Lavoro Schizofrenici

© Med Sci Monit, 2005; 11(8): CR366-375 PMID: 16049378

WWW. M ED S CI M ONIT.COM

Clinical Research

Received: 2005.01.10 Accepted: 2005.05.30 Published: 2005.08.01

Identification of schizophrenic patients by examination of body odor using gas chromatography-mass spectrometry and a cross-selective gas sensor array

Authors’ Contribution: A Study Design B Data Collection C Statistical Analysis D Data Interpretation E Manuscript Preparation F Literature Search G Funds Collection

Corrado Di Natale1,2 CDE, Roberto Paolesse2,3 AD, Giuseppe D’Arcangelo3 B, Paolo Comandini1 BF, Giorgio Pennazza1 BE, Eugenio Martinelli1 CF, Santo Rullo4 AF, Maria Claudia Roscioni5 AF, Claudio Roscioni6 A, Alessandro Finazzi-Agrò7 A, Arnaldo D’Amico1,2 AD 1

Department of Electronic Engineering, University of Rome “Tor Vergata”, Rome, Italy CNR-IMM, Rome, Italy 3 Department of Chemical Science and Technology, University of Rome “Tor Vergata”, Rome, Italy 4 Italian Hospital Group, Guidonia, Rome, Italy 5 A.S.L. Roma, Rome, Italy 6 Azienda Ospedaliera S. Camillo-Forlanini, Rome, Italy 7 Department of Experimental Medicine, University of Rome “Tor Vergata”, Rome, Italy 2

Source of support: Departmental sources

Summary Background:

Material/Methods:

Previous findings have shown that the body odor of patients affected by schizophrenia contains some specific compounds. Chemical sensor technology has proved to be able to classify different odours. We investigated the possibility of using a chemical sensor array to detect body odor alteration in schizophrenic patients. The sweat of subjects was sampled and analysed by Gas Chromatography-Mass Spectrometry (GC-MS) and by an array of cross-selective gas sensors. A total of 27 individuals were involved in the experiment: 9 schizophrenics, 9 with other mental disorders, and 9 controls.

Results:

GC-MS analysis showed a richer composition for the sweat of schizophrenic patients. Nevertheless, the individuation of specific markers was unsuccessful. On the other hand, statistical analysis of cross-selective gas sensor data provided a complete classification of schizophrenic patients with respect to the other three groups.

Conclusions:

The alteration of body odor in schizophrenic patients was confirmed by GC-MS and chemical sensor array. Results show that the alteration is complex and cannot be limited to a single compound, but rather to a global variation of the body odor.

key words: Full-text PDF: Word count: Tables: Figures: References: Author’s address:

CR366

schizophrenia • body odor • gas-chromatography • gas sensors

http://www.medscimonit.com/fulltxt.php?IDMAN=6860 4102 5 3 27 Professor Corrado Di Natale, Department of Electronic Engineering, University of Rome “Tor Vergata”, Via di Tor Vergata 110, 00133 Roma, Italy, e-mail: [email protected]

Current Contents/Clinical Medicine • SCI Expanded • ISI Alerting System • Index Medicus/MEDLINE • EMBASE/Excerpta Medica • Chemical Abstracts • Index Copernicus

Med Sci Monit, 2005; 11(8): CR366-375

BACKGROUND Schizophrenia is generally defined as a complex pathology not immediately connected to specific genetic and environmental causes. The adopted protocol for the diagnosis of this disease is based on the observation of a number of simultaneously occurring behavioral, emotional, and cognitive disturbances [1]. For this reason the introduction of objective diagnostic tools for schizophrenia is highly desirable. Categorical systems of classification of mental disorders provide clear descriptions of diagnostic classes to enable clinicians and investigators to diagnose, communicate about, study, and treat people with mental disorders. Nevertheless, most of the reported clinical presentations are not clearly classified according to the available etiopathogenic or pathophysiological criteria. Since a number of psychotic symptoms, present in both schizophrenia and other mental disorders (i.e. bipolar disorders, personality disorders, etc.), are found to be continuously distributed phenomena without clear boundaries, the adoption of a dimensional system able to classify clinical presentations on the basis of the quantification of attributes, rather than on the assignment to categories, seems more promising. Nevertheless, this is a common problem whenever decisions have to be taken on the basis of qualitative observations. In recent years genetic and environmental influences on schizophrenia have been studied, providing evidence that besides any neuronal, neuroanatomical, or neurochemical factors, schizophrenia has familial and genetic component, and also that environmental factors together play a significant role in the pathophysiology of the disease. From the biochemical point of view, some hypotheses connect symptoms of schizophrenia to a functional hyper-activity of the cerebral dopaminergic system. High levels of dopamine may result from excessive biosynthesis of the amino acid tyrosine, the precursor of dopamine. Dopamine growth has been linked to a faulty gene that codes for the enzyme dopamine-b-hydroxylase, which converts dopamine to norepinephrine. As a partial demonstration of this, Angst [2] showed that this enzyme could be blocked with the drug disulfiram. In alcoholics who overdosed on disulfiram the treatment resulted in symptoms indistinguishable from schizophrenia. In the late sixties, the presence of an odorous substance (known as trans-3-methyl-2-hexenoic acid, a metabolic product of 6-hydroxydopamine) in the sweat of schizophrenics supported the dopamine hypothesis [3,4]. Later, Stein and Wise [5] argued that since trans-3-methyl-2-hexenoic acid is a metabolic product of 6-hydroxydopamine, its presence in schizophrenic patients could be explained through an auto-oxidization mechanism of excess dopamine to 6-hydroxydopamine. As a further consequence this mechanism also causes degeneration of peripheral sympathetic nerve terminals, resulting in a marked and long-lasting depletion of norepinephrine. Further evidence supporting 6-hydroxydopamine as a neural degenerative agent of noradrenergic nerve endings comes from the known hallucinogenic activity in humans of phenethylamine derivatives with the same 2,4,5-substitution pattern of 6-hydroxydopamine.

