- Email: [email protected]
Disaster victim investigation recommendations from two simulated mass disaster scenarios utilized for user acceptance testing CODIS 6.0
Forensic Science International: Genetics 5 (2011) 291–296
Contents lists available at ScienceDirect
Forensic Science International: Genetics journal homepage: www.elsevier.com/locate/fsig
Disaster victim investigation recommendations from two simulated mass disaster scenarios utilized for user acceptance testing CODIS 6.0 Laurie Bradford *, Jennifer Heal, Jeff Anderson, Nichole Faragher, Kristin Duval, Sylvain Lalonde National DNA Data Bank, Royal Canadian Mounted Police, 1200 Vanier Parkway, Room 46, Laboratory Building, Ottawa, Ontario, K1A 0R2, Canada
A R T I C L E I N F O
A B S T R A C T
Article history: Received 4 January 2010 Received in revised form 6 May 2010 Accepted 15 May 2010
Members of the National DNA Data Bank (NDDB) of Canada designed and searched two simulated mass disaster (MD) scenarios for User Acceptance Testing (UAT) of the Combined DNA Index System (CODIS) 6.0, developed by the Federal Bureau of Investigation (FBI) and the US Department of Justice. A simulated airplane MD and inland Tsunami MD were designed representing a closed and open environment respectively. An in-house software program was written to randomly generate DNA profiles from a mock Caucasian population database. As part of the UAT, these two MDs were searched separately using CODIS 6.0. The new options available for identity and pedigree searching in addition to the inclusion of mitochondrial DNA (mtDNA) and Y-STR (short tandem repeat) information in CODIS 6.0, led to rapid identification of all victims. A Joint Pedigree Likelihood Ratio (JPLR) was calculated from the pedigree searches and ranks were stored in Rank Manager providing confidence to the user in assigning an Unidentified Human Remain (UHR) to a pedigree tree. Analyses of the results indicated that primary relatives were more useful in Disaster Victim Identification (DVI) compared to secondary or tertiary relatives and that inclusion of mtDNA and/or Y-STR technologies helped to link family units together as shown by the software searches. It is recommended that UHRs have as many informative loci possible to assist with their identification. CODIS 6.0 is a valuable technological tool for rapidly and confidently identifying victims of mass disasters. Crown Copyright ß 2010 Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: CODIS 6.0 Mass disaster Disaster Victim Identification DNA Joint Pedigree Likelihood Ratio Pedigree tree searching
1. Introduction Mass fatality incidents (MFIs), also known as mass disasters, are caused by accidental catastrophes such as airplane crashes, natural occurrences such as Tsunamis and intentional acts such as terrorist attacks. These incidents generally lead to a high death toll and mass destruction of property [1,2]. On September 2nd 1998, Swiss Air Flight 111 crashed in the ocean off the coast of Peggy’s Cove, Nova Scotia, Canada [3,4]. Two hundred and twenty nine passengers and crew members perished leaving investigators with 1277 human remains to identify [3,4]. This is an example of a closed mass disaster where the victims were potentially known by the airplane manifest. Open environment mass disasters like the Asian Tsunami that occurred on December 26th 2004 and the
Abbreviations: DVI, Disaster Victim Identification; MD, mass disaster; MFI, mass fatality incident; CODIS, combined DNA index system; DNA, deoxyribonucleic acid; NDDB, National DNA Data Bank of Canada; RCMP, Royal Canadian Mounted Police; FBI, Federal Bureau of Investigation; UAT, user acceptance testing; STR, short tandem repeat; Y-STR, Y chromosome short tandem repeat; mtDNA, mitochondrial DNA sequence; UHR, unidentified human remain; MP, missing person; bp, base pair; LR, likelihood ratio; CLR, combined likelihood ratio; JPLR, joint Pedigree Likelihood Ratio. * Corresponding author. Tel.: +1 613 993 8906; fax: +1 613 993 9944. E-mail address: [email protected] (L. Bradford).
World Trade Center attack in New York City on September 11th 2001, were associated with a larger number of victims who were not known [2,5–9]. DNA typing is a robust, consistent and reliable technique ideally suited for identifying victims of mass disasters [3]. Even highly degraded body remains can be typed using STRs (100–400 base pairs (bp)) [7]. For the above-mentioned mass disasters unidentified human remains and reference samples from biological relatives were typed using AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM kits (Applied Biosystems, Foster City, CA) [2–4,6,8]. In recent years, Y-STR and mtDNA techniques have assisted in victim identifications. Y-STRs and mtDNA are haplotypes that are not subject to recombination events [6]. Barring mutations, they are consistent across generations. Y-STR inheritance is via the paternal side and only occurs in males. MtDNA is inherited maternally and is passed on to all offspring [7,10,11]. The disadvantage to using only Y-STRs and mtDNA is that the genetic information is not discriminating at an individual level. Entire families will have the same Y-STR profile for the paternal lines and mtDNA sequences for all maternal lines, with the exception of mutations and heteroplasmy [10,11]. Mass fatality incidences pose a unique challenge for scientists attempting to identify the victim’s remains for their return to family members. The proper identification of victims is dependent
1872-4973/$ – see front matter . Crown Copyright ß 2010 Published by Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.fsigen.2010.05.005
292
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
on: the number of UHRs; cause of death and extent of body fragmentation; DNA degradation; recovery of the remains; integrity of the samples and the type of familial DNA reference samples available to aid the investigation [1,2,11]. Direct reference samples (DRS), also referred to as exemplar samples or personal effects, are samples directly taken from the victim or their personal belongings [2–7]. In the past, direct reference samples could not always be confidently assigned to a victim [4]. Other considerations in identification include the occurrence of genetic core repeat slip mutations between parents and offspring where the slippage is by one or two repeat units (e.g. 4–8 bp), the presence of rare alleles and the more common variants where the slippage may be off by only 1 or 2 bp [3,4,7]. Additionally, high impact mass disasters like airplane crashes can lead to the occurrence of comingled remains, resulting in victim profiles that appear as mixtures requiring interpretation [2–4,6,7,11]. CODIS 6.0 is a software program designed with new features for identifying victims of mass fatality incidences and on-going missing persons programs. One feature of CODIS 6.0 is the redesigned, userfriendly interface. Navigation through the many software modules is improved and more than one window is accessible at one time. Identity searches are still available and a new search feature referred to as Pedigree searching has been added. Pedigree trees can be generated in the new module Pedigree Tree Designer where relatives are associated to the pedigree tree and can be saved in Pedigree Tree Manager. The entire pedigree tree is searched against all of the UHRs to find a suitable rank with the associated relative DNA information. The ranks are displayed in the Rank Manager module of CODIS 6.0 where all the search results can be reviewed. Additionally Y-STRs and mtDNA can be added to assist in victim identification. Metadata (unique non-DNA identifying pieces of information specific to a victim) can also be added for a victim in the new software program and used to assist in their identification. The Pedigree searching feature uses the kinship calculations and produces a Joint Pedigree Likelihood Ratio (JPLR). The JPLR is the ratio of the probability of the missing individuals DNA profile being related to the pedigree tree (containing at least one relative’s DNA information) compared to their DNA profile not being related to the pedigree tree. This gives the CODIS 6.0 user greater confidence in assigning a UHR to a specific pedigree tree and leads to a reduction in fortuitous matches. The statistical calculations, inclusion of new identification technologies and data management capabilities were shown to be assets in victim identification. 2. Materials and methods 2.1. Simulated airplane and Tsunami MD design Two simulated mass disasters, an airplane crash representing a closed environment mass disaster and a Tsunami representing an open environment mass disaster, were designed by members of the National DNA Data Bank (NDDB) of the Royal Canadian Mounted Police (RCMP) for user acceptance testing (UAT) of a CODIS 6.0 pre-release package. Actual mass disasters were used as a guide in designing data that would challenge both the software program and the users during the testing phase. These mass disaster scenarios were created to test the features of CODIS 6.0, and specifically the pedigree searching feature and the outcome of the JPLR calculations. The data files are available to law enforcement agencies only. The simulated airplane mass disaster was designed first. The airplane model used for the mock mass disaster was a Boeing 737 capable of carrying 189 passengers plus a pilot, co-pilot and 6 crew members [12]. A person was assigned to each seat on the manifest in addition to a full crew complement. Names were created for each person and familial relations were delineated and pedigree
Table 1 A breakdown of the pedigree tree statistics for the victims of the airplane and Tsunami mass disasters. Mass disaster
Relative type
Percentage of total
Airplane
Primary only Two primary Primary and secondary Secondary only
84.1 44.7 11.2 4.7
Tsunami
Primary only Two primary Primary and secondary Secondary only
81.7 47.2 11.3 7
