Vol. 57, No. 4
APPLIED AND ENVIRONMENTAL MICROBIOLOGY, Apr. 1991, p. 1207-1212 0099-2240/91/041207-06$02.00/0 Copyright © 1991, American Society for Microbiology
Comparative Study
on
the Identification of Food-Borne Yeasts
T. TOROK AND A. D. KING, JR.* Western Regional Research Center, U.S. Department of Agriculture, Albany, California 94710 Received 7 September 1990/Accepted 24 January 1991
Morphologically distinct yeast colonies from partially and fully processed fruits and vegetables were isolated 3-year period. Identification of 239 strains was achieved by using standard methods, commercial identification kits (API 20C and API YEAST-IDENT), and a simplified system for food-borne yeasts. The identified strains of fruit origin represented 36 species belonging to 19 genera. Among strains of vegetable origin, 34 species representing 17 genera were identified. The simplified identification system and the conventional method provided the same results in 80% of the cases. The commercial identification kits were easy to use but were not appropriate for food-borne yeast species. Computer-assisted identification was helpful.
over a
Yeast identification has always been a key issue in applied microbiology. It is of paramount importance to have accurate identification results because of the increasing involvement of these microorganisms in research and development and in the technological and medical fields. Also, the identification methods should be rapid and inexpensive. Within the food industry, yeasts play important roles as both production and spoilage microorganisms. Their routine identification is essential, but time-consuming conventional methods, testing 80 to 100 morphological and physiological characteristics, frustrate food microbiologists. Several commercially available identification kits and automated computerized systems have been developed for medically important yeasts. However, their widespread use in the food industry is restricted because their data bases include only 19 to 43 species and there are over 200 food-borne yeast species. In addition, there are several software packages now available (3, 5, 9, 18, 26) for the computer-assisted identification of yeasts. They use probabilistic approaches, a form of Bayesian analysis (6). Their advantages and disadvantages are discussed below. By dramatically reducing the number of tests, Deak (8) proposed a simplified identification key for yeast species associated with food. The scheme was improved by Deak and Beuchat (10) and then tested and compared with commercial systems (11). The major advantage of the simplified scheme is that it usually requires only two petri dishes, three test tubes, a microscope, and occasionally up to eight additional tests for species identification, in contrast to the 80 to 100 morphological and physiological tests required by the conventional schemes of Kreger-van Rij (19) and Barnett et al. (4). Rohm and Lechner (27) strongly criticized the scheme as unreliable. They claimed that Deak and Beuchat (10) did not consider the variability of some characteristics. Using the simplified scheme, Rohm and Lechner (27) failed to identify certain reference strains and over 50% of the species they isolated from cultured milk products. Also, they examined the species in all 16 subgroups of the simplified scheme and concluded that only 19 of some 200 species listed could be unequivocally identified by the scheme. The purpose of our investigation was to compare conventional yeast identification methods (4, 19), commercially available identification kits (API 20C and API YEAST*
Corresponding author.
IDENT), and Deak and Beuchat's simplified identification scheme (10), by using a wide range of yeast strains isolated from partially and fully processed fruits and vegetables. (Throughout this paper we use the term partially processed food, in accordance with Huxsoll et al. [15]. Fully processed food samples and spoiled products were sent to our laboratory for examination. We made no attempt to track technological, packaging, storage, or retail features.) We also tested computer programs for the identification of the characterized yeast isolates.
MATERIALS AND METHODS Yeast strains. Different types of partially and fully processed fruits and vegetables from processing plants and from wholesale and retail outlets were examined over a 3-year period by using plate count agar (Difco) and chloramphenicol-containing glucose-yeast extract agar (Oxoid). After preliminary morphological screening, we selected 239 strains for identification. The pure cultures were grown on medium containing 1% yeast extract, 1% peptone, 1% dextrose, and 2% agar. Almost 80% of the 127 selected yeast strains of fruit origin were sent to us for identification. These strains were isolated from fresh fruits and fruit-processing plants at various geographical locations. Of these, 14 strains caused spoilage of processed fruit products. We selected 112 strains from fresh and partially processed vegetables. The wide variety of food products assured a good selection of yeast strains, which in turn allowed a more general comparison of the identification methods. Strain characterization. The yeast strains were characterized by using methods described in standard taxonomic manuals (4, 19) and according to the instructions for the commercial identification kits (API 20C, API YEASTIDENT). Strain characterization by the method of Deaik and Beuchat was as described previously (10), with the exception of the nitrate assimilation test, which was modified as described by Pincus et al. (25). Identification procedures. In one conventional scheme, the yeasts were identified with the dichotomous keys in the manual of Kreger-van Rij (19). Yeast identification by the method of Barnett et al. (4) was achieved by using the computer program designed by the same authors (5). The identification of yeasts within the simplified scheme was carried out by using the master key and the detailed identification keys of Deak and Beuchat (10). To evaluate the test results collected with the commercial 1207
1208
APPL. ENVIRON. MICROBIOL.
TOROK AND KING
TABLE 1. Samples of partially and fully processed fruits and vegetables included in this study No. of sample samples 1 Apple juice ............................................ 1 Apricot pouch ............................................ 1 Blueberry yogurt ..................... ....................... Cabbage, fresh and partially processed ............. .............. 4 Cauliflower, fresh and partially processed ........................ 3 Carrot, fresh and partially processed ................ .............. 6 Cherry frozen dessert with yogurt .................................. 2 1 ............................ Dried apricots ................ Fruit product processing equipment ................................ 28 1 Grapefruit juice ............................................ Lettuce, fresh and partially processed ............................. 77 5 Orange juice ............................................ 2 ................... Peach concentrate ......................... 20 Pear ............................................ 6 Pineapple ............................................ 15 Pineapple concentrate ............................................ 12 Pineapple juice ............................................ 10 Pineapple-orange juice ............................................ 1 Raspberry puree ............................................ 1 Red grape concentrate ............................................ Unknown origin ................... 3 .........................
