicate a seronegative population that transcends ILAR boundaries, encompassing both polyarticular and oligoarticular subsets without regard for the number of joints involved.

Recent studies suggest that lumping should go even further. Hinks and colleagues characterized HLA associations in over 5,000 JIA patients and 14,000 controls (25). Oligoarticular and RF-negative polyarticular JIA were similar, down to specific amino acids within the antigen-binding grove of HLA-DRB1. By contrast, psoriatic JIA and enthesitis related arthritis (ERA) appeared somewhat different, and RF-positive polyarticular JIA was clearly distinct, resembling instead adult seropositive RA. Importantly, polygoJIA resembled adult seronegative RA, at least with respect to HLA associations, strongly suggesting that this subgroup crosses the pediatric/adult boundary (Figure 1).

These considerations do not imply that seronegative arthritis is homogeneous. The strongest case for a splinter group is found at the early end of the age spectrum. Pediatric-onset arthritis exhibits a remarkable incidence peak between the ages of 2 and 4 (39). In these early-onset patients, typically considered to encompass children presenting before the age of 6 years, girls outnumber boys 3:1 and most present with relatively few inflamed joints, often only a single swollen knee. Many develop chronic anterior uveitis, an extra-articular manifestation of arthritis with no adult equivalent, particularly in children positive for anti-nuclear antibody (ANA) (39, 40). Early-onset patients exhibit another remarkable feature: many will enter long-term drug-free remission, a rarity in adult rheumatology (41–43). Thus, early-onset arthritis appears phenotypically distinct from other arthritides across the age spectrum.

Despite these differences, it is not yet clear that genetics draws the same boundary. As noted, Hinks and colleagues found that oligoarticular JIA, seronegative polyarticular JIA, and seronegative RA patients shared many HLA associations (Figure 1) (25). While early-onset arthritis may be a distinct disease, an alternative (if less appealing) hypothesis is that the “same” disease is skewed by early-childhood physiology or exposures in favor of fewer joints, predilection for uveitis, and re-capture of immune tolerance. For example, the chronic anterior uveitis that is so archetypical of these children is also seen in children without JIA, but almost never in adults (44, 45). We suspect that further study will show this early-onset group to be distinct, including in HLA associations as per earlier studies (34, 38) as well as by associations outside the HLA (as suggested by Figure 1). However, more work is required before young children can be split from the seronegative population with confidence, not least because the borders of this subgroup remain to be defined.

How does seronegative arthritis relate to seropositive disease? Statistical considerations indicate that some seronegative patients are “really” seropositive, perhaps reflecting ACPAs missed by standard assays (46). Further, many GWAS risk loci overlap, including loci shared commonly among autoimmune diseases (e.g. PTPN22, STAT4 and TYK2), loci shared among forms of arthritis but not by other autoimmune diseases (RUNX1 and potentially IL2/IL21, IL2RB and IL6R), and loci that have so far been observed selectively in polygoJIA (IRF1, FAS, IL6 and PPRL5) (Figure 1). However, biological differences (summarized above) and contrasting HLA associations remain, as do shared and discordant genetic associations beyond the HLA, exemplified by the closer genetic similarity of polygoJIA than RA to type 1 diabetes, another early-onset disease (47, 48). There are again at least two possibilities. First, seropositive and seronegative arthritis could be distinct entities, arising in different ways. Second, in patients “destined” to develop arthritis, HLA could determine whether arthritis will be seronegative or seropositive. Since citrullination increases with inflammation, citrulline autoreactivity would then serve as an amplification loop, accounting for the greater severity and persistence of seropositive disease. Both alternatives have been seen in animal models. In some arthritic mice, citrullinated antigens represent the key autoantigen, while in others anti-citrulline reactivity amplifies arthritis arising through reactivity to collagen (49, 50). Within families, seropositive or seronegative arthritides tends to cluster with arthritis of the same type, but each also increases the risk of the other (28, 51). As shown in Figure 1, more than half of polygoJIA GWAS risk loci also confer risk for (mostly seropositive) RA (16, 17). Thus seronegative and seropositive arthritis are related, but whether as siblings or cousins remains to be established.

Spondyloarthritis

Perhaps nowhere is the gulf between pediatric and adult classification as large as in the family of conditions referred to with terms including spondyloarthropathy, ankylosing spondylitis, axial spondylarthritis, PsA, seronegative enthesopathy and arthropathy, juvenile psoriatic arthritis, psoriatic JIA, and enthesitis related arthritis (ERA). Broadly, these syndromes share a propensity for involvement of the sacroiliac joints and spine; reactive new bone formation in addition to bone erosion; arthritis favoring the large joints of the lower extremity and the distal interphalangeal joints; an intermittent course; dactylitis; inflammatory lesions of bone, colon and skin; acute anterior uveitis (rather than the chronic anterior uveitis of polygoJIA); and genetic association with HLA-B*27 (Figure 1) (52–54). A further hallmark, and in all likelihood a shared pathologic theme, is inflammation of entheses as well as synovium (55).

