Systemic sclerosis (SSc) or scleroderma is an idiopathic, autoimmune disorder resulting from the abnormal production and deposition of collagen and other extracellular matrix proteins by inappropriately activated fibroblasts^1,2. Deposition of extracellular matrix proteins in the skin results in skin hardening, thickening and tethering, whilst deposition in blood vessels leads to a fibroproliferativevasculopathy and the cardinal symptoms of Raynaud’s phenomenon and digital ulceration^1,3. Deposition in internal organs leads to pernicious sequelae which include scleroderma renal crisis and pulmonary fibrosis^1,3,4.

Clinically, the severity of SSc is classified according to the extent of skin involvement, either limited to the face and extremities distal to the elbows/knees (limited cutaneous systemic sclerosis, lcSSc) or extended to the upper arms, thighs and/or trunk (diffuse cutaneous systemic sclerosis, dcSSc). These changes may occur with or without internal organ involvement ranging from subclinical, non-progressive involvement to severe fibrosis leading to the development of multiple organ failure^1,4. dcSSc is associated with more frequent internal organ involvement, and a faster progression of fibrosis compared to lcSSc⁵.

The prevalence of SSc has a high geographical variation, between 7 and 489/million population, and an incidence between 0.6 and 122/million per year, highest in the United States and Australia⁶. Morbidity and mortality rates for SSc are the highest amongst rheumatologic conditions. The main determinants of prognosis are: the extent of skin and internal organ fibrosis, rate of progression of fibrosis and type of organ involvement¹. These determinants are currently not predictable, either for individual patients or patient subgroups before significant organ involvement has occurred^1,7.

SSc represents an ideal candidate for a stratified approach to patient management as it is highly heterogeneous in terms of severity and prognosis^1,7. This stratification requires the development and validation of appropriate clinical biomarkers for diagnosis, disease classification, identification of organ involvement and evaluation of therapeutic response¹.

The modified Rodnan skin score (MRSS) is a fully validated clinical biomarker, used as the primary outcome measure in most clinical trials⁸. However, MRSS is entailed by a 25% inter-observer variability shortcoming⁹, so for accurate and reliable patient stratification novel approaches are required^7,10.

Although there have been recent reviews published on the subject of clinical biomarkers in systemic sclerosis, these reviews did not focus on the stratified medicine approach and failed to consider the economic incentive for the development of clinical biomarkers^1,7,11. This review aims to address both of these criticisms and also provide a more up to date interpretation of the current approaches to identify clinical biomarkers in systemic sclerosis.

Stratified medicine and clinical biomarkers

Stratified medicine is an approach, which put simply, treats different patients differently. It forms part of a continuum of treatment approaches, ranging from individualised medicine at one extreme, through stratified medicine to the norm of empirical medicine¹².

Individualised medicine is a process which uses the patient’s genotype or phenotype to create an individual treatment, which works for one particular patient. An example of this is the cancer vaccine Oncophage which is derived from the individual patient’s cancer cells¹². At the opposite end of the spectrum, empirical treatments (which are the norm) are medicines which either work for most patients, or work for some patients but there is currently no way of distinguishing patients who would benefit from those who would not. Examples of empirical treatments are non-steroidal anti-inflammatory drugs (NSAIDs) which work for most patients and selective serotonin-reuptake inhibitors (SSRIs) in which responses vary¹².

A stratified treatment is one that has been historically shown to correlate with a beneficial therapeutic response in a specific cohort of patients^12,13. An example of a stratified approach to treatment is trastuzumab (Herceptin) which benefits only patients stratified by the presence of the clinical biomarker: HER2 positivity in breast cancer^14,15.

The definition of a clinical biomarker is quite broad, being any measurement which allows for the stratification of patients into different subgroups by the objective measurement of a normal biological process, pathogenic process or response to therapeutic intervention¹. Examples include: a patient’s genotype, the presence of a specific protein or proteomic expression, a particular metabolite, histology, imaging or simply a clinical sign¹². Ideally, newly developed clinical biomarkers should be easily obtainable, rapidly measureable and must be validated in clinical studies¹.

Health economics of stratified medicine

For stratified medicine to become widely adopted there needs to be additional incentives other than the optimisation of patient treatment. The use of pravastatin is far less effective when associated with a certain genetic subset¹⁶, however patients are not stratified as empirical prescription is far more efficient in terms of cost and time¹².

