Metabolic Responses to Change in Disease Activity during Tumor

Article pubs.acs.org/jpr

Metabolic Responses to Change in Disease Activity during Tumor Necrosis Factor Inhibition in Patients with Rheumatoid Arthritis Rasmus Madsen,† Solbritt Rantapaä -̈ Dahlqvist,‡ Torbjörn Lundstedt,§ Thomas Moritz,∥ and Johan Trygg*,† †

Computational Life Science Cluster (CLiC), Department of Chemistry, Umeå University, 90187 Umeå, Sweden Department of Public Health and Clinical Medicine, Rheumatology, Umeå University Hospital, Umeå, Sweden § AcureOmics AB, Tvistevägen 48, 907 36 Umeå, Sweden ∥ Umeå Plant Science Center, Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences, SE-90183 Umeå, Sweden ‡

S Supporting Information *

ABSTRACT: Assessment of disease activity in patients with rheumatoid arthritis (RA) is of importance in the evaluation of treatment. The most important measure of disease activity is the Disease Activity Score counted in 28 joints (DAS28). In this study, we evaluated whether metabolic profiling could complement current measures of disease activity. Fifty-six patients, in two separate studies, were followed for two years after commencing anti-TNF therapy. DAS28 was assessed, and metabolic profiles were recorded at defined time points. Correlations between metabolic profile and DAS28 scores were analyzed using multivariate statistics. The metabolic responses to lowering DAS28 scores varied in different patients but could predict DAS28 scores at the individual and subgroup level models. The erythrocyte sedimentation rate (ESR) component in DAS28 was most correlated to the metabolite data, pointing to inflammation as the primary effect driving metabolic profile changes. Patients with RA had differing metabolic response to changes in DAS28 following anti-TNF therapy. This suggests that discovery of new metabolic biomarkers for disease activity will derive from studies at the individual and subgroup level. Increased inflammation, measured as ESR, was the main common effect seen in metabolic profiles from periods associated with high DAS28. KEYWORDS: anti-TNF treatment, rheumatoid arthritis, OPLS, metabolomics

■

INTRODUCTION Rheumatoid arthritis (RA) is an autoimmune disease in which the synovial joint tissue, primarily in the hands and feet, is targeted by the immune system.1 The disease is lifelong and causes a gradual destruction of the affected joints, often leading to severe disability in the patient.2 There is no causal treatment currently available for use in patients with RA, albeit a number of disease modifying antirheumatic drugs (DMARDs) are available.3 When the response to traditional DMARDs is insufficient, treatment with biological immune modulator, e.g., tumor necrosis factor-alpha (TNF-alpha) inhibitors, is recommended.4 TNF-alpha has a regulatory effect on a range of other proinflammatory cytokines and is, therefore, an effective pharmacological target in patients with RA.5 TNF-alpha activity can be blocked, which leads, in most cases, to a reduction in rheumatoid disease activity. The main drawbacks of TNF blockade are the potential risk of infection and development of malignancies,6,7 and the cost, which is substantial compared © 2012 American Chemical Society

with traditional treatment regimens. Additionally, the response of patients to TNF blockade is variable, with some patients entering remission while others achieve limited benefits.8 This necessitates a careful survey of treated patients and makes identification of biomarkers associated with patient disease activity highly desirable. For the diagnosis of RA, particularly within the latest criteria, immunological markers, such as rheumatoid factor (RF) and antibodies against citrullinated proteins/peptides (ACPA), have been assigned an important role.9 ACPA in particular have been shown to have a very high specificity for RA.10 Furthermore, ACPA have been suggested to be a prognostic marker for a more severe disease course.11,12 The presence of RF and/or ACPA has been shown to be associated with a reduced response to anti-TNF drugs,13 but neither RF nor ACPA are Received: March 27, 2012 Published: May 10, 2012 3796

