Prediction of the Chemiluminescent Behavior of Pharmaceuticals

Anal. Chem. 2001, 73, 4301-4306

Prediction of the Chemiluminescent Behavior of Pharmaceuticals and Pesticides L. Lahuerta Zamora,† Y. Fuster Mestre,† M. J. Duart,† G. M. Anto´n Fos,† R. Garcı´a Dome´nech,‡ J. Ga´lvez A Ä lvarez,‡ and J. Martı´nez Calatayud*,§

Departamento de Ciencias Quı´micas, Universidad Cardenal Herrera-CEU, 46113 - Moncada, Valencia, Spain, Unidad de Investigacio´ n de Disen˜o de Fa´ rmacos y Conectividad Molecular. Departamento de Quı´mica Fı´sica, Facultad de Farmacia, Universidad de Valencia, 46100, Burjassot, Valencia, Spain, and Departamento de Quı´mica Analı´tica, Facultad de Ciencias Quı´micas, Universidad de Valencia, Burjassot, Valencia, Spain

Analytical interest in liquid-phase chemiluminescence (CL) has increased considerably over the last two decades, and the analytical literature has contained a certain number of applications for the determination of both inorganic (metal ions, inorganic anions) and organic (biomolecules, drugs) compounds in a variety of industrial, clinical, biotechnology, and environmental matrixes.1

The advantages of CL analysis include low limits of detection and a wide linear range; both can generally be achieved with simple, robust, and relatively inexpensive commercial instrumentation. The analytical applications of liquid-phase chemiluminescence can be divided into those procedures that involve well-known CL reactions (luminol, acridinium esters, peroxyoxalates, dioxetanes, and tris(2,2’-bipyridyl)ruthenium(II) are responsible for the vast majority of analytical applications) and those that involve light emission from the reaction oxidant-analyte, which is the substrate in the CL reaction and corresponds to the determinations identified as “direct CL determinations”. In the first modality, the analyte interacts with the CL reaction, usually as a reagent, a catalyst, a quencher, and even an enhancer. There is a limited number of such types of systems and they have been broadly exploited. In the second and simpler approach, the previous experimental work is based on searching chemiluminescent reactions between analyte and a wide range of oxidants (sometimes reductants) in different media, which result in time-consuming trial-and-error screening procedures. There are empirical rules for predicting chemiluminescent behavior: for instance, when one compound or its oxidation product is fluorescent, there is a high probability the oxidation of the molecule resulted in CL emission mainly with strong oxidants, exothermic reaction (130-340 kJ/mol) and the reaction pathway is simple. Nevertheless, there are many exceptions to this principle, and very often, CL reactions cannot be predicted. Although a number of papers have already been published for pharmaceuticals,1-4 dealing with the couple FIA-direct CL, direct CL emission is a growing area in analytical chemistry. 3,4 Experimental efforts (screening procedures) are being made to find direct chemiluminescent behavior with different kinds of substances, mainly owing to the analytical advantages arising from the tandem, mainly low detection limits, wide linear dynamic ranges, speed of response, and reproducible mixing of sample and reagents near the detector.

* Corresponding author: (e-mail) [email protected]. † Universidad Cardenal Herrera-CEU (e-mail contact: [email protected]). ‡ Departamento de Quı´mica Fı´sica, Universidad de Valencia (e-mail contact: [email protected]). § Departamento de Quı´mica Analı´tica, Universidad de Valencia. (1) Worsfold, P. J.; Robards, K. Anal. Chim. Acta 1992, 266, 147.

(2) Martı´nez Calatayud, J. Flow Injection Analysis of Pharmaceuticals. Automation in the Laboratory, Taylor and Francis: Basingstoke, Hants., U.K., 1996. (3) Fuster Mestre, Y.; Lahuerta Zamora, L.; Martı´nez Calatayud, J. Luminescence 2000, 15, 1-23. (4) Worsfold, P. J.; Bowie, A. R.; Sanders, M. G. J. Biolumin. Chemilumin. 1996, 11, 61.

