Use of Fourier-transform infrared spectroscopy in the diagnosis of rheumatoid arthritis: a pilot study

Rheumatoid arthritis is an autoimmune inflammatory disease leading to joint cartilage, bone degradation and limitation of mobility. Diagnosis of RA is difficult and complex. There are also no effective methods for clear discrimination between RA patients and non-RA individuals. In this work we use IR spectroscopy to differentiate RA patients and blood donors’ sera. We found differences between investigated sera (RA and non-RA) in range of 3000–2800 and 1800–800 cm−1 (W1–W5 regions). Based on mathematical analysis we developed a K-NN model characterized by 85 % of sensitivity and 100 % of specificity. Also we found that, wavenumber 1424 cm−1, comprising in W3 region, was the most effective in human sera distinguishing. We conclude that IR spectroscopy may serve as a fast and easy method useful in RA serology.


Introduction
Rheumatoid arthritis (RA) is one of the common inflammatory disease. I t leads to bone and cartilage destruction. Etiology of RA is still unclear, but there is no doubt about its autoimmune background [1]. Early diagnosis is crucial in RA. Presence of autoantibodies in serum is one of the known factors observed in preclinical phase of RA, in some cases even 4 years before disease onset. But their composition may be different and patient dependent [2]. Rheumatoid factor (RFs) and anti-citrullinated peptide antibodies (ACPAs) are the most frequently markers used in disease diagnosis. However, RFs have low specificity [3]. ACPAs are considered as more specific, but repertoire of peptides recognized by ACPAs increase with disease progress. Moreover, before RA symptoms onset, reaction of antibodies may be characteristic also for healthy ACPA-positive individuals [2]. RA patients possess also higher level of osteoprotegerin, soluble receptor activator of nuclear factor kappa B ligand (RANKL) and hyaluronan, as well as chondroitin sulphate in serum, but those molecules are also characteristic for other diseases [4][5][6].
That, fast and correct diagnosis (sometimes also define of a future disease) is difficult and in clinical practice usually impossible. Spectroscopic analysis, especially infrared spectroscopy (I R), seems to be promising tool in a diagnostic process. IR has been used in the study of biological samples. The I R spectrum of biological samples can be divided into windows correspond to groups of bioorganic compounds: fatty acid (W1 3000-2800 cm −1 ), peptides and proteins (W2 1800-1500 cm −1 ), proteins, phosphatecarrying compounds and fatty acid (W3 1500-1200 cm −1 ), carbohydrates (W4 1200-900 cm −1 ). The fragment W5 (900-750 cm −1 ) corresponds to specific peaks unique for the sample [7,8].
Abstract Rheumatoid arthritis is an autoimmune inflammatory disease leading to joint cartilage, bone degradation and limitation of mobility. Diagnosis of RA is difficult and complex. There are also no effective methods for clear discrimination between RA patients and non-RA individuals. I n this work we use I R spectroscopy to differentiate RA patients and blood donors' sera. We found differences between investigated sera (RA and non-RA) in range of 3000-2800 and 1800-800 cm −1 (W1-W5 regions). Based on mathematical analysis we developed a K-NN model characterized by 85 % of sensitivity and 100 % of specificity. Also we found that, wavenumber 1424 cm −1 , comprising in W3 region, was the most effective in human sera distinguishing. We conclude that I R spectroscopy may serve as a fast and easy method useful in RA serology.

Keywords Diagnosis · Rheumatoid arthritis · Serum · Spectroscopy
The derivatives were used for mathematical analysis. To develop dendrograms we used Manhattan metric and Ward's method.

Mathematical model developing for RA patients' differentiation
For developing of prediction model we used K nearest neighbour (K-NN) algorithm. The set of 207 spectra was randomly divided into two subsets: learning subset (157 cases) and validating subset (50 cases). The K-NN model was based on the spectral windows W1-W5. The calculations were performed by using the Statistica 12. The model quality was evaluated on the basis of quality indicators presented in Table 1.