Di Natale C et al – The relationship of body odour and schizophrenia

A peculiar skin odor associated with schizophrenia was observed in some subjects several years ago [6]. As mentioned before, the odd odor was related to the presence of trans-3methyl-2-hexenoic acid, and this compound was proposed as a chemical marker for the diagnosis of schizophrenia. This assumption was not confirmed, however, by later gas chromatography-mass spectrometry (GC-MS) investigation [4]. Studies revealed a large variability of the concentration among individuals, and also pointed to the fact that no absolute relation between the presence of this compound and the occurrence of the disease was observed. However, the original idea was not completely abandoned, and more recently, anomalies in the chemical composition of the expired breath of individuals with schizophrenia have also been reported [7]. It is widely accepted that some pathologies may be associated to skin odor alterations [8]. In general, since body odor results from the combined action of skin glands and bacterial populations, it is expected to be sensitive to a large number of biochemical processes, and to any alteration of them induced by pathological states. The chemical analysis of human body odor requires the adoption of very careful protocols concerning sweat sampling, the extraction of the relevant species, and their qualitative and quantitative evaluation [9]. In this paper, the term odor is used to indicate a blend of several compounds. Since human olfaction may be rather insensitive to some of them, body odor should not be confused with the smell perceived by human olfaction. Because of the great complexity of sweat composition, these studies should be carried out by expensive GC-MS equipment, and thus they cannot be used for large scale screening tests. Nevertheless, in the last two decades the development of chemical sensors has provided the opportunity to reconsider these seminal studies in order to determine whether novel diagnostic tools can be developed, based on the chemical information contained in the exhaled volatile substances. In the 1980s, the absence of selectivity, one of the major drawbacks of chemical sensors, was taken into consideration as the basis for a novel instrument, the “electronic nose”, able to provide global information about samples. This somewhat resembles human olfaction [10]. These instruments are basically arrays on non-selective sensors, characterized by a wide spectrum of sensitivity to many volatile compounds, with a large overlap of response towards several compounds. This fundamental characteristic of artificial sensors is similar to that found in natural olfaction receptors [11], and this similarity is the basis on which artificial olfaction systems are developed. In these systems, sensor response is not unequivocally correlated with the concentration of a single compound, but rather with a combination of the whole chemical information contained in the sample. The performance of natural olfaction in molecular discrimination and recognition results from the complex sensory signal processing carried out in the olfactory bulb and cortex. In the same way, most of the features of artificial olfaction are revealed after a proper application of statistical multicomponent data analysis, ranging from classical statistics, to chemometrics and neural networks [12].

CR367

CR

Clinical Research

Since the beginning, the arrays of non-selective sensors, widely known as electronic noses, have been applied to many different fields [13]. Recently, medical applications have also been taken into consideration, and the application of electronic noses to detect diseases has been proposed [14]. In this context some researchers, following the results obtained by applying standard analytical chemistry approaches [15], have shown the ability of these devices to identify bacteria to be a first step towards the identification of infections in living tissues [16,17]. On this basis, the possibility of detecting gynecological infections [18] and helicobacter pylori infection have been verified [19]. More recently, the possibility to identify lung cancer by analyzing the breath of patients with an electronic nose was demonstrated [20]. In this paper, the problem of identification of schizophrenia-affected individuals through the analysis of their body odor was considered. The experiment involved a number of patients affected by schizophrenia, a population of patients affected by different mental disorders, and healthy people as reference controls. The body odor of each individual participating in the experiment was sampled and analyzed in parallel with the artificial olfaction system and GCMS equipment. The procedures of collection human body odor were reported in another paper, where the capability of chemical sensors to detect compounds secreted by the human body was investigated [21]. The simultaneous use of an analytical instrument, such as GC-MS, made it possible to follow the chemical composition of the body odor and to individuate any eventual anomalous compounds due to errors in the experimental procedure.