The present paper deals with the first attempt to apply molecular connectivity calculations to predict a chemical property with analytical usefulness: the chemiluminescent behavior of substances when reacted with common oxidants in a liquid phase. Preliminary evidence when searching for new direct CL methods consisted of the examination of analyte reaction with a wide range of oxidants and media. This task, which results in timeconsuming and trial-and-error expensive procedures, is necessary due to ensure empirical or theoretical rules for CL prediction are available. On the other hand, in quantitative structure-activity relationship studies, molecular connectivity is a topological method capable of describing the structure of a molecule by means of numbers named indices; subsequent regression in relation to the experimental values of the physical, chemical, or biological properties yields a series of functions called connectivity functions. Discriminant analysis was applied to 200 either chemiluminescent or nonchemiluminescent substances found either bibliographically or in an experimental screening. The method used for the selection of descriptors was a stepwise linear discriminant analysis from the Snedecor F-parameter. The classification criterion used was the minimum value of Mahalanobis. The quality of the discriminant function was calculated through the Wilks U-statistical parameter. Finally, the function was applied to a database including of more than 50 000 structurally heterogeneous compounds. The theoretical predictions were faced with the empirical evidence obtained through a continuous-flow manifold.

10.1021/ac010133i CCC: $20.00 Published on Web 07/27/2001

© 2001 American Chemical Society

Analytical Chemistry, Vol. 73, No. 17, September 1, 2001 4301

This paper is the first attempt to introduce connectivity studies into the practical work of analytical chemistry reactions; this intent has been focused on predicting the chemiluminescent behavior of a molecule when it reacts with common oxidants in liquid phase. For this purpose, the molecular connectivity indices of 200 substances were calculated. The behavior of the substances was found both in the literature and as a result of empirical screening in continuous-flow chemiluminescence. We faced the problem of finding molecules that did or did not result in chemiluminescent emission in liquid phase as a consequence of reaction with strong oxidants by searching in the analytical literature. It was relatively easy to find chemiluminescent (76)5-10 molecules, but it was a problem to find nonchemiluminescent molecules in reaction with a strong oxidant, because researchers working in the field of direct chemiluminescence never reported the molecules found as nonchemiluminescent (negative results). The search was restricted to some (8) found in a book5 and the main group (24) found in some papers6,8 and a personal communication.11 On the other hand, with the goal of studying a certain number of substances, we carried out an experimental screening of 92 compounds, most of them pharmaceuticals and a few pesticides. With the obtained results, the molecular connectivity studies were performed. A relatively simple formalism such as molecular connectivity is able to easily and quickly characterize molecular structures through the so-named topological indices or topological descriptors (TDs).12-14 Each molecule is assimilated to a graph, where atoms are represented by points, called vertexes, and bonds are represented by segments, called edges, between vertexes. Graphs can be analytically represented by matrices from which may be derived one single TD or a set of them. These indices, whether well chosen, are a unique characterization of the molecular structure.15 Furthermore, they can be correlated with many physical, chemical, and biological properties of molecules; we obtained the relationships (QSARs) and they can be even used to find new drugs,16,17 if we are able to discriminate the topological characteristics related to a given activity. By this means, new antivirals,18,19 cytostatics,20 hypoglycemics,21 β-blockers,22 analge(5) Schulman, S. G., Ed. Molecular Luminescence Spectroscopy; John Wiley & Sons: New York, 1993. (6) Nakagama, T.; Yamada, M.; Suzuki, S. Anal. Chim. Acta 1989, 217, 371376. (7) Schroeder, H. R.; Yeager, F. M. Anal. Chem. 1978, 50, 1114-1120. (8) Abbot, R. W.; Townshend, A. Analyst 1986, 111, 635-640. (9) Koukli, I.; Calokerinos, A. C. Analyst 1990, 115, 1553-1557. (10) Syropoulos, A. B.; Calokerinos, A. C. Anal. Chim. Acta 1991, 255, 403411. (11) Calokerinos. A. C. Personal communication. (12) Balaban, A. T. SAR QSAR Environ. Res. 1998, 8, 1-21. (13) Kier, L. B.; Hall, L. H. J. Chem. Inf. Comput. Sci. 1997, 37, 548-552. (14) Galvez, J.; Garcia-Domenech, R.; Salabert-Salvador, M. T.; Soler, R., J. Chem. Inf. Comput. Sci. 1994, 34, 520-525. (15) Kier, L. B.; Hall, L. H. Molecular connectivity in chemistry and drug research; Academic Press: London, 1976; pp 46-79. (16) Galvez, J.; Garcia, R.; de Julian-Ortiz, J. V.; Soler, R. J. Chem. Inf. Comput. Sci. 1995, 35, 272-284. (17) Weinstein, H.; Osman, R.; Green, J. P. In Computer-assisted drug design; Olson, E. C., Cristoffersen, R. E., Eds.; ACS Symposium, Ser. 112; American Chemical Society: Washington, DC, 1979; pp 161-170. (18) Mun ˜oz, C.; De Julian-Ortiz, J. V.; Gimeno, C.; Catala´n, V.; Ga´lvez, J. Rev. Esp. Quimioter. 1994, 7, 279-280.

