Following recent reviews on the setting and the meeting of standards for the efficient replacement of pronutrient antibiotics in pig and poultry nutrition (Rosen, 2003b; 2003c), this review concentrates attention on ways and means of improving and optimizing the use of pronutrient antibiotics in broiler and turkey feeds, with particular reference to the multiplicity of proffered candidates and to the multiplexity and interactivity of influential genetic, environmental, managemental and dietary variables.
Why replace pronutrient antibiotics?
Contentiously but concretely, the search for and validation of antibiotic replacements stems from the attitudes and demands of consumers and their retail suppliers and from legislation based on precautionary principles. Soundly-based estimates of the effects on production costs and on retail prices are sparse. But the number of products offered as alternatives and the volume of literature thereon have risen steeply since 1999. Many hundreds, possibly thousands, of products are on offer. One can envisage several years’ work ahead to sort the wheat from the chaff.
Irrespective of questions of right or wrong in the proscription of prescription-free usage of antimicrobials in food production, the need to deploy replacements is expanding rapidly, creating a formidable task for scientists and producers in their search for fully-effective alternatives. The potential value of feed antibiotics was first demonstrated in the US by Moore et al. (1946) before commercialization in 1949. Potential problems caused by bacterial resistance phenomena were early sources of concern in human medicine, veterinary medicine and farm animal nutrition. In the early 1950s, opponents of antibiotic routines in poultry feeds initiated resistance research. Concerns were heightened by the Japanese discovery in 1959-1960 of infectious or transferable (as against natural) antibiotic resistance involving natural selection, mutation and DNA fragment transfer. The Swann Committee (1969) reported on the use of antibiotics in animal husbandry and veterinary medicine and initiated legislative restrictions banning the use of penicillin and tetracyclines without veterinary prescription. The European Economic Community followed this lead. The US Food and Drug Administration did not. In 1998, the European Union cancelled its approvals of six feed antibiotics, with intent to ban the remainder not later than 31 December, 2005.
Which are the candidates?
Table 1 contains a short list of the seven main categories of antibiotic replacement candidates. Whatever their nature or chemical composition, and no matter how numerous and multifarious are their known or hypothesized modes of action, the common thread in the present context resides in their abilities to enhance poultry performance at least as efficiently as the replaced pronutrient antibiotics in terms of feed conversion efficiency, mortality, liveweight, animal product yield and environmental depollution, measured integrally as net return on investment. The plethora of proffered candidates is evidenced, by way of example, in the 2000-01 Direct-Fed Microbial, Enzyme and Forage Additives Compendium (Miller Publishing Co.), by its lists of 222 candidate products, 90 enzymes, 72 microbials, 40 yeasts, 13 moulds and acids and seven oligosaccharides for use in poultry production alone.
Table 1. Major antibiotic replacements.
Nutrients are also included because supplementary amino acids, minerals, purified energy sources and vitamins overlap performance-wise (Rosen, 1997) and may well overlap with pronutrients in nutritional improvements, irrespective of differences in modes of action.
Is nomenclature satisfactory?
More meaningful, standardized, transparent terminology in this field would benefit scientists, legislators and, above all, consumers. Feed additive antibiotics have been variously named as growth promoters, growth permitters, performance promoters, performance enhancers, production aids, feed economizers, nutrient balancers, efficiency enhancers, digestive enhancers and nutrition improvers, which descriptors have often been regarded in several senses as inaccurate, incomplete or inappropriate. Notwithstanding long-term, nearuniversal usage, the word ‘additive’ is nonetheless ill-conceived, being insufficiently descriptive. In practice the term ‘additive’ can generate auras of minority, subsidiarity, afterthought, optional extra and contaminant. It may be suitable for fuels, but not for food or feed; but it is doubtful whether legislators would ever countenance a change.
The term ‘pronutrient’ has been introduced to replace additive. A pronutrient is defined as a substance that improves the value of nutrients. As anticipated, a survey of 100 consumers has confirmed it as unknown to them, but 85 interpreted it as ‘something good for nutrition’. Therefore it is suggested that pronutrient could usefully replace the term “non nutrient additive” used by the National Research Council to distinguish an additive from a nutrient. However, “non nutrient additive” is unsatisfactory because it fails to convey the nutritional benefits of additives, which can be tantamount to those of nutrients.
The role of a pronutrient is portrayed in the schema in Figure 1. This shows the functional relationship of a pronutrient in nutrition relative to that of an antinutrient.
Figure 1. Effects of pronutrients and antinutrients on diets.