Bernatsky et al. performed an analysis of the direct (medical provision) and indirect (lost productivity) costs of systemic sclerosis¹⁷. Although systemic sclerosis is rare, it represents a condition with considerable morbidity, disability and mortality and a social cost per patient exceeding that of rheumatoid arthritis¹⁷. Two thirds of these costs are incurred in the last 6 months of life. On the other hand, spontaneous remission often occurs and is unpredictable. And so by both under and over-treating patients, additional costs are incurred which could be reduced if it was possible to stratify patients for treatment.

The development of novel therapeutics is an extremely risky enterprise, as a general rule predicted peak annual sales of over $500 million are required to justify their development¹². Currently, withoutbiomarkers which measure disease responseit is impossible to determine whether beneficial effects are due to treatment or the natural course of disease. This problem is further compounded by the rarity of systemic sclerosis and the small potential market size. However, the potential to reduce the size of clinical trials needed to attain significance and therefore cost, as well as the ability to maximize the therapeutic response and minimize side effects may lead to rapid adoption and reimbursement, as was the case of the blockbuster drug tratuzumab¹².

The OMERACT filter

The OMERACT (outcome measures in rheumatology) filter, was developed to identify the validity of a potential clinical biomarker^7,18. The OMERACT filter demands that potential clinical biomarkers must comply with the concepts of truth, discrimination and feasibility (table 1).

Table 1. Components of the OMERACT filter used to assess validity of a novel biomarker, adapted from^7,18.

The modified Rodnan skin score

There are a number of methods for measuring skin involvement in SSc. The modified Rodnan skin score (MRSS) is currently the gold standard^2,8,19,20. However, reviews consistently suggest that MRSS is a flawed clinical biomarker^7,11,19. There are relatively few studies of MRSS in the literature, none of which conclude that MRSS is significantly unreliable to refute its gold standard status^2,5,8,20,21.

MRSS is a measure of skin thickness and has become a validated method of assessing skin involvement in SSc. I it was originally shown to correlate well with skin biopsy sample mass and thus assumed to be an accurate reflection of the fibrotic process¹. It involves the clinician palpating the patient’s skin at 17 specific points (face, anterior chest, anterior abdomen and bilaterally at the upper arm, forearm, dorsum of hand, fingers, thigh, lower leg and the dorsum of foot) and rating these on a four point scale: 0 = no thickening, 1 = mild thickening, 2 = moderate thickening, unable to pinch, 3 = severe thickening, unable to move[8, 20]. Skin thickness can range from 0 (no thickening in any area) to 51 (severe thickening in all 17 areas)⁸. MRSS is cost effective and non-invasive so it meets the OMERACT criteria of feasibility. However, MRSS has three main weaknesses:

1. It is only approximately 75% reproducible between observers^9,20. OMERACT criteria of discrimination - reproducible¹⁸.

2. It may be unable to detect small but clinically significant changes in skin thickness^2,8. OMERACT criteria of discrimination - change over time¹⁸.

3. It may not actually measure skin involvement as it claims to¹. OMERACT criteria of truth - face and construct validity¹⁸ .

MRSS has high inter-observer variability

In a recent study, Ionescu et al. argue that MRSS is a reliable and accurate measure of skin thickness between “inexperienced” observers²⁰. Their study aimed to investigate the accuracy of inexperienced rheumatologists, by offering repeated teaching on conducting MRSS. It also provided useful raw data on MRSS obtained by four “experts” in rheumatology who acted as teachers in this experiment²⁰.

The authors analysed the variation of total MRSS score obtained by “experts” in a sample of 13 patients. The outcome measure was that an intraclass correlation coefficient (ICC) of 0.4-0.6 is “moderate”, 0.6-0.8 is “good” and >0.8 is “excellent” agreement. The “experts” attained an ICC of 0.743 and so the authors concluded that MRSS was a reliable clinical biomarker²⁰. This correlates with a previously observed correlation between observers of approximately 75%⁹. However, the reliance upon this statistical test, and the nullifying effect of averaging the results meant that meaningful differences between observers were overlooked by the authors.

Not only is a “good” ICC in “experts” disappointing, particular attention should be paid to patients 1 and 6 where the highest value obtained is more than twice that of the lowest value (>2 fold difference) and patients 3 and 7 where there is a >9 fold and >4 fold difference respectively (Table 2). Particular attention should also be paid to patients 5, 6, and 7 where the absolute difference in MRSS is >10 points as this represents a large change in disease activity out of a possible of 51 points²⁰.