4302 Analytical Chemistry, Vol. 73, No. 17, September 1, 2001

Figure 1. Continuous-flow manifold. Key: Q1, oxidant, 0.02 M; Q2, medium, 1 M; Q3, medium, 1 M; Q4, tested substance, 400 ppm (or deionized water blank). Flow rate, 2.8 mL min-1 each channel. Tested oxidants: potassium permanganate or cerium(IV), both of them merging with a sulfuric acid stream, and potassium hexacyanoferrate(III) or hydrogen peroxide, both of them merging with a sodium hydroxide solution. HV, high-voltage power supply; R, recorder; P, peristaltic pump: W, waste.

Figure 2. Pharmacological distribution diagram for chemiluminiscence activity from the connectivity function DF: upper, training group; lower, testing group; white line, nonchemiluminiscence; black line, chemiluminiscence.

sics,23,24 bronchodilators,25 antitoxoplasma,26 and antibacterials27 have been obtained and most of them can be considered as new lead drugs. (19) de Julian-Ortiz, J. V.; Ga´lvez, J.; Mun ˜oz-Collado, C.; Garcı´a, R.; GimenoCardona, C. J. Med. Chem. 1999, 42, 3308-3314. (20) Ga´lvez, J.; Garcı´a-Domenech, R.; Gomez-Lecho´n, M. J.; Castell, J. V. J. Mol. Struct. (THEOCHEM) 2000, 504, 241-248. (21) Anto´n-Fos, G. M.; Garcı´a-Domenech, R.; Perez-Gimenez, F.; Peris-Ribera, J. E.; Garcı´a-March, F. J.; Salabert-Salvador, M. T. Arzneim-Forsch./Drug Res. 1994, 44 (2), 821-826. (22) Garcı´a-Domenech, R.; De Gregorio-Alapont, C.; De Julia´n-Ortiz, J. V.; Ga´lvez, J.; Popa. L. Bioorg. Med. Chem. Lett. 1997, 7 (5), 567-572. (23) Ga´lvez, J.; Garcia, R.; De Julian-Ortiz, J. V.; Soler, R. J. Chem. Inf. Comput. Sci. 1994, 34, 1198-1203. (24) Garcı´a-Domenech, R.; Garcı´a-March, F. J.; Soler, R.; Galvez, J.; Anto´n-Fos, G. M.; De Julian-Ortiz, J. V. Quant. Struct.-Act. Relat. 1996, 15, 201-207. (25) Rios-Santamarina, I.; Galvez, J.; Garcı´a-Domenech, R.; Cortijo, J.; Santamaria, P.; Morcillo, E. Bioorg. Med. Chem. Lett. 1998, 8, 477-482.