TABLE 2. Yeast species isolated and identified from partially and fully processed fruits and vegetables
Sample source
kit API 20C, we used the API analytical profile index (2). The API YEAST-IDENT results were evaluated by using the API YEAST-IDENT directory (1). Computer-assisted identification software packages from the American Society for Microbiology Computer Users Group (3), Deaik (9), and Reichart (26) were tested. Since there are no strain characterization methods described for any of these software packages, we used the test results gathered in the other identification schemes. RESULTS The fruit and vegetable samples from which the yeast strains were isolated are listed in Table 1. Among the fruit samples, pineapple and pineapple products appeared most frequently. Lettuce was the major partially processed vegetable product tested (77 samples). The number of samples of each type shows large variation due to special interests and projects. Low sample numbers often represent cases of spoilage. Based on the identifications, more ascomycete-type species were isolated from fruits (106 strains) than from vegetables (77 strains). In contrast, more basidiomycete-type species were isolated from vegetables (35 strains) than from fruits (21 strains). The detailed identification results for the yeast species are listed in Table 2. Although Candida tropicalis and Saccharomyces cerevisiae were the most frequently identified yeasts isolated from fruits, weak or nonfermenting species (Candida lambica [anamorph for Pichia fermentans], Cryptococcus albidus and Trichosporon cutaneum) were predominant among isolates from vegetables. We based our comparison of species names, resulting from the various identification schemes, mainly on the work of Kreger-van Rij (19). Her conventional system of strain characterization and identification gave significantly the highest percentage of correct results. However, in some individual cases, other schemes provided more plausible results. For example, the black yeasts could not be identified with the Kreger-van Rij standard system, and we turned to other sources (12, 14). We calculated the mean value of
Species
Arthroascus javanensis Aureobasidium pullulans Candida ciferrii Candida colliculosa Candida diversa Candida famata Candida haemulonii Candida humicola Candida intermedia Candida krusei Candida lambica Candida magnoliae Candida parapsilosis Candida sake Candida tropicalis Candida versatilis Clavispora lusitaniae Cryptococcus albidus Cryptococcus curvatus Cryptococcus flavus Cryptococcus hungaricus Cryptococcus laurentii Cryptococcus luteolus Cryptococcus macerans Cryptococcus neoformans Debaryomyces castellii Debaryomyces hansenji Debaryomyces marama Hanseniaspora guilliermondii Hanseniaspora valbyensis Hansenula anomala Issatchenkia orientalis Kloeckera apis Kloeckera japonica Kluyveromyces marxianus var. lactis Metschnikowia pulcherrima Metschnikowia reukaufii Pichia membranaefaciens Rhodosporidium infirmo-miniatum Rhodotorula glutinis Rhodotorula minuta Rhodotorula rubra Saccharomyces cerevisiae Stephanoascus ciferrii Torulaspora delbrueckii Trichosporon adeninovorans Trichosporon cutaneum Trichosporon pullulans Trichosporonoides sp. Wickerhamiella domercqii Yarrowia lipolytica Zygosaccharomyces bailli Zygosaccharomyces rouxii Unidentified isolate
Frequency among the identified strains in:
Fruits
Vegetables
1
1 1 1 6
5 1
5
3 2
3 8 3 3 2 10 1 1 3
2
1 1 22
1 6 3 10 1 3 1 7 2
1 1
1 6 1 1 4 1 2
2 1 1
1 1 1
5 1 5 3 4 18 3 1 4 4
1 2 6 1 1 7 1 5 7
1 1 2 4 6 1
reliable identifications for all compared schemes. The results are compiled in Fig. 1A for strains isolated from fruits and in Fig. 1B for strains isolated from vegetables. DISCUSSION With the identification of the 239 yeast strains, we observed variations of the fungus type that could be accounted
VOL. 57, 1991
IDENTIFICATION OF FOOD-BORNE YEASTS
1209
100 c
80
._
0
0
60
C) -o
40
a, -o
0
20 0 100
0
80
._
0 ._
0 C._ he-
60
4-
-o
40
a0)
-i
20 0
FIG. 1. Yeast identification results comparing different identification schemes for strains isolated from partially and fully processed vegetables (A) and fruits (B). **, Significant (P = 0.99); *, significant (P = 0.95); SD95, standard deviation for a P value of 0.95.
1210
APPL. ENVIRON. MICROBIOL.