Table 2. Highest and lowest values of MRSS recorded by “experts” in patients with SSc; Fold difference = Highest value/Lowest value. NA = not applicable. Adapted from Ionescu et al.²⁰.

The authors concede that: “the high inter-rater variations seen in some of the investigated patients demand that in clinical studies the same investigator should assess the skin score in the same patient on each visit”²⁰. This causes practical difficulties in large clinical trials. This recommendation does not resolve the difficulty of managing patients over the course of their disease, where they may not be able to see the same physician. Neither does it resolve the difficulty in recommending different treatments stratified according to MRSS.

One possible answer to this weakness of MRSS is that, instead of stratifying patients by MRSS score, they be stratified instead by the rate of change of MRSS score. This would provide a much better insight into the fibrotic activity, and the need for intervention. However, there is evidence that MRSS may lack the sensitivity to detect small, but clinically meaningful changes in skin thickness over time⁸.

MRSS is not sensitive to change

Kaldas et al. compared data obtained from two randomisedcontrolled trials^22,23 which used MRSS as the primary outcome measure for patients with dcSSc to assess whether MRSS was sensitive to change over time⁸ .

The first compared the therapeutic effect of patients (n = 231) who were blindly randomised to receive either continuous subcutaneous infusion of recombinant human relaxin, or placebo. The second randomised patients (n = 124) to receive either oral bovine type I collagen or placebo. In both trials, there was no statistically significant difference between treatment arms.

Kaldas et al. used patient global assessment (PGA) as an anchor to compare to MRSS, which is a quantification of patients’ reported symptoms and is frequently used as an outcome measure in rheumatology^8,24. The authors concluded that: “Our analysis validates the total MRSS score as an outcome measure”, as they found that the total MRSS correlates well with the PGA, with an effect size (ES) of 0.85 and 0.98 in the respective studies⁸. Again, this is an oversimplification, as the study found that measurements at the lower extremities, abdomen, fingers and face are not sensitive to change over time. The authors concluded that these results may be due to the difficulty in assessing these areas and possibly the large inter-observer variation in these areas⁸.

The trials took place over 24 and 52 weeks respectively⁸, providing further evidence that MRSS is insensitive to change as it failed to detect change at multiple body points, despite changes in PGA over such a long treatment period⁸. However, it must be conceded that changes in PGA may be non-specific due to: vasculopathy, respiratory and gastrointestinal pathology, arthritis and cold weather. So the apparent insensitivity of MRSS may in fact but be due to the ineffectiveness of these two treatments, although the use of double-blinded placebo control reduces this probability.

MRSS may lack truth validity

The final weakness of MRSS is that it may lack the OMERACT criteria of truth. There is evidence to support the idea that MRSS may be unable to differentiate between changes in skin due to fibrosis and changes due to oedema, inflammation, vascular bed engorgement and skin tethering¹. Milano et al. demonstrated that the there was significant overlap in the gene expression in affected and unaffected skin (discussed subsequently)²⁵.

MRSS in summary

In conclusion, MRSS is a subjective outcome measure, which may be described as semi-quantitative at best¹⁹. It relies on the quality of training of the observer and is open to disagreement between even experienced observers²⁰. It also requires that in clinical trials, a large sample of patients are included to obtain statistical significance¹⁹ and that one observer be used to measure all patients’ MRSS at all time points²⁰.

This creates a problem, as the OMERACT filter demands that novel clinical biomarkers have criterion validity, that is they correlate well with the gold standard (Table 1)¹⁸. This presents potential difficulties because if novel clinical biomarkers are compared to MRSS for validation, then there is a potential that the truth and discrimination problems with MRSS will be translated to the new measure.

Other physical measurements of skin involvement

The durometer is a hand-held device which measures skin hardness (as opposed to thickness in MRSS) and has been assessed in a multi-centrerandomised controlled trial for validity and reliability against MRSS²⁶ and so meeting the OMERACT criteria of truth - criterion validity (Table 1)². Crucially, the durometer has been shown to be more reliable, and sensitive to changes over time than MRSS, particularly at the fingers, hands, forearms, upper arms, thighs and feet².