MATERIAL AND METHODS The analyses were performed with a Gas Chromatograph Model HRGC 5160 (Carlo Erba, Italy) coupled to a Mass Spectrometer model Quattro (VG Micromass, UK). A NUKOL capillary column (30 m × 0.25 mm i.d.) from Supelco (USA) was used in split mode (100:1 ratio) at a temperature programmed from 70°C (for 2 minutes) to 190°C at 15°C/min, using Helium (P=100 kPa) as carrier and with the injector maintained at 250°C. An array of Quartz Microbalance (QMB) sensors developed and manufactured at the University of Rome Tor Vergata and CNR-IMM was used. The principles and application of these sensors is based on the variation (D¦) of the fundamental resonance frequency of a thin quartz crystal as a consequence of the adsorption of molecules from the gas phase [22]. The adsorption or desorption of molecules results in a variation of the oscillating mass, which induces a variation of the frequency of the electric signal generated by an electronic oscillator. The chemical sensitivity is provided by coating the surface of quartz with appropriate sensing material. Different kinds of metalloporphyrins were used here. The reason for the use of metalloporphyrins as sensing material for artificial olfaction systems derives from the observation that since odorous compounds are, in general, ligands for metal ions, metal complexes are good candidates for odor sensing. Among organometallic structures, metalloporphyrins are the most

CR368

Med Sci Monit, 2005; 11(8): CR366-375

attractive, offering a wide variety of ways to modify their structure. The sensing properties of sensors based on metalloporphyrins (in terms of selectivity and sensitivity) depend on the nature of both the central metal and peripheral substituents of the porphyrin complex [23,24]. In this way, even small variations in the porphyrin structure give sensors with different adsorption properties. This flexibility makes this class of compounds extremely attractive for building electronic noses, where sensors of different selectivity are required [25]. In this work an electronic nose formed by seven of such sensors was used. The response of each sensor is expressed in Hz as difference between the signal frequency when the sensors are exposed to reference air and to the sample. More details about the instrument can be found elsewhere [26]. This electronic nose is currently being utilized in other medical applications, such as the diagnosis of lung cancer from breath analysis [20]. These sensors have shown a good sensitivity towards aromatic compounds, amines, alcohols, ketones, and carboxylic acids. Because trans-3-methyl-2-hexenoic acid belongs to the latter class of volatile compounds, these sensors are particularly promising for this particular application. The experiment here described involved a total of twentyseven individuals recruited in a small therapy unit in Rome. Nine individuals were affected by schizophrenia, according to diagnostic criteria of the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV, American Psychiatric Association). In particular, three patients were affected by paranoid type (295.30), three patients by residual type (295.60), two patients by disorganized type (295.10), and one patient by undifferentiated type (295.90). Nine individuals were affected by other mental disorders. In particular, three patients were affected by bipolar I disorder (296.40), six patients by personality disorders (one paranoid type: 301.22, one schizotypal type: 301.22, and three borderline types: 301.83). All patients were treated with anti-psychotic drugs (four patients with typical neuroleptics, and fourteen patients with atypical neuroleptics), and twelve in association with antiepileptic drugs and/or benzodiazepins. The main characteristic of each subject participating in the experiment are listed in Table 1. The study was approved by the local ethics committee. Nine individuals, recruited among physicians and nursing staff of the therapy unit, were chosen as a reference population. The body odor was sampled on the upper side of the forearms. The sampling was performed with a cotton compress applied for 30 minutes on the skin. One half of each compress was then analyzed with the electronic nose, and the other one with the GC-MS. In order to remove the influence of detergents and perfumes, the forearms were washed no less than two hours before the sampling with the same non-perfumed, neutral soap. After washing, the forearms were dried with sterile

Med Sci Monit, 2005; 11(8): CR366-375

Di Natale C et al – The relationship of body odour and schizophrenia

Table 1. Characteristics of the subjects participating in the experiment. For schizophrenic and mentally disordered subjects, diagnoses and DSM-IV category are provided. Group

