Dynamic Component Chemiluminescent Sensor ... - ACS Publications

May 30, 2008 - Nephrology, Soroka University Medical Center, Department of Information Systems Engineering, National Institute of. Biotechnology in th...
1 downloads 0 Views 1MB Size
Anal. Chem. 2008, 80, 5131–5138

Dynamic Component Chemiluminescent Sensor for Assessing Circulating Polymorphonuclear Leukocyte Activity of Peritoneal Dialysis Patients Daria Prilutsky,†,‡ Boris Rogachev,§ Marina Vorobiov,§ Moshe Zlotnik,§ Mark Last,| Leslie Lobel,‡ and Robert S. Marks*,†,⊥,# Department of Biotechnology Engineering, Department of Virology, Faculty of Health Science, Department of Nephrology, Soroka University Medical Center, Department of Information Systems Engineering, National Institute of Biotechnology in the Negev, and The Ilse Katz Center for Meso and Nanoscale Science and Technology, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel Recurrent bacterial peritonitis is a major complication in peritoneal dialysis (PD) patients, which is associated with polymorphonuclear leukocyte (PMN) functional changes and can be assessed by a chemiluminescent (CL) reaction. We applied a new approach of a dynamic component chemiluminescence sensor for the assessment of functional states of PMNs in a luminol-amplified whole-blood system. This method is based on the evaluation of CL kinetic patterns of stimulated PMNs, while the parallel measurements of intracellular and extracellular production of reactive oxygen species (ROS) from the same sample can be conducted. Blood was drawn from diabetic and nondiabetic patients during follow-up, and during peritonitis. Healthy medical personnel served as the control group. Chemiluminescence curves were recorded and presented as a sum of three biological components. CL kinetic parameters were calculated, and functional states of PMNs were assessed. Data mining algorithms were used to build decision tree models that can distinguish between different clinical groups. The induced classification models were used afterward for differentiating and classifying new blind cases and demonstrated good correlation with medical diagnosis (84.6% predictive accuracy). In conclusion, this novel method shows a high predictive diagnostic value and may assist in detection of PD-associated clinical states. Peritoneal dialysis (PD) has been recognized as a form of treatment for end-stage renal disease.1,2 A major complication associated with this procedure is acute peritonitis,3–5 and its * Corresponding author. Phone: +972 8 6477182. Fax: +972 8 6472857. E-mail: [email protected]. † Department of Biotechnology Engineering. ‡ Department of Virology, Faculty of Health Science. § Department of Nephrology, Soroka University Medical Center. | Department of Information Systems Engineering. ⊥ National Institute of Biotechnology in the Negev. # The Ilse Katz Center for Meso and Nanoscale Science and Technology. (1) Peterson, P. K.; Matzke, G.; Keane, W. F. Rev. Infect. Dis. 1987, 9, 604– 612. (2) Dalaman, G.; Haklar, G.; Sipahiu, A.; Ozener, C.; Akoglu, E.; Yalcin, A. S. Clin. Chem. 1998, 44, 1680–1684. (3) Gould, I. M.; Casewell, M. W. J. Hosp. Infect. 1986, 7, 155–160. 10.1021/ac800330h CCC: $40.75  2008 American Chemical Society Published on Web 05/30/2008