Results and discussion
IR spectroscopy is very convenient tool in the analysis of biological materials, like tissue sections, cytologic and histologic specimen or biofluids. Types of sample determine methodology of measurement, however there is a manual useful in standardization of the analysis [9].
The aim of this work was to determine the fragments of IR spectra of human sera, characteristic for RA patients.

Experimental Sera
In the experiment 40 sera of blood donors (BD) from Swietokrzyskie Blood Center in Kielce (age average 37.83 ± 14.96), and 29 sera of RA patients (age average 62 ± 13.95) from Swietokrzyskie Rheumathology Center in Konskie were used. Based on age of the individuals, samples were divided into two groups: the young group-50 years old or less (26 blood donors; 5 RA patients); old group-50 years old and more (14 blood donors; 24 RA patients). Sera were stored at −20 °C until measurement of IR spectra. Samples were collected with the approval of the Ethics Committee of the Regional Chamber of Physicians in Kielce.

The measurement of infrared spectra and its processing
Measurements were made using a Perkin Elmer Spectrum 400 spectrometer in Attenuated Total Reflection (ATR) technique. The IR spectra were measured in range of 3000-750 cm −1 with a resolution of 1 cm −1 . Measurements were performed at a constant temperature and air humidity. The human sera were thawed at room temperature and carefully shaken immediately before the measurement. The sample (1 µl of human serum) was dropped on the crystal spectrometer, and then allowed to stand for 5 min for water evaporate. For each serum three independent replication were made and the average spectrum was calculated. Then, the first derivative was calculated using a five-point stencil: where ′ f k is value of 1st derivative for the k wavenumber, f k +2 is the absorbance value of the k + 2 wavenumber, f k +1 is the absorbance value of the k +1 wavenumber, f k −1 is the absorbance value of the k −1 wavenumber, f k −2 is the absorbance value of the k − 2 wavenumber.
The first derivatives were normalized to the range of {0, 1} using the formula:

Optimizing the experiment
Before performing the proper experiment, a series of measurements were made to determine optimum conditions. We optimized volume and time of drying of investigated samples. The most efficient proved to be the use of 1 µl of human serum, and leaving it to dry for 5 min on the crystal of I R spectroscopy coupled with advanced mathematical analysis have big potential as a screening tool in medical diagnosis. It is a useful method in identification of normal, pre-disease and disease states. Biofluids like blood, serum or plasma seems to be good specimen in regard of many protein biomarkers presence [10]. Lima et al. proved that ATR-FTI R [with genetic algorithm (GA) combined with linear discriminant analysis (LDA)] may be used in early detection of ovarian cancer and differentiation of disease stages [11,12]. Moreover, ATR-FTI R spectroscopy show higher classification rate than other (Raman) spectroscopic methods [13]. However, most of data are focused on cancer, Fig. 1 Infrared spectra of human sera. The influence of water content in sample to IR spectrum quality; before and after water evaporationblack and blue respectively (a). First derivatives of patients and a control group spectra (b). Fragment of IR spectra most differentiating RA patients and a control group: based on visual observation (c)-the red color indicates RA patients, while the green color indicates the control group. IR spectra misclassified by K-NN model: serum BD.07 (d), serum BD.09 (e), serum BD.159 (f)-the red color indicates misclassified spectra, while the green color indicates the correctly classified spectra. (Color figure online) control individuals and RA patients in regions corresponding to proteins, lipids and immunoglobulins (1600-1700 and 1430-1480 cm −1 ) [14]. They used the second derivative of spectra for the mathematical analysis. Khanmohammadi et al. suggested that this range contains medium band of C-N stretching and strong signal due to C-C-N bending of creatinine (1250-1000 and 1230-1100 cm −1 , respectively) [15].
We also noted some differences between investigated groups in 1600-1700 and 1430-1480 cm −1 regions (data not showed). We performed detailed analysis of the first derivative of sera IR spectra. We noted evident distinct discrepancy between RA and blood donors' groups of human the apparatus. The use of a larger volume of serum resulted in prolonged water evaporation time and did not result in the improvement of the quality of the IR spectra (data not showed). Presence of water cause reduction of IR light and obscure spectral details [9]. During water evaporation we observed increase of visibility of peaks in different spectrum regions (Fig. 1a). Analogic procedure recommended Baker et al. [9].