dx.doi.org/10.1021/pr300296v | J. Proteome Res. 2012, 11, 3796−3804

Journal of Proteome Research

Article

sensitive to fluctuations in disease activity. C-reactive protein and erythrocyte sedimentation rate (ESR) are routinely measured to assess disease activity. In addition to these, interleukin-6 (IL-6), TNF-alpha,14 IL-18,15 matrix metalloproteinase-3 (MMP-3), IL-2 and oncostatin M16 have been established as being related to disease activity but to date have not achieved general clinical application. We recently reported that patients with RA can be diagnosed on the basis of the metabolic profile measured in serum samples.17 Other groups have shown that a measured metabolic profile changes during treatment.18 We, therefore, hypothesized that metabolic profiling could be a suitable method for finding biomarkers correlated with disease activity as measured using the Disease Activity Score in 28 joints (DAS28).19 This would provide an “objective” complement to current measures of disease activity. This approach could aid the early assessment of a patient’s response to treatment and provide a readily available feedback on the treatment strategy. Another area in which new biomarkers could be of value is the early stages of drug development, for instance, in animal models, when it is often not possible to use the traditional measures of disease activity.20 The purpose of the present investigation was to establish whether serum metabolites, measured using gas and liquid chromatography coupled with mass spectrometry, are linked to disease activity in patients with RA, and thus whether metabolic profiling or potential biomarkers found using this method could compliment current measures of disease activity. The performed longitudinal studies assessed DAS28 on individual patients undergoing treatment with TNF-alpha inhibitors. The use of a series of samples from the same patient allows reduction of effects concerning individual differences among patients and aids the assessment of how the metabolic profile develops in individual patients during changes in DAS28. Two independent cohorts of patients were used in order to investigate the general applicability of the findings.

■

Study 2: Patients

Study 2 included 29 patients (26 female and 3 male) using the same inclusion criteria as described for study 1. Blood samples were collected according to the same protocol as study 1, except an additional sampling time point at 102 weeks for some patients (n = 10) and with some samples lacking baseline sample (n = 17). All patients were treated with Infliximab (3 mg/kg) at baseline and at 2 weeks, 6 weeks and thereafter every 8 weeks up to 78 weeks after initial treatment. All patients continued on DMARDs for the duration of the study with 22 on Metotrexate and five on Azathioprine; two did not receive any DMARDs. Fourteen patients were treated with corticosteroids at inclusion, three discontinued this treatment during the study, and another one initiated it during the study period. Blood sampling and clinical assessment were as in study 1. All individuals in studies 1 and 2 gave a written informed consent for the study protocol. The study was approved by the Regional Ethics Committee at the University Hospital, Umeå, Sweden Metabolic Profiling

Details of metabolic profiling analysis and metabolite identification and quantification are presented in the Supporting Information. The GC−MS protocols closely follow the methods detailed by A et al.22 and Jonsson et al.23 For amino acid profiling, LC−MS analysis was done using the AccQ-Tag kit obtained from Waters (Millford, MA) as specified in the manufacturer’s instructions. In study 1, one sample was omitted prior to modeling because of an extreme DAS28 score being recorded. Multivariate Statistics

All multivariate analysis in this study was based on the OPLS model by Trygg and Wold.24 OPLS was used to relate the metabolite profiling data to clinical parameters such as DAS28 and its subcomponents, i.e., ESR, the number of tender and swollen joints and patient’s global assessment, as well as to other known possible sources of variation such as gender, age, BMI and pharmaceutical treatment. The discriminant version OPLS-DA25 was used for modeling discrete responses, such as gender, smoking or adherence to a certain treatment regimen. The amount of modeled variation in the selected response variable was provided by the R2 value as follows:

PATIENTS AND METHODS

Study 1: Patients

Twenty-seven patients with established RA (American College of Rheumatology 1987 criteria21) were consecutively recruited into the study at Umeå University Hospital, Umeå, Sweden. The inclusion criterion for TNF inhibitor therapy was a failure to control disease activity by DMARDs as evaluated by the individual patient’s physician. Twenty-three patients (15 females and 8 males) were treated with Infliximab (3 mg/kg) at baseline and at 2 weeks, 6 weeks, and thereafter every 8 weeks up to 78 weeks after initial treatment. All except one patient continued treatment with DMARDs for the duration of the study, and 13 continued with corticosteroids for the duration of the study. Four patients (3 females and 1 male) received 50 mg of etanercept weekly and continued their therapy with DMARDs. All patients were assessed at baseline and after 2, 6, 8, 14, 30, 54, and 78 weeks before infusion using the 28-joint count for tender and swollen joints, a global visual analogue scale (patient’s global VAS), erythrocyte sedimentation rate (ESR, mm/h), and the DAS28 score was calculated.19 Blood samples were collected at the stated time points before the infusion or injection. All samples were collected in the morning after a light breakfast according to standard practice at the clinic. Resulting serum samples were stored at −80 °C until analyzed.