diagnosis and effective treatment relies on clinical evaluation of the patient and correlation with laboratory examination of dialysate, including total leukocyte count and identification of microorganisms by culture techniques.2,4,5 However, previous reports have also demonstrated problems associated with the diagnosis of acute peritonitis.3–5 Therefore, alternative diagnostic techniques are needed. Polymorphonuclear leukocytes (PMNs), or phagocytes, play a major role in antimicrobial response of the host,6,7 and these circulate in an already pretuned state for future tasks, which can act as a disease marker.8 During phagocytosis of microbial intruders, phagocytes increase their consumption of molecular oxygen. Activation of PMNs, induced by stimuli, results in the production of various reactive oxygen metabolites, including superoxide anion, hydrogen peroxide, hypochlorous acid, and hydroxyl radical, a process collectively termed “respiratory burst”.9–11 This process is accompanied by light emission (chemiluminescence, CL) in the presence of luminol. CL, therefore, is a sensitive measure of the oxidative potential of phagocytes, which correlates well with antimicrobial activity, and can be detected as luminol-dependent CL (LCL).11,12 Despite sensitivity and convenience of this approach, few clinical applications have been so far implemented.13–15 Peritoneal defense to bacterial intruders in patients with endstage renal failure is associated primarily with two basic cellular (4) Ludlam, H. A.; Price, T. N.; Berry, A. J.; Phillips, I. J. Clin. Microbiol. 1988, 26, 1757–1762. (5) Males, B. M.; Walshe, J. J.; Amsterdam, D. J. Clin. Microbiol. 1987, 25, 2367–2371. (6) Segal, A. W. Annu. Rev. Immunol. 2005, 23, 197–223. (7) Mayer-Scholl, A.; Averhoff, P.; Zychlinsky, A. Curr. Opin. Microbiol. 2004, 7, 62–66. (8) Van Dyke, K.; Van Dyke, C. Methods Enzymol. 1986, 133, 493–507. (9) Rosen, G. M.; Pou, S.; Ramos, C. L.; Cohen, M. S.; Britigan, B. E. FASEB J. 1995, 9, 200–209. (10) Magrisso, M.; Marks, R. S. Handbook of Biosensors and Biochips; Marks, R. S., Cullen, D. C., Karube, I., Lowe, C. R., Weetall, H. H., Eds.; 2007; pp 511-529. (11) Dahlgren, C.; Karlsson, A. J. Immunol. Methods 1999, 232, 3–14. (12) Vilim, V.; Wilhelm, J. Free Radical Biol. Med. 1989, 6, 623–629. (13) Stevens, D. L.; Bryant, A. E.; Huffman, J.; Thompson, K.; Allen, R. C. J. Infect. Dis. 1994, 170, 1463–1472. (14) Allen, R. C. Methods Enzymol. 1986, 133, 449–493. (15) Zgliczynski, J. M.; Kwasnowska, E.; Stelmaszynska, T.; Olszowska, E.; Olszowski, S.; Knapik, J. M. Acta Biochim. Pol. 1988, 35, 331–342.

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

5131

mechanisms:16 influx of PMNs into the peritoneal cavity17,18 and alteration in normal functionality of peripheral PMNs.18 The hypothesis of the present study links the participation of PMNs in the antimicrobial response of PD patients and the monitoring of their altered oxidative burst through CL measurement, thus providing us with a timely diagnostic tool. To gain as much information as possible from the CL kinetics pattern (related to the production of reactive oxygen species (ROS)), the extracellular, as well as intracellular components (phagocytosis- or nonphagocytosis-related), of the phagocyteemitted whole-blood luminol-dependent CL were assessed simultaneously.10,19,20 We used these components, and a previously described methodology,21 to derive CL kinetic parameters that allow quantitative comparison of the dynamic functional state of circulating blood PMNs.10 It was shown that to obtain even more kinetic information10,20,22 from the PMNs, one can prime them artificially in order to obtain a dynamic picture of their circulating functional state. To do this we performed a laboratory-based ex vivo controlled shift in the phagocyte functional state by using two different priming systems (formylmethionyl-leucyl-phenylalanine (fMLP) chemoattractant priming and aging of blood) and comparing their dynamic change when measured in these two different scenarios. In addition, we also measured whole blood without the addition of an artificial primer, and this measurement is called a standard system in this manuscript. In this work, we characterized the specific functional states of PMNs from PD patients, applying the aforementioned approach of dynamic component CL sensing in a luminol-amplified wholeblood system, by evaluating kinetic patterns of CL emission from stimulated PMNs. Different clinical groups were classified by the use of data mining algorithms on the basis of calculated kinetic parameters with training classification error of 0%. These mathematical rules were evaluated afterward in differentiating new blind cases and classifying them into most probable clinical groups, and these demonstrated a good correlation with medical diagnosis (84.6% predictive accuracy). The proposed method of combining whole-blood CL, kinetics component analysis and classification of clinical groups using decision trees demonstrated a high predictive diagnostic value and may assist in detection and prediction of PD-associated clinical states. EXPERIMENTAL SECTION Pilot Study Population. Thirty PD patients were recruited from the Nephrology Department of Soroka University Medical Center. Patients were examined at routine follow-up visits during their infection-free period, and these same patients were monitored for clinical signs of peritonitis for a period of 6 (16) Duwe, A. K.; Vas, S. I.; Weatherhead, J. W. Infect. Immunol. 1981, 33, 130–135. (17) Daniels, I.; Bhatia, K. S.; Porter, C. J.; Lindsay, M. A.; Morgan, A. G.; Burden, R. P.; Fletcher, J. Clin. Diagn. Lab. Immunol. 1996, 3, 682–688. (18) Hirabayashi, Y.; Kobayashi, T.; Nishikawa, A.; Okazaki, H.; Aoki, T.; Takaya, J.; Kobayashi, Y. Nephron 1988, 49, 305–312. (19) Magrisso, M. J.; Alexandrova, M. L.; Bochev, P. G.; Bechev, B. G.; Markova, V. I.; Benchev, I. C. J. Biochem. Biophys. Methods 1995, 30, 257–269. (20) Magrisso, M.; Etzion, O.; Pilch, G.; Novodvoretz, A.; Perez-Avraham, G.; Schlaeffer, F.; Marks, R. Biosens. Bioelectron. 2006, 21, 1210–1218. (21) Magrisso, M. Y.; Alexandrova, M. L.; Markova, V. I.; Bechev, B. G.; Bochev, P. G. Luminescence 2000, 15, 143–151. (22) Magrisso, M.; Marks, R. Patent No. in Submission: 22611-WO-07, 2007.