Schizophrenia

Mental disorders

Reference

Individual

Age

Sex

Diagnoses

DSM-IV category

Therapy

1

44

M

Paranoid

295.30

TN; AE; B

2

62

F

Paranoid

295.30

AN

3

37

M

Paranoid

295.30

TN; AE

4

45

F

Residual

295.60

AN

5

54

M

Residual

295.60

AN; AE; B

6

43

F

Residual

295.60

AN; AE; B

7

54

F

Disorganized

295.10

TN; AE

8

46

F

Disorganized

295.10

TN; AE; B

9

61

F

Undifferentiated

295.90

AN; AE

10

63

M

Bipolar I

296.40

AN

11

55

M

Bipolar I

296.40

TN

12

51

F

Bipolar I

296.40

AN; AE; B

13

40

M

Personality disorders

301.22

AN; AE

14

38

F

Personality disorders

301.22

AN

15

35

M

Personality disorders

301.22

AN; AE; B

16

46

M

Personality disorders

301.83

AN; AE

17

67

F

Personality disorders

301.83

AN

18

74

M

Personality disorders

301.83

AN; AE

19

35

F

20

57

M

21

49

M

22

47

F

23

57

F

24

45

M

25

55

F

26

51

F

27

44

M

General indications about therapy are also listed with the following abbreviations: TN – typical neuroleptic; AN – atypical neuroleptic; AE – antiepileptic; B – benzodiazepins. cotton and kept in open air, not in contact with clothes. The staff at the therapy unit supervised the correct observance of the procedure. The whole experiment lasted for two weeks; during this time, the sample sequence was properly randomized. For the electronic nose measurements, the cotton compress was closed in a sealed vial, held at 25°C for 30 minutes. The headspace of the vial was then inducted at 40 sccm, using the peristaltic built-in pump of the electronic nose, into the sensor chambers. The sensor signal was evaluated as the difference with respect to a reference. As reference, synthetic air (containing 25% of O2 in N2) was used. The sensor

response was achieved in about 3 minutes with a total consumption of 120 ml of sample. Each person was measured three times during the experiment, and the mean signal was reported in the Results section. For the GC-MS measurements the cotton compress was immersed in a 20 ml volume of acetone for 5 minutes. The extract was then filtered, concentrated under vacuum and then diluted again in acetone in order to obtain a constant extract volume of 1 ml. The extract was then analyzed by GC-MS as described above. To control the efficiency of the trans-3-methyl-2-hexenoic acid recovery following this protocol, 10 ml of a 10–5 M physiological solution of commercially available 2-hexenoic acid was absorbed by a cotton

CR369

CR

Clinical Research

A1BIS 100

Med Sci Monit, 2005; 11(8): CR366-375

Scan EI+ TIC 6.64e7 RT

6.038

acetone A 7.388

%

B

D

C

8.513 10.200 9.030 0 4.000

B2

6.000

100

8.000

14.228

rt 10.000 12.000 14.000 16.000 18.000 20.000 22.000

5.880

Scan EI+ TIC 4.53e7 RT

acetone

%

A

3.068

B 10.111

12.271

7.298

D 14.116 14.993

16.838

rt

0 4.000

6.000

8.000

C2Z 5.768

100

acetone

10.000

12.000

14.000

16.000

18.000

Scan EI+ TIC 2.21e7 RT

%

8.378

17.086 22.126 rt 4.000 6.000 8.000 10.000 12.000 14.000 16.000 18.000 20.000 22.000 24.000 26.000

3.518 0

Figure 1. GC-MS spectra of one sample for each class of individuals: a) schizophrenic (subject 1), b) mental disordered (subject 17), c) healthy reference (subject 23). Spectra of the body odor of the schizophrenic individual (a) show a richer assortment of compounds than in the other two cases. In particular, the healthy reference is characterized by few compounds exceeding the threshold of the GC-MS equipment. compress, and then the compress was treated as above; under these conditions more than 99% of the acid was recovered in the acetone solution. Data from individual sensors were analyzed with simple statistics to calculate the mean of three replications for each sample. The correlations between sensor responses and the three classes of individuals measured in the experiment were analyzed by simple box and whiskers plots. The data of the whole array of sensors (electronic nose) were analyzed with a multivariate approach aimed at classifying individuals in the three classes (schizophrenia, other mental disorders, and reference). The main goal of this study was to test the capability of both GC-MS and the electronic nose to classify the individuals to groups according to the mental disorders. For this reason

CR370

a supervised classification technique was chosen. The simplest technique that can be used is the discriminant analysis; in this approach, a linear model between sensor data and classes is built. The use of Partial Least Squares (PLS) helps avoiding problems due to sensor correlation. Furthermore, PLS provides a decomposition of the sensor data in latent variables that can be plotted to provide a visual representation of the classification properties [27]. The number of variables has been optimized in order to minimize the prediction error in a “leave-one-out” cross-validation procedure. All calculations were performed in MatlabÔ 6.