TOROK AND KING
for by characteristic differences in the chemistry and the ecological niches supplied by the two commodities (e.g., the sugars in fruits are more easily accessible than those in vegetables; vegetables are closer to the soil, their pH is different, their carbohydrates are more complex, etc.). However, the numbers of yeast species and genera were similar for both food groups, and there were 18 species and 12 genera in common. This is an indication of the relatively few yeast species and genera described as being associated with food (10). C. lambica and S. cerevisiae were the most frequently isolated species in the whole set of experiments. This is remarkable because, although C. lambica can hardly be misidentified because of its overwhelming number of negative characteristics, S. cerevisiae can be very difficult to identify because of the large number of variable reactions in standard descriptions (4, 19). The simplified identification scheme of Deaik and Beuchat (10) provided significantly more correct identifications than did the other systems in this study when they were compared with the standard identification scheme of Kreger-van Rij (19). Deaik and Beuchat employed the standard descriptions for yeast strains of Barnett et al. (4). However, the Barnett et al. manual does not specify how data were gathered, and Deak and Beuchat had difficulties with variable, delayed, or weak test results and with a lack of information (e.g., on sexual reproduction). Thus, some important food-borne yeasts either cannot be identified or will be continuously misidentified in certain subgroups of the Deak and Beuchat scheme. Some specific examples from our experience are listed below. (i) Within the standard descriptions (4, 19), the responses of the following food-borne species can vary in certain characteristics important for the master key: Candida magnoliae, growth on galactose and cellobiose; Candida tropi-
calis, growth on cellobiose; Candida versatilis, growth on cellobiose; Cryptococcus albidus, growth on galactose, melibiose, raffinose, and erythritol; Debaryomyces hansenii, fermentation of glucose; Torulaspora delbrueckii, growth on galactose, maltose, raffinose, and xylose. (ii) In other cases, standard descriptions (4, 19) disagree about particular characteristics, such as growth of Candida castelli on trehalose, growth of Cryptococcus gastricus on lactose, growth of Sporobolomyces roseus on nitrate, and growth of some Cryptococcus species on inositol. In all of these and similar cases, yeasts may be misidentified by the simplified identification scheme because Deaik and Beuchat (10) failed to include both positive and negative test results in their subgroups. The identification scheme of Barnett et al. (4), with the help of their computer-assisted key (5), can be used successfully for food-borne yeasts. The software identifies or selects yeasts by using the same set of tests as the manual by the same authors (4). The data base was updated for new or amended taxa and now contains information on 497 yeast species. Its main disadvantage lies in the large number of tests required. The commercial API kits were easy to use, and the biochemical tests proved to be highly reproducible. Never-
theless, there were major problems. With the YEASTIDENT kit, the last 11 tests out of a total of 20 were almost always positive and so provided no selection power. Also, due to the limited data base, API 20C and API YEASTIDENT may result in the same but incorrect species identification, e.g., Candida guilliermondii instead of Debaryomyces hansenii. Thus, we cannot recommend API kits for
TABLE 3. Typical printout for comparison of the test results for an unknown with those of a known species in the data base of the Deak Yeast-ID test (9) Test result
Testa S. kluyverib
Ure Ery Nit Ce Mt
Unknown
Vc +
R
1 1 1 50 80 80
G
99
+
Tre Mz Me L Rm Xyl
80 1 99
+
M
It d
1 10 1 99 1
vit 37 50
60 99 50 1 1 1 1 1
hy psm
pell pink
+ +
+
1
cyc
art
+ +
+ + + + +
a Tests are abbreviated on the screen as follows: Ure, urea hydrolysis; Ery, growth on erythritol; Nit, growth on nitrate; Ce, growth on cellobiose; Mt, growth on D-mannitol; M, growth on maltose; R, growth on raffinose; G, growth on D-galactose; Tre, growth on a,c-trehalose; Mz, growth on melezitose; Me, growth on melibiose; L, growth on lactose; Rm, growth on Lrhamnose; Xyl, growth on D-xylose; It, growth on myo-inositol; d, fermentation of D-glucose; cyc, 0.1% (wt/vol) cycloheximide; vit, growth without vitamins; 37, growth at 37°C; 50, 50% (wt/vol) D-glucose; art, arthroconidia; hy, septate hyphae; psm, pseudohyphae; pell, pellicle formation; pink, pink colonies. b For comparison, S. kluyveri was randomly chosen (computed values for the unknown compared with S. kluyveri: frequency, 0.07902718; modal frequency, 1.145122 x 10-7). c Positive reaction in percentage cited in the literature (for computation purposes, a value of 1, instead of 0, was assigned to negative reactions).
effective identification of food-borne yeasts, especially because some diagnoses may seem reasonable to the taxonomically inexperienced microbiologist. Attempts to combine API 20C with appropriate microscopic morphological characteristics were successful only for a limited number of genera (22). The Yeast program of the American Society for Microbiology Computer Users Group (3) (diskette 1106) is one of the probabilistic approaches for yeast identification. It needs to be updated with approved names of yeast species (4, 5, 19, 23). Also, the continuous reloading of the data base required during program operation causes some frustration. The vulnerability of the program lies in the common use of probability assessment for identifying microorganisms. The Yeast-ID program of Deak (9) was based on the same concept as the simplified identification scheme of Deak and Beuchat for yeasts (10). However, some tests have been changed or added (e.g., assimilation of erythritol). Twentyfive test results must be entered into the program (Table 3). The program calculates the frequency of occurrence of a profile in the taxon (21) and the modal frequency for each of the taxa (13), making possible a comparison of the results
VOL. 57, 1991
from an unknown with those of a single species or with those of all species entered into the data base. The user must decide whether to rely more on the frequency of occurrence, on the modal frequency, or on both. The program offers two levels of comparison (all known food-borne yeasts and commonly occurring food-borne yeasts). The weakness of the program is that it adopts literature descriptions of species without considering how those characteristics were gathered. The yeast identification program of Reichart (26) uses the same data base as that used for the Deatk software. The user has the option of setting the confidence values for the diagnosis, providing an intelligent challenge for the experienced zymologist. The interactive program offers additional tests from the data base for better separation of the species. The basic algorithm of the computer-assisted identification programs is a version of the Bayesean model of prior likelihoods (6). The test results of the unknown isolate are compared with entries in the data base, and the likelihood for each species is computed (13, 16, 17, 21, 24, 28). Even if all strain characterization results are correct and reproducible and all data base entries are reasonable, there are still at least five reasons for concern when using probabilistic systems. (i) The tests should have enough credibility and separation power to disclose existing differences between strains. Also, they have to be scientifically sound to differentiate between species. (ii) The user does not know how positive reaction results were collected and expressed mathematically. As many strains as possible from different locations of the same species should be tested. This is especially important when assessing strains with so-called variable, delayed, or weak test results. (iii) There is no information available that estimates the statistical reliability of a computer-assisted diagnosis. Analyses are needed to determine the necessary and sufficient number of data points entered, the redundancy of the data input, and the tolerable level of input errors. (iv) Probabilistic identification systems have a rigid arrangement, and so they ignore the prevalence or rarity of a given microbial species (7). Depending on how the identification characteristics in the data bases were collected and how probability was computed, each software package has "favorites." An unknown isolate may have the same positive test results as several species in the data base. The program, however, will identify the isolate as belonging to the species with the highest positive test percentage. Very often higher positive test percentages are, in turn, due to the frequency of occurrence of a given species. (v) None of the probabilistic identification programs has an option for recognizing new species, and the unknown isolate will be identified (misidentified), although at a low score. We learned from our experience with yeast identification that, for taxonomic purposes or for definitive identification (20), standard yeast identification methods have to be used. At this point, we prefer the work of Kreger-van Rij (19). For routine or presumptive (20) yeast identification in the food industry, we recommend highly the simplified identification scheme developed by Deak and Beuchat (10). Adapting the tests from Deatk's newly released software and revising it based on the available standard works (4, 19) will make both the simplified identification key and the computer program more accurate and user friendly.