5132

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

months. Background diseases and medication histories were retrieved from the patients’ medical records. When physical examination findings including fever, abdominal pain, and clinical laboratory findings, such as cloudy peritoneal fluid, and elevated white blood cell count were present, a diagnosis of peritonitis was defined, and later confirmed by a positive bacterial culture in the peritoneal fluid. Patients were then followed until resolution of infection with a full recovery or treatment failure (i.e., catheter Tenckhoff removal or patient death). The cases recruited formed three clinical groups, namely, the follow-up group, follow-up patients with diabetes, and those afflicted by peritonitis. The additional control group included six age-matched healthy volunteers from the hospital staff with presumably a normal neutrophil respiratory burst. For evaluation of the derived classification model, two patients were evaluated as peritonitis cases, four patients as a follow-up cases, and two patients as a diabetic follow-up cases. Several patients were tested more than once. The study protocol was approved by the Helsinki committee of Soroka University Medical Center (certification no. 4240). All patients gave informed consent for the study. Reagents. Zymosan-A (Sigma Chemical Co., Z4250), used as a stimulating agent, was opsonized for 30 min at 37 °C in each corresponding patient serum sample (20 mg/mL). A zymosan suspension in Krebs-Ringer phosphate medium (KRP) was prepared immediately prior to use.20 Final pH of the buffer was adjusted to 7.4 with 1 M NaOH. Luminol (Bio-Rad HRP substrate kit 170-5040) was used to amplify CL activity. A luminol stock solution was stored in the dark at 4 °C. In some experiments, fMLP (Sigma Chemical Co.) was used (10-9 M final concentration) for priming of the phagocytes. Chemiluminescence Assays. Peripheral whole blood (1:100 (v/v) final dilution) was used to avoid the appearance of artifacts due to the isolation of PMNs and to preserve conditions that are close to the in vivo cellular environment.10 It was collected from patients in heparinized tubes (20 U/mL). Samples, at a total volume of 200 µL, contained whole fresh blood diluted in KRP, luminol, and zymosan in KRP. The whole fresh blood was diluted with KRP immediately prior to use. Before measurement of the signal, all reagents, with the exception of zymosan, were preincubated in the luminometer at 37 °C for 5 min. After addition of opsonized zymosan already diluted in KRP, the sample contents were mixed and CL was measured. The CL kinetics curve represented CL as a function of time measured in seconds. The measurements were taken every minute for a period of 50 min and were recorded using a luminometer (Luminoskan Ascent luminometer, Thermo Labsystems) operating in the photoncounting mode. Each of the curves shown is representative of at least three independent experiments. Cellular luminescence measured by a luminometer is dependent on the red blood cells number and is directly proportional to the number of phagocytes. Therefore, for normalization of CL results, corrections of the CL responses to PMNs and RBC counts were applied to diluted wholeblood samples. Three CL systems were investigated: 1. Standard system (SS): 20 µL of diluted whole blood 1/10 in KRP, 20 µL of luminol, and 20 µL of opsonized zymosan (20 mg/mL) were diluted to a total volume of 200 µL with the

