sep ofs-s1 TUE 13:42 5186435163 F.at UNIVERSITY OF CALIFORNIA, BERKELEY BERKELEY « DAVIS 4 IRVINE * LOS ANGELES « RIVERSIDE ' SAN DIEGO = SAN FRANCISCO SANTA BARBARA * SANTA CRUZ SCHOOL OF PUBLIC HEALTH EARL hake H ALL . - DEPARTMENT OF BIOMEDICAL AND BEBKELEY, CALIFORNIA 94720 ENVIRONMENTAL HEALTH SCIENCES FAX TO: VICTOR McKUSICK / OSKAR ZABORSKY Fax: 301-955-4999 Tel: 301-955-6641 FROM: GEORGE SENSABAUGH Fax: 510-643-5163 Tel; 510-642-1271 6 + 11 pages Oct. 28, 1991 Dear Victor, I regret to inform you that I can not sign off on the report as it stands. My main concern is chapter 3. When I signed off last May, I indicated that I had reservations about this chapter but couldn't add anything to the comments I had already made. Frankly, I had hoped that the reviewers comments might induce some of the changes I had lobbied for. I did not see their comments so I don't know what they said. However the 10/15/91 draft is not that different from the draft I was concerned about in May. My major reservations about chapter 3 are outlined below. 1. The chapter takes a very strong position that the US population is highly substructured and that accordingly any population frequency estimates based on existing data are Meaningless. Although some of the phrasing of this draft is softened relative to previous drafts, the position is not significantly softened. This position is stronger than much of the testimony we heard from other population geneticists, gome of whom questioned the existence and/or relevance of substructuring. It is clear that there is not unanimity of opinion on this point and I think the report should reflect this. At least, there should be some statement that the impact of substructuring on frequency estimation is being actively debated by population geneticists and that there may be legitimate differences of opinion. I appreciate the core concern that extreme frequency estimates have been quoted in court and I know we are all concerned that this might lead to greater weight being placed on the DNA evidence than is warranted. The population substructure argument posits that the general 1 msT-s23a— wo 1 TUE 19:42 S1ISBGée4d¢ssieées P.B2 population contains pockets of individuals who possess certain combinations of genes at high frequency and that this possibility is obscured by the citation of statistics for the general population. In other words, the use of general population statistics may imply absolute identification when, due to these pockets, there is a realistic possibility that other individuals of the same genetic type may exist. I think this concern can be adequately met by providing guidelines for the generation of very conservative frequency estimates. The ceiling approach accomplishes this. We could also recommend a ceiling on the cumulative frequency, e.g., no figures be cited less than one in a million. The population substructure argument, however, I think lays a trap for the future; I comment on this below. The central issue in the statistical interpretation area is the legitimacy of combining individual locus genotype frequencies by multiplication. By multiplying enough single locus frequencies together, one can get infinitely low combination frequencies. The legitimacy of multiplication depends on the statistical independence of allelic combinations of the loci in question, i.e., allelic combinations at different loci cannot be correlated. We propose the ceiling approach to get at the problem of making safe allele frequency estimates. We then state that this conservative approach solves the multiplication problem. However, one element of the population substructure argument is the existence of allelic correlations across loci (the tall, blond, blue-eyed nordic argument); we can't wave this away simply by taking a conservative approach to allele frequency estimation. Moreover, testing small (n=100) homogeneous populations is not likely to reveal any but the most extreme correlations. I suspect our position could be shown to be in error if alleles occur at different loci in a reasonably correlated way although I have no idea what level of correlation it would take. It would be better to address the correlation issue directly: What tests exist for detecting allelic correlations and what are their limits? At what level does allelic correlation begin to significantly erode frequency estimation? Are there correction factors that can be put on frequency estimates at various levels of correlation? What, if anything, can be done with existing databases? The allele correlation quastion is likely to become more of an issue as PCR based typing systems come on line. It will become possible to test a large number of moderately polymorphic loci (heterozygosities 60-90%) and the question of multiplication will inevitably come up. TUE L37243 SilB64s351 nw iy T fi WN The summary recommendation (p. 3-40) that population frequencies be based on counting until appropriate population studies (which we define) be done 1s undesirable on two counts. First, and most important in my mind, it questions genetics; it implies that multiplication of frequencies is mere theory and is not applicable in the real world. To retreat so completely from genetics signals to lay people and lawyers that something is wrong about its principles. (In some places, we sound almost as dismissive of genetics "theory" as creationists are of evolutionary "theory".) Second, the recommendation leaves us in a policy limbo. The N values for the existing databases are on the order of a thousand or two. These sorts of 1/N numbers can be obtained with conventional blood group and protein typing: should we retreat from DNA until the "appropriate" population studies are done? If DNA typing of an evidence sample at one locus matches none in my database, why type additional loci since I can already make my 1/1000 statement. (Studies on the Home Office database show single lecus matches are uncommon and two locus matches are very rare.) This recommendation should be amended to state that this is an absolute upper limit frequency statement and that genetic principles can be used to generate lower frequency estimates. This brings the recommendation more into accord with the text (p. 3-37). What happens if the NCFDT never gets off the ground? It appears to be the final arbiter for the choice of the representative homogeneous population groups from which the allele frequency estimates are to be derived. If there is no NCFTD, then again we end up in a policy limbo where nothing that has been done previously would be considered acceptable. This is not an idle concern; in the past, courts have considered apparent