IDENTIFICATION OF FOOD-BORNE YEASTS
1211
ACKNOWLEDGMENTS We express our gratitude to T. Deak and 0. Reichart for friendly permission to use their then-unpublished software packages. REFERENCES 1. Analytab Products. 1986. YEAST-IDENT directory for the biochemical identification of yeast and yeast-like organisms. Product no. 8886-436016. Analytab Products, Plainview, N.Y. 2. Analytab Products. 1988. API 20C. Analytical profile index yeast, p. 43-103. Analytab Products, Plainview, N.Y. 3. American Society for Microbiology Computer User Group. 1989. Computer-assisted identification of microorganisms (CAIM)yeasts. American Society for Microbiology, Washington, D.C. 4. Barnett, J. A., R. W. Payne, and D. Yarrow. 1983. Yeasts: characteristics and identification. Cambridge University Press, Cambridge. 5. Barnett, J. A., R. W. Payne, and D. Yarrow. 1985. Yeast identification program. Cambridge Micro Software. Cambridge University Press, Cambridge. 6. Bayes, T. 1763. An essay toward solving a problem in the doctrine of chance. Philos. Trans. R. Soc. London 53:269-271. 7. Berger, S. A. 1990. Lack of precision in commercial identification systems: correction using Bayesian analysis. J. Appl. Bacteriol. 68:285-288. 8. Dea4k, T. 1986. A simplified scheme for the identification of yeasts, p. 278-293. In A. D. King, Jr., J. I. Pitt, L. R. Beuchat, and J. E. L. Corry (ed.), Methods for the mycological examination of food. Plenum Publishing Corp., New York. 9. Dealk, T. 1990. Yeast-ID (software available upon request). University of Horticulture and Food Industry, Budapest, Hungary. 10. Deak, T., and L. R. Beuchat. 1987. Identification of foodborne yeasts. J. Food Prot. 50:243-264. 11. Dealk, T., and L. R. Beuchat. 1988. Evaluation of simplified and commercial systems for identification of foodborne yeasts. J. Food Microbiol. 7:135-145. 12. De Hoog, G. S., and E. J. Hermanides-NiJhof. 1977. Survey of the black yeasts and allied fungi, p. 178-223. In Studies in mycology no. 15. Centraalbureau voor Schimmelcultures, Baarn, The Netherlands. 13. Dybowski, W., and D. A. Franklin. 1968. Conditional probability and the identification of bacteria: a pilot study. J. Gen. Microbiol. 54:215-229. 14. Hermanides-Nijhof, E. J. 1977. Aureobasidium and allied genera, p. 141-177. In Studies in mycology, no. 15. Centraalbureau voor Schimmelcultures, Baarn, The Netherlands. 15. Huxsoll, C. C., H. R. Bolin, and A. D. King, Jr. 1989. Physicochemical changes and treatments for lightly processed fruits and vegetables, p. 203-215. In J. J. Jen (ed.), Quality factors of fruits and vegetables. Chemistry and technology. American Chemical Society, Washington, D.C. 16. Kelley, R. W., and S. T. Kellogg. 1978. Computer-assisted identification of anaerobic bacteria. Appl. Environ. Microbiol. 35:507-511. 17. Kellogg, S. T. 1979. MICRID: a computer-assisted microbial identification system. Appl. Environ. Microbiol. 38:559-563. 18. Kirsop, B. E., K. A. Painting, J. E. Henry, M. Fernandes, E. H. Prescott, I. Braid, and P. Foster. 1986. On-line computer assisted identification of yeasts, p. 75. In XIV International Congress of Microbiology, Abstracts. 19. Kreger-van Rij, N. J. W. (ed.). 1984. The yeasts, a taxonomic study, 3rd ed. Elsevier Science Publisher, B.V., Amsterdam. 20. Kurtzman, C. P. 1988. Identification and taxonomy, p. 99-140. In B. E. Kirsop, and C. P. Kurtzman (ed.), Living resources for biotechnology. Yeasts. Cambridge University Press, Cambridge. 21. Lapage, S. P., S. Bascomb, W. R. Willcox, and M. A. Curtis. 1973. Identification of bacteria by computer: general aspects and perspectives. J. Gen. Microbiol. 77:273-290. 22. Lin, C. C. S., and D. Y. C. Fung. 1987. Conventional and rapid methods for yeast identification. Crit. Rev. Microbiol. 14:273289. 23. Moore, W. E. C., and L. V. H. Moore (ed.). 1989. Index of the
1212
TOROK AND KING
bacterial and yeast nomenclatural changes. American Society for Microbiology, Washington, D.C. 24. Pankhurst, R. J. (ed.). 1975. Biological identification with computers. Systematics association special vol. 7. Academic Press, Inc., New York. 25. Pincus, D. H., I. F. Salkin, N. J. Hurd, I. L. Levy, and M. A. Kemna. 1988. Modification of potassium nitrate assimilation test for identification of clinically important yeasts. J. Clin. Microbiol. 26:366-368.
APPL. ENVIRON. MICROBIOL.
26. Reichart, 0. 1990. Yeast identification program (software available upon request). University of Horticulture and Food Industry, Budapest, Hungary. 27. Rohm, H., and F. Lechner. 1990. Evaluation and reliability of a simplified method for identification of food-borne yeasts. Appl. Environ. Microbiol. 56:1290-1295. 28. Wilcox, W. R., S. P. Lapage, S. Bascomb, and M. A. Curtis. 1973. Identification of bacteria by computer: theory and programming. J. Gen. Microbiol. 77:317-330.