addition of 140 µL of KRP. Measurement is performed within 1 h from venipuncture. 2. Priming (or primed) system (SP): the same reagents as in the SS were used, but before the dilution of blood, phagocytes were primed using fMLP (incubation for 5 min, 10-9 final concentration in measuring probe). 3. Aging (or aged) system (SA): the same reagents as in the SS were used, but the test was performed 2 h after the SS test. Data Analysis. Component analysis of CL kinetics was used to quantitate at a given time the status of the nonadaptive immune system contained in the collected blood sample. Incubation of PMNs with the fMLP chemoattractant assumed a change in PMN functionality termed priming factor.10 Moreover, blood storage has been described as modulating the phagocyte respiratory burst, as measured by CL, and was used as a kind of priming factor as well.10,15,21 Thus, fMLP priming and blood storage (i.e., aging of blood) were chosen as factors of differential component assessment, reflecting well-controlled shifts in the phagocyte functional status. Data analysis was performed in consecutive steps. First, the experimental CL curves were recorded, then full sets of kinetic parameters were calculated based on decomposition of the experimental CL curves to the three biological components.19,20,23 Next, patients with similar clinical states were placed into defined groups (such as healthy controls, follow-up patients, diabetic follow-up patients without infection, and those with peritonitis). As a last step, a decision tree algorithm was used to build a classification model for discriminating between different clinical groups and determine which set of chemiluminescent variables plays an important role in discriminating between the specific groups. To check model compliance, blind cases were measured and classified. Component Analysis of the CL Curve. The recorded kinetic CL curves were presented as a sum of three biological components representing following processes:19,20 • First component: processes that take place near the plasma membrane, that are connected with phagocytosis, and which cause extracellular CL. • Second component: processes that are located inside the cell, are connected with phagocytosis, and which cause intracellular CL. • Third component: mainly processes that lead to intracellular CL but are not directly connected with phagocytosis. Priming by fMLP was checked at the same time as a standard (control) system, while the aging system was performed as a second measurement, 2 h after the first standard system measurement. This triple-system set was used to obtain differential CL components through which, after analysis, kinetic parameters could be derived. To decompose the chemiluminescent curve into the three aforementioned components, the PeakFit program was employed, using Poisson-type distribution equations for each peak23 and considering the boundary conditions for the time appearance of each peak. Matching the sum of the CL components obtained by modeling to that of the experimental curve was achieved by determining the minimum sum of the squared (23) Magrisso, M. J.; Bechev, B. G.; Bochev, P. G.; Markova, V. I.; Alexandrova, M. L. J. Biolumin. Chemilumin. 1995, 10, 77–84.

differences.22,24,25 Each component contributes to the total intensity, depending on its own kinetics. For each patient, three sets of parameters were derived from the standard, priming, and aging systems. For each patient, kinetic parameters were calculated, with the main parameters presented in Table 1. Decision Tree Construction. In the differentiation of clinical groups, a decision tree was constructed using the J48 (C4.5) algorithm26 available with the Weka 3.5.627 Machine Learning Software. Weka is a collection of machine-learning algorithms implemented in the Java programming language. We have not changed any default settings of the software in our experiments. The final tree was based on 9 parameters chosen by the algorithm from the 82 parameters that were calculated by an earlier described component approach. In order to relate the specific parameters to a given disease, classification rules were found based on the parameters that were chosen by the program as the most differentiating and defining for the formation of a specific clinical group. The induced model was then evaluated using data obtained from blind patient samples. With the use of this procedure, we classified particular unknown cases, based on chemiluminescent measurements, into known clinical groups. The phagocyte function in patients was subjected to classification rules to assign them to the most appropriate group. RESULTS AND DISCUSSION Patient Characteristics. Thirty PD patients (15 males and 15 females) were recruited for the study after giving their informed consent. The mean ages of the healthy control group and the PD patient population were 54.2 ± 5.4 years and 62.57 ± 13 years, respectively. End-stage renal disease was attributed to diabetic nephropathy in 12 cases and to nondiabetic etiology in the other 18. Table 2 shows the baseline characteristics of the PD population and control groups. Twenty-one patients were examined on the day of their regular follow-up, and no infections or inflammatory processes were observed. The group of patients with acute peritonitis included 11 people, 3 of whom were also examined during an infectionfree period (self-control), 2 of which developed acute peritonitis after a previous period of a tunnel infection, and 6 who were examined during the course of acute peritonitis. For the latter two groups, no preinfection self-control values were available. Three patients developed acute peritonitis twice, caused by different microbes. Several peritonitis patients were tested more than once. The peritonitis bacterial agents were Streptococcus group B, Klebsiella pneumonia, and Staphylococcus epidermitidis. The final numbers of all training cases used for the model induction were 12 follow-up (FUP) cases, 10 follow-up diabetic cases, 22 peritonitis cases, and 6 healthy control cases. The final numbers of blind cases evaluated by the induced model were 5 FUP cases, 3 FUP diabetic cases, and 5 peritonitis cases. (24) Magrisso, M.; Marks, R. Publication No. WO/2006/092787, International Application No. PCT/IL2006/000272, 2005. (25) Magrisso, M.; Marks, R. Publication No. WO/2006/092788, International Application No. PCT/IL2006/000273, 2006. (26) Quinlan, J. R. C4.5: Programs for Machine Learning; Morgan Kaufmann Publishers Inc.: San Francisco, CA, 1993. (27) Witten, I. H.; Frank, E. Data Mining: Practical Machine Learning Tools and Techniques, 2nd ed.; Morgan Kaufmann: San Francisco, CA, 2005.