trees were then developed for each person on the plane. In total, 79 pedigree trees were drawn. They were designed to include primary and/or secondary relatives with some containing a lot, or very little family information in order to test the software capabilities (Table 1). For the purpose of the user acceptance testing, primary relatives included biological parents, children and full-siblings. Secondary relatives included grandparents, grandchildren, halfsiblings, aunts, uncles, nieces and nephews. In some cases the entire family was on the airplane and all were victims. There were pedigree trees where the biological father was not known and where a child was adopted so no biological relative information was made available. The airplane was broken apart in a simulated crash and a grid was drawn over the crash site and labelled in meters-squared based on an Interpol gridding system [13,14]. The UHR naming convention was based on the World Trade Center mass fatality incident and Interpol guidelines and required a maximum of 25 characters for CODIS 6.0. It consisted of a unique mass disaster number, a unique investigator’s number, the UHR number assigned by that investigator and the grid location of the body or part based on the X and Y coordinates [9,14]. The relatives were named based on the mass disaster number followed by the relationship to the victim(s) and the manifest seat number [8]. The manifest was numbered consecutively in order of the seat numbers. To aid in developing the number of STR profiles required for a project of this scale, the NDDB information technology support member created an Access database pedigree-generating software program. The program is able to calculate up to 15 DNA profiles within multiple generations of a pedigree. The AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM (Applied Biosystems, Foster City, CA) STR profiles came from a database of over 100,000 randomly generated profiles that were formed using a weighted random number generator created by Dr. B. Leclair (Myriad Genetics, Salt Lake City, UT) (Dr. B. Leclair, personal communication). The allele frequencies were representative of a standard Caucasian population and these frequencies came from the validation studies conducted at the RCMP for the AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM kits (Applied Biosystems, Foster City, CA). An additional feature of the program is that all generated profiles are editable and customizable to suit project requirements, for example creating partial profiles or adding mutations. STR DNA profiles were generated for each victim and available relative(s) in the pedigree tree. In total there were 245 UHR profiles generated from whole bodies and body parts and 215 profiles from relatives. Some of the UHR DNA profiles were made to be partial profiles consistent with true findings from airplane mass disasters (Table 2). The partial profiles were created only from complete CODIS 13 loci profiles and were very reasonable having at least 9 loci present with some allele dropout present in some of the loci, and exceeding the NDDB’s current requirement of having at least 7 loci for CODIS searching. For the purpose of the UAT, we did not differentiate between partial and incomplete profiles. The mock
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296 Table 2 DNA profile breakdowns for the UHRs in the airplane and Tsunami mass disasters after building representative specimens. Mass disaster
Profile type
Percentage of total profiles
Airplane
Complete Partial Profiler PlusTM Mixtures/comingled
73.30% 17.30% 7.30% 2.1
Tsunami
Complete Partial
98.9 1.1
mass disaster included variants, rare alleles, mutations, comingled remains and direct reference samples presumed to be for a victim when they were not [4]. Additionally, the disaster data included a trisomy and a chimera to test the software capabilities. Y-STR profiles were created based on the loci from the AmpF‘STR1 YFilerTM kit (Applied Biosystems, Foster City, CA). Two Y-STR databases were used to create a total of 36 simulated YSTR profiles for the victims and relatives of the mass disaster [10,15,16]. The Health Science Center from the University of North Texas supplied mtDNA sequences for 58 victims and their relatives and they also reviewed the Y-STR profiles confirming that they were representative profiles (Dr. J. Planz, personal communication). Seventy-nine direct reference sample profiles were created for the victims. The exemplar sources included toothbrushes, hairbrushes, personal hygiene items (i.e. feminine hygiene products, diapers, etc.), underwear and an ‘‘other’’ category (i.e. tissue biopsy samples) consistent with Interpol guidelines and the source types used in the Swiss Air flight 111 mass disaster investigation [4,13,14]. These direct reference samples were named based on the order families and close acquaintances of the victims reported to the investigators. Metadata was also utilized in victim identification. The metadata that included: tattoos; piercings; X-rays of broken bones; dental X-rays and any unique identifying piece of information specific to a MP, was used to narrow down the identification to one individual. Metadata was generated for 24 victims and linked to their direct reference sample profiles. The Tsunami mass disaster data was generated in a similar manner to the Airplane mass disaster however it was representative of an open environment. A total of 82 pedigree trees were created for the Tsunami mass disaster consisting of primary, secondary and tertiary relatives. Not all of the persons reported missing were victims and not all victims of this disaster were reported missing, indicative of what might occur in an actual open environment mass disaster. A grid was drawn over satellite photos of the disaster site, Tofte, Minnesota, USA, based on a scale of 30 m squared [17]. The naming convention for the relatives and direct reference samples differed due to the grid format and absence of a known victim list. Similar to the airplane mass disaster, the UHR naming convention composed of four parts included a mass disaster number, investigator number, whole body or body part number found by the investigator and the grid location based on the x and y coordinates. Consequently, relatives were named based on mass disaster number, the order the relatives reported to the mass disaster investigative team and by their relationship to the victim(s). Once more the STR DNA profiles were generated for each victim and available relative(s) in the pedigree trees. In total there were 228 UHR profiles generated and 132 relative profiles. Very few of the UHR DNA profiles were partial in the Tsunami mass disaster. The principles for variants, rare alleles and mutations were applied here as in the airplane mass disaster. Included were 38 victims with Y-STR profiles created in the same fashion as the airplane
293
mass disaster and 62 victims and associated relatives, with mtDNA sequences supplied by the University of North Texas. In this case, 55 exemplar profiles were created to help identify the victims. These were produced in a similar fashion to the airplane mass disaster. Metadata was added for 26 victims for further assistance in their identification. Finally, Coroner’s Reports and death certificates were designed and generated for each victim for both disasters. The reports included: all body parts associated with a victim and the associated STR DNA profile(s); Y-STR profile data and mtDNA data if available; relative STR DNA profiles including their Y-STR and mtDNA data if available and the exemplar(s) STR DNA profile(s) associated with that victim if provided. In an attempt to best simulate a true mass disaster, the scientists who performed the data analysis were different scientists than those who developed the mass disaster scenarios and created the data. The data developers acted as coroners for the mock mass disasters. Their role was to confirm the identity of the victims and issue a death certificate as per a genuine mass disaster [11]. 2.2. Statistical analysis After completion of the UAT, the data from both simulated mass disaster scenarios were combined for statistical analysis as the searching protocols for each were the same. The statistical analysis was performed using a Kruskal–Wallis test or two-sample Kolmogorov–Smirnov test. Chi-square tests were performed to test the significance between variable groups. The statistical calculations based on the analyses performed were expected to reflect the importance of using primary relatives in victim identification. Having more primary relatives available was expected to show a significant impact in victim identification. Secondary relatives, in conjunction with primary relatives, were expected to contribute towards victim identification. Additionally the Combined Likelihood Ratios (CLR) which combine the JPLR with mtDNA and/or Y-STR likelihood ratios were predicted to improve confidence when included. Comparisons of complete profiles, partial profiles and Profiler PlusTM only profiles demonstrated the importance of having as many loci typed for each victim when possible. 2.3. CODIS data imports and searching protocols For each mass disaster scenario, CODIS import files were created for the UHR, relative and exemplar profiles. For the purpose of our user acceptance testing, separate import files were created for the mtDNA profiles and the Y-STR profiles, however, the STR and Y-STR can be contained in the same file. All mass disaster profiles provided were imported into CODIS 6.0, using a server/ workstation computer. The server was a Dell PowerEdge 1900 (4 GB Ram) with a Dual Processor Intel Xeon E5310. It contained a 160 GB hard drive, a gigabit Ethernet and ran on a Microsoft1 Windows Server 2003 and SQL Server 2005. For further information detailing configurations and searching procedures please see the supplemental information. 3. Results and discussion 3.1. CODIS 6.0 statistical calculations CODIS 6.0 software provided a combined likelihood ratio (CLR), JPLR and mtDNA and/or Y-STR DNA likelihood ratios (LRs) separately resulting from the Pedigree searching feature. The combined likelihood ratio includes the JPLR and likelihood ratios from the inclusion of mtDNA and/or Y-STR profiles for a pedigree
294
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
Table 3 A breakdown of the average JPLR, mtDNA LR, Y-STR LR and CLR for both mass disaster scenarios.