RESULTS Figure 1 shows three examples of GC-MS spectra of one individual sample for each class (schizophrenic, mental disordered, reference). Acetone (the first peak of the spectra) was used as solvent to extract compounds from the cotton compress, and was then used as reference. These spectra show a different array of compounds, both qualitatively and quantitatively. In order to collectively evaluate the GC-MS spectra, the chromatograms were synthetically represented, extracting the outstanding peaks and expressing the concentrations as relative to the amount of solvent (acetone) used to extract the volatile compounds from the cotton compress. These results show that the body odor changes are complex, and none of the compounds can be considered as a sure and reliable identifier of the disease; rather, it is their combination that could be useful for diagnostic purposes. In order to explore this possibility, it is necessary to analyze the data in a multivariate approach. For this reason a GC-MS response pattern was built, considering the relative abundance of peaks appearing in the GC. Fourteen variables were selected, whose values are listed in Table 2. Only compounds showing the same MS spectra at the same elution time were included in the pattern. For the gas sensors, the net frequency shift between the steady signal conditions when the sensors are exposed to the sample and to reference air were considered as the sensor response. Table 3 lists the values (in Hz) recorded for each measured subject. These values were obtained as the average of three measurements of the same sample. Gas sensors were sensitive to the body odor, and in particular, the sensitivity seems to be enough to ascertain with some degree of accuracy the presence of the diseases. Figure 2 shows, as a box and whiskers plot, the statistics of the data of the seven sensors utilized in the experiment. Figure 2 shows, for each sensor, the statistical distribution for each of the three classes. In particular, the box has lines at the lower quartile, median, and upper quartile values. The whiskers are lines extending from each end of the box to show the extent of the rest of the data. Outliers are data with values beyond the ends of the whiskers. This box and whisker plot has been drawn in the Matlab programming environment using an embedded function. Except for sensor 6, all sensors are, in some degree, sensitive to the different chemical composition of the body odor of subject grouped in the three populations; however, single sensors are not enough to classify the samples correctly.

Med Sci Monit, 2005; 11(8): CR366-375

Di Natale C et al – The relationship of body odour and schizophrenia

Table 2. Relative abundance of the most recursive peaks found in GC-MS for each measured subject. 7.36*

8.51*

9.03

10.20*

11.41

12.87

12.92

13.00

13.68

14.09

14.22§

15.03

16.79

16.83

1

0.29

0.18

0.07

0.1

0

0

0

0

0

0

0.16

0

0

0

2

0

0

0

0.075

0

0

0

0

0

0

0

0

0

0

3

0.17

0

0

0.09

0

0.075

0

0

0

0

0.125

0

0

0

4

0.15

0

0

0

0

0.125

0

0

0

0

0

0.08

0.08

0

5

0

0

0

0

0

0.82

0

0

0

0

0

0

0

0

6

0

0

0

0.12

0

0.075

0

0

0

0

0.1

0.12

0

0

7

0.09

0

0

0.17

0.06

0

0

0

0

0

0.12

0

0

0

8

0.10

0.1

0

0

0

0

0

0.2

0.13

0

0

0

0

0

9

0.05

0.06

0

0

0.05

0

0

0.14

0

0

0

0

0

0

10

0

0

0

0.1

0

0.1

0

0

0

0

0.06

0.075

0.06

0

11

0

0

0

0.1

0

0

0

0

0

0

0.09

0

0

0

12

0.7

0

0

0.05

0

0

0

0

0

0

0.03

0

0.06

0

13

0

0.06

0

0

0

0.04

0

0

0

0

0

0.06

0

0

14

0

0.05

0

0

0

0

0

0

0

0

0

0

0

0

15

0

0

0

0.15

0

0

0

0

0

0

0.12

0

0

0

16

0.08

0

0

0.12

0.09

0

0

0

0

0

0.08

0

0

0

17

0.05

0

0

0.05

0

0.18

0

0

0

0.04

0

0

0

0.06

18

0

0.1

0

0

0.09

0.12

0

0

0

0

0

0

0

0

19

0

0

0

0.03

0

0

0.04

0

0

0

0.02

0

0

0

20

0

0

0

0.08

0

0.12

0

0.09

0

0

0.07

0

0

0

21

0

0

0

0.05

0

0

0

0

0

0

0.05

0

0

0

22

0.1

0

0

0

0

0

0

0

0

0.05

0

0

0

0

23

0

0.06

0

0

0

0

0

0

0

0

0

0

0

0

24

0

0

0

0.15

0

0.05

0

0

0

0

0.12

0

0

0

25

0

0

0

0.12

0

0

0

0

0

0

0.05

0

0

0

26

0

0.12

0

0

0.05

0.16

0

0

0

0

0

0

0

0

27

0

0.08

0

0

0.05

0.09

0

0

0.2

0

0

0

0

0

Each column is related to a different elution time (in minutes) of the GC measurement; the correspondence between elution time and compound has been verified by comparing the MS spectra, also if only in a few cases compound identification was achieved. Elution times marked with * correspond to compounds identified as products of lipid and protein fragmentation. The elution time marked with § is the hexenoic acid relative compound. These compounds are described in the text.