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

5133

Table 1. List of the Parameters Used To Construct a Classification Model for Differentiating Clinical Groups (*Also Applies to Priming and Aging Systems) abbreviation S SP SA Extra Intra nonPhago parameter PtimeS CapS AmpS SlopeS RelPtimeSP RelCapSP S_eff S_vel RelEff_SP RelVel_SP RelPhagS TimeP CapP ExtraS TimeExS nonPhagoS RelTimeExSP RelExtraSP RelIntranoPhagSP

definition standard sample primed sample aged sample extracellular phagocytosis emission intracellular phagocytosis emission nonphagocytosis-related emission definition peak time of standard* sample capacity of standard* samplea amplitude of standard* sample peak of standard* sample divided by time required to reach it peak time of primed* sample divided by peak time of standard sample capacity of primed* sample divided by capacity of standard sample effectiveness of standard* sampleb velocity of standard* samplec effectiveness of primed* sample divided by effectiveness of standard sample velocity of primed* sample divided by the velocity of standard sample phagocytosis capacity of standard* sample divided by total capacity of standard sample time peak of specific peak of specific sample capacity of specific peak of specific sample extracellular phagocytosis-related emission of standard* sample time peak of extraphago emission of standard* sample nonphago-related CL of standard* sample time of extracellular phagocytosis-related emission of primed* sample divided by time of extracellular phagocytosis-related emission of standard sample extracellular capacity of primed sample divided by extracellular capacity of standard sample intracellular nonphagocytosis capacity of primed* sample divided by intracellular nonphagocytosis capacity of standard sample

ref

19 19 19 ref 33 21, 33 10, 10, 10, 21 21 22

33 24, 25 24, 25 24, 25

22 22 22, 22, 10, 22, 10, 22

33 33 24, 25 33 24, 25

22 10, 24, 25

a Capacity is presented by the total CL capacity of unit cells, which reflects their ability to generate ROS; calculated as the total area under the CL curve (ref 21). b Effectiveness is presented as a ratio of the capacity of the second component to that of the first; shows the effectiveness of ROS generated during phagocytosis (ref 21). c Velocity is presented as a ratio of the sum of the capacities of the first and second components to the capacity of the third component of CL kinetics; shows the velocity to achieve respiratory burst (ref 21).

Table 2. Baseline Characteristics of the Peritoneal Dialysis Patients Baseline Characteristics of the Patients (No. of Patients ) 30) sex (M/F) average age (years) average duration of dialysis (months) continuous cycle PD patients continuous ambulatory PD patients

15:15 62.57 (27-86) 38.2 (4-112) 16 14

Diagnosis (No. of Cases (%)) diabetic nephropathy 12 (40%) glomerular disease 2 (7%) tubulointerstitial disease 3 (10%) adult polycystic kidney disease 2 (7%) unknown 5 (16%) others 6 (20%) Control Group (No. of Patients ) 6) sex (M/F) 6:0 average age (years) 54.2 (49-62)

Whole-Blood Chemiluminescence. Data was analyzed and compared to values obtained from healthy, noninfected controls, demonstrating the earlier described “resting” pattern.21 PMNs in the “resting” state showed a small capacity to generate ROS, which they produce relatively slowly, mainly by processes indirectly associated with phagocytosis. This functional state is distinguished by a low readiness for defense and killing activity.21 The described 5134

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

behavior was demonstrated in the three systems checked, i.e., in standard, priming, and aging conditions, and these in comparison to other clinical states (Figures 1 and 2). CL kinetics of follow-up (FUP) cases, in general, showed moderate respiratory burst capacity, good velocity, and effectiveness (Figure 1A). This corresponds to the “restoring state” (CLEHVH ) capacityLoweffectivenessHighvelocityHigh).21 In this state, ROS directly associated with phagocytosis is predominant and ROS generated mainly inside the cells (i.e., high effectiveness). In this state, the PMNs do not have sufficient capacity for ROS generation. The ability of PMNs to generate ROS both rapidly and effectively without being activated shows a high readiness for function. The clinical state of follow-up patients is associated with some background nephrologic disorder; however, during the measurements, patients were in a noninfectious state and overcoming general supportive PD treatment so that the phagocytes were in the restoring state. This state assignment finds theoretical support in the assigned functional state of PMNs from the patients, consistently provoked by recurrent infections associated with continuous PD processes. In other words, the defense cells of the immune system are ever-prepared to act as the organism is pretuned for this task by recurrent infections. Next, the PMNs CL pattern of the PD group can be further subdivided in relation to the presence or absence of diabetes. The impaired phagocytic capacity of cells in patients with diabetes is

Figure 1. Description and comparison of derived respiratory burst parameters (capacity, effectiveness, and velocity) in four clinical states (FUP, diabetic FUP, peritonitis, and control) in the standard (A), priming (B), and aging (C) systems. Functional states of PMNs are assigned according to the calculated parameters as “resting” state in the case of control, “restoring” state in the case of FUP, “alternatively activated” state in diabetic FUP, and “effective” state in peritonitis.