Airplane Tsunami
Average JPLR
Average mtDNA LR
Average Y-STR LR
Average CLR
7.55E+11 N = 163 9.39E+17 N = 176
884 N = 56 902 N = 54
3450 N = 32 2520 N = 25
3.35E+16 N = 172 1.16E+21 N = 177
tree. The JPLR alone does not include any additional information from these DNA technologies. Average JPLR, mtDNA LR, Y-STR LR and CLR values were calculated for each mass disaster scenario to demonstrate the JPLR values obtained in the simulated mass disaster searches (Table 3). Initially, an UHR to UHR Identity search was done to be able to add together samples from the same source (representative sample). Then missing person vs. UHR Identity AutoSearches were performed to match the victim’s direct reference sample(s) with their UHR STR DNA profiles from the mass disaster. The Identity AutoSearch compares the profiles at each locus and returns matches based on how many alleles at each locus matched. This is useful for identifying some victims prior to pedigree searching. The pedigree trees were searched individually and without the use of the Pedigree AutoSearcher feature. This was done to solve the mass disasters more efficiently by avoiding the high number of ranks that would have occurred from an AutoSearch on this large of a scale. The AutoSearch feature would have taken hours to complete in terms of checking each match for the mass disasters and these were mass disasters with fewer body remains than realistic mass disasters therefore the pedigree trees were searched individually. This can be partially attributed to the fact that all of the UHRs were entered into CODIS 6.0 simultaneously, compared to an approach where they are entered in batches. In a continuous missing persons program, the Pedigree AutoSearcher would be used on a routine basis however the challenges a mass disaster search imposes has specific differences compared to a missing person search mainly that the victims are frequently known and there will expectedly be a high number of matches resulting from searching. After a pedigree search, the JPLR calculations are returned in Pedigree Searcher and viewed in Rank Manager in the descending order. A true top rank can be defined as the expected result, i.e. the victim, being returned and having the highest ranking JPLR for the node being searched. In the case of the airplane mass disaster scenario, 89.3% of the time the true rank was returned in the top 5 ranked candidates and 90.2% of the time the true rank was returned in the top 10 ranked candidates. The average number of ranks returned for a searched node was 3.8. For the Tsunami mass disaster scenario 95.9% of the time the true rank was within the top 5 ranked candidates and 96.7% of the time the true rank was in the top 10 ranked candidates. The average number of ranks returned for a searched node in the Tsunami was 5.8. For both simulated mass disaster searches, 99% of the true ranks had JPLRs above 4.89E+1 and 95% of the true ranks were above 7.15E+02. Setting a minimum JPLR threshold from these numbers would be possible, however the consequences of threshold settings must be carefully weighed for each individual mass disaster. In total 72 ranks were returned without a JPLR value and 6 of these were the true ranks. This occurred due to the existence of extremely partial profiles not fitting the mutation model set within the software. Every UHR was successfully identified who could be identified. These identifications were made with direct reference samples, metadata, Pedigree Searches or Identity Searches. There were 6 UHRs remaining unidentified in the Tsunami, which were incorporated into the mass disaster design.
There were several factors that affected the presence of true ranks during the Pedigree searches. Very degraded profiles, comingled samples, or UHRs having mutations beyond the allowance of the mutation model were not returned as ranks upon performing a Pedigree search. The mutation model accounts for two core repeat unit mutations and minor slip mutations in a pedigree tree. In these above cases, or where no ranks were returned at all (for example, where a non-biological father was associated to the pedigree tree), identity search(es) were subsequently performed. It is therefore recommended that all Pedigree searches be complemented with Identity searches, to verify and confirm the ranks and to ensure that the true rank was not missed. It is also important to note that some ranks will not be returned if their JPLR values are below the set JPLR threshold of the Pedigree search configurations. Pedigree searches that return ranks must be viewed differently from conventional search results using pair-wise likelihood ratios. For instance, if two parents are associated to a pedigree tree and three of their children are victims of a mass fatality incident, then the returned ranks should include all three children, even though only one node is being searched. The rank returned for the node that is being searched is referred to as the expected rank by the authors. Subsequently, the other two children are not fortuitous ranks as they are part of the pedigree tree being searched and are equivalent to the searched node. They are also true ranks and referred to as equivalent ranks by the authors. Associated ranks refer to ranks returned for a related, non-equivalent node that belong to a different generation of the pedigree tree with the exception of cousins who are from the same generation. Take for example a pedigree tree with two children typed, with their parents and grandparents unknown. If a Pedigree Search is performed searching the father node then the father will be returned as the expected rank and the mother will be returned as an equivalent rank (Pedigree searches are not Amelogenin gender dependent however gender can be set in metadata and used as a filter) and the grandparents will be returned as associated ranks. A fortuitous rank is any rank that is returned and does not fall under the category of expected, equivalent or associated rank for that pedigree tree. They occur whereby a victim appears to be related to the kin but in fact they are not [1,2,6]. It is an STR DNA profile incorrectly associated with the pedigree tree being searched. It has been suggested that inclusion of more relative’s information may prevent or reduce the occurrence of true fortuitous ranks [4]. In many cases, these can also be attributed to common alleles in the profiles. These matches can be avoided by using more DNA information such as mtDNA or Y-STRs and having more primary relatives available for kinship analysis. The existence of mutations must be considered in data searching and the leniency allowed in search parameters can lead to more fortuitous matches [6]. The occurrence of fortuitous matches is greatly reduced in CODIS 6.0. 3.2. Contribution of primary relatives Both primary and secondary relatives contribute to victim identification. Primary relatives consisting of parents, offspring and full-siblings are ideal contributors for identifying victims. Having at least two primary relatives available in a pedigree tree is the current recommendation for missing persons and mass Disaster Victim Identification (Dr. D. Hares, personal communication) [11,18]. The highest contributing primary relatives are the parents and children of a missing individual [11]. Siblings are not ideal primary contributors due to varying degrees of relatedness between each other. This was supported by the combined statistical analysis of the mass disaster data sets. The number of primary contributing relatives was tested over both mass disaster
295
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296 Table 4 Results from a Kruskal–Wallis statistical analysis demonstrating the mean ranks and mean JPLR values compared for victims having 1, 2 or 3 plus primary relatives available in their pedigree tree. No. primary relatives
Mean JPLR
1
1.80E+11 N = 100 1.06E+18 N = 156 3.74E+14 N = 55
2 3+
Total
Mean rank 78.69 189.75 200.84
N = 311