∑i (xî − x ̅ )2 R = ∑i (xi − x ̅ )2 2

where xi is the measured value, x̂i is the modeled value, and x̅ is the average value. Seven-fold cross-validation was used to decide number of model components (denoted as number of predictive + orthogonal components) and to assess the predictive ability of the OPLS model.26 All samples from a patient were collectively excluded in the cross-validation procedure to avoid overfitting. Cross-validation results were recorded as the rootmean-square error of cross-validation (RMSECV) as follows: N

RMSECV =

Q2 = 1 − 3797

∑1 (yi ̂ − yi )2 N

∑i (yi ̂ − yi )2 ∑i (yi − y ̅ )2 dx.doi.org/10.1021/pr300296v | J. Proteome Res. 2012, 11, 3796−3804


Article

where ŷi is the response value calculated by the model with the observation excluded, yi is the observed response value, y ̅ is the average response value, and N is the total number of samples. Significance testing was performed using cross-validated analysis of variance CV-ANOVA.27

metabolite’s correlation to an increase in DAS28 score. The OPLS correlation scaled loadings of the predictive component constitutes the common metabolite profile in each study. As shown in Table 1, factors that differentiated between both studies were disease duration, which was significantly longer in study 2 (p = 0.014), and a higher proportion of females in study 2. Also, four of the patients in study 1 received Etanercept instead of Infliximab. In order to compensate for these discrepancies, modeling was repeated using only women receiving infliximab treatment. This resulted in a deterioration of the models for both studies (study 1: 1 + 2 components, R2 = 0.291, Q2 = 0.152, p = 0.036; study 2: 1 + 0 components, R2 = 0.128, Q2 = −0.0052, p = 1). The metabolites characteristic for high and low DAS28 scores in each study were quite different. In study 1, samples from periods associated with high disease activity were characterized by high content of hippuric acid and aspartic acid, with lower amounts of citrulline, tryptophan, citric acid, and glutamine as illustrated in Figure 1C. In study 2, samples from periods with high disease activity were characterized by high content of cystine, L-hydroxy-proline, alanine, 2,5diaminovalerolactam, and threonic acid, with lesser amounts of cysteine as shown in Figure 1D. By plotting the model correlation-scaled loadings (Figure 2) against each other, the metabolites with similar behavior in both studies were identified.28 As shown arginine, guanosine, homoserine, and 2,5-diaminovalerolactam were increased in samples associated with high DAS28 scores, while several other amino acids were consistently decreased. Most other metabolites had less reproducible trends. It should be remembered that these common effects were not the most apparent features in either of the studies when evaluated separately. Note also that the loading values are slightly different from those presented in Figure 1, since only metabolites quantified in both studies were included in the calculations. The described differences in metabolite profiles indicated that patients from both cohorts could not be modeled as a homogeneous group. Since each patient in the study was represented by up to eight sampling points longitudinally, it was possible to model DAS28 on an individual patient level. In study 1, 23 of the patients had sufficient sampling points. The mean prediction error (RMSECV) of DAS28 in the individual models in study 1 was 0.73 (range: 0.32−1.79) in comparison to an error in the global model of 1.366 units. In study 2, 26 patients were modeled. The mean prediction error of DAS28 in the individual models in study 2 was 0.60 (range: 0.14−1.60). Using the OPLS model, it was possible to assess the amount of variation in the metabolic profiles that could be attributed to changes in DAS28. In the global models, the amount of variation related to DAS28 was 3.7 and 4.4% of the total variation in the metabolite data in study 1 and 2, respectively. In individual models, the changes in DAS28 varied between 12.3 and 43.6% (mean = 23.6 ± 7.7%, median 23.05%). The subcomponents of DAS28 (ESR, # tender joints, # swollen joints, and patient’s global VAS) were also modeled in relation to the metabolite data. As shown in Table 2, ESR was the factor best modeled by the metabolite data, with global health assessment as a distant second. Swollen joints showed very weak correlations in study 1 but not in study 2, while tender joints did not produce a valid model. Thus inflammation is considered to be the main common component that was reflected in the metabolic profile.

Univariate Statistics

All statistical tests were performed using two-sided Student’s t tests for samples with equal variance.

■

RESULTS

Modeling of DAS28 and Its Subcomponents

The purpose of this study was to investigate whether a metabolic profile of patients with RA correlated to changes in DAS28. Patients were followed up to 2 years of anti-TNF therapy, with up to 8 sampling points for each patient, each characterized by their unique disease activity as assessed by the DAS28 score. Two independent cohorts were studied. Study 1 included 143 samples from 27 patients with 65 metabolites measured. Study 2 included 163 samples from 29 patients with 78 metabolites measured. The patient demographics of the two studies are described in Table 1. The studies were matched as for sampling conditions, sample storage time, patient age, BMI and treatment. Table 1. Demographic Characteristics of the Patients in the Two Studiesa % female age (mean ± SD) disease duration (mean ± SD) DAS28 baseline (mean ± SD) DAS28 (mean ± SD, all samples from follow-up visits) BMI (mean ± SD) % smokers % infliximab % etanercept % metotrexate % azathioprine % prednisolone

study 1

study 2

p-value

66.7 (n = 18) 52.4 (±12.0) 11.2 (±8.1) 6.05 (± 1.00) 3.78 (± 1.39) 23.8 (± 4.1) 14.8 (n = 4) 85.2 (n = 23) 14.8 (n = 4) 77.8 (n = 21) 0

89.7 (n = 26) 56.5 (± 12.6) 16.7 (± 8.0) 5.76 (±1.17) 3.79 (± 1.32) 24.7 (± 3.4) 10.3 (n = 3) 100 (n = 29) 0

0.036

75.8 (n = 22) 17.2 (n = 5) 48.3 (n = 14)

0.74

66.7 (n = 18)

0.22 0.014 0.32 0.73 0.40 0.34

0.90

a

It is seen that study 2 has significantly more women and that the average disease duration here is significantly longer. No other significant differences were identified between the two cohorts.