DISCUSSIONS GC spectra reveal that a compound with retention time of 14.1–14.2 minutes (peak D in Figure 1A) is effectively a relevant compound in the body odor of these subjects. Because previous studies reported trans-3-methyl-2-hexenoic acid as a compound with an abnormal concentration in patients affected by schizophrenia [2], we prepared a pure sample of this acid according to the literature [3], in order to have an internal standard. The superimposition of its retention time with that exhibited by the standard, the molecular peak, and the fragmentation pattern revealed by the mass spectrum of compound D strongly supported its identifica-

tion as trans-3-methyl-2-hexenoic acid. Nevertheless, such a compound cannot be considered an absolute marker for schizophrenia, since it is not always present in all patients, and moreover it is also present, although at lower concentrations, in the body odors of patients affected with other mental illnesses, and in references. Although it cannot be considered a reliable marker, there is a tendency for schizophrenic individuals to produce an abnormal amount of this compound, since its concentration is higher for schizophrenic than for other individuals. Beside the hexenoic acid parent compound (peak D in Figure 1), other compounds are found with important quan-

CR371

CR

Clinical Research

Med Sci Monit, 2005; 11(8): CR366-375

Table 3. Sensor responses for each measured subject. Values are given in Hz and result from an average of three measurements. Individuals

Sensor 1

Sensor 2

Sensor 3

Sensor 4

Sensor 5

Sensor 6

Sensor 7

1

136

198

129

351

143

281

116

2

172

155

100

378

73

237

137

3

133

157

116

561

90

143

111

4

94

191

47

358

150

239

111

5

140

184

134

403

122

271

69

6

142

245

124

394

122

246

115

7

149

170

102

586

59

275

148

8

155

257

100

447

117

273

134

9

178

233

88

343

131

184

135

10

114

92

128

300

91

111

116

11

138

101

145

529

80

234

125

12

130

140

117

545

104

192

105

13

113

153

110

521

78

181

99

14

194

160

129

290

108

168

135

15

122

189

123

373

118

134

107

16

88

135

92

374

151

187

127

17

183

135

182

415

115

139

91

18

140

153

148

344

94

155

116

19

56

99

144

212

103

64

95

20

109

110

55

249

87

93

62

21

128

137

24

356

75

81

83

22

63

122

120

446

83

30

75

23

94

39

82

296

60

114

78

24

86

124

42

261

87

71

73

25

55

128

76

329

75

126

38

26

126

122

42

246

82

34

77

27

90

146

134

410

49

47

108

tities. These compounds, with retention times from 7.3 to 10.2 minutes (peaks A, B, and C in Figure 1) were identified by analyzing their mass spectra as molecules obtained from the fragmentation of lipids and proteins, such as for example lactones. GC spectra demonstrate the complex composition of body odor, and suggest that although trans-3-methyl-2-hexenoic acid is sometimes present, it cannot be considered as an indicator of the disease. Rather, it is the whole composition of the body odor that may be correlated with the presence of schizophrenia. In order to investigate the global composition it is necessary to transform the GC spectra in patterns as discussed in the previous section. The pattern listed in Table 2 was then used as the input of a multivariate classifier.

CR372

Beside the compounds discussed above, none of the peaks composing the pattern was identified by the MS library. Although the identification of such compounds is of clear interest and could indicate which biochemical processes underlie the facts here revealed, this task would require a deeper analytical chemistry effort, which was considered to fall outside the scope of the present study. As multivariate classifier, PLS-DA was used. It should be borne in mind that this method is one of the simplest, and for this reason the results may underestimate the actual performance of the measurement systems here described. PLS-DA was cross-validated by the leave-one-out technique. Table 4 shows the confusion matrix of the PLS-DA calculated with the GC-MS data. The prediction error was minimized, considering eight latent variables.

Med Sci Monit, 2005; 11(8): CR366-375

Di Natale C et al – The relationship of body odour and schizophrenia

Table 4. Results of discriminant analysis applied to the GC-MS data. Misclassifications occur between schizophrenia and mental disorders classes. Schizophrenia

Mental disorder

Reference

Schizophrenia

6

1

0

Mental disorder

2

7

0

Reference

1

1

9

Percentage of correct identifications

66%

77%

100%

CR

Table 5. Results of the discriminant analysis of gas sensors array data. The totality of schizophrenia patients is correctly classified, while misclassifications are restricted to the classes of reference controls and non-schizophrenic mental disorders. Schizophrenia