Table 1. Results Obtained in the LDA Study and Classification of the Compounds from Pattern of Chemiluminiscence Activity Proposed: Training Active and Inactive Groups compound

DF

prob

class

compound

luminol diethylisoluminol isoluminol derivate isoluminol derivate acriflavine derivate acriflavine derivate acriflavine derivate DMAL TDE flufenamic acid epinephrine skatole amoxicillin dihydromorphine ninhydrin sulfadiazine sulfamethoxypyridazine naloxone tetracycline tryptamine Resorcinol Phloroglucinol 3-Indolyacetic acid acridine derivate 2 adrenaline

1.787 1.814 0.851 1.376 0.775 0.502 2.677 1.279 0.728 0.117 2.490 -0.442 0.158 -0.825 -3.847 2.098 0.321 1.438 2.915 1.246 0.184 0.090 0.687 -0.385 2.490

0.857 0.860 0.701 0.798 0.685 0.623 0.936 0.782 0.674 0.529 0.923 0.391 0.540 0.305 0.021 0.891 0.580 0.808 0.949 0.777 0.546 0.522 0.665 0.405 0.923

+ + + + + + + + + + + + + + + + + + + + +

glutamic acid menthol xanthine carnitine inositole neomicine biguanide vitamin K3 sorbitol saccharin proline pyrazinamide nicotinamide moroxydine methylnicotinate metformin lysine phenylalanine dextromethorphan chloramphenicol caffeine alprazolam nitrazepam

-3.611 1.133 -4.207 1.468 -4.162 -0.859 -1.293 -1.778 -0.996 -0.475 -0.380 -1.102 -2.048 -0.538 -2.416 -6.111 -1.385 -2.515 -0.108 -0.588 -4.733 -3.561 -3.412

0.955 0.244 0.985 0.187 0.985 0.703 0.785 0.855 0.730 0.617 0.530 0.751 0.886 0.631 0.918 0.998 0.800 0.925 0.527 0.643 0.991 0.972 0.968

Inactive Group tetradifon + methomyl lindane + phenylpropanolamine nortriptyline nicotinic acid levamisole phenobarbital cystine cyclohexylamine atropine ampicillin oxalic acid muconic acid pyruvic acid tartaric acid vitamin B6 vitamin K3 ciprofloxacine xanthine uric acid triazolam lormetazepam

Active Group codeine oxytetracycline doxycicline syringaldazine lucigenin lophine normorphine morphine ondansetron TCPO DNPO 2,3-dimethylindole tryptophan L-Dopa trimetoprim antipyrine benzophthalazine-1,4-dione eskacine chlorpromazine thioridazine hidroquine catechol dopamine probucol normetanephedrine

EXPERIMENTAL SECTION Reagents. All solutions were prepared from analytical reagent grade materials using reverse osmosis and deionized water. Apparatus and Procedure. The continuous-flow manifold for the screening the chemiluminescent substances is depicted in Figure 1. It consisted of a peristaltic pump (Gilson Minipuls 2, Middleton, WI) which pumped oxidant (flow rate Q1, 2.8 mL min-1, 0.02 mol L-1), medium (flow rated Q2 and Q3, both 2.8 mL min-1, 1 mol L-1), and tested substance (flow rate Q4, 2.8 mL min-1, 400 ppm) solutions through PTFE tubes (0.8-mm i.d.). All solutions merged finally in a T-shaped piece positioned 2 cm (26) Gozalbes, R.; Galvez, J.; Garcı´a-Domenech, R.; Derouin, F. SAR QSAR Environ. Res. 1999, 10, 47-60. (27) Gregorio-Alapont, C.; De Garcı´a-Domenech, R.; Galvez, J.; Ros, M. J.; Wolski, S.; Garcı´a, M. D. Bioorg. Med. Chem. Lett. 2000, 10, 2033-2036.