Pronutrients function via a hundred or more different modes of action in extending the value of the limiting nutrient in a diet. There is thus an overlap between the provision of a pronutrient and a ‘topping-up’ of a limiting nutrient. Hence, nutrients are, as aforementioned, included as a Table 1 category for consideration herein. The choice of a pronutrient or a nutrient in replacing feed antibiotics is simply a question of relative cost-effectiveness.
The terms ‘probiotic’ and ‘prebiotic’ are both plagiarisms, appropriated by scientists and suppliers. The US Food and Drug Administration and the European Union Commission, as regulatory authorities, both refused them as vagaries, opting respectively for direct-fed ‘microbial’ and ‘microorganism’. The term probiotic was, as a matter of fact, first coined by Winter (1955) in his research on botanical antibiotics found in cruciferous plants, when he stated that ‘we can call these substances probiotics; they are antibiotic against pathogenic microbes and they are therefore probiotic for the infected organism’. Gibson and Roberfroid (1994) took the ‘ prebiotic concept’ from its original and literally-correct meaning and long-standing usage to define a chemical involved in the origin of life. Prebiotic is incorrect as an appellation for a biosynthesized molecule, which is biofunctional, which nourishes a live microorganism and which inhabits a live host, in no sense whatsoever, before life. The adoption of simple, realistic duplex descriptors to simultaneously impart nature and function could be a useful forward step in this field. For example, we could specify (a) pronutrient cellulase, formate, B. cereus, capsicum, etc.; (b) prophylactic narasin, nifursol, P. acidilactici, mushroom polysaccharide, etc., (c) therapeutic tylosin phosphate, lincomycin hydrochloride, penicillin, trimethoprim, etc. and (d) pro-environmental phytase, carbohydrase, protease, etc. Furthermore, it would be advantageous to discontinue reference to ‘antibiotic alternatives’ or ‘antibiotic substitutes’, thereby avoiding a rekindling of consumers’ scientifically-unproven, anti-antibiotic prejudices.
How are antibiotic replacements compared?
All available properly-controlled test data need to be taken into account. The use of a handful of tests can illustrate the potential of a product as a starting point, but five tests cannot take account of the wide range of genetic, environmental, managemental and dietary factors affecting response in praxis. A recent survey on the current status and future needs of replacements based on the view of 50 suppliers, users, consultants, educators, communicators and academics revealed a large number of problems met in comparing candidate efficacies (Rosen, 2003c). Of the 92 nominated, the main problem areas are variation in response, use of uncontrolled tests, inadequate test designs, missing feed compositions, invalid ‘field’ tests and unjustified dosage recommendations.
Serious shortcomings in commonly-used averaging procedures are illustrated in Table 2. For example, the comparison of the effects on feed conversion ratio (FCReff) of enzymes and antibiotics, as such or in percentage terms, cannot be meaningful due to their 11-day mean duration difference. It is known that FCReff diminishes through starter to finisher phases. The same applies to time span (year) differences, averaging 1972 and 1987. The coefficients of variation in response of 129-1,449% constitute a warning against comparisons based on small test numbers.
For some researchers, direct comparisons within tests are fundamental, but they are, more often than not, too expensive. They require much larger, more costly experiments to manifest statistically significant differences between pairs of candidates, which differences are normally much smaller than those between candidates and negative controls.
Table 2. Superficial comparison of the effects of antibiotics, enzymes and microbials (FDIeff, LWGeff, FCReff and MORTeff) on control feed intake (FDIC), control liveweight gain (LWGC), control feed conversion ratio (FCRC) and control mortality (MORTC) with their coefficients of variation (CV) in broiler nutrition.
2Percentage of tests with improved FCReff and LWGeff.
*DUR = duration of test.
Comparative tests versus a positive control alone are contraindicated because they provide no evidence of a beneficial response to either candidate. Table 2 also starkly indicates a crucial gap in the large majority of tests, which fail to report mortality, viz. 71%, 83% and 76%, respectively for antibiotics, enzymes and microbials. Interestingly, the response improvement frequencies for these three candidate categories are virtually equal in the range 70-75%. Frequencies of beneficial response and magnitudes of variation therein merit greater attention than hitherto.
How should we determine efficacy?