unresolved issues to be grounds for exclusion. This concern extends as well to the other recommendations that depend on the NCFTD. I am not opposed to additional population studies; indeed I welcome them. Rather, I would like to see more consideration given to the studies that have been done as at least providing first approximations. The 10% gene frequency celling is unnecessarily high. We know from classical genetics that there are genes that are uncommon (< 1%) in all populations and there is no need to set an artificially high platform for them. In any case, it should be clearly stated that the ceiling principle ylelds an estimate for a worst case scenario - that the frequency estimates used are higher than would be expected in any real population. OeT-29-31 TUE t3:3:44 5166435163 PP. 9. The chapter focuses primarily on the problems associated with tha VNTR loci with a very large number of "continuous" alleles. We should be providing guidance, however, on the upcoming PCR based systems that detect discrete allales for here some of the issues are different. For loci with smaller numbers of discrete alleles (5-15), the assessment of allele frequencies becomes more straightforward and the issue of correlation becomes more tractable: we should provide guidance. With all due respect to Lewontin, the main message of his 1972 apportionment of diversity paper was that the greatest proportion of diversity was at the individual level. Within yace population variation and between race variation accounted for only about 15% of the total variation, 8% and 7% respectively, as I recall. We should make this point clear since as the draft now implies that the major proportion of variation is between populations. It should also be stated up front rather than as a coda that the example illustrating variation between Poles and Italians represents the extreme values for these two groups. Also, and unfortunately from a blood grouping perspective, these are not terribly good examples since both the Rh and Kell allele frequencies are inferred from results obtained by testing for the dominant antigen; both are dominant/recessive systems and the allele frequency estimates may not be very good. It is suggested that laboratory error rates be included as part of the overall population frequency statement. Although I am a strong advocate for open access to proficiency test results, it must be appreciated that these results have limited relevance to particular cases, First, when labs make errors, it is incumbent upon them to try to identify the source of the error and to take corrective actions. Thus proficiency test error rates should not reflect a steady state rate. Second, and more important, each case is different. In some cases, internal consistency checks can be incorporated in the testing scheme that virtually guarantees the accuracy of the result: in other cases, this may not be possible. None of this can be reduced to a simple number) the guality of the work product must be evaluated individually in each case. Finally, we have skirted the ultimate question: at what point does a population frequency become a statement of identification. Is it at 1/million, l/billion, 1/trillion? Or do we want to state that identification requires a subjective leap? We make a number of statements that absolute identification is on the horizon; how will we know when we are there. -T-29-91 TUE 13:45 5146435162 F. Some other general comments, by page and line: ES-9, 23 and 2-38, 20: delete too; if the sample is too small, nothing can ba done with it. ES-20, 7-10 and corresponding section in ch 4: I object to the presumption that any non-accredited laboratory is not meeting standards. It would be better to say that any non- accredited laboratory should be expected to demonstrate that it is operating at the same level of standards as accredited laboratories. This makes it harder but not impossible for non-accredited labs to do work, for example, as in the case of a non-forensic lab repeating the analysis of a forensic lab. Our objective is to ensure high standards, not to discriminate on the basis of accreditation. ES=22, 3-4: "list of genetic types" instead of "list of fragment sizes", 1-16, 18: "numeric value" => "digital code" 1-16, 22-23: It is premature to predict that this system might lead to absolute identification; it needs to undergo the game validation as other DNA typing systems. Moreover, since the work is not yet published in a peer reviewed Journal (although I have no doubt it will ba soon) we should not wax too enthusiastic. 2-1, 11-16: "patterns" -> "types", The patterns are the basis for assigning type designations but are not the types themselves. 2-4, 19-20: Lab results have both qualitative and quantitative elements, e.qg., bands on an autoradiogram are characterized both by position and intensity. 2-10, 11-12: Where did the recommendation for 10 or fewer easily distinguished alleles come from; this is new to me. This takes out all the currently used VNTR systems. Do we really want to say this? 2-22, 11-12: The yield gel is used to assess both the quantity and quality of DNA in a sample; properly done, that is, with appropriate quantitative standards, it will give an adequate estimate (within 3X) of the quantity of high molecular weight DNA. Fluorescence assays can't distinguish degraded from undegraded DNA. There is no real gain in information with a pure quantitative assay. 2-40, refs: If there is to be a reference in this chapter to my writing on the analysis and interpretation of blood group and protein markers, the better reference is Sensabaugh, GF. 5 et a OCT-29-31 THE 19:46 S1GB6435162 FP. Blochemical markers of individuality, in: Forensic Science Handbook (R. Saferstein, ed.) Prentice Hall, 1982, pp. 338- 415. 3-generally: The term "odds" is used throughout the chapter where "estimated frequency of occurrence" is intended. "Odds" has a defined meaning in probability theory, specifically a ratio of probabilities as in log odds scores. 3-6, Sff: Combined frequency estimates for the conventional markers often are less than 1/100. These frequencies have not been verified by simple counting but by multiplication. The justification for multiplication is the lack of correlation of alleles at different loci as indicated by pairwise comparisons. With the conventional markers, no evidence of significant allele correlations has been noted. As a final note, shouldn't we have seen all the reviewers comments? I got 7 pages of comments faced by a page labeled "general", For your information, I include the Sept. 1991 report of the DNA Commission of the Society for Forensic Haemogenetics.