Mental disorder

Reference

Schizophrenia

9

0

0

Mental disorder

0

6

1

Reference

0

3

8

Percentage of correct identifications

100%

66%

88%

It is interesting to note that none of the control samples were identified as pathological, while only two samples from patients, one per each class, were incorrectly classified as healthy. Furthermore, a good identification of those belonging to the other two classes was obtained. Gas sensor data were analyzed using a multivariate approach. In this regard, all sensor signals were utilized to build a model to predict whether a sample would belong to one of the three classes. Several methods are available to solve this problem; here, a simple linear discriminant analysis was adopted. There are two reasons to keep the data analysis simple. The first is the relatively scarce number of data. More complex data analysis indeed requires largest data sets to achieve reliable results. As a consequence, with respect to linear classifiers, the extension to general cases of the results obtained with non-linear methods (e.g. neural networks or fuzzy logic) is less reliable. For this reason PLS-DA was applied to the electronic nose data. Prediction error was minimized in a leave-one-out cross validation by three latent variables. The model provided more than 80% correct classifications. The results of classification are shown in Table 5. It is worth noting that, according to this classification, all the schizophrenic patients were correctly identified, and misclassification occurred between the other two classes. The results may also be visually displayed by plotting the score plot of the PLS. Figure 3 shows the score plots of the first versus the second latent variables (A) and the first versus the third latent variables (B). The best segregation of class data points is visible in Figure 3B. It should be recalled that the classification performance is then obtained by the use of three latent variables. Considering the score plots in Figure 3, no relationships were found between the therapy profiles and the distribution of

data points in Figure 3, and inside each class a negligible correlation with the DSM-IV score was also observed. Asymmetries in the box and whisker plots of electronic nose data (Figure 2) suggest that the data distribution is far from Gaussian. This indicates that the population of samples does not form straightforward classes. In some sense, this characteristic suggests the application of non-linear data analysis systems, such as neural networks, as the most suitable classification technique.

CONCLUSIONS This paper reports a combined GC-MS and electronic nose study of a population of patients, living in the same therapy unit. The study was performed in order to find out whether or not the body odor is a potential source of information on which to base a possible instrumental diagnosis and monitoring of schizophrenia disease. The results presented here show that the chemical composition of the odor emanating from the skin of individuals has a certain relation with the occurrence of mental disorders, and schizophrenia in particular. GC-MS analysis revealed that trans-3-methyl-2-hexenoic acid is not a reliable indicator, because it is not always present in the odor of schizophrenic subject, and sometimes it is found in the body odor of healthy patients. However, when it is present, its concentration is higher is schizophrenics than in other individuals. On the other hands, GC-MS spectra show a richer composition of the odor of schizophrenic patients. Gas sensors exhibited higher responses to schizophrenic patients in comparison to the others, confirming the trend evidenced by GC-MS. Both GC-MS and sensor array data were analyzed with a classification algorithm (PLS-DA) in order to ascertain the possibility to discriminate, from these measurements, patients affected by schizophrenia from either those affected

CR373

Clinical Research

Med Sci Monit, 2005; 11(8): CR366-375

18 0

sensor 1

sensor 2 200 Delta F [Hz]

Delta F [Hz]

16 0 14 0 12 0 10 0

100 1

2 class

3

1

sensor 2

120 100

2 class

3

sensor 4

500 Delta F [Hz]

Delta F [Hz]

140

400

300

80 1

2 class

3

1

Delta F [Hz]

120

2 class

300

sensor 5

140 Delta F [Hz]

150

3

sensor 6

200

100 100 80 1

2 class

3

1

2 class

3

sensor 7

Delta F [Hz]

140 120 100 80 1

2 class

3

Figure 2. Box and whisker plot of each gas sensor data related to the three population of measured samples. Clear indications about the sensitivity of sensors to the different chemical composition of the body odors are observed for most of the sensors. by different mental disorders or the population of healthy references. Identification performance was about 80% for GC-MS and the sensor array.

such a simple approach provides enough accuracy to demonstrate that the body odor contains some information that can be correlated with the presence of schizophrenia.

It is important to point out that the number of collected samples did not allow for a reliable use of more sophisticated non-linear data analysis approaches, such as neural networks. As a consequence of the simple analysis we adopted, the results here presented may be considered an underestimation of the real potentiality of the method. Nevertheless,

Findings indicate that body odor is a source of information that can be correlated with the presence of schizophrenia. Furthermore, for their relative simplicity of use, arrays of cross-selective gas sensors (electronic noses) can be considered as a viable technique to be pursued for the development of future diagnostic instruments. Finally, it is worth remark-

CR374

Med Sci Monit, 2005; 11(8): CR366-375

Di Natale C et al – The relationship of body odour and schizophrenia 5. Stein, Wise: Possible etiology of schizophrenia: Progressive damage to the noradrenergic reward system by 6-hydroxydopamine. Science, 1971, 171; 1032–36

Scores Plot

2.5

6. Smith K, Sines J: Demonstration of a peculiar odor in the sweat of schizophrenic patients. Arch Gen Psychiat, 1960; 212: 184

1

2 1.5

2

LV 2 (12.28%)

1 0.5

3 3

2 2 1 2 1 2 1

0 -0.5

1 3 2 3

1

-1.5 -5

-4

-3

10. Persaud K, Dodds G: Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature, 1982; 299: 352–55 2

11. Sicard G, Holley A: Receptor cell responses to odorants: similarities and differences among odorants. Brain Research, 1984; 292: 283–96

2

1

-2 -1 0 LV 1 (84.08%)