In the light of the aforementioned shortcomings of few tests and superficial averaging, the answer to this question resides in the formulation of optimal multifactorial empirical algebraic models for nutritional effects of each candidate. Such models are used to calculate a requirement for any given set of circumstances and conditions, in order to compare an antibiotic and a candidate or to compare two or more candidates. Essentially, all published data are accepted. Manifestly-uncontrolled tests are excluded. Computer filing of all relevant data for the hundred or more potentially-important variables includes routines for elimination of errors in the original reports or in data abstraction and repeats. Performance must be measured from start to finish. If second or later phase data are required they are obtained, e.g., by subtracting 0-21 day from 0-42 day values. Standard statistical packages, e.g. Nie et al. (1975) and multiple regression methodology, e.g. Draper and Smith (1981) are used to determine best-fit models.
Best-fit algebraic models have statistically significant regressions, maximum multiple correlation coefficient squares, minimum standard deviations about regression and significant partial regression coefficients for all independent variables. After exclusion of aberrant (>3 x SD) responses, the emergent models are used to estimate nutritional responses and 95% confidence limits for any required set of values of the component independent variables. Differences in these predicted responses are tested for statistical significance. Nutritional response estimates are then used to compute and compare net profits to target liveweight and/or target duration.
These procedures are then used in feed formulation to quantify specific requirements for pronutrients or nutrient counterparts for target production. Thus one can assess whether a nutrient, e.g. L-methionine, would be a better choice than a pronutrient, e.g. endoxylanase or Bacillus subtilis. In other words, should one top up a limiting nutrient per se or increase the amount of the dietary limiting nutrient?
Which are key variables?
The 48 factors listed in Table 3 exemplify the range of variables potentially relevant in the elaboration of working models. There are also subsets of these factors to be considered. Routinely dose is considered as linear, quadratic, logarithmic or exponential. For broilers there are four sex types: male, female, mixed (50/50) and as-hatched. Ten to 20 antibiotics and five to 10 anticoccidials may be relevant, alone or in admixtures. Different disease challenges can be identified. The data emanate from more than 100 countries, though 10 or so usually furnish most of the test data. A few individual brands have numbers worthy of test. Types of oils and fats, animal protein, and vegetable proteins, including admixtures, total more than 200. For example, in Brozyme (Rosen, 2000; 2002) there are, including mixtures of each type, 71 animal proteins, 15 vegetable proteins, 81 oils and fats and 25 carbohydrate sources.
Based on experience to date, level of negative control performance, duration of feeding and year of test are basic, accounting for 10-50% of variation in effects. Contrastingly, some highly statistically significant factors can account for 1% or less. Interactions as product terms are normally relatively small contributors. In models containing 10 or more significant variables, dosage tends to be less important, often accounting for 7.5% or less of total variation.
Table 3. Independent genetic, environmental, management, dietary and nutrient variables used in the elaboration of multifactorial nutritional models.
1Second antibiotic/enzyme/acid/microbial/other pronutrient/nutrient
2Dietary concentrations (columns 3 and 4)
3Digestible or metabolizable energy are alternatives.
Some independent variables are vital for the assessment of praxis values, compared with research conditions, namely, feed form, use of not-day-olds, males, cages, part-purified diets and specific disease challenges.
If the treated value is used as the dependent variable in modelling instead of the effect, multiple correlation coefficient squares (R2) are grossly inflated by the high correlation of control values and effects, e.g. for virginiamycin models (n=306), LWGeff and FCReff R2 values of 0.14 and 0.44 respectively, become 0.97 for LWG and 0.99 for FCR as treated dependent variables. Such inflation of R2 is misleading in variation accountancy and for some it harbours a delusion that R2 less than 0.4-0.6 is unworthy.
Incomplete test reports limit the accountancy of variation. Recent data on phytase responses in broilers offer a useful perspective in which models containing 20 significant independent variables account for 64- 72% of the sizeable variations in nutritional responses with time and place (Rosen, 2003a). Further progress, however, is unlikely until editors of peer-reviewed journals take a lead, for others to follow, in ensuring routine publication of fundamental variables such as temperature, altitude, lighting pattern, disease status and mortalities.
As examples, Figure 2 contains basal models for a group of five important broiler antibiotics in order to illustrate the magnitudes of the effects of key variables, with algebraic signs in accordance with nutrition science and practice.
These models contain 19 significant independent variables. For LWGeff and FCReff, negative coefficients quantify inferior and superior respective response contributions, with increase in LWGC and FCRC. Their positive DURs afford better and worse effects respectively with age increase. Diagnosed or endemic disease enhances LWGeff and FCReff. The partial regression coefficients of LWGC and FCRC mean that each 100 g better control performance would reduce LWGeff by 1.1 g and that each 10 point lower value for FCRC would reduce FCReff by 1.6 points.