In each of the two studies, significant models were obtained linking the metabolic profile to DAS28 scores (study 1: 1 predictive +2 orthogonal components, R2 = 0.273, Q2 = 0.219, p = 6.40 × 10−6; study 2: 1 predictive +0 orthogonal components, R2 = 0.106, Q2 = 0.057, p = 0.0092; Figure 1A,B). The predictive errors of the models (RMSECV) were 1.366 and 1.324 DAS28 units in study 1 and 2, respectively. In the OPLS model all metabolites were characterized by a correlation-scaled OPLS loading (p(corr)) describing the 3798



Article

Figure 1. Correspondence between DAS28, measured at a clinic visit, and DAS28 predicted from metabolic profiling data in study 1 (A) and study 2 (B). There was a reasonable linear relation in both studies. Metabolite correlation coefficients indicating how each metabolite was affected during increase or decrease in DAS28. It is seen that the metabolites that describe changes in DAS28 were very different in studies 1 (C) and 2 (D). Metabolites are sorted according to correlation coefficient.

Investigation of the Relationship between Individual Metabolic Profiles and Common Metabolic Profile

in their metabolic profiles, resulting in a positive correlation to other patients. For example, patient 7 in study 1 showed strong similarities to patients 3, 8, 9, 10, 11, 12, 16, 17, and 26. Conversely patients 2, 4, 19, and 27 in study 1 showed only very weak similarities to any of the other patients. These commonalities can also be assessed by each patient’s correlation (Pearson) to the global metabolic profile in each study as plotted in Figure 3A. It is concluded from this analysis that the larger part of the patients in study 1 exhibited similarities (positive correlation) to the global metabolic profile

The above results indicate that changes in the DAS28 score are, in most instances, accompanied by changes in metabolic profile. The changes observed, however, appear to be highly specific for the individual patients. The correlation (Pearson) between each patient’s metabolic profiles (OPLS correlation-scaled loadings) was plotted (Figure 3A,B) to understand these metabolic changes. It was apparent that some patients had commonalities 3799



Article

Figure 2. Correlations of metabolites to DAS28 in each of the two studies plotted against each other. This identifies that the metabolites behaved similarly in both studies. These are marked green and with identity.

Table 2. P-values for Models of Subcomponents of the DAS28 Calculated Using CV-ANOVAa erythrocyte sedimentation rate (ESR) tender joints swollen joints patient’s global visual analogue scale

study 1

study 2

2.13 × 10−6 1 0.025 7.61 × 10−4

2.92 × 10−7 1 0.87 7.77 × 10−4

study. Summarized in Table 3 are the metabolic effects of the patients’ physical attributes, their medical treatment, and smoking habits. The magnitude of these effects in terms of amount of explained variation in the metabolic profile was in the range 3−10%, i.e. comparable to the effect seen from changes in DAS28. The study was not specifically designed to assess these effects, but what is important is that none of them could be found responsible for the DAS28 effect seen in either study. In the Supporting Information, plots of model loading of the DAS28 model against models of other possibly confounding factors are provided.

a

It is noteworthy that ESR was well predicted from the metabolite data, whereas the number of tender and swollen joints could not be predicted. This indicates that inflammation is an important contributing factor to the changes in metabolite profiles.

■

of this study, whereas the patients in study 2 were more heterogeneous. Furthermore, it was seen that approximately half of the patients in study 2 showed similarities to the global metabolic profile of study 1. It should be noted that these correlations are relatively weak, and it was only possible to obtain meaningful predictions for a few of the patients in study 2 using the study 1 model. On the other hand, models were much stronger when modeling those patients with similar metabolic profiles together.