In the adult context, this spectrum is reflected well in common nomenclature (54, 56, 57). The term “spondyloarthritis” is recognized as a unifying overarching category, while individual clusters under this umbrella receive more specific names on the basis of clinical manifestations, severity, and concomitant conditions (e.g. psoriasis). Discrete criteria are defined for research, but in practice overlap is recognized as the norm rather than the exception.

Within pediatric rheumatology the state of affairs is more confused. In part this is because children often present ambiguously (58). For example, overt psoriasis may lag behind arthritis by a decade or more. A family history of psoriasis is non-specific; anti-arthritic medications such as methotrexate can suppress skin disease; and the possibility of TNF inhibitor-induced psoriasis clouds classification in children who develop psoriasis later. To complicate matters, psoriatic JIA echoes the older/younger-onset divide seen in JIA generally. Younger patients (onset age <6 years) tend to be ANA positive and to develop a mixed large joint/small joint oligoarthritis and chronic anterior uveitis, while older patients more closely resemble adult PsA – an observation that could lend credence to the possibility mentioned above, namely that early childhood physiology skews disease presentation, even if the underlying process is similar (59, 60). Thoughtful observers have questioned whether there is any value to distinguishing psoriatic arthritis in children. We do not share this hesitation, not least because the prevalence of overt psoriatic manifestations in patients with JIA (at least 7% in most North American and European series) so greatly exceeds the prevalence of psoriasis among children in general (~1–2%) that a chance association seems implausible, while manifestations such as nail pits and dactylitis are exquisitely specific for adult PsA (58, 61). Nevertheless, the utility of a psoriatic category remains an open debate within pediatric rheumatology.

Some of the confusion also represents self-inflicted injury. In adopting the term ERA, ILAR created a yawning nomenclature gap with adult rheumatology. To make room for ERA, criteria were defined that render it more difficult for patients with enthesitis to be termed psoriatic JIA. This is a choice that might puzzle adult rheumatologists, not only because enthesitis is a canonical manifestation of PsA but more generally because it attempts to resolve by definition a question that should really be addressed with data, namely the role of enthesitis in psoriasis-associated arthritis in children (61).

Here again, genetic studies begin to shed some welcome objective light (25). While the number of patients with ILAR-defined psoriatic JIA and ERA is small even in the largest series, Hinks and colleagues found that psoriatic JIA exhibited HLA associations that seemed at least partially distinct from polygoJIA but resembled those of adult PsA (Figure 1) (25). Both ERA and to a lesser extent psoriatic JIA exhibited associations with HLA-B*27 (findings confounded by the fact that this allele is an inclusion criterion for ERA and a potential exclusion criteria for psoriatic JIA, while the number of psoriatic JIA patients was too low for statistical confidence). These results support the suggestion of Colbert and colleagues that spondyloarthritis transcends the pediatric/adult divide, even if in practice children within this family are sometimes difficult to discriminate clinically (62).

Systemic arthritis

A final disease cluster that exemplifies the nomenclature challenges facing pediatric and adult rheumatology includes systemic JIA (sJIA) and AOSD. These conditions are sundered from one another by distinct diagnostic criteria (1, 63). Here genetics cannot yet guide us. Recent studies in sJIA found an association with HLA-DRB1*11 and distinguished sJIA genetically from polygoJIA (note complete lack of overlap in Figure 1) (64, 65). However, no comparable studies are available for AOSD. Nevertheless, clinical similarity between these disorders is widely recognized. Both are characterized by fevers and evanescent rashes; striking elevation in markers of systemic inflammation including ferritin and D-dimer; development of intractable arthritis in a subset of patients; risk for macrophage activation syndrome; a roughly even ratio of males to females; low uveitis risk; and remarkable response to IL-1 and/or IL-6 inhibition in many patients (66). Unlike seropositive arthritis, which is epidemiologically an adult disease with a pediatric “tail,” systemic arthritis is most prevalent in early childhood, while most affected adults present in the younger end of the age spectrum (67–70). Like sJIA, AOSD is characterized by sky-high levels of IL-6 and IL-18, and peripheral blood gene expression signatures are similar (71–73). Finally, both sJIA and (perhaps) AOSD resolve in a substantial subset of patients (74). Thus, although genetically-defined subcategories will likely emerge, the preponderance of evidence suggests that sJIA and AOSD should be considered part of the same systemic (Still’s) spectrum (75).