DF

prob

class

4.429 3.453 3.453 6.703 2.333 1.579 2.245 3.709 0.866 0.283 0.602 -3.185 -0.546 -0.627 3.455 1.995 1.899 2.308 2.061 1.819 1.742 0.347 1.556 5.806 1.661

0.988 0.969 0.969 0.999 0.912 0.829 0.990 0.976 0.704 0.570 0.646 0.040 0.367 0.348 0.969 0.880 0.870 0.910 0.887 0.860 0.851 0.586 0.826 0.997 0.840

+ + + + + + + + + + + + + + + + + + + + + +

1.152 -3.920 -3.001 -0.315 1.223 -2.209 -8.765 -1.961 -4.292 -2.248 -4.290 -0.247 -3.735 5.973 -7.159 -2.620 0.121 -1.778 4.575 -4.207 -4.278 -2.828 -1.892

0.240 0.981 0.953 0.578 0.227 0.901 1.000 0.877 0.987 0.904 0.986 0.561 0.977 0.003 0.999 0.932 0.470 0.855 0.010 0.985 0.986 0.944 0.869

+ + + + + -

before entering the flow cell. The cell consists on a flat spiralcoiled quartz tube (1.0-mm i.d., 3-cm total diameter of the flow cell, without gaps between loops). The flow cell was placed 2 mm from the photomultiplier tube (end window, type 9902, from Electron Tubes. Co., Middlesex, U.K.) and backed by a mirror for maximum light collection. The T-piece, flow cell, and photomultiplier were placed in a homemade, absolutely light-tight box. The photomultiplier was operated at -1273 V supplied by the PHV-40 programmable photomultiplier high-voltage power supply, from Acton Research Corp. (Acton, MA). The output was fed to a computer provided with a suitable board and software developed ad hoc for this system. Linear Discriminant Analysis. The objective of the linear discriminant analysis (LDA), which is considered as an heuristic Analytical Chemistry, Vol. 73, No. 17, September 1, 2001

4303

Table 2. Results Obtained in the LDA Study and Classification of the Compounds from Pattern of Chemiluminiscence Activity Proposed: Test Active and Inactive Groups compound

DF

prob

pyrogallol noradrenaline sulfamerazine captopril metoclopramide isoluminol tyrosine proflavine isoniazid 6-monoacetylmorphine nalorphine buprenorphine chlortetracycline gallic acid indole 3-methylindole isoluminol derivate (17) isoluminol derivate (19) isoluminol derivate (20) mefenamic acid indomethacine hydrochlorothiazide folic acid phenoxymethyl penicillinic acid fluphenazine emetine

0.373 1.314 2.975 -1.995 2.817 1.410 -1.350 1.883 -0.498 4.605 6.178 0.412 3.229 0.458 3.912 -0.442 1.735 1.919 1.718 0.164 0.634 -0.299 2.159 -0.017 2.522 5.093

0.592 0.788 0.951 0.120 0.944 0.804 0.206 0.868 0.378 0.990 0.998 0.602 0.962 0.613 0.980 0.391 0.850 0.872 0.848 0.541 0.653 0.426 0.896 0.496 0.926 0.994

aspartic acid arginine alanine uracil thymine lorazepam temazepam flunitrazepam isoluminol derivate (2) isoluminol derivate (8) tolazamide estearate chloramine T riboflavine

-2.912 -1.659 -7.557 -3.814 -5.170 -2.668 -2.641 -2.419 2.224 2.193 -4.833 -0.191 -0.483 0.452

0.948 0.840 0.999 0.978 0.994 0.935 0.933 0.918 0.098 0.100 0.992 0.548 0.619 0.389

class Active Group + + + + + + + + + + + + + + + + + + + + +

compound

DF

prob

class

chlorpheniramine bromhexine acetylcysteine amitriptyline isoluminol derivate (18) isoluminol derivate (26) oxytetracycline pipemidic acid guaifenesin acridine derivate 5 acridine derivate 1 histamine benzydamine isoxicam hydrocortisone propyphenazone penicillin G tetracaine sulfanilamide prometazine piramidon? penicillamine novocaine loprazolam tartrazine sulfamethoxazole

