JOURNAL OF CLINICAL MICROBIOLOGY, Aug. 1984, p. 209-213 0095-11371841080209-05$02.00/0 Copyright 1984, American Society for Microbiology Vol. 20, No. 2 Specific Enzyme-Linked Immunosorbent Assay for Detection of Bovine Antibody to Brucella abortus L. B. TABATABAI* AND B. L. DEYOE National Animal Disease Center, Agricultural Research Service, U.S. Department of Agriculture, Ames, Iowa 50010 Received 9 January 1984/Accepted 11 May 1984 Six soluble antigens prepared from BruceUa abortus were compared with a salt-extractable protein (CSP) antigen in an enzyme-linked immunosorbent assay for the detection of antibody to B. abortus in cattle sera. Of seven preparations tested, antigens from B. abortus soluble antigen (prepared from an autoclaved cell suspension) and CSP produced typical sigmoidal saturation titration kinetics. Only dialyzed B. abortus soluble antigen and CSP were stable on frozen storage. Enzyme-linked immunosorbent assay with CSP antigen under optimal conditions was from 100- to 700-fold more sensitive than the standard agglutination, card, Rivanol precipitation-plate agglutination, and the complement fixation tests in detecting immunoglobulin G antibody. From a practical point of view, however, using the most stringent criteria for determining an "upper negative" value, the enzyme-linked immunosorbent assay with CSP was at least 12-fold more sensitive than the standard agglutination test and any of the other serological tests. Furthermore, the enzyme-linked immunosorbent assay with CSP was specific for antibody to B. abortus. Application of the enzyme-linked immunosorbent assay (ELISA) for the detection of bovine serum antibody to Brucella abortus has been reported. Smooth lipopolysaccharide (LPS) (4, 9), whole sonicated B. abortus antigen (13, 14), or B. abortus soluble antigen (BASA; National Veterinary Services Laboratories, U.S. Department of Agriculture, Ames, Iowa [7]) served as the solid-phase antigen in the above-mentioned ELISA methods. All of the antigens employed, however, contained LPS, which, according to Lamb et al. (9), may result in nonspecific binding of immunoglobulin M (IgM) to LPS. Consequently, an assay system based on LPS or LPS-rich antigens may be difficult to interpret. To overcome some of the problems associated with LPS, we used an ELISA based on a salt-extractable, partially purified protein antigen (CSP) (15). In our experiments we demonstrated that this solid-phase antigen detects specific antibody and that the test is sensitive compared with conventional serological tests currently in use for detection of bovine brucellosis. MATERIALS AND METHODS Antigens. Salt-extractable protein antigens were prepared from aqueous methanol-inactivated, washed B. abortus 19 as previously described (15). Briefly, washed cells (0.2 g of 1 M NaCI-0.1 M Na-citrate per ml) were shaken with an equal volume of washed 0.1-mm glass beads in a Mickle tissue disintegrator (Brinkmann Instruments, Westbury, N.Y.) at 5 C for 1 h. After settling of the beads, the cell suspension was centrifuged at 20,000 x g, and the supernatant was freeze-dried and dialyzed against 5 mm NH4HCO3. The CSP was concentrated by precipitation with ammonium sulfate to 70% saturation, and the dialyzed protein was stored at 1 mg/ ml at -70 C. BASA was lot 14 from the National Veterinary Services Laboratories. BASA preparations have been characterized by Berman et al. (2). BASA was dialyzed extensively against 5 mm NH4HCO3 and clarified at 20,000 x g (a copious precipitate resulted), the supernatant was freeze-dried, and the residue was reconstituted to 1 mg of protein per ml. Both * Corresponding author. 