How many tests are needed for a working model?
Hitherto, the notional n = c. 50 controlled tests has been thought of as a minimum required to produce a useful model to quantify the magnitudes of nutritional responses in feed formulation. A recent project addressed the question of a minimum platform via random fragmentation of a large 1,709/mortality 708 test resource of five of the most important feed antibiotics (i.e. two bacitracins, two tetracyclines and virginiamycin) into smaller subsets down to 34-43 for the elaboration of progressively smaller-based models (Rosen, 2004). The parent models are those in Figure 2. The analysis of a total of 704 fragmented resource models revealed that (a) no model at all was afforded in 23% of the random fragments; (b) progressive subset fractionation of the data set sharply reduced the number of significant variables from 79% in the parent models to 17% in the smallest, which averaged 1.2 significant independent variables; (c) the chance of three or more significant variables in subset models was one in eight; and (d) the smallest sets, averaging n = 38/mortality 15, did in toto reveal all the significant variables, albeit very patchily, found in the parent models.
Figure 2. Antibiotic models relating feed intake, liveweight gain, feed conversion ratio and mortality effects to level of control performance, duration, date, dosage, anticoccidial use and disease status.
How are requirement models used?
Requirements are needed for each and every set of circumstances for a defined target objective, for nutrients per se or for pronutrients. In other terms, what response might we expect within what confidence limits for a given dosage of supplementary methionine compared with an optimal dosage of bacitracin, methylene disalicylate, or 6-phytase or a mannan oligosaccharide, etc.? The conjoint (overlap) consideration of the value of a limiting nutrient supplement or the improved efficiency of a limited nutrient supply with a pronutrient is of particular interest for the comparison of multifunctional substances. This approach could well resolve the ongoing, decades-old dispute on the relativities of DL-methionine (DLM) and methionine hydroxyanalogues (HMTBA and CaHMTB). The use of requirement models based on all-available, unselected control test data for DLM, HMTBA and CaHMTB with dose expressed simply as product weight (not molar or other equivalencies in malnutrition tests) should be decisive. Such methodology is imperative because DLM is a mixture of a nutrient and a prenutrient and HMTBA and CaHMTB are bifunctional as prenutrient and pronutrient.
The mode of application and value of multifactorial models for antibiotics and their replacements can next be illustrated in four examples as follows.
Using zinc bacitracin models based on 1,164 tests and first-generation phytase models (n=296), Table 4 shows that zinc bacitracin at 80 ppm affords a five point conversion improvement compared with the isocost level of phytase with no effect. In addition, a huge increase in phytase dosage (x 20) gives a three point conversion improvement, for which 42 ppm of a zinc bacitracin would suffice.
Table 4. Iso-input cost and iso-feed conversion effect comparisons of zinc bacitracin (n=1164) and first-generation phytases (n=296) for as-hatched 56 day-old broilers of LWGC=3,266g and FCRC=2.079.
Secondly, Table 5 provides a model-based perspective on the results of a test from 7 to 28 days of age on female broiler chicks, which compared a Chinese herbal preparation, virginiamycin and a negative control (Guo et al., 2000). The herbal at 938 ppm afforded an FCReff four points inferior to virginiamycin at 20 ppm, even though the latter manifested, in this test, a response three points below its predicted result.
Table 5. Application of LWGeff and FCReff models to assess the results of a comparative test on a Chinese herbal formulation vs virginiamycin for LWGC=1,098 g and FCRC=1.554 on 7-28 dayold female birds.
* 306-test requirement models
The import and value of a negative control is further evidenced in a 39-day broiler test by Messikommer on 50 ppm zinc bacitracin versus a rhubarb supplement (Wenk, 2000). The predicted zinc bacitracin response of -0.074 ± 0.026 illustrates that the observed value of -0.049 is as expected; and that the botanical has potential value at 2,500 ppm (-0.038), but not at 5,000 ppm (+0.018).
Thirdly, truncated comparative tests without a negative control are preferred by some researchers, even though they are unable to confirm efficacy for either test product. An assessment of the results of a positive control only test can be made using a model for the test antibiotic. Syrvidis et al. (2003) reported a test on Digestarom 1317 Poultry Premium (a socalled phytobiotic) versus Flavomycin-80, but failed to include test dosages. Hence, a notional negative control performance of LWGC = 2,125 g and FCRC = 1.970 was calculated from flavomycin models (n=394) for a presumed dosage of 2 ppm. The computed negative control values reveal very large Digestarom responses of LWGeff = 326 g (15.3%) and FCReff = -0.220 (-11.2%). At today’s production standards, however, a feed conversion as-hatched for a 40-day 2,125 LWGC would be about 1.730, i.e., 12.2% less than the computed FCRC of 1.97 at 42 days old in this trial. Such a result against a positive control alone should be treated with reserve. The availability of further tests against negative controls would afford a better view on the potential of Digestarom. The same applies to any product if it claims value from one or more tests solely against a positive control(s).