scenarios. The more information made available in a pedigree tree from first degree relatives, the fewer number of ranks would be returned. Only the UHRs consistent with that pedigree tree were returned as ranks from the search, further reducing the number of fortuitous ranks. It was noted that the JPLR values increased significantly as more primary relatives were made available in a pedigree tree (Table 4). However there was not a significant difference seen when three or more primary relatives were made available compared to two (Table 4). An exception is when the parents are unavailable and siblings are able to provide DNA information. In this circumstance it is desirable to have more siblings available showing contribution of common alleles from the parents. 3.3. Relationship of relatives to victims Secondary relatives, such as grandparents, grandchildren, halfsiblings, aunts, uncles, nieces and nephews can provide DNA information in a pedigree tree and having both maternal and paternal secondary relatives is beneficial [2,4,5,8,9,13,19]. This was investigated statistically when the simulated data was analyzed to determine if the primary relatives provided more useful information when identifying victims compared to secondary relatives. Overall there was a significant difference seen between victims with primary relatives only, victims with secondary and tertiary relatives and victims with all relative types available (Table 5). There was also a significant difference seen between victims with primary relatives only and victims with all relative types available (Table 5). The number of ranks generated during a Pedigree Search was mostly dependent on the degree of relationship between the relatives in the pedigree tree and the UHR. For example, a pedigree tree with only secondary and/or tertiary relative(s) would generate more ranks, mostly fortuitous ranks, than a pedigree tree incorporating primary relatives. Siblings also seemed to generate more ranks compared to parents who generated fewer ranks overall most likely attributed to varying degrees of relatedness between each other. Our results indicate that primary relatives alone are better than a combination of primary, secondary and in very few cases tertiary relatives. Again this may be caused by a Table 5 Results from a Kruskal–Wallis statistical analysis demonstrating the mean ranks and mean JPLR values compared to victims having only primary relatives, secondary and tertiary relatives and all relative types available in their pedigree tree. Group
Mean JPLR
Mean rank
1 Primary relatives only
6.01E+17 N = 275 1.57E+5 N = 20 5.96E+10 N = 36
182.04
2 Secondary and tertiary relatives 3 Primary, secondary and tertiary relatives
Total
N = 331
25.4 121.56
confounding effect when using secondary and/or tertiary relatives with primary relatives. The addition of secondary and/or tertiary relatives to a pedigree tree seems to negatively contribute to confidence in assigning a UHR to a missing person node. This may explain the trend seen when the JPLR values are lower for pedigree trees with all relative types compared to pedigree trees with only primary relatives available (Table 5). The results strongly suggest that the primary relatives provide the most genetic information for victim identification compared to secondary and tertiary relatives. Alleles from primary relatives will be in a victim’s STR DNA profile while additional non-shared alleles present in secondary and tertiary relatives are not often present in the victim’s profile. They may only serve to expand the allelic pool of possibilities, thereby potentially decreasing the JPLRs or confidence in the relatedness. The type of primary relative is probably a factor as parents would potentially provide more useful genetic information even compared to full-siblings, who may or may not share a high genetic relatedness. 3.4. Contribution of mtDNA and Y-STR As expected, the secondary and/or tertiary relatives were not strong contributors for victim identification unless additional technologies, such as mtDNA and Y-STRs, were provided. This was noted by the higher combined likelihood ratio values assigned to UHRs with the addition of one or both of these technologies. There was a significant difference seen when comparing victims with additional DNA technologies available and victims with no additional information available (Table 6). Therefore the inclusion of mtDNA and/or Y-STR technologies improved the CLR for victims and in turn contributed positively to victim identification. The addition of more DNA technologies increases discrimination and certainty in assigning a victim to a pedigree tree. 3.5. Importance of profile type Additionally, it is recommended to have complete CODIS 13 core loci profiles for UHRs. The partial profiles contained a minimum of 8 complete loci and 2 partial loci where an allele had dropped out. There was no significant difference between the complete 13 CODIS core loci profiles, partial profiles based from the complete profiles and AmpF‘STR1 Profiler PlusTM profiles, however it is always beneficial to have more loci available in victim identification (Table 7). Alternative methods may be used to further identify UHRs from the same family. Direct reference samples can be used to clarify Pedigree Search results. For example when two siblings of the same gender are both returned as high ranks to a pedigree tree, they can be distinguished by their exemplar DNA profiles [2]. It is also recommended that more than one exemplar sample be submitted for each victim to counter the possibility of shared personal effects and partial profiles [4]. Metadata entered in CODIS 6.0 can be referenced for further individual identification.
Table 6 Results from a Kolmogorov–Smirnov statistical analysis showing the mean LRs for victims who had a JPLR value with no additional DNA information compared to victims who had a CLR with mtDNA and/or Y-STR DNA information available for their pedigree searches. Group
Mean LR value
JPLR with no additional information
2.59E+12 N = 183 1.21E+19 N = 164
CLR (JPLR plus mtDNA and/or Y-STR information)
Total
N = 347
296
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
Table 7 Results from a Kruskal–Wallis statistical analysis comparing mean ranks from victims with complete 13 CODIS loci profiles, victims with partial profiles and victims with only Profiler PlusTM DNA profiles used in their identifications. Group
N value
Mean rank
1 Complete profile 2 Partial profile 3 AmpF‘STR1 Profiler PlusTMprofile
143 22 12
90.33 89.68 71.83
Total
177
Table 8 Breakdown of the average JPLR and CLR values for all profile types in the airplane and Tsunami mass disasters scenarios.
JPLR
CLR
Complete profiles
Partial profiles
Profiler PlusTM profiles
Mixed profiles
1.58E+14 N = 127 9.55E+17 N = 173 4.53E+16 N = 127 1.18E+21 N = 174
4.16E+10 N = 20 2.12E+07 N=2 2.64E+13 N = 28 7.30E+10 N=2
1.97E+09 N = 12
1.59E+05 N=3
3.91E+12 N = 12
1.20E+05 N=4
processes; Dr. John Butler (NIST) for Y-STR databases and references; Insp. Kevin Miller (RCMP) for providing invaluable Disaster Victim Identification information; the NDDB CODIS administrators and Dave Morissette (Acting OIC NDDB) for their assistance and support; Centre of Forensic Science (Toronto, ON, CA); Laboratoire de sciences judiciaires et de me´decine le´gale (Montre´al, QC, CA); RCMP National Forensic Services CODIS administrators; Dr. Darrell Williams (RCMP) for his analysis of our data; Dr. George Carmody for his review of our database and statistical interpretation; and Tim Zolandz, Dr. Douglas Hares, Scott Carey and Jennifer Luttman for their assistance in reviewing a draft of this paper. We appreciate the strong support and keen interest from all involved in making our contribution to the FBI UAT testing of CODIS 6.0 a success. Appendix A. Supplementary data
N.B. The Tsunami mean values are shown in bold.
The creation of a victim list and use of a manifest when available is a valuable option as has been recommended by International Society for Forensic Genetics (ISFG) [11]. The victims can be crossreferenced with the list as they are identified creating a record of which victims remain to be identified. Even in the case of an open mass disaster a list can be made of reported missing individuals from the relative and investigator interviews. 4. Conclusions CODIS 6.0 offers many new features useful for the identification of mass disaster victims and missing persons. The new user interface allows for multiple tabs to be open simultaneously, allowing for smoother navigation while searching and ease of associating DNA information to pedigree trees in Pedigree Tree Designer. In addition CODIS 6.0 includes the option of using other DNA technologies to assist identifications. Identity searching and the new feature Pedigree Searching, are both available as options for investigating missing persons and mass disaster identification processes. The new features increase the efficiency of identifying UHRs from mass disasters and reduce the occurrence of fortuitous matches that cause delays in mass disaster investigations. The UHRs are identified rapidly and with confidence, provided in the calculation of a JPLR or combined likelihood ratio (CLR) value (Table 8). Acknowledgements The authors wish to thank Dr. John Planz, Steve Gammon and Melody Josserand from the University of North Texas for their generous review of our Y-STR profiles and for supplying our anonymous mtDNA sequences. We thank Dr. Benoıˆt Leclair (Myriad Genetics Inc., Salt Lake City) for the extensive database of DNA profiles; Dr. Chantal Fre´geau and Dr. Ron Fourney (RCMP) for consultation on SwissAir Flight 111 victim identification
Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.fsigen.2010.05.005.