High frequency ultrasound (10-30MHz) has been shown to be more sensitive to small, but meaningful changes in skin fibrosis compared to MRSS¹¹. Ultrasound may also be able to distinguish between changes in skin due to oedema and induration or sclerosis, a known weakness in MRSS¹¹. Ultrasound is far more reliable and reproducible between observers than MRSS, however, it lacks the OMERACT criteria of feasibility (Table 1) as it requires specialist equipment not readily available in most centres and demands the time consuming exact determination of several body sites to achieve accuracy¹¹.

The use of magnetic resonance imaging (MRI) scanning for skin thickness, is again limited by the OMERACT criteria of feasibilityas it is expensive, prohibiting its use in large clinical trials and routine patient follow up¹¹.

Problems with all physical measurements of skin involvement

A recent gene expression study by Milano et al. has suggested that the fibrotic process which results in thickening skin in SSc patients may be present in unaffected as well as clinically affected, thickened skin^7,25. Gene expression did not therefore correlate with MRSS, there are two interpretations to this finding:

1. There was a failure of criterion validity (Table 1) of genome profiling; that is, the gene profiling study did not correlate well with the gold standard and so was incorrect

2. There was a failure of the construct validity (Table 1) of MRSS; that is, MRSS does not accurately reflect the underlying pathology in SSc⁷.

However, the most significant failing of MRSS and other physical measurements of skin involvement is that they are retrospective. They give information of disease activity after clinical changes have occurred, whereas an appropriate clinical biomarker which allows treatment stratification and prognosis, must predict clinical changes⁷.

Molecular approaches to identify clinical biomarkers

Although the dermatological aspect of SSc is the most obvious, it is not necessarily the most relevant in the search for clinical biomarkers. SSc is a systemic disease, with approximately two thirds of deaths due to internal organ complications, so the development of biomarkers which reflect internal organ involvement is appropriate^3,7,19.

Biomarkers which measure skin involvement, rely on the assumption that skin thickness/hardness is correlated well with internal organ involvement⁸, and although there is evidence to support this notion²⁷, there is an urgent clinical need to identify biomarkers which better reflect the fibrotic activity of patients with SSc¹⁹.

The traditional approach to identify (molecular) clinical biomarkers in SSc is by a ‘hypothesis driven’ approach, where by a molecule which was known, or postulated to be related to SSc pathogenesis was measured in patients with SSc and attempts at stratifying patients by these measures were made⁷. In unbiased, non-hypothesis driven approaches, large scale analysis of the molecular footprint of SSc patients, whether this be proteins, DNA or other molecules are analysed in an attempt to find molecular biomarkers which correlate with disease activity⁷.

Genetic risk factors in SSc

Genome wide studies of SSc patients to identify shared polymorphisms have generated limited and frequently contradictory results⁷. The exception to this rule has been the identification of specific human leukocyte antigen (HLA) haplotypes²⁸. Recent genome wide genetic profiling has revealed several loci aside from the HLA alleles. Radstake et al. performed a genome wide study in over 5000 patients and identified loci (CD247, STAT4 and IRF5) associated with SSc, however there was considerable overlap with similar conditions such as systemic lupus erythematous and rheumatoid arthritis^29-31. Gorlova et al. performed a meta analysis of data previously collected by Radstake et al.³². They identified three non-HLA genes: IRF8, GRB10 and SOX5 associated with SSc, as well as specific HLA loci confined to specific auto-antibodies subgroups (discussed subsequently)³². However, at this point these associations do not offer sufficient discrimination for use in isolation as clinical biomarkers. Although there is potential for several genetic loci to contribute to a clinical biomarker which may be used for the stratification of patients into lcSSc/dcSSc subgroups³². Moreover, these studies have provided crucial insights into the pathogenesis of SSc.

By analyzing not for the presence of polymorphisms, but the transcriptome of a particular tissue – the total RNA produced, a more accurate picture of the molecular characteristics of SSc patients can be obtained. This technique is well established in cancer research, but is a novel approach in SSc⁷. Milano et al. have shown that a set of genes are over expressed in patients with SSc, in affected, and unaffected skin, which possibly invalidates the truth criteria of MRSS (discussed previously)²⁵.

These studies, although extremely promising do not pass the OMERACT filter, as this technique may not have feasibility for use in a clinical setting, due to both financial and time constraints⁷, although the rapid progress in this area will perhaps make this possible in the future.

Autoantibodies

Alongside the fibroproliferativevasculopathy which occurs in SSc, there are severe alterations in both cellular and humoral immunity, leading to the production of numerous antinuclear antibodies (ANA)¹. There are several validated biomarkers in use for the early diagnosis of SSc, however due to their varying specificity and sensitivity, they are not used currently in clinical diagnostic criteria, but may be used as adjunct for diagnosis¹.