APPLIED AND ENVIRONMENTAL MICROBIOLOGY, Apr. 1991, p. 1207-1212 0099-2240/91/041207-06$02.00/0 Copyright © 1991, American Society for Microbiology
Comparative Study
on
the Identification of Food-Borne Yeasts
T. TOROK AND A. D. KING, JR.* Western Regional Research Center, U.S. Department of Agriculture, Albany, California 94710 Received 7 September 1990/Accepted 24 January 1991
Morphologically distinct yeast colonies from partially and fully processed fruits and vegetables were isolated 3-year period. Identification of 239 strains was achieved by using standard methods, commercial identification kits (API 20C and API YEAST-IDENT), and a simplified system for food-borne yeasts. The identified strains of fruit origin represented 36 species belonging to 19 genera. Among strains of vegetable origin, 34 species representing 17 genera were identified. The simplified identification system and the conventional method provided the same results in 80% of the cases. The commercial identification kits were easy to use but were not appropriate for food-borne yeast species. Computer-assisted identification was helpful.
over a
Yeast identification has always been a key issue in applied microbiology. It is of paramount importance to have accurate identification results because of the increasing involvement of these microorganisms in research and development and in the technological and medical fields. Also, the identification methods should be rapid and inexpensive. Within the food industry, yeasts play important roles as both production and spoilage microorganisms. Their routine identification is essential, but time-consuming conventional methods, testing 80 to 100 morphological and physiological characteristics, frustrate food microbiologists. Several commercially available identification kits and automated computerized systems have been developed for medically important yeasts. However, their widespread use in the food industry is restricted because their data bases include only 19 to 43 species and there are over 200 food-borne yeast species. In addition, there are several software packages now available (3, 5, 9, 18, 26) for the computer-assisted identification of yeasts. They use probabilistic approaches, a form of Bayesian analysis (6). Their advantages and disadvantages are discussed below. By dramatically reducing the number of tests, Deak (8) proposed a simplified identification key for yeast species associated with food. The scheme was improved by Deak and Beuchat (10) and then tested and compared with commercial systems (11). The major advantage of the simplified scheme is that it usually requires only two petri dishes, three test tubes, a microscope, and occasionally up to eight additional tests for species identification, in contrast to the 80 to 100 morphological and physiological tests required by the conventional schemes of Kreger-van Rij (19) and Barnett et al. (4). Rohm and Lechner (27) strongly criticized the scheme as unreliable. They claimed that Deak and Beuchat (10) did not consider the variability of some characteristics. Using the simplified scheme, Rohm and Lechner (27) failed to identify certain reference strains and over 50% of the species they isolated from cultured milk products. Also, they examined the species in all 16 subgroups of the simplified scheme and concluded that only 19 of some 200 species listed could be unequivocally identified by the scheme. The purpose of our investigation was to compare conventional yeast identification methods (4, 19), commercially available identification kits (API 20C and API YEAST*
Corresponding author.
IDENT), and Deak and Beuchat's simplified identification scheme (10), by using a wide range of yeast strains isolated from partially and fully processed fruits and vegetables. (Throughout this paper we use the term partially processed food, in accordance with Huxsoll et al. [15]. Fully processed food samples and spoiled products were sent to our laboratory for examination. We made no attempt to track technological, packaging, storage, or retail features.) We also tested computer programs for the identification of the characterized yeast isolates.
MATERIALS AND METHODS Yeast strains. Different types of partially and fully processed fruits and vegetables from processing plants and from wholesale and retail outlets were examined over a 3-year period by using plate count agar (Difco) and chloramphenicol-containing glucose-yeast extract agar (Oxoid). After preliminary morphological screening, we selected 239 strains for identification. The pure cultures were grown on medium containing 1% yeast extract, 1% peptone, 1% dextrose, and 2% agar. Almost 80% of the 127 selected yeast strains of fruit origin were sent to us for identification. These strains were isolated from fresh fruits and fruit-processing plants at various geographical locations. Of these, 14 strains caused spoilage of processed fruit products. We selected 112 strains from fresh and partially processed vegetables. The wide variety of food products assured a good selection of yeast strains, which in turn allowed a more general comparison of the identification methods. Strain characterization. The yeast strains were characterized by using methods described in standard taxonomic manuals (4, 19) and according to the instructions for the commercial identification kits (API 20C, API YEASTIDENT). Strain characterization by the method of Deaik and Beuchat was as described previously (10), with the exception of the nitrate assimilation test, which was modified as described by Pincus et al. (25). Identification procedures. In one conventional scheme, the yeasts were identified with the dichotomous keys in the manual of Kreger-van Rij (19). Yeast identification by the method of Barnett et al. (4) was achieved by using the computer program designed by the same authors (5). The identification of yeasts within the simplified scheme was carried out by using the master key and the detailed identification keys of Deak and Beuchat (10). To evaluate the test results collected with the commercial 1207
1208
APPL. ENVIRON. MICROBIOL.
TOROK AND KING
TABLE 1. Samples of partially and fully processed fruits and vegetables included in this study No. of sample samples 1 Apple juice ............................................ 1 Apricot pouch ............................................ 1 Blueberry yogurt ..................... ....................... Cabbage, fresh and partially processed ............. .............. 4 Cauliflower, fresh and partially processed ........................ 3 Carrot, fresh and partially processed ................ .............. 6 Cherry frozen dessert with yogurt .................................. 2 1 ............................ Dried apricots ................ Fruit product processing equipment ................................ 28 1 Grapefruit juice ............................................ Lettuce, fresh and partially processed ............................. 77 5 Orange juice ............................................ 2 ................... Peach concentrate ......................... 20 Pear ............................................ 6 Pineapple ............................................ 15 Pineapple concentrate ............................................ 12 Pineapple juice ............................................ 10 Pineapple-orange juice ............................................ 1 Raspberry puree ............................................ 1 Red grape concentrate ............................................ Unknown origin ................... 3 .........................