References [1] A. Alonso, P. Martin, C. Albarra´n, P. Garcia, L. Ferna´ndez de Simon, M. Iturralde, A. Ferna´ndez-Rodriguez, I. Atienza, J. Capilla, J. Garcia-Hirschfeld, P. Martinez, G. Vallejo, O. Garcia, E. Garcia, P. Real, D. A´lvarez, A. Leo´n, M. Sancho, Challenges of DNA profiling in mass disaster investigations, Croatian Medical Journal 46 (4) (2005) 540–548. [2] B. Leclair, R. Shaler, G. Carmody, K. Eliason, B. Hendrickson, T. Judkins, M. Norton, C. Sears, T. Scholl, Bioinformatics and human identification in mass fatality ncidents: the World Trade Center disaster, Journal of Forensic Sciences 52 (4) (2007) 806–819. [3] B. Leclair, Large-scale comparative genotyping and kinship analysis: evolution in its use for human identification in mass fatality incidents and missing persons databasing, International Congress Series 261 (2004) 42–44. [4] B. Leclair, C. Fre´geau, K. Bowen, R. Fourney, Enhanced Kinship Analysis and STRbased DNA typing for human identification in mass fatality incidents: the Swissair Flight 111 disaster, Journal of Forensic Sciences 49 (5) (2004) 939–953. [5] L. Biesecker, J. Bailey-Wilson, J. Ballantyne, H. Baum, F. Bieber, C. Brenner, B. Budowle, J. Butler, G. Carmody, P. Michael Conneally, B. Duceman, A. Eisenberg, L. Forman, K. Kidd, B. Leclair, S. Niezgoda, T. Parsons, E. Pugh, R. Shaler, S. Sherry, A. Sozer, A. Walsh, DNA Identifications after the 9/11 world trade center attack, Science 310 (2005) 1122–1123. [6] C. Brenner, B. Weir, Issues and strategies in the DNA identification of World Trade Center victims, Theoretical Population Biology 63 (2003) 173–178. [7] B. Budowle, F. Bieber, A. Eisenberg, Forensic aspects of mass disasters: strategic considerations for DNA-based human identification, Legal Medicine 7 (2005) 230–243. [8] B. Leclair, S. Niezgoda, G. Carmody, R. Shaler, Kinship analysis and human identification in mass disasters: the use of MASS DISASTERKAP for the world trade center tragedy, in: 13th International Symposium on Human Identification, 2002. [9] Lessons Learned from 9/11: DNA Identification in Mass Fatality Incidents. U.S. Department of Justice, Washington, DC, 2006. [10] J. Butler, B. McCord, Y-STRs and mtDNA, in: 58th Annual Meeting of the American Academy of Forensic Sciences, Seattle, Washington, 2006. [11] M. Prinz, A. Carracedo, W.R. Mayr, N. Morling, T.J. Parsons, A. Sajantila, R. Scheithauer, H. Schmitter, P.M. Schneider, DNA Commission of the International Society for Forensic Genetics (ISFG): recommendations regarding the role of forensic genetics for disaster victim identification (DVI), Forensic Science International Genetics 1 (2007) 3–12. [12] KLM Website, www.klm.com/travel/gb_en/travel_information/on_board/sea ting_plans/737-900.htm. [13] http://www.interpol.int/Public/DisasterVictim/. [14] Interpol Disaster Guidelines. CBRN. Emergency Management. [15] NIST Website, http://www.cstl.nist.gov/div831/strbase/NISTpop.htmm. [16] Yhrd Website, www.yhrd.org/rcms/navigation/1000010.html. [17] http://www.mapquest.com/maps?city=Tofte&state=MN. [18] SWGDAM Subcommittee on Missing Persons and Mass disasters (2008) (Dr. John Planz, UNT, Committee Chair). [19] N. Ritter, Identifying remains: lessons learned from 9/11, NIJ 256 (2007).
Contents lists available at ScienceDirect
Forensic Science International: Genetics journal homepage: www.elsevier.com/locate/fsig
Disaster victim investigation recommendations from two simulated mass disaster scenarios utilized for user acceptance testing CODIS 6.0 Laurie Bradford *, Jennifer Heal, Jeff Anderson, Nichole Faragher, Kristin Duval, Sylvain Lalonde National DNA Data Bank, Royal Canadian Mounted Police, 1200 Vanier Parkway, Room 46, Laboratory Building, Ottawa, Ontario, K1A 0R2, Canada
A R T I C L E I N F O
A B S T R A C T
Article history: Received 4 January 2010 Received in revised form 6 May 2010 Accepted 15 May 2010
Members of the National DNA Data Bank (NDDB) of Canada designed and searched two simulated mass disaster (MD) scenarios for User Acceptance Testing (UAT) of the Combined DNA Index System (CODIS) 6.0, developed by the Federal Bureau of Investigation (FBI) and the US Department of Justice. A simulated airplane MD and inland Tsunami MD were designed representing a closed and open environment respectively. An in-house software program was written to randomly generate DNA profiles from a mock Caucasian population database. As part of the UAT, these two MDs were searched separately using CODIS 6.0. The new options available for identity and pedigree searching in addition to the inclusion of mitochondrial DNA (mtDNA) and Y-STR (short tandem repeat) information in CODIS 6.0, led to rapid identification of all victims. A Joint Pedigree Likelihood Ratio (JPLR) was calculated from the pedigree searches and ranks were stored in Rank Manager providing confidence to the user in assigning an Unidentified Human Remain (UHR) to a pedigree tree. Analyses of the results indicated that primary relatives were more useful in Disaster Victim Identification (DVI) compared to secondary or tertiary relatives and that inclusion of mtDNA and/or Y-STR technologies helped to link family units together as shown by the software searches. It is recommended that UHRs have as many informative loci possible to assist with their identification. CODIS 6.0 is a valuable technological tool for rapidly and confidently identifying victims of mass disasters. Crown Copyright ß 2010 Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: CODIS 6.0 Mass disaster Disaster Victim Identification DNA Joint Pedigree Likelihood Ratio Pedigree tree searching
1. Introduction Mass fatality incidents (MFIs), also known as mass disasters, are caused by accidental catastrophes such as airplane crashes, natural occurrences such as Tsunamis and intentional acts such as terrorist attacks. These incidents generally lead to a high death toll and mass destruction of property [1,2]. On September 2nd 1998, Swiss Air Flight 111 crashed in the ocean off the coast of Peggy’s Cove, Nova Scotia, Canada [3,4]. Two hundred and twenty nine passengers and crew members perished leaving investigators with 1277 human remains to identify [3,4]. This is an example of a closed mass disaster where the victims were potentially known by the airplane manifest. Open environment mass disasters like the Asian Tsunami that occurred on December 26th 2004 and the
Abbreviations: DVI, Disaster Victim Identification; MD, mass disaster; MFI, mass fatality incident; CODIS, combined DNA index system; DNA, deoxyribonucleic acid; NDDB, National DNA Data Bank of Canada; RCMP, Royal Canadian Mounted Police; FBI, Federal Bureau of Investigation; UAT, user acceptance testing; STR, short tandem repeat; Y-STR, Y chromosome short tandem repeat; mtDNA, mitochondrial DNA sequence; UHR, unidentified human remain; MP, missing person; bp, base pair; LR, likelihood ratio; CLR, combined likelihood ratio; JPLR, joint Pedigree Likelihood Ratio. * Corresponding author. Tel.: +1 613 993 8906; fax: +1 613 993 9944. E-mail address: [email protected] (L. Bradford).