1.591 1.573 -0.199 1.169 1.530 2.879 3.453 1.459 -0.027 2.358 2.231 2.289 0.008 2.250 2.551 1.569 0.317 2.570 0.173 2.967 3.185 1.040 2.352 1.562 -1.569 0.151

0.831 0.828 0.450 0.763 0.822 0.947 0.969 0.811 0.493 0.914 0.903 0.908 0.502 0.905 0.928 0.828 0.579 0.929 0.543 0.951 0.960 0.739 0.913 0.827 0.172 0.538

+ + + + + + + + + + + + + + + + + + + + + + +

-0.196 -4.671 -1.514 -1.350 5.622 -0.848 -5.028 0.292 -1.302 -1.919 -3.055 -0.931 1.407 1.755

0.549 0.991 0.820 0.794 0.996 0.700 0.993 0.428 0.786 0.872 0.955 0.717 0.197 0.147

+ + +

Inactive Group urea p-benzoquinone glucose tyrosine DMCTC vitamin B1 veronal chlorhexidine + sucrose + treonine asparagine valine isoluminol derivate (7) + isoluminol derivate (5)

algorithm able to distinguish between two or more cathegories of objects, is to find a linear function able to discriminate between the active and nonactive compounds. Two large sets of compounds, the first one with a proven activity (in the present paper, chemiluminiscence) and the second one composed of inactive compounds, are considered for analysis. The discriminant ability is tested by the percentage of correct classifications into each group. LDA was performed by using the BMDP 7M package.28 The selection of the descriptors was based on the Snedecor Fparameter, and the classification criterion was the shortest Mahalanobis distance (distance from any individual value to the global average appearing in the regression equation). 7M chose the variables used in computing the linear classification functions in a stepwise manner: at each step, the variable that adds the most to the separation of the groups is entered into (or the variable that adds the least is removed from) the discriminant function. The quality of the discriminant function is evaluated by the Wilk’s

λ or U-statistic parameters, the later being a multivariate analysis of variance statistic which tests the equality of group means for the variable(s) in the discriminant function. Pharmacological Activity Distribution Diagrams. In a recent paper,29 we realized that connectivity functions may be used not only for the prediction of properties or classification of the activity, but they can also be used to evaluate the expentancy of activity for a given compound in a chosen pharmacological (or chemical) activity. This was achieved through the pharmacological distribution diagrams (PDDs), which allowed prediction of the probability of activity for a given molecule. These diagrams have demonstrated their usefulness in revealing the pharmacological activity of several types of drugs, such as bronchodilators, antibacterials, and antimalarials. They consist of applying the connectivity function with the property P to the group of active compounds and to a representative group of nonactive molecules. The structures are grouped into the predicted value of the property P intervals; the frequency

(28) Dixon, W. J. BMDP Statistical software, University of California, Berkeley, 1990.

(29) Galvez, J.; Garcı´a-Domenech, R.; De Gregorio-Alapont, C.; Julia´n -Ortiz, J. V.; De Popa, L. J. Mol. Graph. 1996, 14, 272-276.

4304 Analytical Chemistry, Vol. 73, No. 17, September 1, 2001

Table 3. Theoretical Classification and Experimental Results for Tested Compounds compound

DF

prob

theor class

exptl result

oxidant/medium for direct CL

salicylamide maltol captan iproniazid phenylephrine ergonovine acriflavine permethrin cypermethrin phenvalerate quinalphos lidocaine propifenazone minoxidil pilocarpine bifentrhin bromohexine hydrochloride picric acid 3-tert-butyl-4-hydroxyanisole spironolactone ampicillin trihydrate triamterene cimetidine cetrimide chlorpheniramine maleate sodium glycocholate hydrate griseofulvin diphenylhydantoin hydroxyprogesterone nicardipine nifedipine probenecid piroxicam clotrimazole metoclopramide chlorobenzorex oximetazoline sulisobenzone hydroxicin tiazolin dicarboxilic acid cloxacillin

