A new strategy to approach multiresponse optimization in conjunction to a D-optimal design for
simultaneously optimizing a large number of experimental factors is proposed. The procedure is applied
to the determination of biogenic amines (histamine, putrescine, cadaverine, tyramine, tryptamine, 2-
phenylethylamine, spermine and spermidine) in swordfish by HPLC-FLD after extraction with an acid
and subsequent derivatization with dansyl chloride. Firstly, the extraction from a solid matrix and the
derivatization of the extract are optimized. Ten experimental factors involved in both stages are studied,
seven of them at two levels and the remaining at three levels; the use of a D-optimal design leads to
optimize the ten experimental variables, significantly reducing by a factor of 67 the experimental effort
needed but guaranteeing the quality of the estimates. A model with 19 coefficients, which includes those
corresponding to the main effects and two possible interactions, is fitted to the peak area of each amine.
Then, the validated models are used to predict the response (peak area) of the 3456 experiments of the
complete factorial design. The variability among peak areas ranges from 13.5 for 2-phenylethylamine to
122.5 for spermine, which shows, to a certain extent, the high and different effect of the pretreatment on
the responses. Then the percentiles are calculated from the peak areas of each amine. As the experimental
conditions are in conflict, the optimal solution for the multiresponse optimization is chosen from
among those which have all the responses greater than a certain percentile for all the amines. The
developed procedure reaches decision limits down to 2.5 mg L 1 for cadaverine or 497 mg L 1 for histamine in solvent and 0.07 mg kg 1 and 14.81 mg kg 1 in fish (probability of false positive equal to
0.05), respectively.