World Trade Center attack in New York City on September 11th 2001, were associated with a larger number of victims who were not known [2,5–9]. DNA typing is a robust, consistent and reliable technique ideally suited for identifying victims of mass disasters [3]. Even highly degraded body remains can be typed using STRs (100–400 base pairs (bp)) [7]. For the above-mentioned mass disasters unidentified human remains and reference samples from biological relatives were typed using AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM kits (Applied Biosystems, Foster City, CA) [2–4,6,8]. In recent years, Y-STR and mtDNA techniques have assisted in victim identifications. Y-STRs and mtDNA are haplotypes that are not subject to recombination events [6]. Barring mutations, they are consistent across generations. Y-STR inheritance is via the paternal side and only occurs in males. MtDNA is inherited maternally and is passed on to all offspring [7,10,11]. The disadvantage to using only Y-STRs and mtDNA is that the genetic information is not discriminating at an individual level. Entire families will have the same Y-STR profile for the paternal lines and mtDNA sequences for all maternal lines, with the exception of mutations and heteroplasmy [10,11]. Mass fatality incidences pose a unique challenge for scientists attempting to identify the victim’s remains for their return to family members. The proper identification of victims is dependent
1872-4973/$ – see front matter . Crown Copyright ß 2010 Published by Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.fsigen.2010.05.005
292
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
on: the number of UHRs; cause of death and extent of body fragmentation; DNA degradation; recovery of the remains; integrity of the samples and the type of familial DNA reference samples available to aid the investigation [1,2,11]. Direct reference samples (DRS), also referred to as exemplar samples or personal effects, are samples directly taken from the victim or their personal belongings [2–7]. In the past, direct reference samples could not always be confidently assigned to a victim [4]. Other considerations in identification include the occurrence of genetic core repeat slip mutations between parents and offspring where the slippage is by one or two repeat units (e.g. 4–8 bp), the presence of rare alleles and the more common variants where the slippage may be off by only 1 or 2 bp [3,4,7]. Additionally, high impact mass disasters like airplane crashes can lead to the occurrence of comingled remains, resulting in victim profiles that appear as mixtures requiring interpretation [2–4,6,7,11]. CODIS 6.0 is a software program designed with new features for identifying victims of mass fatality incidences and on-going missing persons programs. One feature of CODIS 6.0 is the redesigned, userfriendly interface. Navigation through the many software modules is improved and more than one window is accessible at one time. Identity searches are still available and a new search feature referred to as Pedigree searching has been added. Pedigree trees can be generated in the new module Pedigree Tree Designer where relatives are associated to the pedigree tree and can be saved in Pedigree Tree Manager. The entire pedigree tree is searched against all of the UHRs to find a suitable rank with the associated relative DNA information. The ranks are displayed in the Rank Manager module of CODIS 6.0 where all the search results can be reviewed. Additionally Y-STRs and mtDNA can be added to assist in victim identification. Metadata (unique non-DNA identifying pieces of information specific to a victim) can also be added for a victim in the new software program and used to assist in their identification. The Pedigree searching feature uses the kinship calculations and produces a Joint Pedigree Likelihood Ratio (JPLR). The JPLR is the ratio of the probability of the missing individuals DNA profile being related to the pedigree tree (containing at least one relative’s DNA information) compared to their DNA profile not being related to the pedigree tree. This gives the CODIS 6.0 user greater confidence in assigning a UHR to a specific pedigree tree and leads to a reduction in fortuitous matches. The statistical calculations, inclusion of new identification technologies and data management capabilities were shown to be assets in victim identification. 2. Materials and methods 2.1. Simulated airplane and Tsunami MD design Two simulated mass disasters, an airplane crash representing a closed environment mass disaster and a Tsunami representing an open environment mass disaster, were designed by members of the National DNA Data Bank (NDDB) of the Royal Canadian Mounted Police (RCMP) for user acceptance testing (UAT) of a CODIS 6.0 pre-release package. Actual mass disasters were used as a guide in designing data that would challenge both the software program and the users during the testing phase. These mass disaster scenarios were created to test the features of CODIS 6.0, and specifically the pedigree searching feature and the outcome of the JPLR calculations. The data files are available to law enforcement agencies only. The simulated airplane mass disaster was designed first. The airplane model used for the mock mass disaster was a Boeing 737 capable of carrying 189 passengers plus a pilot, co-pilot and 6 crew members [12]. A person was assigned to each seat on the manifest in addition to a full crew complement. Names were created for each person and familial relations were delineated and pedigree
Table 1 A breakdown of the pedigree tree statistics for the victims of the airplane and Tsunami mass disasters. Mass disaster
Relative type
Percentage of total
Airplane
Primary only Two primary Primary and secondary Secondary only
84.1 44.7 11.2 4.7
Tsunami
Primary only Two primary Primary and secondary Secondary only
81.7 47.2 11.3 7
trees were then developed for each person on the plane. In total, 79 pedigree trees were drawn. They were designed to include primary and/or secondary relatives with some containing a lot, or very little family information in order to test the software capabilities (Table 1). For the purpose of the user acceptance testing, primary relatives included biological parents, children and full-siblings. Secondary relatives included grandparents, grandchildren, halfsiblings, aunts, uncles, nieces and nephews. In some cases the entire family was on the airplane and all were victims. There were pedigree trees where the biological father was not known and where a child was adopted so no biological relative information was made available. The airplane was broken apart in a simulated crash and a grid was drawn over the crash site and labelled in meters-squared based on an Interpol gridding system [13,14]. The UHR naming convention was based on the World Trade Center mass fatality incident and Interpol guidelines and required a maximum of 25 characters for CODIS 6.0. It consisted of a unique mass disaster number, a unique investigator’s number, the UHR number assigned by that investigator and the grid location of the body or part based on the X and Y coordinates [9,14]. The relatives were named based on the mass disaster number followed by the relationship to the victim(s) and the manifest seat number [8]. The manifest was numbered consecutively in order of the seat numbers. To aid in developing the number of STR profiles required for a project of this scale, the NDDB information technology support member created an Access database pedigree-generating software program. The program is able to calculate up to 15 DNA profiles within multiple generations of a pedigree. The AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM (Applied Biosystems, Foster City, CA) STR profiles came from a database of over 100,000 randomly generated profiles that were formed using a weighted random number generator created by Dr. B. Leclair (Myriad Genetics, Salt Lake City, UT) (Dr. B. Leclair, personal communication). The allele frequencies were representative of a standard Caucasian population and these frequencies came from the validation studies conducted at the RCMP for the AmpF‘STR1 Profiler PlusTM and AmpF‘STR1 COfilerTM kits (Applied Biosystems, Foster City, CA). An additional feature of the program is that all generated profiles are editable and customizable to suit project requirements, for example creating partial profiles or adding mutations. STR DNA profiles were generated for each victim and available relative(s) in the pedigree tree. In total there were 245 UHR profiles generated from whole bodies and body parts and 215 profiles from relatives. Some of the UHR DNA profiles were made to be partial profiles consistent with true findings from airplane mass disasters (Table 2). The partial profiles were created only from complete CODIS 13 loci profiles and were very reasonable having at least 9 loci present with some allele dropout present in some of the loci, and exceeding the NDDB’s current requirement of having at least 7 loci for CODIS searching. For the purpose of the UAT, we did not differentiate between partial and incomplete profiles. The mock
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296 Table 2 DNA profile breakdowns for the UHRs in the airplane and Tsunami mass disasters after building representative specimens. Mass disaster
Profile type
Percentage of total profiles
Airplane
Complete Partial Profiler PlusTM Mixtures/comingled
73.30% 17.30% 7.30% 2.1
Tsunami
Complete Partial
98.9 1.1
