Bruce Budowle and Keith L. Monson
Forensic Science Research and Training Center, Laboratory Division, FBI Academy, Quantico, Virginia 22135, USA

× Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø × Ø

The National Research Council (NRC) published its second report on forensic DNA applications in 1996, concentrating on the conclusions which should be drawn from a coincidental match (1). As discussed previously (2), the second report (or NRC II Report) 1) evaluated extant data and made recommendations for determining the rarity of a DNA profile, 2) rectified misstatements or erroneous interpretations in the first NRC Report (3) on forensic applications, and 3) addressed misapplication of the recommendations of the first NRC Report (3).

Most, if not all, of the recommendations of the NRC II Report (1) already are in practice in the forensic community. In addition, the NRC II Report (1) has been useful in minimizing controversies in the legal setting. However, the NRC II Report (1) deferred full consideration of certain topics to the forensic science community for amplification, clarification, or accumulation of additional data. One of those areas, sample size of PCR-based marker databases, is addressed in this paper.

DNA databases for forensic statistical calculations generally contain 100 (although sometimes less) to several hundred individuals. These sample sizes are adequate for forensic applications (4). Indeed, Nei (5) demonstrated that population substructure analyses could be performed on databases containing less than 50 individuals with loci containing only 2 to 6 alleles. However, the NRC II Report (1) stated "The empirical studies show that the differences between frequencies of the individual profiles estimated by the product rule from different adequate subpopulation databases (at least several hundred persons) (emphasis added) are within a factor of about 10 of each other, and that provides a guide to the uncertainty of the determination for a single profile" (p. 160). It is clear from the context of the NRC II Report (1) that this statement refers to RFLP/VNTR markers, which are loci that contain a large number of alleles. For more information supporting the NRC II Committee's conclusion on differences in profile estimates, one may consult Budowle, et al. (6,7).

Regardless, one could misconstrue that the same sample size requirement should be imposed on databases of many of the currently employed PCR-based markers. However, these loci (e.g., LDLR, GYPA, HBGG, D7S8, Gc, and HLA-DQA1) contain far fewer alleles, approximately 2-7 alleles. Thus, it is obvious that database sample sizes smaller than several hundred individuals would permit reliable estimation of DNA profile frequencies. Rather than specify a minimum database size, uncertainty in estimates for DNA profiles can be described by the use of confidence intervals (e.g., refs. 1 and 8), which can be implemented readily and would be the statistically preferable approach. Alternatively, the use of the 10-fold rule, advocated by the NRC II Committee (1), could be invoked even when using databases containing fewer than 100 individuals if empirical studies for PCR-based markers show results/trends similar to those performed on RFLP/VNTR loci.

Monson and Budowle (9) have performed such empirical studies on multiple PCR-based locus DNA profiles. They explored the consequences on profile frequency estimates of group or subgroup misassignment and the forensic significance of using various reference databases for several PCR-based loci: LDLR, GYPA, HBGG, D7S8, Gc (these five loci collectively known as PM), HLA-DQA1, and D1S80. The databases included the general groups: African-American, Caucasian, and Hispanic. Subgroups included: regional United States databases, Australian aborigines, Chileans, Dubaian Arabs, French Antillians, Haitians, Hungarians, Israelis, Japanese, Koreans, Mexicans, Navajos, Pueblos, Sioux, Eskimos, northern and southern Croatians, Palestinian Arabs, Spanish Basques, and Vietnamese. The database sample sizes ranged from 72 to 212. Logarithmic scatter plots, where the inverse frequency (probability of occurrence) evaluated for each target profile in its source database of the remaining target profiles is plotted against that in a different reference database, were used to evaluate the consequences of using a different database. A difference greater than a factor of ten between multiple locus probability estimates was set to identify forensically significant frequency estimates (and would be informative for the 10-fold rule). As is the case with RFLP/VNTR markers, it was found that subdivision, either by ethnic group or by U.S. geographic region, did not substantially affect forensic estimates of the likelihood of occurrence of a DNA profile within a major population group. As expected, the greatest variation in statistical estimates occurs across major population groups. Thus, in most cases, there will be no unfair bias applying general population database estimates. Profile frequencies estimated from a geographic or ethnic subgroup data base are usually within a factor of ten of those derived from the associated general U.S. group, when frequency estimates are less than 1 in 10⁵. Moreover, results using smaller databases, even less than 100 individuals, were consistent with those from larger databases of similar groups. Based on empirical data, there is no demonstrable need for employing larger sample size databases than currently exist.