Fourthly, even prior to the availability of working models for individual product brands of replacements, test data collections of the latter can be reviewed as to their potential value as antibiotic replacements.
Bio-Mos® mannan oligosaccharide collections of 34 tests for broilers (Hooge, 2003a) and 27 tests for turkeys (Hooge, 2003b) are used to provide an example. Table 6 compares mean liveweight gain and feed conversion responses in broilers and turkeys with corresponding predictions for equivalent circumstances, using comprehensive multifactorial models for the antibiotic products virginiamycin (n=306) in broilers and zinc bacitracin (n=226) in turkeys.
Table 6. Comparison of mean LWGeff and FCReff in 34 mannan oligosaccharide (Bio-Mos®) broiler tests with predicted responses for virginiamycin for LWGC=2,149 g, FCRC=1.879 at 42.2 daysold and in 27 turkey tests with predicted responses for zinc bacitracin for LWGC=5,643 g, FCRC=1.981 at 68.7 days.
The Table 6 data are encouraging for Bio-Mos® as a replacement when used at a dosage of 1 g/kg in broiler and turkey feed. More searching comparisons, however, must now await the availability of requirement models for Bio-Mos®, quantifying the influences of dosage, level of bird performance, presence or absence of other pronutrients, especially anticoccidials, bird sex, feed form and limiting nutrients.
By virtue of extensive 15-year test programmes, exogenous enzymes would appear at present to be the best characterized replacement category. The data available for organic acids in pigs, microbials and oligosaccharides in poultry and pigs may already suffice for the elaboration of initial working models. But all available data should be used to avoid possible selection bias, as in the organic acid studies of Partanen and Mroz (1999) and Partanen (2001).
How should we test admixtures?
Problems in the use of admixtures of antibiotic replacements arise from shortcomings in nomenclature and posology (dosage science). ‘Additivity’ and ‘synergism’ are often misused, usually for sub-additivity. The possibility of antagonism should always be borne in mind. For purposes of definition, the possible effects of admixture are classified herein as sub-additive, additive, synergistic, ineffective or antagonistic, as defined for the admixture of 2+3 of A and B providing 4, 5, 6, 2 or 3 and 1 unit(s) of response respectively. A 2 x 2 factorial test is a good starting point, shown in Table 7, which also includes other iso-cost admixtures for A or B alone or for higher single dosages of A or B. The acid test for maximal results uses A+B each at its economic optimum.
Table 7. Scope of admixture tests.
* MU money units
Nutritional models can provide a useful guide towards maximum admixture efficiency by pre-determination of the economic optimal doses of Replacements A and B. These admixture guidelines are also pertinent in situations where pronutrient antibiotics can still be utilized in admixture with nutrients or pronutrient supplements.
The thesis presented herein essentially advocates that we should optimize the choice and dosage of nutrient or pronutrient antibiotic replacements by taking cognizance of all available data expressed in predictive, empirically-based, multifactorial multiple regression requirement models for feed, gain, conversion and mortality effects, at least. Such models should be updated annually to take account of the accelerating current flow of scientifically-controlled feeding tests, as seen for example, in the increase of exogenous enzyme publications from 1,422 up to the year 2000 rising to its latest content in the Brozyme resource (Rosen, 2002) of 2,175. It is intended also to extend these studies to table egg and breeder hens and at least ducks among the minor species.
Following the lead of this Symposium in ‘reimagining the feed industry’, should we not also prune and improve its verbiage to better its science and raise its transparency to consumers, setting a good example for all members of the food chain?
In conclusion, it may be apposite, in line with the question and answer format of this review, to conclude with a Seven Question Test with which the user of an antibiotic replacement can assess the potential value of a supplier’s Product X.
1) How many properly-controlled feeding tests do you have on the efficacy of Product X?
2) How many of these have no negative controls?
3) Can you supply a bibliography for 1) and 2)?
4) How many times out of ten does Product X improve liveweight gain and feed conversion?
5) What are the coefficients of variation in the gain and conversion responses?