209 nondialyzed BASA and dialyzed BASA (BASA-d) were used as antigens in the ELISA described below. Brucella endotoxin complex (LPS fraction 5) was prepared from B. abortus 19 (M. Redfearn, Ph.D. thesis, University of Wisconsin, Madison, Wis., 1960). The cell-free supernatant of B. abortus plate antigen (preserved in 0.5% phenol; National Veterinary Services Laboratories) was dialyzed extensively against 5 mm NH4HCO3 to remove the dye, clarified at 20,000 x g, and concentrated by freeze-drying, and the residue was reconstituted to 1 mg of protein per ml. TX-3 was prepared in a manner similar to BASA, except that B. abortus 19 was used, and cells were washed extensively before being autoclaved. The supernatant was dialyzed for 24 h against saline and centrifuged at 20,000 x g. Sera. A positive control serum was collected from a nonvaccinated, culture-positive cow at 7 months post-challenge (4 months post-abortion). The cow had been experimentally infected with B. abortus 2308. A negative control serum was a composite from 73 pooled sera from a brucellosis-free nonvaccinated herd maintained at the animal facilities at the National Animal Disease Center, Ames, Iowa, unless otherwise stated. Sera from cattle infected or vaccinated with various disease agents were donated for interference testing in the ELISA for brucellosis by the following investigators from the National Animal Disease Center: J. H. Bryner (Campylobacter jejuni), H. C. Ellinghausen (Leptospira pomona), G. W. Pugh, Jr. (Moraxella bovis), R. S. Merkal (Mycobacterium avium and Mycobacterium paratuberculosis), G. H. Frank (Pasteurella haemolytica type 1), and Salmonella cholera-suis was from T. T. Kramer, Iowa State University, Ames. Sera from cattle vaccinated and challenged with B. abortus were from experiments conducted at the National Animal Disease Center. Serology. Conventional serological methods used were the card test, the standard agglutination test (SAT), the Rivanol precipitation-plate agglutination test, and the microtiter complement fixation test (1). The ELISA used was essentially that of Voller and Bidwell (16), except that we employed a horseradish peroxi-
210 TABATABAI AND DEYOE dase conjugate. Solid-phase antigens were diluted to 0.1 mg of protein per ml in 5 mm NH4HCO3 and stored frozen in 10- samples). Microtiter plates (Immulon 1; Dynatech Labo- RI ratories, Alexandria, Va.) were coated with 100 Rl of 0.5 M NaCO3, ph 9.6, containing antigens at 10 ng of protein or various designated concentrations and stored overnight at 5 C. Plates were washed five times for 5 min with 200,ul of saline-0.05% Tween 80 (Difco Laboratories, Detroit, Mich.) and blotted between washes. Test and control sera were diluted 1:100 or serially, as designated, in 50 mm phosphatebuffered saline (ph 7.2)-0.05% Tween 80 (PBS-Tween), dispensed in plates in 100-,u quantities and incubated for 2 h at room temperature. The washing procedure was repeated and horseradish peroxidase-conjugated goat anti-bovine IgG (National Veterinary Services Laboratories; heavy- and light-chain specificity not determined) was diluted 1:1,000 in PBS-Tween. Plates were incubated for 2 h at room temperature or overnight at 5 C. The washing procedure was repeated, and the enzyme reaction was initiated by adding 100,ul of substrate solution containing 5 mm H202 (0.015%) and 0.2 mm ABTS [2,2'-azino-di-(3-ethyl-benzthiazoline-6-sulfonic acid)] (Sigma Chemical Co., St. Louis, Mo.) in 50 mm sodium citrate (ph 4.0). After exactly 10 min of incubation at room temperature, the reaction was stopped by the addition of 100 RI of 0.1 M hydrofluoric acid (the enzyme reaction was linear with time for 15 min). The absorbance ratio at 410 nm/450 nm (to minimize plate to plate differences) was recorded with a model 600 plate reader (Dynatech Laboratories). Chemical assays. Protein concentrations of antigen solutions (Table 1) were determined with the Folin phenol reagent (10), with bovine serum albumin (Armour Pharmaceutical Co., Phoenix, Ariz.) as a standard. Total carbohydrate was determined with the phenol-sulfuric acid method (6) with a-d-glucose as a standard and 2-keto-3-deoxy sugar acid measured as described by Warren (17), except that a 10- min preliminary hydrolysis step with 0.25 N H2SO4 at 100 C was included. 