The Monson and Budowle study (9) shows that, when frequency estimates are less than 1 in 10⁵, the ten-fold rule of thumb compensates adequately for uncertainties in profile frequency estimates, whether arising from genetic or mathematical assumptions, from limited data base size, or that a particular person belongs to a subgroup with frequencies differing from those of the population average. Therefore, the current database sample sizes for PCR-based markers provide valid estimates of the likelihood of occurrence of a DNA profile.

The argument that a database containing fewer than several hundred individuals is inadequate for estimating multiple locus DNA profiles should be moot for general United States population groups. Sufficient data exist to achieve databases containing several hundred or more individuals. We describe here two examples - one for HLA-DQA1 and PM databases and one for short tandem repeat (STR) loci databases - where data were pooled, allele frequencies were generated, and tests for independence were performed. These data also can be used when database sample size becomes an issue in the legal setting.

In a previous study, Budowle, et al. (10) compared HLA-DQA1 and PM population data on different samples of African-Americans, U.S. Caucasians, and southwestern Hispanics. Within each major population group comparison, the allele frequencies at all loci were statistically similar, except the D7S8 locus in southwestern Hispanics (p=0.028). Out of eighteen comparisons (i.e., six loci comparisons in each of three population groups), only one comparison was significantly different. This observation is no more than would be expected by chance. Thus, the data within each major population group could be merged. Moreover, since the allele frequencies are similar, the resulting multiple locus profile frequency estimates also should be similar.

For demonstration purposes, profile data on the loci LDLR, GYPA, HBGG, D7S8, Gc (or PM loci), and HLA-DQA1 were compiled for African-Americans, Caucasians, and southwestern Hispanics from three different United States studies (only two studies were merged for southwestern Hispanics) (10-12) to create databases containing several hundred individuals. The distribution of the observed PM and HLA-DQA1 allele frequencies in African-Americans, Caucasians, and southwestern Hispanics are shown in Tables 1-2. Based on the exact test (13), all loci, except the HLA-DQA1 locus in the African-American sample (p=0.017), meet Hardy-Weinberg expectations (HWE) (Tables 1 and 3). However, a portion of these HLA-DQA1 samples were subtyped for the 4 allele (i.e., data from refs. 10 and 12). If considering only those database studies where the HLA-DQA1 locus was subtyped for the 4 allele, then all loci in all three databases meet HWE (Table 4). It is likely that the departure from independence at the HLA-DQA1 locus is due to sampling.

An inter-class correlation test analysis (14) detected no significant departures from expectations of independence between alleles of the HLA-DQA1 and PM loci. Thus, the data support that the loci meet expectations of independence for the pooled African-American, U.S. Caucasian, and southwestern Hispanic populations.

2. STR Loci - CSF1PO, TPOX, TH01 -

Sample sizes also can be increased readily for another class of PCR-based markers - STR loci. Budowle, et al. (15) demonstrated very few differences for allele frequencies between population samples within a major population group (i.e., Caucasian v. Caucasian or African-American v. African-American). In this current paper, allele frequency data are compared for the STR loci CSF1PO, TPOX, and TH01 from two sample sources: Budowle, et al. (15) and genotype data provided by Dr. A. Eisenberg (personal communication). Between-group comparisons indicate that allele frequencies are similar for the STR loci CSF1PO, TPOX, and TH01 within a major population group in the United States (Table 5). Thus, the data between each major population group compared were merged.

The distributions of observed allele frequencies for the CSF1PO, TPOX, and TH01 loci are shown in Tables 6-8. The observed heterozygosities for the loci CSF1PO, TPOX, and TH01 range from 61.8% for the TPOX locus in Hispanics to 78.7% for the TPOX locus in African-Americans. All loci meet HWE and an inter-class correlation test analysis failed to detect any correlations between the alleles at any of the pair-wise comparisons of the three STR loci (Tables 6-9).

Because of the overall similarities in allele frequencies between sample populations within a major population group before and after databases are merged, there should be no anticipated substantial differences in DNA profile frequency estimates if a database containing fewer than several hundred individuals was employed for the loci PM, HLA-DQA1, CSF1PO, TPOX, and TH01. Scatter plot analyses also support this conclusion.

The extant data demonstrate that valid multiple locus PCR-based DNA profile frequency estimates can be made with databases containing fewer than several hundred individuals and that profile frequencies would be similar when using different size reference databases. However, if desired, larger sample size databases are available for general United States population groups.

This is publication number 97-12 of the Laboratory Division of the Federal Bureau of Investigation. Names of commercial manufacturers are provided for identification only, and inclusion does not imply endorsement by the Federal Bureau of Investigation.