6) What dosage of Product X will maximize return on my investment and why?
7) Can you supply me with a model to predict responses to Product X under my conditions?
Receipt of answers of 1) 30; 2) five; 3) yes; 4) seven; 5) 100-200%; and 6) “x ppm because . . .” should be encouraging. A ‘yes’ to 7) would be even better.
Author: GORDON D. ROSEN
Draper, N.R. and H. Smith. 1981. Applied Regression Analysis, 2nd Edition. John Wylie and Sons, Inc., New York.
Gibson, G.R. and M.B. Roberfroid. 1994. Dietary modulation of the human colonic microbiota: introducing the concept of prebiotics. J. Nutr. 125:1401-1412.
Guo, F., R.B. Kwakkel and M.W.A. Verstegen. 2000. The use of Chinese herbs as alternatives for growth promoters in broiler diets. In: Abstracts and Proceedings, XXI World’s Poultry Congress, Montreal, Canada, Aug. 20-24. CD-Rom, pp. 2.
Hooge, D.M. 2003a. Broiler chicken performance may improve with MOS. Feedstuffs, January 6, pp. 11-13.
Hooge, D.M. 2003b. Dietary MOS may have application in turkey diets. Feedstuffs 75(18):11- 13. p.42.
Moore, P.R., A. Evenson, T.D. Luckey, E. McCoy, C.A. Elvehjem and E.B. Hart. 1946. Use of sulphasuccidine, streptothricin and streptomycin in nutrition studies with the chick. J. Biol. Chem. 165:437-441.
Nie, N.H., C. Hadlai-Hull, J.G. Jenkins, K. Steinbrenner and B.H. Bent. 1975. In: Statistical Package for the Social Sciences, 2nd Edition. McGraw Hill Book Company, New York.
Partanen, K. 2001. Organic acids - their efficacy and modes of action. In: Gut Environment of Pigs (A. Piva, K.E. Bach Knudsen and J.E. Lindberg, eds). Nottingham University Press, UK, pp. 79-96.
Partanen, K.H. and Z. Mroz. 1999. Organic acids for performance enhancement in pig diets. Nutr. Res. Rev. 12:117-145.
Rosen, G.D. 1997. In: Proceedings of the 16th Scientific Day, Southern African Branch of the World’s Poultry Science Association (D.G. Poggenpoel, ed). October 16, 1997. pp. 60-79.
Rosen, G.D. 2000. Enzymes for broilers: a multifactorial assessment. Feed Int’l. 21(12): 14-18.
Rosen, G.D. 2002. Exogenous enzymes as pronutrients in broiler diets. In: Recent Advances in Animal Nutrition 2002. (P.C. Garnsworthy and J. Wiseman, eds). Nottingham University Press, UK. 89-103.
Rosen, G.D. 2003a. Effects of genetic, managemental and dietary factors on the efficacy of exogenous microbial phytase in broiler nutrition. Brit. Poultry Sci. 44S1:S25-S26.
Rosen, G.D. 2003b. Pronutrient antibiotic replacement standards discussed. Feedstuffs 75(30):11-13. p.16.
Rosen, G.D. 2003c. Setting and meeting standards for the replacement of pronutrient antibiotics in poultry. Proceedings of the 30th Annual Carolina Poultry Nutrition Conference, Carolina Feed Industry Association, Research Triangle Park, October 30. pp. 69-79.
Rosen, G.D. 2004. How many broiler feeding tests are required to elaborate working models for the prediction of nutritional effects of pronutrients in praxis? Brit. Poultry Sci. In press.
Swann Committee. 1969. Joint Committee on the use of Antibiotics in Animal Husbandry and Veterinary Medicine Report (Cmnd. 4190). Her Majesty’s Stationary Office, London. p. 83.
Syrvidis, V., R. Bobiniene, V. Priudokiene and D. Vencius. 2003. Phytobiotics add value to broiler feed. World Poultry 19(1):16-17.
Wenk, C. 2000. Herbs, spices and botanicals; ‘Old fashioned’ or the new feed additives for tomorrow’s feed formulations? In: Biotechnology in the Feed Industry, Proceedings of Alltech’s 16th Annual Symposium (K.A. Jacques and T.P. Lyons, eds). Nottingham University Press, UK, pp. 79-96.
Winter, A.G. 1955. The significance for therapy and diet of antibiotic substances from flowering plants (with special reference to nasturtiums, cress and horseradish). Die Medizinische Nr. 2:73-80.
Pronutrient Services Ltd., Wimbledon, London, United Kingdom