well-known, as well as the fact that infections remain a serious challenge to diabetic patients.28 CL kinetics of follow-up cases with diabetes showed high capacity, high effectiveness, and moderate velocity, lower than observed in regular FUP cases. This state corresponds to an “alternatively activated state” (CHEHVL),21 relative to the FUP state, and is achieved by the presence of higher levels of glucose in the blood that, in turn, influences the phagocyte state. In this state, PMNs generate ROS relatively slowly, mainly by processes not directly connected to phagocytosis (VL). Although the process of ROS generation during phagocytosis occurs predominantly inside the cells (i.e., high effectiveness), phagocytosis itself is weak. If we assume that ROS generation must proceed mainly by processes directly associated with phagocytosis, this functional state represents an alternative to this (28) Pozzilli, P.; Leslie, R. D. Diabetic Med. 1994, 11, 935–941.

assumption. In general, for follow-up patients with the diabetes, fMLP priming and aging led to a decrease in CL capacity (Figure 1, parts B and C). Peritonitis cases are associated with increased capacity, decreased velocity, and good effectiveness, as compared to FUP cases. This state corresponds to an “effective state” (CHEHVH), very close to the “fighting state” (CHELVH).21 In this state, the capacity of PMNs to generate ROS is significant. During phagocytosis, ROS generation flows predominantly inside the cells. Furthermore, the oxidative potential of the PMNs is achieved mainly by those processes directly associated with phagocytosis. In addition, fMLP priming and aging of this group caused no change in CL kinetics parameters (Figure 1, parts B and C), as well as no visible change in shape of the CL kinetics curves (Figure 2D). It can be seen from the patterns of the curves presented in Figure 2 that in the cases of FUP and FUP with diabetes, both fMLP priming and aging affected the shape of the CL kinetics curves. CL kinetics of the FUP group show good respiratory burst capacity and velocity. Peritonitis cases can be characterized by curves without visible change in the different systems, with decreased velocity. Control curves show low velocity and low capacity. Such characteristic CL shape differences were used to build a description of PD respiratory burst patterns in different groups by component analysis of the CL curve. In the present work, we were able to distinguish between different clinical conditions of PD patients by their distinct patterns of PMNs functionality. PD patients, as a group, have a clearly different CL pattern than that of healthy controls. By contrast, control follow-up and diabetic follow-up cases that are not affected by any kind of infection but did suffer from some kind of nephrologic background disorder demonstrated a completely different ROS generation behavior. As was shown in previous work,21 the CL kinetics of patients with infectious diseases show mainly an effective type of behavior, supported by the behavior of cases of peritoneal bacterial infection considered in this study. Even though the variation of phagocyte functional activity was broad, our approach derives sufficient information from a particular patient CL response to quantitatively place that individual closer to some clinical group and find specific parameters characterizing this group. The different kinetic patterns of phagocyte functionality are indicative of PMNs input in immune defense and can serve as a useful tool in predicting the competency and success of defense. Phagocyte CL Information Can Differentiate between Clinical Groups. The previously described phagocyte parameters (Table 1) were measured in all cases, and these were checked to evaluate to what extent the proposed methodology can serve as a relevant tool to describe clinically primed and challenged phagocyte diversity. The method uses the standard CL system as an internal reference for comparison to primed systems (i.e., fMLP priming and aging). Basic descriptive kinetic parameters (e.g., the area under the CL kinetics curve, initial slope, time to peak), as well as CL kinetics component information, allowed us to calculate more parameters so as to derive more respiratory burst-related information, and those two groups were used to assess phagocyte CL activity. After retrieving information on all parameters, a decision tree model was built using the J48 Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

5135

Figure 2. CL pattern data recorded using standard, priming, and aging systems and averaged by different groups. Four clinical states are represented: control (A), FUP (B), diabetic FUP (C), and peritonitis (D). Different CL patterns are used for distinguishing between different clinical conditions of PD patients by their distinct patterns of PMNs functionality.

algorithm. Such analysis resulted in distinct statistical separation of the clinical groups in the training data. The classification error rate was 0% for all training cases. As expected, different sets of CL parameters can distinguish between four different clinical groups like follow-up, follow-up with diabetes, control, and peritonitis cases according to the induced classification model, presented as a decision tree (Figure 3). Control Group. Three main parameters were chosen to classify the control group, i.e., the capacity of the standard sample (CapS), the slope of the standard sample (SlopeS), and the capacity of the extracellular part of the standard sample (CapExtraS). In general, the capacity of the standard system should be lower in all groups, as compared to that fraction of the cases of the diabetic follow-up group, differentiated on the basis of this parameter alone. This is supported by Figure 1A, where the capacity of the diabetic FUP group has the highest values of all the groups. To separate the control group, SlopeS should be low, that is, the ratio of peak magnitude to time required to reach this point should be low, reflecting slow kinetics. This finding, derived from the decision tree model, is supported by the CL curves presented in Figure 2A, showing low velocity. In addition, the capacity of the extracellular part should be higher than in other groups, with effectiveness being the lowest, but still good, as compared to the other clinical cases. This finding is supported by the derived parameters presented in Figure 1A. The parameters chosen by the decision tree support the estimated functional state of phagocytes in the control group, assigned according to CL kinetics, as a “resting state”. FUP Group. Three paths were followed in classifying this group. The first path includes the influence of the capacity of the 5136