REFERENCES

1. National Research Council. The Evaluation of Forensic DNA Evidence, Washington, D.C., National Academy Press, 1996.

2. Budowle B., Monson K.L. Accepted practices by the forensic DNA community supported by the NRC II Report. In: Seventh International Symposium on Human Identification 1996, Promega Corporation, Madison, Wisconsin pp. 27-38, 1997.

3. National Research Council. DNA typing: Statistical bases for interpretation. In: DNA Technology in Forensic Science, Chapter 3, Washington, D.C., National Academy Press, pp. 74-96, 1992.

4. Chakraborty R. Sample size requirements for addressing the population genetic issues of forensic use of DNA typing. Human Biology 64:141-159, 1992.

5. Nei M. Estimation of average heterozygosity and genetic distance from a small number of individuals. Genetics 89: 583-590, 1978.

6. Budowle B., Monson K.L., Giusti A.M., Brown B. The assessment of frequency estimates of Hae III-generated VNTR profiles in various reference databases. J. Forens. Sci. 39:319-352, 1994.

7. Budowle B., Monson K.L., Giusti A.M., Brown B. Evaluation of Hinf I-generated VNTR profile frequencies determined using various ethnic databases. J. Forens. Sci. 39:988-1008, 1994.

8. Chakraborty R., Srinivasan M.R., Daiger S.P. Evaluation of standard error and confidence interval of estimated multilocus genotype probabilities and their implications in DNA forensics. Amer. J. Hum. Genet. 52:60-70, 1993.

9. Monson K.L., Budowle B. Effect of reference database on frequency estimates of polymerase chain reaction (PCR)–based DNA profiles. J. Forens. Sci. (submitted).

10. Budowle B., Koons B.W., Moretti T.R. Subtyping of the HLA-DQA1 Locus and Independence Testing with PM and STR/VNTR Loci. J. Forens. Sci. (in press).

12. Jankowski L.B., Budowle B., Swec N.T., Pino J.A., Freck-Tootell S., Corey H.W., Schwartz R., LaRue E.J., Rochin W.L., Kearner C.J., Tarver M.L. New Jersey Caucasian, African-American, and Hispanic population data on the PCR-based loci HLA-DQA1, LDLR, GYPA, HBGG, D7S8, and Gc. J. Forens. Sci. (submitted).

13. Guo S.W., Thompson E.A. Performing the exact test of Hardy-Weinberg proportion for multiple alleles. Biometrics 48:361-72, 1992.

14. Karlin S., Cameron E.C., Williams P.T. Sibling and parent-offspring correlation estimation with variable family size. Proc. Natl. Acad. Sci. USA 78:2664-2668, 1981.

15. Budowle B., Smerick J.B., Keys K.M., Moretti T.R. United States population data on the multiplex short tandem repeat loci - HUMTHO1, TPOX, and CSF1PO and the variable number tandem repeat locus D1S80. J. Forens. Sci. (in press).

17. Lewontin R.C., Felsenstein J. The robustness of homogeneity tests in 2 X N tables. Biometrics 21:19-33, 1965.

Table 1. HLA-DQA1 Observed Allele Frequencies In Combined United States General Population Groups (Refs. 10-12)

Table 2. Observed Allele Frequency Distributions For PM Loci in Pooled United States General Population Groups (Refs. 10-12)

Table 3. Tests for Independence on PM Loci

Table 4. HLA-DQA1 Observed Allele Frequencies in Pooled United States General Population Groups (refs. 10 and 12)

Table 5. G Statistic Test (p values+S.D.) for Homogeneity (16,17) on TH01, TPOX, and CSF1PO Allele Distributions Between the Two United States Caucasian Sample Populations, Between the Two African-American Sample Populations, and Between the Two Hispanic Sample Populations (ref 15 and personal communication data - Dr. A. Eisenberg)

^a   Observed Homozygosity = 0.268; Expected Homozygosity (unbiased) = 0.258; HWE - Homozygosity Test (p=0.638), Likelihood Ratio Test (p=0.564), and Exact Test (p=0.698)

^b   Observed Homozygosity = 0.220; Expected Homozygosity (unbiased) = 0.213; HWE - Homozygosity

Test (p=0.701), Likelihood Ratio Test (p=0.630), and Exact Test (p=0.603)

^c   Observed Homozygosity = 0.227; Expected Homozygosity (unbiased) = 0.245; HWE - Homozygosity

Test (p=0.397), Likelihood Ratio Test (p=0.907), and Exact Test (p=0.970)

^d   N = number of individuals

Table 7. TPOX Allele Frequencies in Pooled United States Sample Populations (ref 15 and personal communication data - Dr. A. Eisenberg

Table 8. CSF1PO Allele Frequencies in Pooled United States Sample Populations (ref 15 and personal communication data - Dr. A. Eisenberg)