DISCUSSION

DAS28 Response and Common Metabolic Effects

Significant models for DAS28 were obtained in each study and in most cases when modeling individual patients, whereas the attempt to transfer a model capable of predicting DAS28 for new patients from one study to the other was unsuccessful. This is technically challenging using present GC−MS technology; thus, the main aim was comparing the response to changes in DAS28 in the two studies. Clearly responses were not common for all patients. Rheumatoid arthritis is correctly classified as a heterogeneous disease, and it is well-known that patients can respond very differently to identical treatments. This study of metabolic profiles associated with changes in DAS28 in patients with RA further supports the hypothesis that different underlying mechanisms may account for individual patients’ symptoms. Studies of gene expression have revealed

Other Effects Noted in the Metabolic Profiles

As discussed above, the common metabolic effects of changes in the DAS28 score are small when compared with those seen when modeling individuals. It is, furthermore, clear that there are other factors that can influence the metabolic profile, which may indicate why metabolites that did not have a reproducible effect could still appear to be candidate markers in a single 3800



Article

Figure 3. Degree of correlation (Pearson) between the metabolic profile (correlation-scaled loadings) connected with DAS28 for individual patients in studies 1 (A) and 2 (B). Red indicates positive correlation, meaning that the compared patients had similar metabolic profile changes accompanying changes in DAS28. These plots demonstrate that metabolic effects were shared by groups of patients but were not common to all. (C) Degree of correlation (Pearson) between each patient’s metabolic profile to the common metabolic profile (correlation-scaled loadings) in studies 1 and 2, respectively. Patients with red color had similar metabolic behavior to the common profile of the compared study accompanying DAS 28 changes. It is also seen that patients in study 2 showed correlation to the metabolic profile found in study 1.

Table 3. List of Factors, Other than DAS28, Assumed to Impact the Metabolic Profile and the Estimated Metabolic Effects Seen metabolites affected: study 1 increased

metabolites affedted: study 2 decreased

increased

substance (infliximab) prednisolone use methotrexate use BMI (high)

Eicosanoic acid, Inositol-2-phosphate, Heptadecanoic acid and S-methyl-cystine Nonsignificant

Maleic acid

Cysteine and Alanine Uric acid, Valine and Glutamic acid

Threonic acid and Glyceric acid Asparagine and Threonine

Phenylalanine, glycerol-3phosphate and cholesterol Valine and Heptanoic acid

age (high)

Inositol and Valine.

Guanosine and Inosine

sex (male) smoking

Most amino acids Tryptophan and Glutamic acid

Dihydroxybutyric acid and Threonic acid Amino acids Phosphoric acid and threonic acid

Histidine and Citric acid

Threonic acid and Glycerol-3-phosphate

that RA patients have different profiles of active genes,29 and it is possible that this can also result in differences in metabolic networks leading to different metabolic profiles, as seen in this study. Unfortunately, this study was not designed to assess subtypes of RA disease, and consequently it is difficult to draw any firm conclusions about their relation to subtypes reported in other studies, but it is something to be aware of in future studies. A special complication in studies of this type is that patients receive TNF inhibition and other treatments, which are known to have metabolic effects. TNF inhibitors, for instance, have

decreased

n/a Beta-alanine and Dihydroxy-butanoic acid Citric acid and Glucose 2,5-Diaminovalerolactam and Glutamine Leucine and Valine Lipids and Phosphoric acid Heptanoic acid

effects in glucose and fatty acid energy metabolism.30 Although this can complicate the picture, it should be noted that all of the patients studied here received similar treatment (with the reservations discussed above) and that all blood samples were drawn prior to the infusion/injection of anti-TNF. The common effects in both studies in periods of high DAS28, i.e., reduced levels of several amino acids and an increase of nucleic acid related metabolites, could be consistent with increased cell turnover in conjunction with increased inflammation. An increase of nucleic acid metabolites could indicate tissue damage secondary to inflammation, since they 3801



Article

are not normally very abundant in plasma. The overall effect of inflammation on plasma amino acids levels is difficult to deduce. Synthesis of acute phase proteins and inflammatory cells could explain the observed reduction in amino acids, whereas a release of glucocorticoids would tend to counteract this effect.31 The observation that ESR was well indicated by the metabolic data would support that the observed metabolic effects could be connected to inflammation. On the basis of the correlation structures presented in Figure 3, some loose metabolic subgroups could be defined. One group (i.e., patients: 1, 4, 8, 9, 10, 11, and 12 in study 1 and 6, 8, 9, 10, 20, and 22 in study 2) were the patients who had a metabolic profile similar to that seen in study 1, and another (patients: 6, 15 16, 17, 26, and 27 in study 1 and 5, 12, 15, 16, and 26 in study 2) had a metabolic profile to that seen in study 2. In both cases, modeling of DAS28 in these subgroups showed better model properties than modeling all patients in the study; also, more metabolites showed similar patterns in study 1 and 2 when studying subgroups, supporting the relevance of regarding RA as a disease with several subgroups of patients. Often RA patients visit the clinic for many years, and in that light, an individual metabolic-profile-based protocol could be a realistic complement to the regular clinical assessment of DAS28. Whether this will actually be useful clinically will depend on whether it will be possible to define more distinct groups of metabolic subtypes and relate these to other phenomena. An interesting addition to this study would be to relate the metabolite data to cytokines, hormones, proteins, and genes in order to increase the understanding of the role played by different metabolites in RA patients. This could possibly contribute to a more complete picture of the diverse pathogenic processes involved in the progression of RA.