TABLE 2. Yeast species isolated and identified from partially and fully processed fruits and vegetables
Sample source
kit API 20C, we used the API analytical profile index (2). The API YEAST-IDENT results were evaluated by using the API YEAST-IDENT directory (1). Computer-assisted identification software packages from the American Society for Microbiology Computer Users Group (3), Deaik (9), and Reichart (26) were tested. Since there are no strain characterization methods described for any of these software packages, we used the test results gathered in the other identification schemes. RESULTS The fruit and vegetable samples from which the yeast strains were isolated are listed in Table 1. Among the fruit samples, pineapple and pineapple products appeared most frequently. Lettuce was the major partially processed vegetable product tested (77 samples). The number of samples of each type shows large variation due to special interests and projects. Low sample numbers often represent cases of spoilage. Based on the identifications, more ascomycete-type species were isolated from fruits (106 strains) than from vegetables (77 strains). In contrast, more basidiomycete-type species were isolated from vegetables (35 strains) than from fruits (21 strains). The detailed identification results for the yeast species are listed in Table 2. Although Candida tropicalis and Saccharomyces cerevisiae were the most frequently identified yeasts isolated from fruits, weak or nonfermenting species (Candida lambica [anamorph for Pichia fermentans], Cryptococcus albidus and Trichosporon cutaneum) were predominant among isolates from vegetables. We based our comparison of species names, resulting from the various identification schemes, mainly on the work of Kreger-van Rij (19). Her conventional system of strain characterization and identification gave significantly the highest percentage of correct results. However, in some individual cases, other schemes provided more plausible results. For example, the black yeasts could not be identified with the Kreger-van Rij standard system, and we turned to other sources (12, 14). We calculated the mean value of
Species
Arthroascus javanensis Aureobasidium pullulans Candida ciferrii Candida colliculosa Candida diversa Candida famata Candida haemulonii Candida humicola Candida intermedia Candida krusei Candida lambica Candida magnoliae Candida parapsilosis Candida sake Candida tropicalis Candida versatilis Clavispora lusitaniae Cryptococcus albidus Cryptococcus curvatus Cryptococcus flavus Cryptococcus hungaricus Cryptococcus laurentii Cryptococcus luteolus Cryptococcus macerans Cryptococcus neoformans Debaryomyces castellii Debaryomyces hansenji Debaryomyces marama Hanseniaspora guilliermondii Hanseniaspora valbyensis Hansenula anomala Issatchenkia orientalis Kloeckera apis Kloeckera japonica Kluyveromyces marxianus var. lactis Metschnikowia pulcherrima Metschnikowia reukaufii Pichia membranaefaciens Rhodosporidium infirmo-miniatum Rhodotorula glutinis Rhodotorula minuta Rhodotorula rubra Saccharomyces cerevisiae Stephanoascus ciferrii Torulaspora delbrueckii Trichosporon adeninovorans Trichosporon cutaneum Trichosporon pullulans Trichosporonoides sp. Wickerhamiella domercqii Yarrowia lipolytica Zygosaccharomyces bailli Zygosaccharomyces rouxii Unidentified isolate
Frequency among the identified strains in:
Fruits
Vegetables
1
1 1 1 6
5 1
5
3 2
3 8 3 3 2 10 1 1 3
2
1 1 22
1 6 3 10 1 3 1 7 2
1 1
1 6 1 1 4 1 2
2 1 1
1 1 1
5 1 5 3 4 18 3 1 4 4
1 2 6 1 1 7 1 5 7
1 1 2 4 6 1
reliable identifications for all compared schemes. The results are compiled in Fig. 1A for strains isolated from fruits and in Fig. 1B for strains isolated from vegetables. DISCUSSION With the identification of the 239 yeast strains, we observed variations of the fungus type that could be accounted
VOL. 57, 1991
IDENTIFICATION OF FOOD-BORNE YEASTS
1209
100 c
80
._
0
0
60
C) -o
40
a, -o
0
20 0 100
0
80
._
0 ._
0 C._ he-
60
4-
-o
40
a0)
-i
20 0
FIG. 1. Yeast identification results comparing different identification schemes for strains isolated from partially and fully processed vegetables (A) and fruits (B). **, Significant (P = 0.99); *, significant (P = 0.95); SD95, standard deviation for a P value of 0.95.
1210
APPL. ENVIRON. MICROBIOL.