0.847 0.665 1.139 0.698 1.673 3.498 1.795 3.512 4.156 2.678 0.163 1.710 1.569 1.697 1.290 3.857 1.570 1.735 4.742 4.397 1.748 4.990 3.555 1.904 1.570 1.941 2.995 7.360 2.166 7.398 3.483 6.180 2.181 2.688 2.814 1.785 3.267 5.210 4.491 3.524 2.415

0.700 0.660 0.757 0.668 0.842 0.971 0.858 0.965 0.985 0.936 0.541 0.606 0.828 0.845 0.784 0.979 0.828 0.850 0.991 0.986 0.852 0.992 0.976 0.868 0.828 0.868 0.956 1.000 0.900 1.000 0.969 0.998 0.896 0.923 0.932 0.857 0.966 0.996 0.988 0.975 0.920

+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +

+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +

KMnO4/H2SO4 KMnO4/H2SO4 KMnO4/H2SO4 Ce(IV)/H2SO4 KMnO4/H2SO4 Fe(CN)63-/NaOH KMnO4/H2SO4 Fe(CN)63-/NaOH Fe(CN)63-/NaOH Fe(CN)63-/NaOH KMnO4/H2SO4 none KMnO4/H2SO4 KMnO4/H2SO4 none none KMnO4/H2SO4 H2O2/NaOH KMnO4/H2SO4 KMnO4/H2SO4 Fe(CN)63-/NaOH Ce (IV)/H2SO4 H2O2/NaOH Fe(CN)63-/NaOH KMnO4/H2SO4 Fe(CN)63-/NaOH H2O2/NaOH Fe(CN)63-/NaOH Fe(CN)63-/NaOH H2O2/NaOH KMnO4/H2SO4 Fe(CN)63-/NaOH Fe(CN)63-/NaOH H2O2/NaOH KMnO4/H2SO4 Fe(CN)63-/NaOH Fe(CN)63-/NaOH KMnO4/H2SO4 Fe(CN)63-/NaOH KMnO4/H2SO4 Ce(IV)/H2SO4

of its appearances along each interval of P is determined, and thus, the expectancy E of finding a molecule with a desired value of P is obtained. For each arbitrary interval of whatever function, we can define an activity expectancy as Ea ) a/(i + 1), where a is the ratio between the number of active compounds in this interval and the number of total active compounds In the same way, i represents the ratio of inactive compounds, and we can also determine the expectancy of inactivity as Ei ) i/(a + 1). In this paper, Ea means the expectancy of the chemiluminiscence activity and Ei the expectancy of the nonchemiluminiscence activity. For a connectivity function, Ea acquires the form of a function of distribution and Ei tends to 0 over the curve; the overlapping is minimum, so this function can be useful for selection and molecular design. This permits us to determine the intervals of the function where the probability of finding new active compounds is maximum.

substance was considered nonchemiluminescent when the signal’s difference (sample minus blank) was minor to twice the average background, with any tested oxidant and medium. Molecular Connectivity Calculations. Calculation of the Topological Descriptors. In this work, we have used charge indices Ji,14 as well as Kier and Hall connectivity indices up the tenth order, mχt,15 and electrotopological indices.13 Linear Discriminant Analysis. In this work, a set of 200 compounds forming a structurally heterogeneous sample with either chemiluminiscence or nonchemiluminiscence activity has been analyzed. Each group was separated in two groups, the training and the test, respectively. Based on this, the obtained discriminant functions were validated. The discriminant function chosen was

DF ) -87.987χCH - 276.127χVCH + 1.224S(sCHd) 35.38J3v + 66.81J4v - 0.20 (1)

RESULTS AND DISCUSSION Experimental Screening in a Continuous-Flow Assembly. The continuous-flow manifold (Figure 1) was used for performing the experimental screening. The chemiluminescence was monitored (sample minus blank) with the homemade luminometer. A