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

extracellular part of the standard sample and the slope of the aged sample. A minor extracellular part reflects good effectiveness (i.e., low slope of the standard sample), with the low slope of the aging curve corresponding to the low extracellular part in the aging. The second path includes the high capacity of extracellular phagocytosis-related emission of the aged sample in some cases, as clearly seen in Figure 2B. This corresponds to low effectiveness in the aging FUP group (Figure 1C), but it is still acceptable, as compared to the other clinical cases. The third path includes the parameter that differentiates between the FUP and peritonitis groups in forms of the peak time of the aging sample. To separate between the two groups, this value should be higher in the FUP group. An additional parameter that plays a role is the capacity of the standard sample, a value that should be lower in all groups than in some cases in the diabetic FUP group (supported by Figure 1A). According to the chosen parameters of the standard system, it can be concluded that the effectiveness of the standard sample is good and the capacity is low, supporting the estimated functional state of phagocytes in this clinical group (i.e., the restoring state). FUP + Diabetes Group. Two main parameters influence differentiation. The first is the time required to reach the peak not connected to the phagocytosis part of the standard sample. This value for the diabetic FUP group should be much lower than in the peritonitis case, even though this parameter does not offer a direct connection in estimating the functional state of the phagocytes. The second parameter is the standard capacity. Higher capacity is presented by the diabetic FUP group (in part of the

Figure 3. Decision tree and confusion matrix of PD patients in the training set. Four clinical groups were successfully classified (0% error): control (6 cases), FUP (12 cases), FUP + diabetes (10 cases), and peritonitis (22 cases). The tree is based on 9 parameters chosen by the algorithm from the 82 parameters that were calculated according component approach.

cases), as compared to other groups (supported by Figure 1A). This is also supported by the chosen functional state, where the capacity of the diabetic FUP is determined as being high. Peritonitis Group. The main group that was differentiated by the model comprises 12 cases characterized by a low capacity of the extracellular part of the aged sample, that is, high effectiveness of the aging system for the peritonitis cases. This is supported by chemiluminescent parameters of the aged kinetics presented in Figure 1C. Another path shows that the capacity of the extracellular part of the standard sample of the peritonitis group should be lower than the control, meaning higher effectiveness, as seen in Figure 1A. As was previously mentioned, the capacity of the standard system should be lower than in part of the diabetic FUP cases. Other parameters presented in different paths did not show specificity for the peritonitis group. Respiratory burst-related information from the chemoattractantprimed systems and different sets of CL parameters were able to separate the four clinical groups, which correlate well with the descriptions and evaluation of the functional phagocyte states. The model that was built was able to classify, with high precision, four clinical groups, while choosing the most unique connections between the various parameters. The model illustrates the fact that the control group was classified on the basis of low velocity and lowest capacity value in the standard system. This feature is also portrayed in the CL kinetics pattern that distinguishes this group. The follow-up group was specified by a lower effectiveness

in the aging system and lower capacity in the standard system than the diabetic follow-up group. It can be concluded that the aging system plays an important role in the classification of this group. In classifying the diabetic follow-up group, the capacity in the standard system is the parameter that plays a role in defining the group and which can be compared to the regular follow-up group. This means that impaired phagocytic activity in the diabetic state and readiness for defense are expressed by the higher capability to generate ROS. The peritonitis group was classified on the basis of higher effectiveness in the aging and standard systems, as compared to controls, and as supported by a clinically implied diagnosis of a more activated immune system in the case of an infection. Blind Case Evaluation. Early proper diagnosis and initiation of the appropriate therapy can greatly affect the outcome of every disease. Hence, an approach for rapid, sensitive, cheap, and simple evaluation of clinical states is needed. In our study, we checked the ability of the proposed approach in combination with a decision tree classification technique, to assess the functional status of circulating phagocytes in different patients and differentiate between diseases. The classification tree is a special mathematical model, according to which differentiation of specific clinical states can be achieved. By calculating the special kinetic parameters that characterize each chemiluminescent curve for each patient and substituting the values of those parameters into the model provided by the program, a clinical state of the patient can be Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

5137

Table 3 (A) Classification of 13 Peritoneal Dialysis Blind Cases According to the Induced Model classified as f FUP diabetes peritonitis total percent correct FUP diabetes peritonitis

