Sources of variance in the double normalized value: an evaluation of its reproducibility as a measure on HLA-D locus identity.

Two different sets of mixed lymphocyte culture (MLC) experiments were performed using HLA-D homozygous typing cells (HTC). In all stimulator-responder combinations the median cpm's of four replicate cultures were reduced into double normalized values (DNV). In the first experimental set the responder panel was unselected, whereas the responders for the second set were chosen on the basis of either sharing an HLA-D determinant with the stimulator (ID) or not (non-ID). The experiments were designed: Set 1: To estimate the technical variability in the DNV's and to observe the distribution of this variability. The standard deviation of an observation on a stimulator-responder combination was approximately 24 DNV units. Thus, by running 6 experiments on each responder we would have a mean DNV with a standard error of 10 DNV units. Set 2: To determine whether stimulators typing for the same Dw specificity had the same distribution of DNV's and to investigate the variability between responders, between experiments and within experiments. Although the mean DNV is the same for all HTC's, the variability in observations was greater for some HTC's than for others. The variability may be completely technical for some HTC's, whereas for other HTC's there is evidence of responder variability and between experiment variability. Important implications of these results are: (1) that in using a single cut-off value of 60 for all HTC to define typing responses one will have a very high misclassification rate for a large percentage of ID responders; (2) for some HTC this error rate can be reduced through repeating the experiment 4 times and raising the cut-off point; (3) this error, together with the technical and/or experimental variance can be further reduced by using 2 or more HTC's of the same specificity in each experiment and by combing their data; (4) misclassification can be reduced in every situation by doing 4 experiments, using 2 HTC per responder and computing a cut-off which gives a misclassification rate for each Dw type of 10% in ID's and in non-ID's. Thus the best approach to achieving Dw locus typing with a desirable low rate of misclassification would be to do similar control studies of every HTC, to estimate technical experimental and responder variance and then use this information to determine a cut-of value and the number of HTC's and experiments per responder required to keep the error rate at a satisfactory level.