Other approaches to disease categorization

ANA as a marker of subtypes within juvenile arthritis

The high frequency of ANA in children with arthritis has prompted the proposal that non-ERA, non-sJIA seronegative juvenile arthritis should be divided into ANA-positive and ANA-negative subsets (76, 77). For this purpose ANA positivity is often defined as a titer ≥ 1:160 on two occasions at least 3 months apart, and correlates not only with higher risk of chronic anterior uveitis but also female predominance, younger age of onset, a lower number of involved joints, and other distinguishing features (78–80). However, ANA titer tends to wax and wane, and a positive ANA is not uncommon even in healthy children (81). Uveitis risk is not restricted to children with a persistently high ANA titer, and ANA positivity does not correlate with peripheral blood gene expression or disease course independent of age of onset (37, 39, 42, 82). Whereas children with early-onset arthritis are more commonly ANA positive, differences between ANA-positive and ANA-negative JIA could simply reflect enrichment for early-onset seronegative arthritis. Nevertheless, autoantibodies are an important mirror of pathogenesis (4), and a pathogenic understanding of childhood arthritis must ultimately explain why some children that cluster phenotypically also exhibit substantial levels of ANA.

“Big data” to define disease categories

One promising approach to identifying disease subcategories uses bioinformatics to find patterns within complex datasets. Recently, Yeung and colleagues employed this strategy in children with non-systemic JIA (83). As input variables, they selected patient demographics; time to diagnosis; clinical and laboratory data as used for ILAR categorization; ANA; standard measures of disease activity; laboratory parameters; and cytokine and chemokines as measured by multiplex assay. Using machine learning, they identified 5 patient clusters, which were superior to ILAR categories in defining homogeneous groups (per the same variables). These clusters replicated in a validation cohort and correlated with disease trajectory over 6 months. To summarize the features of each cluster briefly: (Cluster I) older age at diagnosis, low cytokines, Th1 skew; (Cluster II) low disease activity, low circulating cytokines, Th1 skew; (Cluster III) any age at diagnosis, high disease activity, moderate levels of Th1-, Th17-, and macrophage-associated cytokines; (Cluster IV) young age at diagnosis, short interval between onset and diagnosis, elevated platelets, lower hematocrit, and tendency to express macrophage and Th2 cytokines; and (Cluster V) older age at diagnosis, longer interval between onset and diagnosis, lower disease activity but higher levels of circulating cytokines.

The complexity of these clusters highlights the challenges intrinsic to such studies. Investigators choose what to include, what to leave out, and how to value one feature against another. Here, if the investigators had used different variables (e.g. genetic risk alleles), prioritized some over others, or included adult as well as pediatric patients, then they would have arrived at a different set of clusters, and these would also likely have replicated at validation. Further, the variables tested must be the important ones – imagine an attempt to classify pneumonias without including the causative pathogen. The goal of big data studies is to define arthritis subtypes that differ in pathogenesis, and therefore in prognosis and therapy. Which big data will best address this goal is a not a trivial question.

Thesis: There is no such thing as juvenile idiopathic arthritis

The considerations raised in this review suggest that the adult/child divide is a poor foundation for the classification of arthritis. Put more forcefully, there is very little evidence that there is such a thing as juvenile idiopathic arthritis, any more than there is juvenile cellulitis or juvenile pneumonia. Of course, each form of arthritis will have its own age predilection; yet even the most narrowly-defined pediatric phenotype, ANA-positive persistent oligoarthritis with chronic anterior uveitis, is not a disease of children per se but rather of young children, sparing the 12 year old almost as completely as the 40 year old. Conversely, seropositive polyarthritis in a 14 year old girl is not a rare form of JIA but rather the most common idiopathic inflammatory arthritis in the Western world. Almost all forms of arthritis cross the adult/pediatric divide. Distinctions baked into current nomenclature simply reflect the fact that adult and pediatric rheumatologists have historically addressed disease classification separately rather than together.

We are left therefore not with forms of juvenile arthritis and forms of adult arthritis, but rather with forms of arthritis, period. Erring on the side of lumping rather than splitting, we highlight four clusters: (1) seropositive arthritis; (2) seronegative arthritis, encompassing seronegative RA and polygoJIA, likely with a distinct subgroup with onset typically before the age of 6 years; (3) spondyloarthritis, including ankylosing spondylitis and PsA; and (4) systemic arthritis (Figure 2). Of these, all but early-onset seronegative arthritis present to both pediatric and adult rheumatologists. Growing understanding will no doubt allow us to split these clusters into smaller, biologically-discrete entities, as exemplified by the recent identification of what appears to be a form of systemic arthritis mediated by loss of function at LACC1 (84, 85). However, we need let the data generate the splits, rather than a priori assumptions about differences between children and adults.

Figure 2. Four major clusters of arthritis as informed by human genetics.

ILAR categories of JIA are depicted as boxes. Early-onset arthritis is depicted as a subcategory within seronegative arthritis.

It is important to emphasize that even if the diseases are similar across the age spectrum, it does not follow that management is the same. There are many ways in which kids are really not little adults. These include the growing skeleton; drug metabolism and toxicity; differential diagnosis; exposures and comorbidities; psychosocial development; and where and how we care for patients. For some purposes it will still make sense to study children and adults separately, not because the diseases are different but because the patients are. We need rheumatologists who are pediatricians and rheumatologists who are internists, and perhaps some who are both. If we recognize the similarities as well as the differences among our diseases, then we will all be better positioned to learn from each other and from our patients.