mass disaster included variants, rare alleles, mutations, comingled remains and direct reference samples presumed to be for a victim when they were not [4]. Additionally, the disaster data included a trisomy and a chimera to test the software capabilities. Y-STR profiles were created based on the loci from the AmpF‘STR1 YFilerTM kit (Applied Biosystems, Foster City, CA). Two Y-STR databases were used to create a total of 36 simulated YSTR profiles for the victims and relatives of the mass disaster [10,15,16]. The Health Science Center from the University of North Texas supplied mtDNA sequences for 58 victims and their relatives and they also reviewed the Y-STR profiles confirming that they were representative profiles (Dr. J. Planz, personal communication). Seventy-nine direct reference sample profiles were created for the victims. The exemplar sources included toothbrushes, hairbrushes, personal hygiene items (i.e. feminine hygiene products, diapers, etc.), underwear and an ‘‘other’’ category (i.e. tissue biopsy samples) consistent with Interpol guidelines and the source types used in the Swiss Air flight 111 mass disaster investigation [4,13,14]. These direct reference samples were named based on the order families and close acquaintances of the victims reported to the investigators. Metadata was also utilized in victim identification. The metadata that included: tattoos; piercings; X-rays of broken bones; dental X-rays and any unique identifying piece of information specific to a MP, was used to narrow down the identification to one individual. Metadata was generated for 24 victims and linked to their direct reference sample profiles. The Tsunami mass disaster data was generated in a similar manner to the Airplane mass disaster however it was representative of an open environment. A total of 82 pedigree trees were created for the Tsunami mass disaster consisting of primary, secondary and tertiary relatives. Not all of the persons reported missing were victims and not all victims of this disaster were reported missing, indicative of what might occur in an actual open environment mass disaster. A grid was drawn over satellite photos of the disaster site, Tofte, Minnesota, USA, based on a scale of 30 m squared [17]. The naming convention for the relatives and direct reference samples differed due to the grid format and absence of a known victim list. Similar to the airplane mass disaster, the UHR naming convention composed of four parts included a mass disaster number, investigator number, whole body or body part number found by the investigator and the grid location based on the x and y coordinates. Consequently, relatives were named based on mass disaster number, the order the relatives reported to the mass disaster investigative team and by their relationship to the victim(s). Once more the STR DNA profiles were generated for each victim and available relative(s) in the pedigree trees. In total there were 228 UHR profiles generated and 132 relative profiles. Very few of the UHR DNA profiles were partial in the Tsunami mass disaster. The principles for variants, rare alleles and mutations were applied here as in the airplane mass disaster. Included were 38 victims with Y-STR profiles created in the same fashion as the airplane
293
mass disaster and 62 victims and associated relatives, with mtDNA sequences supplied by the University of North Texas. In this case, 55 exemplar profiles were created to help identify the victims. These were produced in a similar fashion to the airplane mass disaster. Metadata was added for 26 victims for further assistance in their identification. Finally, Coroner’s Reports and death certificates were designed and generated for each victim for both disasters. The reports included: all body parts associated with a victim and the associated STR DNA profile(s); Y-STR profile data and mtDNA data if available; relative STR DNA profiles including their Y-STR and mtDNA data if available and the exemplar(s) STR DNA profile(s) associated with that victim if provided. In an attempt to best simulate a true mass disaster, the scientists who performed the data analysis were different scientists than those who developed the mass disaster scenarios and created the data. The data developers acted as coroners for the mock mass disasters. Their role was to confirm the identity of the victims and issue a death certificate as per a genuine mass disaster [11]. 2.2. Statistical analysis After completion of the UAT, the data from both simulated mass disaster scenarios were combined for statistical analysis as the searching protocols for each were the same. The statistical analysis was performed using a Kruskal–Wallis test or two-sample Kolmogorov–Smirnov test. Chi-square tests were performed to test the significance between variable groups. The statistical calculations based on the analyses performed were expected to reflect the importance of using primary relatives in victim identification. Having more primary relatives available was expected to show a significant impact in victim identification. Secondary relatives, in conjunction with primary relatives, were expected to contribute towards victim identification. Additionally the Combined Likelihood Ratios (CLR) which combine the JPLR with mtDNA and/or Y-STR likelihood ratios were predicted to improve confidence when included. Comparisons of complete profiles, partial profiles and Profiler PlusTM only profiles demonstrated the importance of having as many loci typed for each victim when possible. 2.3. CODIS data imports and searching protocols For each mass disaster scenario, CODIS import files were created for the UHR, relative and exemplar profiles. For the purpose of our user acceptance testing, separate import files were created for the mtDNA profiles and the Y-STR profiles, however, the STR and Y-STR can be contained in the same file. All mass disaster profiles provided were imported into CODIS 6.0, using a server/ workstation computer. The server was a Dell PowerEdge 1900 (4 GB Ram) with a Dual Processor Intel Xeon E5310. It contained a 160 GB hard drive, a gigabit Ethernet and ran on a Microsoft1 Windows Server 2003 and SQL Server 2005. For further information detailing configurations and searching procedures please see the supplemental information. 3. Results and discussion 3.1. CODIS 6.0 statistical calculations CODIS 6.0 software provided a combined likelihood ratio (CLR), JPLR and mtDNA and/or Y-STR DNA likelihood ratios (LRs) separately resulting from the Pedigree searching feature. The combined likelihood ratio includes the JPLR and likelihood ratios from the inclusion of mtDNA and/or Y-STR profiles for a pedigree
294
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
Table 3 A breakdown of the average JPLR, mtDNA LR, Y-STR LR and CLR for both mass disaster scenarios.
Airplane Tsunami
Average JPLR
Average mtDNA LR
Average Y-STR LR
Average CLR
7.55E+11 N = 163 9.39E+17 N = 176
884 N = 56 902 N = 54
3450 N = 32 2520 N = 25
3.35E+16 N = 172 1.16E+21 N = 177
tree. The JPLR alone does not include any additional information from these DNA technologies. Average JPLR, mtDNA LR, Y-STR LR and CLR values were calculated for each mass disaster scenario to demonstrate the JPLR values obtained in the simulated mass disaster searches (Table 3). Initially, an UHR to UHR Identity search was done to be able to add together samples from the same source (representative sample). Then missing person vs. UHR Identity AutoSearches were performed to match the victim’s direct reference sample(s) with their UHR STR DNA profiles from the mass disaster. The Identity AutoSearch compares the profiles at each locus and returns matches based on how many alleles at each locus matched. This is useful for identifying some victims prior to pedigree searching. The pedigree trees were searched individually and without the use of the Pedigree AutoSearcher feature. This was done to solve the mass disasters more efficiently by avoiding the high number of ranks that would have occurred from an AutoSearch on this large of a scale. The AutoSearch feature would have taken hours to complete in terms of checking each match for the mass disasters and these were mass disasters with fewer body remains than realistic mass disasters therefore the pedigree trees were searched individually. This can be partially attributed to the fact that all of the UHRs were entered into CODIS 6.0 simultaneously, compared to an approach where they are entered in batches. In a continuous missing persons program, the Pedigree AutoSearcher would be used on a routine basis however the challenges a mass disaster search imposes has specific differences compared to a missing person search mainly that the victims are frequently known and there will expectedly be a high number of matches resulting from searching. After a pedigree search, the JPLR calculations are returned in Pedigree Searcher and viewed in Rank Manager in the descending order. A true top rank can be defined as the expected result, i.e. the victim, being returned and having the highest ranking JPLR for the node being searched. In the case of the airplane mass disaster scenario, 89.3% of the time the true rank was returned in the top 5 ranked candidates and 90.2% of the time the true rank was returned in the top 10 ranked candidates. The average number of ranks returned for a searched node was 3.8. For the Tsunami mass disaster scenario 95.9% of the time the true rank was within the top 5 ranked candidates and 96.7% of the time the true rank was in the top 10 ranked candidates. The average number of ranks returned for a searched node in the Tsunami was 5.8. For both simulated mass disaster searches, 99% of the true ranks had JPLRs above 4.89E+1 and 95% of the true ranks were above 7.15E+02. Setting a minimum JPLR threshold from these numbers would be possible, however the consequences of threshold settings must be carefully weighed for each individual mass disaster. In total 72 ranks were returned without a JPLR value and 6 of these were the true ranks. This occurred due to the existence of extremely partial profiles not fitting the mutation model set within the software. Every UHR was successfully identified who could be identified. These identifications were made with direct reference samples, metadata, Pedigree Searches or Identity Searches. There were 6 UHRs remaining unidentified in the Tsunami, which were incorporated into the mass disaster design.