2-Keto-3-deoxyoctulosonic acid and 2-deoxyribose (Sigma Chemical Co.) were used as standards in the 2- keto-3-deoxy sugar acid assay. RESULTS Of the seven antigen preparations used initially (see Table 1 for composition) with the positive and negative bovine sera, only three (CSP, BASA, and BASA-d) showed typical saturation kinetics in plots of absorbance-versus-protein concentration (Fig. 1). Even though endotoxin fraction 5 appeared to be a sensitive reagent, i.e., gave positive ELISA values at 3 pg of protein per well, this antigen did not give typical saturation kinetics under the conditions used. In TABLE 1. Composition of B. abortus antigens evaluated for the ELISA methoda Antigen Carbohydrate KDob (mg/ml) (,ug/ml) BASA 0.81 17.1 BASA-d 1.19 16.3 CSP 0.05 0.05 LPS fraction 5 0.12 8.68 Plate antigen-dc 0.10 3.82 TX-3 1.04 15.9 TX-3-dd 0.57 14.6 a For solutions containing 1 mg of protein per ml. bkdo, 2-Keto-3-deoxyoctolusonic acid. c Plate antigen, extensively dialyzed. d TX-3-d, extensively dialyzed TX-3. 1.0 0.8 a 0.6 0.4 J. CLIN. MICROBIOL. 0.2i, 0 ~ T- I I lo-, lo-, 02 0lo- I 10 lo, lo ng PROTEIN/WELL FIG. 1. Titration kinetics of various antigens employed in ELISA for detecting antibody to B. abortus. The following antigens were employed at the designated protein concentrations: CSP (0), BASA-d (A), dialyzed TX-3 (A), dialyzed plate antigen (*), BASA (not dialyzed) after one freeze-thaw cycle (A), BASA after repeated freeze-thaw cycles (V), TX-3 (not dialyzed) (O), LPS fraction 5 (0). A solid line indicates ELISA absorbance values with a known positive control serum (3,200 IU of antibody per ml, diluted 1:100); a dashed line (Oi---O) indicates average ELISA absorbance values with a negative serum (SAT negative, at a 1:100 dilution) and with CSP as the antigen. general those antigens containing relatively high amounts of LPS (endotoxin fraction 5, plate antigen, TX-3, and dialyzed TX-3) produced atypical titration kinetics, i.e., shallow, almost concentration-independent curves with high background values, with the negative serum. BASA was ca. 10- fold less sensitive than CSP and IBASA-d (Fig. 1). Furthermore, after repeated freeze-thaw cycles BASA produced atypical titration curves similar to that for endotoxin fractions. Therefore, the antigens considered for further evaluation with cattle sera were CSP and BASA-d antigens. The optimal protein concentration for CSP and BASA-d was 10 ng of protein per well. This protein concentration was chosen for conducting the ELISA testing, as it was near the upper end of the linear portion of the titration curve (Fig. 1). Mainly because production, reproducibility in the ELISA of various preparations, and characterization of CSP by polyacrylamide electrophoresis and isoelectrofocusing could be readily accomplished (unpublished results), this antigen was exclusively used in testing for brucellosis with the ELISA method. Optimal dilution of serum for routine testing was chosen from a titration conducted at constant antigen and 10-fold serial dilutions of positive and negative sera. The negative serum employed in the three-dimensional plot shown in Fig. 2 was negative by conventional serological techniques but had a low titer with the ELISA method. Subsequently, all negative sera were first screened with the ELISA before being used as negative controls. The optimal working dilution of serum giving the greatest difference between positive and negative control values at a designated antigen concentration was found to be a 1:100 dilution of the sera in PBS- Tween. Furthermore, consideration was also given to ease of pipetting (20-I,l quantities) and volume of diluent (2.0 ml per dilution used) without sacrificing accuracy and precision of pipetting. Comparison of ELISA to conventional serological tests. Results of the titration of a positive control serum serially diluted in normal bovine serum are shown in Table 2. For comparison, each dilution was considered a separate sample