Analysis of IR spectra of human sera
Little data about use of IR spectroscopy in RA serology has been published. Carvalho et al. observed differences between Fig. 2 The cluster analysis based on the first derivative of IR spectra of human sera. Up arrows-individual younger that 50 years old, Down arrows-individuals 50 years old and more. Dendrogram was calculated using Ward`s method and Manhattan length collagen degradation [18]. On the other hand, increased level of tryptophan was noted for RA patients [19]. However, using only one wavenumber for differentiating of sera samples is discussing. In such cases there is a risk of loose of efficacy when investigated group is changed. Much more proper seems to be model taking into account a whole IR spectrum range (W1-W5).

Conclusion
RA is one of the most common systemic diseases leading to joint deformation and increased mortality. It concerns up to 1 % of world population, mostly women. Early and precise diagnosis of RA is crucial. However, diagnostic tests are still insufficient (conventional disease markers are present also in healthy population or are not specific only for RA) [14].
In this work we used ATR-FTIR technique to differentiate human sera of RA patients and non-RA blood donors. Our results show, that FTIR spectroscopy may be promising tool in RA investigation. It may serve as a fast, low cost and sensitive sera discriminatory method or as a prognostic test. sera in many fragments of the IR spectra. These fragments were uniformly dispersed throughout the analyzed spectra (Fig. 1b) and were found in each of the 5 regions (W1-W5) associated with biological compounds, but in W4 region (range of 1180-1080 cm −1 ) the difference between controls and patients was more clear (Fig. 1c). In this fragment, the shapes of I R spectra for patients and control individuals were different.

Chemometric analysis
Cluster analysis (CA), including W1-W5 regions, based on the first derivative of IR spectra revealed the existence of two distinct clusters (Fig. 2). Cluster I contains blood donors, while cluster II contains RA patients. Age (as well as therapy in RA patients, data not shown) has no effect on clustering. The attempts to differentiate RA patients from the control group on the basis of each region separately were unsuccessful (data not shown).
Because of this reason the whole IR spectra range (W1-W5) for developing a predictive mathematical model (the K nearest neighbor algorithm) for discriminating RA patients and blood donors has been used. Our model is characterized by moderate sensitivity (85 %) and very high specificity (100 %)- Table 1. All incorrect classified IR spectra sera belong to blood donors (Fig. 1d-f). The incorrectly classified IR spectra (DB09 and BD159) are characterized by a lower absorbance band of amide I and amide II. Most likely, the water content in the sera was too high and resulting absorbance decrease in the entire range of the spectra. I n case of one serum (BD07) higher absorbance has been observed and reason for such discrepancy in measurements is unknown. Nevertheless, the proposed model is a promising tool for detecting rheumatoid arthritis based on IR spectra of sera.
Regardless of this model we also noted that one wavenumber in I R spectra of analyzed sera is also useful in samples differentiation. The average values of wavenumber 1424 cm −1 (W3 region) differ significantly (p < 0.0001, t-test) in RA patients and control group. For this value distribution in both groups was normal (p = 0.58 for RA patients and p = 0.35 for blood donors) based on Shapiro-Wilk test.
Differentiating between RA patients and control group was created by following rule: if the value of the first derivative of 1424 cm −1 is <0.013115, then the person is RA patient; otherwise person does not suffer from RA.
Value 1424 cm −1 comprise in mixed region W3 carrying peaks of proteins, phosphate molecules and fatty acid. It was shown that this wavenumber is connected with H-C-H and O-C-H in-plane bending vibration characteristic for proteins (especially for two amino acids: proline and tryptophan) [16,17]. Proline is one of amino acid which occurs in collagen type I and hydroxyproline is a marker of bone