N ) 96; F ) 19.5; U-statistics (Wilks λ) ) 0.427 Although it is not our aim here to reach a complete interpretation of the chemiluminescent phenomena just from topological Analytical Chemistry, Vol. 73, No. 17, September 1, 2001

4305

approaches, and more work must be developed in the future, it is clear that the χi connectivity indices are taking into account pure structural features, while the Ji charge indices evaluate the distributions of intramolecular charge. Finally, the Si atomic sumelectrotopological index encodes electronic information related to the sCHd group. As expected, chemiluminescence is related to both, pure structural as well as electronic features. Tables 1 and 2 summarize the classification of the results obtained with DF discriminant functions for each group. A compound will be selected as chemiluminiscence if DFi > 0 or as nonchemiluminiscence if DFi < 0. As outlined in the tables, in both training and test groups, overall accuracy is higher than 90%. Moreover, the most cases, we work within a success probability higher than 80% (see column 3 in the Tables 1 and 2). When we applied the DF function to the previous compounds, a pharmacological distribution diagram (PDD) could be constructed representing the expectancy for each classification group in any range of DF. The graphs depict only a very small overlapping region, which is indicative of the discriminant power of DF function. Despite using a big group of molecules, the profiles of PDD for both training and test groups were very similar. The discriminant ability obtained from DF is corroborated when DF was applied the benzodiazepine group. In fact, it is known11 that the benzodiazepines, with the exception of loprazolam, are nonchemiluminescent compounds. The DF function classifies them correctly (see table). Success of Theoretical Predictions. Then, the DF function was applied to structures that could be found in a personal databank with more than 50 000 heterogeneous structures. We selected as potential chemiluminescent compounds those which, tested through the discriminant function, yielded DF > 0. From all of them we selected 11 that were easily available and 30 by means of a random procedure. The former predictions were experimentally tested by means of the continuous-flow manifold and conditions depicted in Figure

1. The reactions were monitored with the tested substance in different solvents (Q1; substance solution), in different media (Q2 and Q3; H2SO4 or NaOH), and with different oxidants (cerium(IV), IO4-, S2O82-/Ag+, H2O2 and ClO4-, MnO4-, and Fe(CN)64-) (Q4). A substance was considered chemiluminescent when the signals difference (sample minus blank) was more than 5-fold the background. Results are shown in Table 3.

(30) Fuster Mestre, Y.; Ferna´ndez-Band, B.; Lahuerta Zamora, L.; Martı´nez Calatayud, J. Analyst 1999, 124, 413-416. (31) Fuster Mestre, Y.; Lahuerta Zamora, L.; Martı´nez Calatayud, J. Anal. Chim. Acta 1999, 394, 159-163.

Received for review January 30, 2001. Accepted June 7, 2001.

4306

Analytical Chemistry, Vol. 73, No. 17, September 1, 2001

CONCLUSIONS From the 41 compounds whose chemiluminescent behavior was predicted, 38 were indeed found experimentally as chemiluminescent. It represent a success of 92.7% in the prediction. Moreover, the present results demonstrate an adequate choice of topological descriptors. It is possible to predict the chemiluminescent activity of a compound and, therefore, to corroborate its applicability in the search for new chemiluminescent compounds. In this work, the usefulness of molecular topology in the search for new chemiluminescent compounds is clearly demonstrated. The interest of these results is even higher considering that all topological descriptors used in this work are easily obtained through a simple algorithm from the topological matrix, beginning with the obtained results that are clearly positive. Using multilinear regression and the LDA, a pattern of topological similarity of chemiluminiscence activity has been obtained. Finally, it can be concluded that molecular topology is a very interesting formalism for the prediction of chemiluminiscence property. An additional advantage is that a high-throughput screening into large databases can be performed with a high efficiency. To date, we have already published some papers dealing with FIA-direct CL determinations of some of the chemiluminescent well predicted compounds.30,31

AC010133I