effect, however, was not statistically significant for any individual compound over both studies. The classical solution to these problems is an increase in sample size and thus statistical power. In this study we have demonstrated the feasibility of an alternative approach, i.e., following patients over a period of time. Following patients over time allows identification and modeling of different effects in various patients. Currently there is an increasing understanding that not all patients are similar just because they share a common diagnosis, and consequently studies designed around disease dynamics in individual patients will contribute to increased understanding of these differences and their possible clinical implications. The fact that both the different studies detailed in this work produced valid models according to all normal tests and yet identified different metabolic profiles demonstrates some of the problems discussed above. This highlights the importance of ensuring thorough validation of any results from profiling studies and of using transparent techniques for data analysis and visualization.33 It was realized that the problems in predicting DAS28 are that the metabolic changes accompanying variation in DAS28 are, in fact, distinct for different patients and that this could easily have been missed had the two separate studies not been performed.

■

CONCLUSIONS The feasibility of predicting DAS28 in RA patients embarking on TNF inhibition therapy was assessed in two independent studies. It was found that individual patients had different metabolic responses to changes in DAS28 and should be assessed on an individual or subgroup basis. Laboratory parameters of inflammation were identified as the most influential common factor in the metabolic profiles. Additionally, the studies highlighted the importance of validating results of profiling studies, since biological variation can easily dominate the factors that are being studied.

Extraction of Individualized Information

This study demonstrated that extracting relevant information from profiling studies is not an easy task. Modern profiling techniques aimed at, for instance, genes, proteins, hormones, cytokines, or metabolites, hold great promise for increasing the understanding of complex pathological states such as RA and for the development of new diagnostic and prognostic tools. Furthermore, dynamic compound groups, such as metabolites, hold the promise of being able to follow developments in the pathological state(s). This could be useful for assessment of the patient’s response to various treatment regimens. Caution is necessary, however, since it is well-known that measuring many variables from a limited number of patients can lead to false positive results.32 Another important property that we consider to be the primary challenge in metabolic profiling studies is that metabolites are very dynamic and affected by many factors in addition to the one being studied (in the present study, changes in DAS28). In this study it was possible to demonstrate that age, sex, BMI and medical treatment influence the metabolic profile. Certain circadian rhythms are also well-known, as is the effect of diet, exercise, and probably several other factors that are not easily controlled for or measured. This produces two problems: confounding variation might overshadow the desired effects so that they are missed, and spurious observations may be mistaken for pathologically related effects. The most reproducible effect in these two studies was a reduction of amino acids in periods accompanying low DAS28 scores; the

■

ASSOCIATED CONTENT

S Supporting Information *

Experimental methods and supporting figures. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Telephone: +46 90 7866917. Fax: +46 90 7867655. Author Contributions

RM participated in data acquisition and analysis, the conception and design of the study, and wrote the manuscript. SRD participated in the conception and design of the study, acquisition of the data, and drafting of the manuscript. TM participated in the acquisition of the data. JT participated in the conception and design of the study, data analysis, and revision of the manuscript. TL participated in the conception and design of the study and revision of the manuscript. Notes

The authors declare the following competing financial interest(s): TM, TL and JT are shareholders of AcureOmics AB. The authors declare no other competing interests. No financing has been received from this company. 3802