TOROK AND KING
for by characteristic differences in the chemistry and the ecological niches supplied by the two commodities (e.g., the sugars in fruits are more easily accessible than those in vegetables; vegetables are closer to the soil, their pH is different, their carbohydrates are more complex, etc.). However, the numbers of yeast species and genera were similar for both food groups, and there were 18 species and 12 genera in common. This is an indication of the relatively few yeast species and genera described as being associated with food (10). C. lambica and S. cerevisiae were the most frequently isolated species in the whole set of experiments. This is remarkable because, although C. lambica can hardly be misidentified because of its overwhelming number of negative characteristics, S. cerevisiae can be very difficult to identify because of the large number of variable reactions in standard descriptions (4, 19). The simplified identification scheme of Deaik and Beuchat (10) provided significantly more correct identifications than did the other systems in this study when they were compared with the standard identification scheme of Kreger-van Rij (19). Deaik and Beuchat employed the standard descriptions for yeast strains of Barnett et al. (4). However, the Barnett et al. manual does not specify how data were gathered, and Deak and Beuchat had difficulties with variable, delayed, or weak test results and with a lack of information (e.g., on sexual reproduction). Thus, some important food-borne yeasts either cannot be identified or will be continuously misidentified in certain subgroups of the Deak and Beuchat scheme. Some specific examples from our experience are listed below. (i) Within the standard descriptions (4, 19), the responses of the following food-borne species can vary in certain characteristics important for the master key: Candida magnoliae, growth on galactose and cellobiose; Candida tropi-
calis, growth on cellobiose; Candida versatilis, growth on cellobiose; Cryptococcus albidus, growth on galactose, melibiose, raffinose, and erythritol; Debaryomyces hansenii, fermentation of glucose; Torulaspora delbrueckii, growth on galactose, maltose, raffinose, and xylose. (ii) In other cases, standard descriptions (4, 19) disagree about particular characteristics, such as growth of Candida castelli on trehalose, growth of Cryptococcus gastricus on lactose, growth of Sporobolomyces roseus on nitrate, and growth of some Cryptococcus species on inositol. In all of these and similar cases, yeasts may be misidentified by the simplified identification scheme because Deaik and Beuchat (10) failed to include both positive and negative test results in their subgroups. The identification scheme of Barnett et al. (4), with the help of their computer-assisted key (5), can be used successfully for food-borne yeasts. The software identifies or selects yeasts by using the same set of tests as the manual by the same authors (4). The data base was updated for new or amended taxa and now contains information on 497 yeast species. Its main disadvantage lies in the large number of tests required. The commercial API kits were easy to use, and the biochemical tests proved to be highly reproducible. Never-
theless, there were major problems. With the YEASTIDENT kit, the last 11 tests out of a total of 20 were almost always positive and so provided no selection power. Also, due to the limited data base, API 20C and API YEASTIDENT may result in the same but incorrect species identification, e.g., Candida guilliermondii instead of Debaryomyces hansenii. Thus, we cannot recommend API kits for
TABLE 3. Typical printout for comparison of the test results for an unknown with those of a known species in the data base of the Deak Yeast-ID test (9) Test result
Testa S. kluyverib
Ure Ery Nit Ce Mt
Unknown
Vc +
R
1 1 1 50 80 80
G
99
+
Tre Mz Me L Rm Xyl
80 1 99
+
M
It d
1 10 1 99 1
vit 37 50
60 99 50 1 1 1 1 1
hy psm
pell pink
+ +
+
1
cyc
art
+ +
+ + + + +
a Tests are abbreviated on the screen as follows: Ure, urea hydrolysis; Ery, growth on erythritol; Nit, growth on nitrate; Ce, growth on cellobiose; Mt, growth on D-mannitol; M, growth on maltose; R, growth on raffinose; G, growth on D-galactose; Tre, growth on a,c-trehalose; Mz, growth on melezitose; Me, growth on melibiose; L, growth on lactose; Rm, growth on Lrhamnose; Xyl, growth on D-xylose; It, growth on myo-inositol; d, fermentation of D-glucose; cyc, 0.1% (wt/vol) cycloheximide; vit, growth without vitamins; 37, growth at 37°C; 50, 50% (wt/vol) D-glucose; art, arthroconidia; hy, septate hyphae; psm, pseudohyphae; pell, pellicle formation; pink, pink colonies. b For comparison, S. kluyveri was randomly chosen (computed values for the unknown compared with S. kluyveri: frequency, 0.07902718; modal frequency, 1.145122 x 10-7). c Positive reaction in percentage cited in the literature (for computation purposes, a value of 1, instead of 0, was assigned to negative reactions).
effective identification of food-borne yeasts, especially because some diagnoses may seem reasonable to the taxonomically inexperienced microbiologist. Attempts to combine API 20C with appropriate microscopic morphological characteristics were successful only for a limited number of genera (22). The Yeast program of the American Society for Microbiology Computer Users Group (3) (diskette 1106) is one of the probabilistic approaches for yeast identification. It needs to be updated with approved names of yeast species (4, 5, 19, 23). Also, the continuous reloading of the data base required during program operation causes some frustration. The vulnerability of the program lies in the common use of probability assessment for identifying microorganisms. The Yeast-ID program of Deak (9) was based on the same concept as the simplified identification scheme of Deak and Beuchat for yeasts (10). However, some tests have been changed or added (e.g., assimilation of erythritol). Twentyfive test results must be entered into the program (Table 3). The program calculates the frequency of occurrence of a profile in the taxon (21) and the modal frequency for each of the taxa (13), making possible a comparison of the results
VOL. 57, 1991
from an unknown with those of a single species or with those of all species entered into the data base. The user must decide whether to rely more on the frequency of occurrence, on the modal frequency, or on both. The program offers two levels of comparison (all known food-borne yeasts and commonly occurring food-borne yeasts). The weakness of the program is that it adopts literature descriptions of species without considering how those characteristics were gathered. The yeast identification program of Reichart (26) uses the same data base as that used for the Deatk software. The user has the option of setting the confidence values for the diagnosis, providing an intelligent challenge for the experienced zymologist. The interactive program offers additional tests from the data base for better separation of the species. The basic algorithm of the computer-assisted identification programs is a version of the Bayesean model of prior likelihoods (6). The test results of the unknown isolate are compared with entries in the data base, and the likelihood for each species is computed (13, 16, 17, 21, 24, 28). Even if all strain characterization results are correct and reproducible and all data base entries are reasonable, there are still at least five reasons for concern when using probabilistic systems. (i) The tests should have enough credibility and separation power to disclose existing differences between strains. Also, they have to be scientifically sound to differentiate between species. (ii) The user does not know how positive reaction results were collected and expressed mathematically. As many strains as possible from different locations of the same species should be tested. This is especially important when assessing strains with so-called variable, delayed, or weak test results. (iii) There is no information available that estimates the statistical reliability of a computer-assisted diagnosis. Analyses are needed to determine the necessary and sufficient number of data points entered, the redundancy of the data input, and the tolerable level of input errors. (iv) Probabilistic identification systems have a rigid arrangement, and so they ignore the prevalence or rarity of a given microbial species (7). Depending on how the identification characteristics in the data bases were collected and how probability was computed, each software package has "favorites." An unknown isolate may have the same positive test results as several species in the data base. The program, however, will identify the isolate as belonging to the species with the highest positive test percentage. Very often higher positive test percentages are, in turn, due to the frequency of occurrence of a given species. (v) None of the probabilistic identification programs has an option for recognizing new species, and the unknown isolate will be identified (misidentified), although at a low score. We learned from our experience with yeast identification that, for taxonomic purposes or for definitive identification (20), standard yeast identification methods have to be used. At this point, we prefer the work of Kreger-van Rij (19). For routine or presumptive (20) yeast identification in the food industry, we recommend highly the simplified identification scheme developed by Deak and Beuchat (10). Adapting the tests from Deatk's newly released software and revising it based on the available standard works (4, 19) will make both the simplified identification key and the computer program more accurate and user friendly.
