Objectives: The use of biologicals such as infliximab has dramatically improved the treatment of rheumatoid arthritis (RA). However, factors predictive of therapeutic response need to be identified. A proteomic study was performed prior to infliximab therapy to identify a panel of candidate protein biomarkers of RA predictive of treatment response.
Methods: Plasma profiles of 60 patients with RA (28 non-responders (as defined by the American College of Rheumatology 20% improvement criteria (ACR20)) negative and 32 responders (ACR70 positive) to infliximab) were studied by surface enhanced laser desorption/ionisation time-of-flight mass spectrometry (SELDI-TOF MS) technology on two types of arrays, an anion exchange array (SAX2) and a nickel affinity array (IMAC3-Ni). Biomarker characterisation was carried out using classical biochemical methods (purification by ammonium sulfate precipitation or metal affinity chromatography) and identification by matrix assisted laser desorption/ionisation time-of-flight (MALDI-TOF) MS analysis.
Results: Two distinct protein profiles were observed on both arrays and several proteins were differentially expressed in both patient populations. Five proteins at 3.86, 7.77, 7.97, 8.14 and 74.07 kDa were overexpressed in the non-responder group, whereas one at 28 kDa was increased in the responder population (sensitivity>56%, specificity>77.5%). Moreover, combination of several biomarkers improved the sensitivity and specificity of the detection of patient response to over 97%. The 28 kDa protein was characterised as apolipoprotein A-I and the 7.77 kDa biomarker was identified as platelet factor 4.
Conclusions: Six plasma biomarkers are characterised, enabling the detection of patient response to infliximab with high sensitivity and specificity. Apolipoprotein A-1 was predictive of a good response to infliximab, whereas platelet factor 4 was associated with non-responders.
Statistics from Altmetric.com
Prevalence of rheumatoid arthritis (RA) varies between 0.3 and 0.8% in European countries.1 Joint invasion and destruction are mainly mediated by inflammatory cells, cytokines and proteolytic enzymes. The heterogeneity of pathological manifestations, disease course and response to treatments suggests that several subtypes of RA exist, especially at a molecular level.
Early referral, diagnosis and initiation of treatments to control tissue damage and structural progression are highly recommended, considering the disease’s high propensity towards joint destruction even during periods of clinical remission.2 Although classical treatments are still used,3 4 a radical new approach to RA treatment is now available with the use of the biological therapies, such as tumour necrosis factor α (TNFα) inhibitors (etanercept, infliximab, adalimumab), anti-CD20 antibodies (rituximab), interleukin 1 (IL1) receptor antagonists (anakinra) and an inhibitor of T lymphocyte costimulation (abatacept). Almost 70% of treated patients respond to these therapies, but for unknown reasons approximately a third of patients with RA do not respond.5 Indeed, the currently used markers reflecting bone and cartilage turnover6 are few and inaccurate, and no predictive factor of treatment resistance has been identified. Discovery of biomarkers for diagnosis, prognosis and disease monitoring7 is necessary.
In order to identify proteins predictive of a treatment response, we chose to study the plasma proteome (ie, the whole protein content) of patients with RA by surface enhanced laser desorption/ionisation-time of flight mass spectrometry (SELDI-TOF MS) developed by Ciphergen Biosystems (Fremont, California, USA).8 9 This method has allowed characterisation of numerous biomarkers for many diseases,10–12 particularly cancer.13–16
Concerning RA, a first SELDI-based study on 12 synovial fluid samples characterised an RA biomarker absent in patients with osteoarthritis, the proinflammatory S100A8 protein.17 More recently the same protein was identified in a larger survey comparing serum protein profiles of RA and non-inflammatory control samples.18
In our study, we used SELDI-TOF MS technology to characterise biomarkers predictive of an anti-TNFα therapy response in patients with RA. Protein profiles were performed on plasma specimens taken prior to the start of infliximab treatment, from patients that were further evaluated as non-responders (as defined by being American College of Rheumatology 20% improvement criteria (ACR20) negative) or responders (ACR70 positive) to the therapy. Several biomarkers were isolated and were highly correlated with the infliximab response independently or when associated in a decision tree algorithm. Moreover, corresponding proteins were identified using classical biochemistry techniques.