VOL. 20, 1984 2 er -:I LLJ w Cr m CD) cn -: 0.7 0.6 0.5-04 0. 3-0.2-0.1 0 FIG. 2. Three-dimensional plot of absorbance at 410 nm versus CSP concentration and serum dilution of a known positive serum (upper surface) and negative serum (stipled lower surface). so that a known amount of antibody was present in each tube. Serum containing 32 IU of agglutinating antibody per ml (tube 5) was positive by one conventional test (SAT) at a dilution of 1:25; thus, a little over 1 IU of antibody was detectable by this method. In contrast, the ELISA method was still positive at 0.01 IU of antibody per ml (1:100 dilution of tube 10) in this experiment, i.e., when "positive" was defined as an absorbance value equal to or greater than twice the absorbance value obtained with the "negative" serum. Interestingly, when an identical titration was performed with serum diluted in PBS-Tween, the titration curve was shifted toward the right (Fig. 3), and the apparent sensitivity was increased by an additional sevenfold. Hence, in our system the lower limit of detectable antibody to B. abortus of the bovine IgG class by the ELISA method is 0.0014 IU/ml. Thus, the ELISA method is ca. 100- to 700-fold more sensitive than the SAT method and the other serological tests in detecting anti-brucella IgG. Specificity of the ELISA. The ELISA was specific for anti- B. abortus antibody, and no cross-reactions were detected (Table 3). Although the present test could not distinguish ELISA FOR BOVINE ANTIBODY TO B. ABORTUS 211 between vaccinated and infected (with the virulent field strain 2308) animals, no nonspecific reactions were observed (Table 3). However, low levels of antibody were sometimes observed among known brucellosis-free, nonvaccinated catle (Table 4). Among 103 sera tested, 9 (8.7%) were considered positive if the lower limit for positive was set at twice the mean value of the ELISA absorbance results for 103 negative serum samples obtained from a brucellosis-free herd (i.e., ELISA values of.0.088 absorbance units [AU]); 6 (5.8%) were positive when the lower limit for positive was set at the mean value plus 2 standard deviations (SD) (i.e.,.0.109 AU), and 2 (1.9%) were positive with the lower limit at the mean plus 3 SD (i.e.,.0.142 AU). When these animals were retested 4 months later, these positive sera were negative by all three criteria. Using an approach analogous to that described in the previous section for the comparison of sensitivities of the ELISA method and the SAT, we determined that the ELISA is at least 12-fold more sensitive than the SAT method (and the other serological tests as well), using the 0.142 AU value as the cutoff point. DISCUSSION The effectiveness of the ELISA method described lies mainly in the type of antigen used. The primarily proteincontaining CSP from B. abortus (15) is stable to frozen storage and is similar if not identical biochemically from batch to batch. The ELISA with CSP gives very low background values with negative sera and is about 100- to 700-fold more sensitive than the SAT method in detecting IgG in bovine serum (Table 2). CSP gives very low background values with negative sera and is specific for Brucella antibody (Table 3). Moreover, 100,000 tests, or 50,000 tests in duplicate, can be performed with 1 mg of CSP protein. Even though the ELISA with CSP seems specific for Brucella antibody, about 2% of 103 brucellosis-free cattle tested gave false-positive results with the most stringent criteria; these cattle, however, were negative upon retesting 4 months later. Spurious reactions of this sort are also common with the conventional serological (12) methods as well as with other ELISA methods (3). Although the mechanism for the nonspecific reactions are not completely understood, TABLE 2. Comparison of serological tests on a serially diluted serum from a culture-positive cowa ELISA of additional Tube no. Serum dilution (reciprocal) 1:100 dilution of serum Card SAT RIVb CFc AU Score 1 1 0.428 + + + + + (160) 2 12.5 0.424 + + + I (100)d - 3 25 0.424 + + + (100) 1 (50) - 4 50 0.396 + Tr + (50) 1 (25) 5 100 0.339 + Tr + (25) - 6 200 0.300 + - Tr (25) 7 400 0.142 + - - 8 800 0.072 + - - - - 9 1600 0.064 + - - - - 10 3200 0.037 + - - - - 11 6400 0.029 + - - - - 12 12,800 0.025 - - - - - Negative serume 0.013 a Titer was established as 3,200 IU/ml. Serum was diluted with negative bovine serum. Values in parentheses indicate the highest dilution at which test result was positive. briv, Rivanol precipitation-plate agglutination test. c CF, Complement fixation test. d I, Incomplete reaction at the indicated dilution. ' Serum was from selected pooled serum samples from a brucellosis-free herd. The upper negative value for this experiment was considered to be twice the absorbance value of the ELISA result for the negative serum, i.e., 0.026 AU.