4 0 1

0 3 0

1 0 4

5 3 5

80 100 80

(B) Detailed Accuracy by Class: Blind Peritoneal Dialysis Cases class

TP rate

FP rate

precision

recall

F-measure

ROC area

FUP diabetes peritonitis

0.8 1 0.8

0.125 0 0.125

0.8 1 0.8

0.8 1 0.8

0.8 1 0.8

0.838 1 0.838

offered and the functional state of the phagocytes can be described. On the basis of the decision tree model in Figure 3, blind cases can be evaluated and their group membership determined. As such, 13 blind cases were examined. The cases were assigned by our collaborating physician in the following classification that included five FUP cases, three diabetic FUP cases, and five peritonitis cases. Eleven of the 13 cases were correctly classified (84.6%), with 80% of the blind FUP, 100% of the blind diabetic FUP, and 80% of the blind peritonitis cases being classified correctly (Table 3, sections A and B). A 95% confidence interval for the true proportion of incorrectly classified cases, based on the binomial distribution, is {1.9%, 45%}. As can be seen in Table 3B, the true positive (TP) rate in diabetic FUP cases was 1 (100%), whereas in the FUP and peritonitis cases, the value was 80%. The false positive (FP) rate was zero in case of diabetic FUP. The FUP and peritonitis groups were built with 80% precision, whereas the diabetic FUP group attained 100% precision. The F-measure and the ROC area were lower than 100% in FUP and peritonitis cases. In this study, the classification model was built on the basis of a relatively small number of cases. As such, the model is less strong than if a bigger sample set population had been used. In spite of this, the approach showed very good compliance and high a percentage of blind cases were classified correctly. According to the model built (Figure 3), the blind cases that were checked gave 84.6% correctly classified instances.

a reflection of the organism’s readiness to defense, has been difficult to evaluate. It is well-known that PMNs circulate in a “primed” state and, therefore, can be of high predictive value.30,31 Attempts have been made to correlate the primed activity of circulating PMNs with the severity of disease and outcome.13,20,32 Recently, the component CL approach21 was reported as a tool for the simultaneous determination of extracellular and intracellular ROS emission from the same sample, as well as for the assessment of the phagocyte functional status. In this work, we have tried to estimate the functional states of phagocytes according to the described CL component approach and to classify the clinical states into appropriate groups. Our approach was shown to differentiate between various intraabdominal pathological processes afflicting PD patients that were undergoing peritoneal dialysis. As a result, the dynamic wholeblood component chemiluminescent method, combined with decision tree classification algorithms, was shown to be precise and informative. The method presented herein, thus, has high predictive diagnostic value and can be implemented in various medical institutions as an adjunct to clinical decision making. In order to become a powerful and efficient clinical tool to aid in the diagnosis and prognosis of any infection, a more expanded database is required with many more clinical cases. Future studies are needed to further characterize the specific CL kinetic patterns generated from the interactions of PMNs with different types of microorganisms and generation of data mining models on those specific organisms. Enlargement of a pool of kinetic parameters used for the evaluation of each patient is needed as well, which can be attained by using additional priming systems. ACKNOWLEDGMENT We thank Moni Magrisso for his participation in the study at its early stage. We thank all the medical staff, especially Elise Hadar and Tatiana Bernshtein from the Nephrology Department, in providing clinical information and blood samples. D.P. thanks the Council for Higher Education for Converging Technologies for their Fellowship support. R.S.M. thanks the Israel Institute of Technology for financial support through the Institute for National Security. Received for review February 17, 2008. Accepted April 14, 2008. AC800330H

CONCLUSIONS For many decades the phenomenon of phagocytosis has been of prime interest to medical scientists as an important factor in the defense of the host in infectious disease.29 However, the functional state of circulating phagocytes, which is known as being (29) Sbarra, A. J.; Karnovsky, M. L. J. Biol. Chem. 1959, 234, 1355–1362.

5138

Analytical Chemistry, Vol. 80, No. 13, July 1, 2008

(30) Zimmerman, G. A.; Renzetti, A. D.; Hill, H. R. Am. Rev. Respir. Dis. 1983, 127, 290–300. (31) Maderazo, E. G.; Woronick, C. L.; Albano, S. D.; Breaux, S. P.; Pock, R. M. J. Infect. Dis. 1986, 154, 471–477. (32) Wakefield, C. H.; Carey, P. D.; Foulds, S.; Monson, J. R.; Guillou, P. J. Arch. Surg. 1993, 128, 390–395. (33) Giltinan, D. M.; Capizzi, T. P.; Abruzzo, G. K.; Fromtling, R. A. J. Clin. Microbiol. 1986, 23, 531–535.