1

2

3

4

1

14. Turner APF, Magan N: ‘Electronic noses and disease diagnostic’, Nature review. Microbiology, 2004; 2: 161–66

1

0.6

1

15. Julak J, Stranska E, Prochazkova-Francisci E, Rosova V: Blood cultures evaluation by gas chromatography of volatile fatty acids. Med Sci Monit, 2000; 6(3): 605–10

1

0.4 LV 3 ( 1.48%)

2

0.2 2 2

1

0

1 11 2

1

-0.2

3

2

3

2

-0.6

18. Chendiok S, Crawley BA, Oppenheim BA et al: Screening for bacterial vaginosis: a novel application of artificial nose. J Clin Pathol, 1997; 50: 790–91

2

-5

-4

-3

-2

-1 0 LV 1 (84.08%)

16. Gardner JW, Craven M, Dow C, Hines EL: The prediction of bacteria type and culture growth phase by an electronic nose with a multi-layer perceptron network. Measurement Science and Technology, 1998; 9: 120–27 17. Holmberg M, Gustafsson F, Hörnsten EG et al: Bacteria Classification Based on Feature Extraction from Sensor Data. Biotechnology Techniques, 1998; 12: 319–24

3

2

-0.8 -6

3 33 3 3

3

2

-0.4

12. Jurs PC, Bakken GA, McClelland HE: Computational methods for the analysis of chemical sensor array data from volatile analytes. Chem Rev, 2000; 100: 2651 13. Pearce T, Schiffmann S, Troy-Nagle H, Gardner J, (eds): Handbook of olfaction machines VCH, 2002

Scores Plot

0.8

8. Smith M: The uses of smell in differential diagnosis. Lancet, 1982; 25: 1452 9. Zlatkis A, Brazell RS, Poole CF: The role of organic volatile profiles in clinical diagnosis. Clin Chem, 1981; 27: 789

3

1

-1

-2 -6

3 3 3 3

2

1

7. Phillips A, Erickson GA, Sabas M et al: Volatile organic compounds in the breath of patients with schizophrenia. J Clin Path, 1993; 49: 466

1

2

3

4

Figure 3. Score plots of PLS-DA calculated with the electronic nose data sets. Figure 3A shows latent variable 1 vs latent variable 2, and Figure 3B shows latent variable 1 vs latent variable 3. ing that the term odor is here used to indicate the machine perception of the global composition of the air surrounding the body, and has no direct relationship with the sensation of smell as perceived by human olfaction.

REFERENCES: 1. Sawa A, Snyder SH: Schizophrenia: diverse approaches to a complex disease. Science, 2002; 296: 692 2. Angst A: Psychoses in disulfiram (anatabus) treatment; review of literature and etiology. Schweiz Med Wochenschrift, 1956; 46; 1304–6 3. Smith K, Thomspon GF, Koster HD: Sweat in schizophrenic patients: identification of the odorous substance. Science, 1969; 166: 398 4. Gordon SG, Smith K, Rabinowitz JL, Vagelos PR: Studies of trans3-methyl-2-hexenoic acid in normal and schizophrenic humans. J Lipid Res, 1973; 14: 495

19. Pavolu AK, Magan N, Sharp D et al: An intelligent rapid odour recognition model in discrimination of Helicobacter pylori and other gastroesophageal isolates in vitro. Biosensors and Bioelectronics, 2000; 15: 333–42 20. Di Natale C, Macagnano A, Martinelli E et al: Lung cancer identification by the analysis of breath by means of an array of non-selective gas sensors. Biosensors and Bioelectronics, 2003; 18: 1209–18 21. Di Natale C, Macagnano A, Paolesse R et al: Electronic nose approach to human skin odour analysis. Sensors and Actuators B, 2000; 65: 216–19 22. Ballantine DS, White RM, Martin SJ et al: Acoustic wave sensors, Academic Press, 1997 23. Brunink J, Di Natale C, Bungaro F et al: The application of metalloporphyrins as coating material for QMB Based chemical sensor. Analytica Chimica Acta, 1996; 325: 53–64 24. Di Natale C, Paolesse R, Macagnano A et al: Sensitivity-selectivity balance in mass sensors: the case of metalloporphyrins. Journal of Material Chemistry, 2004; 14: 1281–87 25. Di Natale C, Paolesse R, Macagnano A et al: Characterization and design of porphyrins-based broad selectivity chemical sensors for electronic nose applications. Sensors and Actuators B, 1998; 52: 162–68 26. D’Amico A, Di Natale C, Macagnano A et al: Technology and tools for mimicking olfaction: status of the Rome Tor Vergata Electronic Nose. Biosensors and Bioelectronics, 1998; 13: 711–21 27. Massart DL, Vendeginste BG, Deming SN et al: Chemometrics: a textbook, in Data Handling in Science and Technology Vol. 2, Elsevier Science, 1988

CR375

CR