PATIENTS AND METHODS
Patients and study protocol
Patients with a RA diagnosis according to ACR criteria were recruited in four university hospitals from the Rhône-Alpes area. All patients had disease duration of at least 6 months, had never received anti-TNFα therapy before inclusion and had no active or latent bacterial, fungal or viral infection at the time of enrolment. Blood was withdrawn on a spray-coated K2EDTA tube before the first infliximab infusion and plasma aliquots were stored at −80°C. Patients then received the same infliximab treatment as previously described19 and clinical outcome was assessed at week 30. Among the 60 recruited patients, 28 were ACR20 negative (non-responders), whereas 32 were ACR70 positive (responders). Differences between both populations of each classical parameter measured at inclusion were evaluated with the Mann–Whitney U test or the χ2 test.
The plasma protein profile was investigated with the SELDI-TOF MS method developed by Ciphergen Biosystems that associates a first phase of protein fractionation on chips with distinct chromatographic surfaces (cationic, anionic, hydrophobic etc) to a second detection step of protein molecular weight by mass spectrometry after laser desorption. We used an anion exchange array (strong anion exchange (SAX)2) and a nickel affinity chip (immobilized metal affinity chromatography (IMAC)3-Ni). Firstly, the sample protein concentration was evaluated using the Bradford method.20 For the SAX2 array, 5 μl of plasma diluted to a final concentration of 0.7 μg/μl in a washing/loading buffer (10 mM Tris, 0.1% Triton X100 pH 7) was loaded twice in duplicate for a 30 min incubation period at room temperature. Then spots were washed twice with 5 μl of washing buffer, twice with 5 μl of the same buffer without Triton X100 and finally with 5 μl of 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), pH 7. For the IMAC-Ni array, 20 μl of plasma was mixed with 30 μl of a denaturation buffer (50 mM Tris, 8 M urea, 1% 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate (CHAPS) pH 7.4) for 15 min at +4°C with frequent agitations, and 5 μl of this protein solution diluted in the washing/loading buffer (phosphate-buffered saline (PBS), 0.5 M NaCl, 0.1% Triton X100 pH 7.3) to 0.7 μg/µl was applied on each spot previously activated with 100 mM NiSO4. Incubation and washing steps were performed as previously described, using the IMAC3 buffer with (first two washes) or without Triton X100 (last two washes). Then 0.8 μl of saturated sinapinic acid in 0.5% trifluoroacetic acid, 50% acetonitrile was applied twice on each spot and chips were analysed on the ProteinChip reader (model PBSII). Mass data processing was performed with the Ciphergen ProteinChip software V 3.1 and intensities, (height from baseline), of characterised proteins were compared in each group using the Mann–Whitney U test. Receiver operating characteristic (ROC) curve analyses were performed to measure the diagnostic potential of each biomarker and correlations between variables (biomarkers and clinical data) were evaluated using the Spearman rank test. Finally, classification and regression tree (CART) analysis allowed data classification in a tree-like structured decision diagram.21
Protein purification and identification
Apolipoprotein A-I was purified from 300 μl of responder plasma by protein precipitation with 50% ammonium sulfate. After centrifugation at 14000 g for 10 min, 5 μl of supernatant (6.35 μg) was loaded on a 15% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel which was further stained with Coomassie blue R-250. The only band revealed at about 28 kDa was excised from the gel, trypsin digested and sequenced by liquid chromatography-tandem mass spectrometry (LC-MSMS).
Platelet factor 4 (PF4) was purified from 5.2 ml of non-responder plasma on a cobalt-based affinity resin. After washings, the matrix-bound proteins were desorbed with 6 ml PBS ×1 containing 20 mM imidazole and 0.5 M NaCl. This specific eluate was dialysed overnight in distilled water, lyophilised and the resuspended powder was finally loaded on an 18% SDS-PAGE gel. A thick Coomassie blue R-250 stained band between 6 kDa and 14 kDa was sequenced by LC-MSMS as previously described. As results were not conclusive, we set up an ELISA for PF4 in order to deplete the column specific eluate from the 7769 Da biomarker. Then, 100 μl of eluate diluted twice in 2% bovine serum albumin (BSA) was incubated on a 96-well plate for 2 h at room temperature in wells previously coated with 1, 2 or 10 μg of monoclonal anti-human PF4 antibody (R&D Systems, Lille, France). Next, 2 μl of supernatant was loaded on a normal phase NP20 chip, which was analysed on the ProteinChip reader. Results were expressed as a ratio of the intensity of the 7769 Da peak to the BSA peak value in order to overcome possible loading discrepancies and significance was assessed using the Wilcoxon test.
Blood protein profiles of patients with RA before the start of infliximab treatment
Characteristics of enrolled patients are summarised in table 1. Both populations were similar in terms of patient identity (age and sex), RA characteristics (disease duration, number of swollen and tender joints) and biological parameters such as rheumatoid factor (RF), C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR), demonstrated that in this group, as well as in the general RA population, no classical marker of RA severity and aggressiveness could predict the patient response to an anti-TNFα therapy.