IDENTIFICATION OF FOOD-BORNE YEASTS
1211
ACKNOWLEDGMENTS We express our gratitude to T. Deak and 0. Reichart for friendly permission to use their then-unpublished software packages. REFERENCES 1. Analytab Products. 1986. YEAST-IDENT directory for the biochemical identification of yeast and yeast-like organisms. Product no. 8886-436016. Analytab Products, Plainview, N.Y. 2. Analytab Products. 1988. API 20C. Analytical profile index yeast, p. 43-103. Analytab Products, Plainview, N.Y. 3. American Society for Microbiology Computer User Group. 1989. Computer-assisted identification of microorganisms (CAIM)yeasts. American Society for Microbiology, Washington, D.C. 4. Barnett, J. A., R. W. Payne, and D. Yarrow. 1983. Yeasts: characteristics and identification. Cambridge University Press, Cambridge. 5. Barnett, J. A., R. W. Payne, and D. Yarrow. 1985. Yeast identification program. Cambridge Micro Software. Cambridge University Press, Cambridge. 6. Bayes, T. 1763. An essay toward solving a problem in the doctrine of chance. Philos. Trans. R. Soc. London 53:269-271. 7. Berger, S. A. 1990. Lack of precision in commercial identification systems: correction using Bayesian analysis. J. Appl. Bacteriol. 68:285-288. 8. Dea4k, T. 1986. A simplified scheme for the identification of yeasts, p. 278-293. In A. D. King, Jr., J. I. Pitt, L. R. Beuchat, and J. E. L. Corry (ed.), Methods for the mycological examination of food. Plenum Publishing Corp., New York. 9. Dealk, T. 1990. Yeast-ID (software available upon request). University of Horticulture and Food Industry, Budapest, Hungary. 10. Deak, T., and L. R. Beuchat. 1987. Identification of foodborne yeasts. J. Food Prot. 50:243-264. 11. Dealk, T., and L. R. Beuchat. 1988. Evaluation of simplified and commercial systems for identification of foodborne yeasts. J. Food Microbiol. 7:135-145. 12. De Hoog, G. S., and E. J. Hermanides-NiJhof. 1977. Survey of the black yeasts and allied fungi, p. 178-223. In Studies in mycology no. 15. Centraalbureau voor Schimmelcultures, Baarn, The Netherlands. 13. Dybowski, W., and D. A. Franklin. 1968. Conditional probability and the identification of bacteria: a pilot study. J. Gen. Microbiol. 54:215-229. 14. Hermanides-Nijhof, E. J. 1977. Aureobasidium and allied genera, p. 141-177. In Studies in mycology, no. 15. Centraalbureau voor Schimmelcultures, Baarn, The Netherlands. 15. Huxsoll, C. C., H. R. Bolin, and A. D. King, Jr. 1989. Physicochemical changes and treatments for lightly processed fruits and vegetables, p. 203-215. In J. J. Jen (ed.), Quality factors of fruits and vegetables. Chemistry and technology. American Chemical Society, Washington, D.C. 16. Kelley, R. W., and S. T. Kellogg. 1978. Computer-assisted identification of anaerobic bacteria. Appl. Environ. Microbiol. 35:507-511. 17. Kellogg, S. T. 1979. MICRID: a computer-assisted microbial identification system. Appl. Environ. Microbiol. 38:559-563. 18. Kirsop, B. E., K. A. Painting, J. E. Henry, M. Fernandes, E. H. Prescott, I. Braid, and P. Foster. 1986. On-line computer assisted identification of yeasts, p. 75. In XIV International Congress of Microbiology, Abstracts. 19. Kreger-van Rij, N. J. W. (ed.). 1984. The yeasts, a taxonomic study, 3rd ed. Elsevier Science Publisher, B.V., Amsterdam. 20. Kurtzman, C. P. 1988. Identification and taxonomy, p. 99-140. In B. E. Kirsop, and C. P. Kurtzman (ed.), Living resources for biotechnology. Yeasts. Cambridge University Press, Cambridge. 21. Lapage, S. P., S. Bascomb, W. R. Willcox, and M. A. Curtis. 1973. Identification of bacteria by computer: general aspects and perspectives. J. Gen. Microbiol. 77:273-290. 22. Lin, C. C. S., and D. Y. C. Fung. 1987. Conventional and rapid methods for yeast identification. Crit. Rev. Microbiol. 14:273289. 23. Moore, W. E. C., and L. V. H. Moore (ed.). 1989. Index of the
1212
TOROK AND KING
bacterial and yeast nomenclatural changes. American Society for Microbiology, Washington, D.C. 24. Pankhurst, R. J. (ed.). 1975. Biological identification with computers. Systematics association special vol. 7. Academic Press, Inc., New York. 25. Pincus, D. H., I. F. Salkin, N. J. Hurd, I. L. Levy, and M. A. Kemna. 1988. Modification of potassium nitrate assimilation test for identification of clinically important yeasts. J. Clin. Microbiol. 26:366-368.
APPL. ENVIRON. MICROBIOL.
26. Reichart, 0. 1990. Yeast identification program (software available upon request). University of Horticulture and Food Industry, Budapest, Hungary. 27. Rohm, H., and F. Lechner. 1990. Evaluation and reliability of a simplified method for identification of food-borne yeasts. Appl. Environ. Microbiol. 56:1290-1295. 28. Wilcox, W. R., S. P. Lapage, S. Bascomb, and M. A. Curtis. 1973. Identification of bacteria by computer: theory and programming. J. Gen. Microbiol. 77:317-330.