212 TABATABAI AND DEYOE E 0.3 4. 0.2 0.1 0- lo-, lo-, 10-4 lo-, lo-6 lo-? SERUM DILUTION FIG. 3. Titration of a known positive serum serially diluted in normal bovine serum (0) and PBS-Tween (0). CSP concentration was 10 ng of protein per well; horseradish peroxidase-conjugated goat anti-bovine IgG was diluted 1:1,000 in PBS-Tween. The dashed line (O---O) denotes ELISA absorbance values with a negative serum. The enzyme reaction was terminated with 0.1 M hydrofluoric acid after 10 min of incubation at room temperature (20 C). TABLE 3. Specificity of the ELISA method for detecting Brucella antibody No. Serum from cattle vaccinated Titer of ELISA or infected with: sera AU t SD tested B. abortus Mo 8 postvaccination <25-50a 7 0.226 ± 0.05 Mo 12 postvaccination <25-50 8 0.081 ± 0.048 Wk 10 postchallenge, 200-1600 8 0.538 ± 0.072 not vaccinated Wk 10 postchallenge, 50-400 8 0.600 ± 0.053 vaccinated Campylobacter fetus Bacterin (2 mg), 5 mo 2,500a 2 0.012 ± 0.020 postvaccination Bacterin (10 mg), 5 2,500 2 0.023 ± 0.031 mo post-vaccination L. pomona Day 2 postinfection 20a 1 0.026 Day 6 postinfection 10,240 1 0.028 Day 15 postinfection 10,240 1 0.030 Moraxella bovis, Wk b 5 0.009 ± 0.010 6 postinfection Mycobacterium avium, b 3 0.041 ± 0.025 Mycobacterium paratuberculosis P. haemolytica Type 1, initial test 2C 5 0.011 ± 0.003 Mo 2 6-8 5 0.016 + 0.008 S. cholera-suis 0 antigen 6,400a 1 H antigen 25,600 1 0.073 a Agglutination titer. b No titer determined, organism isolated. c Indirect hemagglutination titer. TABLE 4. Survey of brucellosis-free, nonvaccinated cattle by the ELISA method No. of sera positive with low limit at: Test. tr1 Negative control Negative control (no. of sera) Negative0control) + 2 SD ± 3 SD x 2 (.0.088 AU) (.0.109 AU) (-0.142 AU) First test (103) 9 (8.7%) 6 (5.8%) 2 (1.9%) Retestb (9) 0 0 0 a Negative control is 0.044 ± 0.032 AU. b Cattle retested 4 months after the first test. it has been observed that E. coli-infected cattle produced IgM antibody that gave false serological reactions to B. abortus (12), presumably through LPS-mediated reactions (9). Sera with high titers (Table 3) from cattle hyperimmunized with C. jejuni bacterin or infected with P. haemolytica or S. cholera-suis, however, did not interfere with our assay. It is therefore possible that the false-positive results described above were due to a low level of anti-igm activity present in the anti-igg conjugate. However, examination of the anti-igg conjugate by immunoelectrophoresis did not reveal anti-igm activity (R. Ranger, personal communication). Alternatively, unpublished results (M. Kehrli, Jr., and L. B. Tabatabai) have shown that ELISA tests with CSP conducted on paired sera from four cattle hyperimmunized with boiled whole cell vaccine prepared from an E. coli mutant (J5) gave absorbance values ranging from 0.089 to 0.136, which roughly corresponded to reciprocal J5 LPS ELISA titers of 100 to 12,800. However, with the "upper negative" value set at 0.142 AU, those sera would be considered negative. Therefore, these data suggest to us that a transient infection of cattle with E. coli could result in low, but nonspecific B. abortus CSP ELISA reactions. Although we chose not to express the ELISA results by any one of the linear transformation methods reviewed by de Savigny and Voller (5), we deduced from both the titration experiments and from tests on the negative cattle population (Tables 2 and 4) that an absorbance value of greater than twice the negative control serum value should be considered positive, rather than using the mean value plus 2 or 3 SD. However, from a practical point of view, using the negative control value plus 3 SD as the cutoff point would eliminate close to 7% of the actually negative cattle to be subjected to retesting. This upper negative limit (0.142 AU) in relation to the parallel titrations performed (shown in Table 2) would still result in a negative status by the SAT method. In fact, the ELISA method would be at least 12-fold more sensitive