ANA are present in greater than 90% of patients with SSc³³. Examples are: anti-Scl-70, anticentromeric anti-RNA polymerase I and III, anti-fibrillarin and anti-PM-Scl. Anti-Scl-70 (more accurately termed antitopoisomerase I³⁴ and anticentromeric antibodies are found in the diffuse and limited forms of SSc respectively, and rarely in both^1,35,36. Anti-Scl-70 has high disease specificity, against healthy controls, other connective tissue diseases, and classifications of SSc: found in 37% of patients with dcSSC but only 10% of patients with lcSSc ^1,7,37. However, anti-Scl-70 is relatively insensitive, overall occurring in only 20% of SSc cases⁷.

Anticentromeric antibodies are specific for lcSSc, occuring in around 20-30% of all SSc cases, but greater than 90% of lcSSC patients³⁸. These are therefore very interesting for the stratification of patients with SSc, however anticentromeric antibodies are also insensitive, occurring in only 20-30% of SSc patients. They are also insensitive to change over time and so limited to diagnosis and classification purposes^7,39.

Other antibodies (anti-RNA polymerase I and III, anti-fibrillarin, anti-PM-Scl) shall not be reviewed here, as they are infrequently observed in patients with SSc, and have been reviewed elsewhere⁷.

Autoantibodies can therefore be thought of as adjunct for diagnosis and stratification, as in the terminology of the OMERACT filter they do not offer sufficient discrimination to be used as clinical biomarkers in isolation^1,7.

Proteomics

Proteomics is the study of the entire complement of proteins produced by an organism, system or cell². To date, there have been four approaches for characterizing the proteome of SSc: bronchoalveolar lavage^40,41,saliva⁴², in explanted lung fibroblasts⁴³ and in the skin^19,44. There are strengths and weaknesses of all four approaches.

The first approach, identifying proteins in bronchoalveolar lavage is interesting but may not be feasible in a clinical setting, as it has large cost and time implications⁷. To overcome this problem, the measure must be significantly better than the current gold standard for measuring lung involvement in SSc: high resolution CT scan, but there is currently no evidence that this is the case⁷.

The second approach, which identified biomarkers in the saliva of patients with SSc has better feasibility, as saliva can be more readily obtained⁴². However, as there was only a comparison between SSc and healthy controls, this study may lack the specificity for SSc due to potential overlap with other autoimmune or fibrotic diseases. Bogatkevich et al. identified a complement of proteins, which are over expressed in the presence of connective tissue growth factor. Their study was carried out with the aim of increasing the knowledge of the pathogenesis of SSc⁴³, but presents a complement of proteins which may have future use as biomarkers of interstitial lung disease. This is a crucial approach, as 42% of death in SSc are due to pulmonary manifestations³. However they may offer little in the way of patient stratification and therapeutic response, other than in regards to lung pathology.

Aden et al. performed proteomic analysis of skin biopsies to identify the profile of protein expression in clinically involved skin⁴⁴. The authors concluded that skin in SSc displays a phenotype which is characteristic of wound healing⁴⁴. It is unlikely, however that candidate biomarkers identified by this approach will be specific for systemic sclerosis, due to this non-specific ‘healing’ phenotype which was observed.

Carlsson et al. conducted an impressive study which compared the serum proteome of SSc and systemic lupus erythematous (SLE), another autoimmune disease⁴⁵. This approach is very promising, as it directly identified proteins in the blood, meaning that the feasibility of identified clinical biomarkers in unquestionable. However, by using another autoimmune disease to compare with SSc, the authors inevitably identify proteins associated with immune system dysregulation. Although the authors demonstrated correlation with disease severity, as well as specificity of SSc proteome against SLE, they concede that the expression in SSc was similar to healthy controls⁴⁵. Despite these problems, a full scale validation of candidate biomarkers is greatly anticipated.

A final study by Del Galdo et al. identified nine proteins (Table 3), candidate biomarkers of the pro-fibrotic phenotype from explanated dermal fibroblasts, by analysis of two dimensional gel electrophoresis and identification by mass spectrometry¹⁹.These candidate biomarkers are of particular interest as they have the potential to meet all three criteria of the OMERACT filter (Table 1).

Table 3. Proteins expressed preferentially in explanted dermal fibroblasts in both SSc and NSF¹⁹Protein