A new method of optimization of the medium composition by biostatistical methods is reported. The well-known BOX-WILSON scheme combined with the method of optimum curves was applied for an exploration of the response surface. The primary goal has been to optimize the concentration of nutrients for the production of gibberellic acid by the fungus Gibberella fujikuroi with a minimum number of experiments. As the result, a nearly fivefold improvement yield of the was achieved performing only two experiments.Gibberellic acid (GA,) and some other biological active gibberellins are produced by the phytopathogenic fungus Gibberella fujikuroi (SAW) WOLL. (Fusarium moniliforme SHELD) when grown in submerged culture in nitrogen-limited media.Gibberellins are typical secondary metabolites. The phases of growth can be discerned, described and related to the nutritional and environmental state of the fermentation (BORROW etal., 1961, 1964, VASS and JEFFERYS, 1979. -Significant production ,of gibberellins starts only after nitrogen exhaustion. On the other hand, production is greater the higher the value of initial nitrogen over a wide range. Further increases in nitrogen concentrations lead to a decrease of the mycelium productivity (BORROW et al., 1964). Therefore, the selection of both the initial concentration of nitrogen and of an optimal C/N ratio are the most important factors.The object of this paper was the improvement of the medium composition by mathematical methods for planning the experiment.In an unconventional manner the Box-WILSON scheme is used for the optimization of a nutrient medium composition consisting of four nutrients -corn steep liquor (CL), ammonium sulfate (AS), sunflower oil (SO), and potassium dihydrogen phosphate (PP). Originally, the BOX-WILSON method was developed for the optimization of chemical processes (Box and WILSON 1951). To optimize a biological response such as the biosynthesis of gibberellins and other secondary metabolites, some modifications according design and analysis are needed to make the BOX-WILSON procedure more effectively. A combination with the method of optimum curves developed by SCHRODER and WEIDE (1973) seems to be very helpful to avoid an overdimensional expansion of experimental effort. In the following by the example of gibberellic acid formation an appropriate methodical basis is given for the application of factorial designs in fermentation. As the rule a consulting statistician solving problems in fermentation is confronted with four problems which are typical for the optimization of biological response. I The great variation of experimental results after repetition; I1 The bad reproducibility of fermentation experiments over the time; I11 The overdimensional effort of the factorial design (2k or 2k-p) as a consequence of the high-dimensional parameter space; IV The violation of the smoothness hypothesis underlying the response surface technique. 16'

If the number of potentially relevant compounds which have to be optimally selected is large, then random balance experimentation provides a useful tool for the development of nutrient media.The object of investigation was a thiostrepton producing Srreptomyces strain from our own collection. The thiostreptone formation was increased although we had insufficient information about the physiological needs or regulation aspects concerning the biosynthesis. The medium development was carried out alongside mutation experiments within a limited temporal interval.Our study shows how a 9-component medium can be improved by a 6-point experimental design. At the end of the paper the basic principles for constructing restricted random balance designs are outlined.One of the most common operations to study the physiology of micro-organisms is the development of a medium, optimal for the parameters in which one is interested, with regard to cell yield, the yield of a metabolic product, or the related enzyme level in the cell, Although our knowledge of the cellular regulation and the mechanisms of biosynthesis is generally good, each special case demands as a rule empirical practice.Since medium-screening is essentially an optimization process, it was considered worthwhile to investigate the use of response surface methodology in this situation.This technique is a collection of mathematical and statistical methods which have been developed and used to aid the solution of particular types of problems, pertinent to scientific and engineering processes. The methods include experimental designs, statistical inference, and mathematical optimization methods, which when combined enable the experimenter to make an efficient empirical investigation of the processes in which he is interested.This approach to exploring the relationship between variables has, to date, found its best application in industry. In fact, its modern development was motivated by problems encountered in the chemical industry (Box and WILSON 1951). Typically, in this area the problem is one of optimizing the yield of a particular process. In such situations, for example, it is known that the yield of the process is related to the levels of the input variables, and it is of interest to determine the level of each of these variables that produces an optimum yield.The essence of a response-surface approach is that a mathematical relationship (model) exists between the levels of treatment and the outcome observed. Thus, one generally works with causal models where the levels of treatment or design variables can be controlled by the researcher.In Fig. 1 a classification of response surface techniques is given. The so-called "systematic approach" is based on the following philosophy: Starting from a base point, corresponding to the best prior estimate of the optimum conditions, the first step is to construct and carry out a small initial group of experiments around this base point, in order to estimate tlie main effects. The best design here is the 2 factorial arrange...

The optimization of a fermentation model is realized by the procedure EVOL based on the principles of the biological evolution. An application of this method for a typical secondary metabolit fermentation model is given. The fermentation process is considered as a black-box. "Mutation" and "selection" steps are carried out. As the objective function the economic profit is calculated. After about 80 "evolution" steps the global optimum will be reached. The scope and limitations of this method are discussed.

Suvz~nary:The met,hod is based on t,he determination of the number of the nonmot~ile sperms and/or the lethally damaged sperms with the aid of the fluorescent dye primuline. Incubation t,akes place a t a suspension densit'y of 10.OOO/minl ... 38.000/nimJ, at 4OoC or 46 "C for 15 ... 60 min.During the test wit'h sublimate t'he loss of motilit'y amounts to 55 oi0 at 4 mgil Hgzf, t,he LC,, is 7.3 mg/l Hg2+. Phenyl-mercuric acetate causes t,he total loss of motility with 20 mg/l, the LCE, is above 168 mg/l (equivalent to 100 mg/l Hg"). The LC,, of sodiumpentadecylmonosulphonate and dodecylpyridinchlcridc are 11.0 and 15.8 mg/l, resp. Filtrates of blooming of cyanophyceae on Baltic, having been digested by repeated freezing and thawing, with 1.5 ... 0.4 g/l dry matter showed moti1it.y losses of 30 ... 100 n, ' o for Nodctlaria spurnigena, whereas no significant, effect was produced by a filtrate of Microcystis aerlrginosa with 28.2 g!l dry matter as well as waterblooming of Noddoria spurnigena from the Small Jasmund Bodden. Einf uhrung

Paper given ;it the 2nd Symposium of Socialist Countries on Biotechnology, L+eip;.ig 2 -5 . 12. 1 9~tThe niethod of steepest ascent, introduced by BOX and WILSON [21, which is frequently used in the field of the optimization of microbial fermentations, is modified in order to overcome several difficulties in connection with planning and analyzing experimerrts (RECKNACEL and BOCKER [3]; BOCKER and RECKNAGEL [ t]).The 2"-factor designs are replaced by the (2n + 1)-spectrum (Tab. I), which has to be tribnsfort>ied in the llsrlill way before applying them to d practical problem (Fig. 1). 'rith. 1. ( 2 . 3 -tl)-factoriiil designFrom the outcome of experiment the coefficients of the linear model ?/ = b, + bpz, + b2x2 + *.. + bsxs are estimated by means of the method of least deviations (MLD, Ll-approximation), instead of least squares (MLQ) (Tab. 2).Some experimental steps in the direction (b,, . . ., b,) of steepest ascent (Fig. 2) allow to screen the essential nutrient conipounds and the concentrations ( x i , . .., N,) of a nutrient niediuni in the vicinity of optimal ones, too (Tab . :3 and 4).