than the SAT, card, Rivanol, and complement fixation tests under the conditions of this experiment, and would detect a considerable number of infected animals that would otherwise be missed. Although IgM antibody has been demonstrated to be important in the early stages of bovine brucellosis (8), the B. abortus LPS ELISA test does not perform well for the detection of IgM antibody (9, 11). The ELISA with CSP for the detection of IgG antibody would be a sensitive and specific test and could be used to complement the standard agglutination test which primarily detects IgM antibody. ACKNOWLEDGMENTS J. CLIN. MICROBIOL. We gratefully acknowledge the expert technical help of Carol Belzer, Kathryn Meredith, and Carol Irvin. We thank Judith Patterson for the preparation of LPS and Gene Hedberg for his help in the preparation of the three-dimensional plot.
VOL. 20, 1984 LITERATURE CITED 1. Alton, G. G., L. M. Jones, and D. E. Pietz. 1975. Laboratory techniques in brucellosis. World Health Organization monograph series no. 55. World Health Organization, Geneva, Switzerland. 2. Berman, D. T., B. L. Wilson, E. Moreno, R. D. Angus, and L. M. Jones. 1980. Characterization of Brucella abortus soluble antigen employed in immunoassay. J. Clin. Microbiol. 11:355-362. 3. Byrd, J. W., F. C. Heck, and R. J. Hidalgo. 1979. Evaluation of the enzyme linked immunosorbent assay for detecting Brucella abortus antibodies. Am J. Vet. Res. 40:891-898. 4. Carlsson, H. E., B. Hurvell, and A. A. Lindberg. 1976. Enzymelinked immunosorbent assay (ELISA) for titration of antibodies against Brucella abortus and Yersinia enterocolitica. Acta Pathol. Microbiol. Scand. Sect. C 84:168-176. 5. de Savigny, D., and A. Voller. 1980. Communication of ELISA data from laboratory to clinician. J. Immunoassay 1:105-128. 6. Dubois, M., K. A. Gilles, T. K. Hamilton, P. A. Rebers, and F. Smith. 1956. Colorimetric method for determination of sugars and related substances. Anal. Chem. 28:350-356. 7. Heck, F. C., B. L. Deyoe, and T. D. Williams. 1982. Antibodies to Brucella abortus in sera from strain 19 vaccinated and nonvaccinated cows as determined by enzyme linked immunosorbent assay and conventional serologic methods. Vet. Immunol. Immunopathol. 3:629-634. 8. Jones, L. M. 1977. Brucella antigens and serologic test results, p. 40-48. In R. P. Crawford and R. J. Hidalgo (ed.), Bovine brucellosis. An international symposium. Texas A&M University Press, College Station, Tex. ELISA FOR BOVINE ANTIBODY TO B. ABORTUS 213 9. Lamb, V. L., L. M. Jones, G. G. Schurig, and D. T. Berman. 1979. Enzyme-linked immunosorbent assay for bovine immunoglobulin subclass-specific response to Brucella abortus lipopolysaccharides. Infect. Immun. 26:240-247. 10. Lowry, 0. H., N. J. Rosebrough, A. L. Farr, and R. J. Randall. 1951. Protein measurement with the Folin phenol reagent. J. Biol. Chem. 193:265-275. 11. Nielsen, K. H., F. C. Heck, J. M. Stiller, and B. Rosenbaum. 1983. Interaction of specifically purified isotypes of bovine antibody to Brucella abortus in the haemolysis in gel test and enzyme-linked immunosorbent assay. Res. Vet. Sci. 35:14-18. 12. Nielsen, K., B. S. Samagh, and B. Stemshorn. 1980. Agglutination of Brucella abortus cells by sera from cattle experimentally infected with Escherichia coli. Vet. Microbiol. 5:123-134. 13. Ruppanner, R., M. E. Meyer, P. Willeberg, and D. E. Behymer. 1980. Comparison of the enzyme-linked immunosorbent assay with other tests for brucellosis, using sera from experimentally infected heifers. Am. J. Vet. Res. 41:1329-1332. 14. Saunders, G. C., E. H. Clinard, M. L. Bartlett, and W. M. Sanders. 1977. Application of the indirect enzyme labeled antibody microtest to the detection and surveillance of animal diseases. J. Infect. Dis. 136:5258-5266. 15. Tabatabai, L. B., B. L. Deyoe, and A. E. Ritchie. 1979. Isolation and characterization of toxic fractions from Brucella abortus. Infect. Immun. 36:668-679. 16. Voller, A., and D. E. Bidwell. 1975. A simple method for detecting antibodies to rubella. Br. J. Exp. Pathol. 56:338-339. 17. Warren, L. 1959. The thiobarbituric acid assay of sialic acids. J. Biol. Chem. 245:1971-1975.