■

Article

(14) Tak, P. P.; Smeets, T. J. M.; Daha, M. R.; Kluin, P. M.; Meijers, K. A. E.; Brand, R.; Meinders, A. E.; Breedveld, F. C. Analysis of the synovial cell infiltrate in early rheumatoid synovial tissue in relation to local disease activity. Arthritis Rheum. 1997, 40, 217−225. (15) Petrovic-Rackov, L.; Pejnovic, N. Clinical significance of IL-18, IL-15, IL-12 and TNF-alpha measurement in rheumatoid arthritis. Clin. Rheumatol. 2006, 25, 448−452. (16) Rioja, I.; Hughes, F. J.; Sharp, C. H.; Warnock, L. C.; Montgomery, D. S.; Akil, M.; Wilson, A. G.; Binks, M. H.; Dickson, M. C. Potential novel biomarkers of disease activity in rheumatoid arthritis patients: CXCL13, CCL23, transforming growth factor α, tumor necrosis factor receptor superfamily member 9, and macrophage colony-stimulating factor. Arthritis Rheum. 2008, 58, 2257−2267. (17) Madsen, R. K.; Lundstedt, T.; Gabrielsson, J.; Sennbro, C. J.; Alenius, G. M.; Moritz, T.; Rantapaa-Dahlqvist, S.; Trygg, J. Diagnostic properties of metabolic perturbations in rheumatoid arthritis. Arthritis Res. Ther. 2011, 13, R19. (18) Lauridsen, M. B.; Bliddal, H.; Christensen, R.; DanneskioldSamsøe, B.; Bennett, R.; Keun, H; Lindon, J. C.; Nicholson, J. K.; Dorff, M. H.; Jaroszewski, J. W.; et al. 1H NMR spectroscopy-based interventional metabolic phenotyping: a cohort study of rheumatoid arthritis patients. J. Proteome Res. 2010, 9, 4545−4553. (19) Prevoo, M. L.; van ’t Hof, M. A.; Kuper, H. H.; van Leeuwen, M. A.; van de Putte, L. B.; van Riel, P. L. Modified disease activity scores that include twenty-eight-joint counts. Development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum. 1995, 38, 44−48. (20) Rosengren, S.; Firestein, G. S.; Boyle, D. L. Measurement of inflammatory biomarkers in synovial tissue extracts by enzyme-linked immunosorbent assay. Clin. Diagn. Lab. Immunol. 2003, 10, 1002− 1010. (21) Arnett, F. C.; Edworthy, S. M.; Bloch, D. A.; McShane, D. J.; Fries, J. F.; Cooper, N. S.; Healey, L. A.; Kaplan, S. R.; Liang, M. H.; Luthra, H. S.; et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum. 1988, 31, 315−324. (22) A, J.; Trygg, J.; Gullberg, J.; Johansson, A. I.; Jonsson, P.; Antti, H.; Marklund, S. L.; Moritz, T. Extraction and GC/MS analysis of the human blood plasma metabolome. Anal. Chem. 2005, 77, 8086−8094. (23) Jonsson, P.; Johansson, A. I.; Gullberg, J.; Trygg, J.; A, J.; Grung, B.; Marklund, S.; Sjostrom, M.; Antti, H.; Moritz, T. High-throughput data analysis for detecting and identifying differences between samples in GC/MS-based metabolomic analyses. Anal. Chem. 2005, 77, 5635− 5642. (24) Trygg, J.; Wold, S. Orthogonal projections to latent structures (O-PLS). J. Chemom. 2002, 16, 119−128. (25) Bylesjö, M.; Rantalainen, M.; Cloarec, O.; Nicholson, J. K.; Holmes, E.; Trygg, J. OPLS discriminant analysis: combining the strengths of PLS-DA and SIMCA classification. J. Chemom. 2006, 20, 341−351. (26) Wold, S. Cross-validatory estimation of the number of components in factor and principal components models. Technometrics 1978, 20, 397−405. (27) Eriksson, L.; Trygg, J.; Wold, S. CV-ANOVA for significance testing of PLS and OPLS® models. J. Chemom. 2008, 22, 594−600. (28) Wiklund, S.; Johansson, E.; Sjostrom, .L; Mellerowicz, E. J.; Edlund, U.; Shockcor, J. P.; Gottfries, J.; Moritz, T.; Trygg, J. Visualization of GC/TOF-MS-based metabolomics data for identification of biochemically interesting compounds using OPLS class models. Anal. Chem. 2008, 80, 115−122. (29) van der Pouw Kraan, T. C.; Wijbrandts, C. A.; van Baarsen, L. G.; Voskuyl, A. E.; Rustenburg, F; Baggen, J. M.; Ibrahim, S. M.; Fero, M.; Dijkmans, B. A.; Tak, P. P.; Verweij, C. L. Rheumatoid arthritis subtypes identified by genomic profiling of peripheral blood cells: assignment of a type I interferon signature in a subpopulation of patients. Ann. Rheum. Dis. 2007, 66, 1008−1014. (30) Sethi, J. K.; Hotamisligil, G. S. The role of TNF alpha in adipocyte metabolism. Semin. Cell Dev. Biol. 1999, 10, 19−29.

ACKNOWLEDGMENTS The authors thank Inga-Britt Carlsson, Krister Lundgren, Annika Johansson, Swedish University of Agricultural Sciences, Umeå, Sweden and Renata Marcinowska for practical assistance. This work was financially supported by Grant 2011-6044 from the Swedish Research Council (to JT and RM).