Two distinct protein profiles were obtained through plasma analysis on the SAX2 and IMAC-Ni arrays (fig 1A,B). Moreover, duplicate loadings of samples showed a similar protein pattern, characterising the reproducibility of our experimental procedure (data not shown). Several proteins were found to be differentially expressed in both patient populations: only one small protein at 3.86 kDa was characterised on the SAX2 anion exchange array whereas five distinct markers at 7.77, 7.97, 8.21, 28.0 and 74.07 kDa were identified from the IMAC-Ni affinity array (table 2). All biomarkers were significantly increased in the non-responder population, except the 28.0 kDa protein, which was overexpressed in responders, and they discriminated both populations with an overall high specificity (>77.5%), a lower sensitivity (>56%) and an area under the curve (AUC) ranging from 0.761 to 0.846, depending on the biomarkers. Correlation analyses between the six biomarkers showed that only the three proteins around 8 kDa isolated on IMAC-Ni were strongly positively related: 7.77 and 7.97 (r = 0.587, p<0.001), 7.77 and 8.21 (r = 0.597, p<0.001), and 7.97 and 8.14 (r = 0.869, p<0.001).
The entire protein profiles of both populations were then compared in a multivariate CART analysis. The best tree, obtained with the IMAC3-Ni proteomic profile, comprises five nodes: following the path according to the node value (protein peak intensity) leads to a terminal node classifying the patient in either of the two populations. This classification tree predicts infliximab response in this population with a specificity of 97.5% and a sensitivity of 97.1% (fig 2). Combining proteomic profiles with clinical data (CRP, ESR, disease duration and number of swollen or tender joints) or using the SAX2 and IMAC3 profiles did not improve specificity and sensitivity. Moreover, three proteins of the tree at 7.77, 8.21 and 28 kDa are individually highly discriminating (table 2). Box plot representation of individual peak intensities of these proteins showed a tiny overlap of values between both populations (fig 3A–C), especially for the 7.77 and 28 kDa biomarkers. We then focused on the characterisation of these two biomarkers.
Purification and identification of the 7.77 and 28.0 kDa biomarkers
The 28 kDa biomarker was first purified by ammonium sulfate precipitation: although part of the protein was precipitated in the pellet, a nearly pure band was recovered at about 28 kDa in the supernatant as shown on an SDS-PAGE gel (fig 4A). Loading supernatant on an IMAC3-Ni chip confirmed this protein to be the previous identified biomarker at m/z 27 976 (data not shown). Finally the sequencing of this protein by LC-MSMS identified it undoubtedly as apolipoprotein A-I (see Supplementary material). Correlation analysis showed that this biomarker was independent of clinical variables.
Identification of the 7.77 kDa required more steps. A first purification step on an IMAC-based column allowed the isolation in the specific eluate of one main protein at m/z 7769 contaminated by faint amounts of three proteins between 8 and 9 kDa, as characterised on the IMAC-Ni and NP20 arrays (data not shown and fig 4B). Fractioning this sample on an SDS-PAGE revealed a thick band around 8 kDa that was sequenced by LS-MSMS: several proteins were characterised, among which PF4 (or CXCL4) was the most represented (see Supplementary material). Based on previous studies identifying this 7769 Da protein as mature PF4 (without the first 31 amino acids of the signal peptide),22 23 we tried to deplete the column eluate from the biomarker with a homemade PF4 ELISA. The decrease of the peak intensity in the supernatant after incubation with anti-PF4 antibodies, evaluated on an NP20 array, was proportional to the increase of the amount of antibodies coated in the wells (fig 4B), with a significant 44% reduction of the signal for the highest antibody concentration (fig 4C). This last experiment formally identified the 7.77 kDa biomarker as platelet factor 4. As previously, this biomarker was not correlated to the clinical parameters.
In this study with SELDI-TOF MS we characterised several biomarkers closely correlated to the response to infliximab therapy in patients with RA. Although other clinical and biological parameters such as number of inflamed joints, human leukocyte antigen (HLA) haplotypes or CRP level, can partially predict treatment efficiency,7 this is the first time to our knowledge that biomarkers with specificity as high as 87.5% independently and 97.5% when associated in a classification tree, have been determined. Indeed, these prediction characteristics were obtained for the discrimination of two extreme patient groups (ACR20 negative and ACR70 positive) and it is likely that specificity and sensitivity of these biomarkers will be lower when used on a more homogeneous RA population. Six biomarkers between 3.86 and 74.07 kDa were isolated on two chromatographic surfaces, suggesting different reactivities of these proteins. The smallest one at 3.86 kDa was characterised on an anion array, meaning that this protein is negatively charged at pH 7. The other five biomarkers isolated on an IMAC-Ni chip have a specific affinity for this metallic ion. Three of these proteins were also recovered in the classification tree built from the IMAC-Ni plasma profile, emphasising their importance in the determination of the infliximab response.
We focused initially on the 28 kDa protein that is overexpressed in the responder population and is also the first node of the decision tree, and we identified it as apolipoprotein A-I. Apolipoprotein A-I is the major protein component of high-density lipoprotein (HDL) particles and the primary acceptor for cholesterol in extrahepatic tissues. In RA, several studies have reported a decrease of circulating levels of apolipoprotein A-I and HDL-cholesterol in patients compared to the general population,24–26 and concentration of both parameters increases significantly in patients responding to disease-modifying antirheumatic drug (DMARD) treatment compared to non-responders.27 By contrast, apolipoprotein A-I was highly expressed in inflamed RA synovial tissues particularly in perivascular areas containing infiltrated T cells and macrophages, but was not detected in normal tissue28 or in the synovium of non-inflammatory RA.29 Moreover, increased levels of apolipoprotein A-I and cholesterol have been measured in RA synovial fluid.30 As apolipoprotein A-I was reported to inhibit the synthesis of the major inflammatory cytokines TNFα and interleukin 1β by blocking direct contact between T lymphocytes and monocytes,31 increased levels of this protein in synovial tissue could modulate inflammation and disease evolution by controlling interactions between immune cells and then cytokine production.28 32 This specific function as a “negative” acute phase protein is also emphasised in a study reporting that patients with lower levels of apolipoprotein A-1 develop more systemic inflammatory response syndrome criteria.33
Finally, it has been recently shown in a retrospective study among blood donors (n = 1078) who later developed RA (n = 79) that the lipid profile was perturbed at least 10 years prior to the onset of symptoms, especially the HDL-cholesterol value that was significantly decreased.34 The authors suggested that this specific lipid profile may be favourable to the onset of RA or the presence of a constant high inflammatory state due to low levels of HDL-cholesterol. In our study, we demonstrated that high pretreatment levels of apolipoprotein A-I were predictive of a good response to infliximab. This increased initial concentration may favour the response to anti-TNFα treatment, via the decrease of chronic inflammation, by inhibiting cytokine release and potentiating the efficiency of therapy in a synergistic way.
The second biomarker we focused on was overexpressed in non-responder patients and was identified as PF4 (or CXCL4). This chemokine specific for platelet α granules exerts multiple functions in several physiological processes.35 In RA, several studies reported an increase of PF4 in patient synovial fluid36 37 and elevated plasma levels in patients with cutaneous vasculitis.38 In our study, the increase in plasma PF4 was correlated to a non-response to infliximab, and the same rise was recently shown in sera of patients with Crohn disease who had not responded to infliximab therapy.39 Two of its coagulation-unrelated functions may explain this result: PF4 was reported to induce degranulation of TNFα-primed neutrophils40 and to inhibit monocyte apoptosis.41 A high PF4 concentration may increase the inflammatory response by potentiating the proinflammatory functions of neutrophils and monocytes, that may counteract the efficiency of an anti-TNFα therapy.
In this proteomic study that investigated the plasma protein profile of patients with RA who were either responders or non-responders to infliximab therapy, we identified two biomarkers (apolipoprotein A-I and PF4) that may be key elements for RA treatment monitoring. Firstly, these proteins are involved in RA physiopathology and may be associated with specific clinical situations such as inflammation or vasculitis and secondly, both are strongly correlated with and without clinical response to infliximab treatment. However, these biomarkers need to be validated in a larger cohort of patients and to be evaluated for determination of intermediate treatment response assessed by the ACR50 score or more linearly by the 28-joint Disease Activity Score (DAS28) index. The identification of the other biomarkers is currently under investigation.
Web only appendix 68:8;1328-1333
Competing interests: None declared.
Funding: This work was supported by grants from the “Ministère de l’Enseignement Supérieur et de la Recherche”, the “Région Rhône-Alpes”, Joseph Fourier University, the GEFLUC (department of Grenoble), the “Fondation pour la Recherche Médicale”, the “Direction Régionale de la Recherche Clinique” (CHU Grenoble) and the “Ligue Nationale contre le Cancer”.
Ethics approval: Obtained.
▸ Additional data (Supplementary tables 1 and 2) are published online only at http://ard.bmj.com/content/vol68/issue8
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.