■

REFERENCES

(1) McInnes, I. B.; Schett, G. Cytokines in the pathogenesis of rheumatoid arthritis. Nat. Rev. Immunol. 2007, 7, 429−442. (2) Welsing, P. M. J.; Gestel, A. M. V.; Swinkels, H. L.; Kiemeney, L. A. L. M.; Riel, P. L. C. M. V. The relationship between disease activity, joint destruction, and functional capacity over the course of rheumatoid arthritis. Arthritis Rheum. 2001, 44, 2009−2017. (3) Kwoh, C. K.; Anderson, L. G.; Greene, J. M.; Johnson, D. A.; O’Dell, R.; Robbins, M. L.; Roberts, W. N.; Simms, R. W.; Yood, R. A.; Amer Coll Rheumatology, S. Guidelines for the management of rheumatoid arthritis2002 update. Arthritis Rheum. 2002, 46, 328− 346. (4) Smolen, J. S.; Landewé, R; Breedveld, F. C.; Dougados, M.; Emery, P.; Gaujoux-Viala, C.; Gorter, S.; Knevel, R.; Nam, J.; Schoels, M.; et al. EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs. Ann. Rheum. Dis. 2010, 69, 964−975. (5) Feldmann, M.; Brennan, F. M.; Elliott, M. J.; Williams, R. O.; Maini, R. N. TNF alpha is an effective therapeutic target for rheumatoid arthritis. Ann. N. Y. Acad. Sci. 1995, 766, 272−278. (6) Bongartz, T.; Sutton, A. J.; Sweeting, M. J.; Buchan, I; Matteson, E. L.; Montori, V. Anti-TNF antibody therapy in rheumatoid arthritis and the risk of serious infections and malignanciesSystematic review and meta-analysis of rare harmful effects in randomized controlled trials. JAMA, J. Am. Med. Assoc. 2006, 295, 2275−2285. (7) Askling, J.; van Vollenhoven, R. F.; Granath, F.; Raaschou, P.; Fored, C. M.; Baecklund, E.; Dackhammar, C.; Feltelius, N.; Coster, .L; Geborek, P.; et al. Cancer risk in patients with rheumatoid arthritis treated with anti-tumor necrosis factor alpha therapies: does the risk change with the time since start of treatment? Arthritis Rheum. 2009, 60, 3180−3189. (8) Geborek, P.; Crnkic, M.; Petersson, I. F.; Saxne, T.; Grp, S. S. A. T. Etanercept, infliximab, and leflunomide in established rheumatoid arthritis: clinical experience using a structured follow up programme in southern Sweden. Ann. Rheum. Dis. 2002, 61, 793−798. (9) Aletaha, D.; Neogi, T.; Silman, A. J.; Funovits, J.; Felson, D. T.; Bingham, C. O.; Birnbaum, N. S.; Burmester, G. R.; Bykerk, V. P.; Cohen, M. D.; et al. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Ann. Rheum. Dis. 2010, 69, 1580− 1588. (10) van Boekel, M. A. M.; Vossenaar, E. R.; van den Hoogen, F. H. J.; van Venrooij, W. J. Autoantibody systems in rheumatoid arthritis: specificity, sensitivity and diagnostic value. Arthritis Res. 2002, 4, 87− 93. (11) Innala, L.; Kokkonen, H.; Eriksson, C.; Jidell, E.; Berglin, E.; Dahlqvst, S. R. Antibodies against mutated citrullinated vimentin are a better predictor of disease activity at 24 months in early rheumatoid arthritis than antibodies against cyclic citrullinated peptides. J. Rheumatol. 2008, 35, 1002−1008. (12) van der Helm-van Mil, A. H.; Verpoort, K. N.; Breedveld, F. C.; Toes, R. E.; Huizinga, T. W. Antibodies to citrullinated proteins and differences in clinical progression of rheumatoid arthritis. Arthritis Res. Ther. 2005, 7, R949−958. (13) Potter, C.; Hyrich, K. L.; Tracey, A.; Lunt, M.; Plant, D.; Symmons, D. P. M.; Thomson, W.; Worthington, J.; Emery, P.; Morgan, A. W.; et al. Association of rheumatoid factor and anti-cyclic citrullinated peptide positivity, but not carriage of shared epitope or PTPN22 susceptibility variants, with anti-tumour necrosis factor response in rheumatoid arthritis. Ann. Rheum. Dis. 2009, 68, 69−74. 3803



Article

(31) Hall, J. E. Guyton and Hall Textbook of Medical Physiology, 12th ed.; Saunders Elsevier: Philadelphia, PA, 2011. (32) Lay, J. O., Jr.; Liyanage, R.; Borgmann, S.; Wilkins, C. L. Problems with the “omics”. TrAC, Trends Anal. Chem. 2006, 25, 1046−1056. (33) Madsen, R.; Lundstedt, T.; Trygg, J. Chemometrics in metabolomicsA review in human disease diagnosis. Anal. Chim. Acta 2010, 659, 23−33.

3804


Metabolic Responses to Change in Disease Activity during Tumor

Recommend Documents