There were several factors that affected the presence of true ranks during the Pedigree searches. Very degraded profiles, comingled samples, or UHRs having mutations beyond the allowance of the mutation model were not returned as ranks upon performing a Pedigree search. The mutation model accounts for two core repeat unit mutations and minor slip mutations in a pedigree tree. In these above cases, or where no ranks were returned at all (for example, where a non-biological father was associated to the pedigree tree), identity search(es) were subsequently performed. It is therefore recommended that all Pedigree searches be complemented with Identity searches, to verify and confirm the ranks and to ensure that the true rank was not missed. It is also important to note that some ranks will not be returned if their JPLR values are below the set JPLR threshold of the Pedigree search configurations. Pedigree searches that return ranks must be viewed differently from conventional search results using pair-wise likelihood ratios. For instance, if two parents are associated to a pedigree tree and three of their children are victims of a mass fatality incident, then the returned ranks should include all three children, even though only one node is being searched. The rank returned for the node that is being searched is referred to as the expected rank by the authors. Subsequently, the other two children are not fortuitous ranks as they are part of the pedigree tree being searched and are equivalent to the searched node. They are also true ranks and referred to as equivalent ranks by the authors. Associated ranks refer to ranks returned for a related, non-equivalent node that belong to a different generation of the pedigree tree with the exception of cousins who are from the same generation. Take for example a pedigree tree with two children typed, with their parents and grandparents unknown. If a Pedigree Search is performed searching the father node then the father will be returned as the expected rank and the mother will be returned as an equivalent rank (Pedigree searches are not Amelogenin gender dependent however gender can be set in metadata and used as a filter) and the grandparents will be returned as associated ranks. A fortuitous rank is any rank that is returned and does not fall under the category of expected, equivalent or associated rank for that pedigree tree. They occur whereby a victim appears to be related to the kin but in fact they are not [1,2,6]. It is an STR DNA profile incorrectly associated with the pedigree tree being searched. It has been suggested that inclusion of more relative’s information may prevent or reduce the occurrence of true fortuitous ranks [4]. In many cases, these can also be attributed to common alleles in the profiles. These matches can be avoided by using more DNA information such as mtDNA or Y-STRs and having more primary relatives available for kinship analysis. The existence of mutations must be considered in data searching and the leniency allowed in search parameters can lead to more fortuitous matches [6]. The occurrence of fortuitous matches is greatly reduced in CODIS 6.0. 3.2. Contribution of primary relatives Both primary and secondary relatives contribute to victim identification. Primary relatives consisting of parents, offspring and full-siblings are ideal contributors for identifying victims. Having at least two primary relatives available in a pedigree tree is the current recommendation for missing persons and mass Disaster Victim Identification (Dr. D. Hares, personal communication) [11,18]. The highest contributing primary relatives are the parents and children of a missing individual [11]. Siblings are not ideal primary contributors due to varying degrees of relatedness between each other. This was supported by the combined statistical analysis of the mass disaster data sets. The number of primary contributing relatives was tested over both mass disaster
295
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296 Table 4 Results from a Kruskal–Wallis statistical analysis demonstrating the mean ranks and mean JPLR values compared for victims having 1, 2 or 3 plus primary relatives available in their pedigree tree. No. primary relatives
Mean JPLR
1
1.80E+11 N = 100 1.06E+18 N = 156 3.74E+14 N = 55
2 3+
Total
Mean rank 78.69 189.75 200.84
N = 311
scenarios. The more information made available in a pedigree tree from first degree relatives, the fewer number of ranks would be returned. Only the UHRs consistent with that pedigree tree were returned as ranks from the search, further reducing the number of fortuitous ranks. It was noted that the JPLR values increased significantly as more primary relatives were made available in a pedigree tree (Table 4). However there was not a significant difference seen when three or more primary relatives were made available compared to two (Table 4). An exception is when the parents are unavailable and siblings are able to provide DNA information. In this circumstance it is desirable to have more siblings available showing contribution of common alleles from the parents. 3.3. Relationship of relatives to victims Secondary relatives, such as grandparents, grandchildren, halfsiblings, aunts, uncles, nieces and nephews can provide DNA information in a pedigree tree and having both maternal and paternal secondary relatives is beneficial [2,4,5,8,9,13,19]. This was investigated statistically when the simulated data was analyzed to determine if the primary relatives provided more useful information when identifying victims compared to secondary relatives. Overall there was a significant difference seen between victims with primary relatives only, victims with secondary and tertiary relatives and victims with all relative types available (Table 5). There was also a significant difference seen between victims with primary relatives only and victims with all relative types available (Table 5). The number of ranks generated during a Pedigree Search was mostly dependent on the degree of relationship between the relatives in the pedigree tree and the UHR. For example, a pedigree tree with only secondary and/or tertiary relative(s) would generate more ranks, mostly fortuitous ranks, than a pedigree tree incorporating primary relatives. Siblings also seemed to generate more ranks compared to parents who generated fewer ranks overall most likely attributed to varying degrees of relatedness between each other. Our results indicate that primary relatives alone are better than a combination of primary, secondary and in very few cases tertiary relatives. Again this may be caused by a Table 5 Results from a Kruskal–Wallis statistical analysis demonstrating the mean ranks and mean JPLR values compared to victims having only primary relatives, secondary and tertiary relatives and all relative types available in their pedigree tree. Group
Mean JPLR
Mean rank
1 Primary relatives only
6.01E+17 N = 275 1.57E+5 N = 20 5.96E+10 N = 36
182.04
2 Secondary and tertiary relatives 3 Primary, secondary and tertiary relatives
Total
N = 331
25.4 121.56
confounding effect when using secondary and/or tertiary relatives with primary relatives. The addition of secondary and/or tertiary relatives to a pedigree tree seems to negatively contribute to confidence in assigning a UHR to a missing person node. This may explain the trend seen when the JPLR values are lower for pedigree trees with all relative types compared to pedigree trees with only primary relatives available (Table 5). The results strongly suggest that the primary relatives provide the most genetic information for victim identification compared to secondary and tertiary relatives. Alleles from primary relatives will be in a victim’s STR DNA profile while additional non-shared alleles present in secondary and tertiary relatives are not often present in the victim’s profile. They may only serve to expand the allelic pool of possibilities, thereby potentially decreasing the JPLRs or confidence in the relatedness. The type of primary relative is probably a factor as parents would potentially provide more useful genetic information even compared to full-siblings, who may or may not share a high genetic relatedness. 3.4. Contribution of mtDNA and Y-STR As expected, the secondary and/or tertiary relatives were not strong contributors for victim identification unless additional technologies, such as mtDNA and Y-STRs, were provided. This was noted by the higher combined likelihood ratio values assigned to UHRs with the addition of one or both of these technologies. There was a significant difference seen when comparing victims with additional DNA technologies available and victims with no additional information available (Table 6). Therefore the inclusion of mtDNA and/or Y-STR technologies improved the CLR for victims and in turn contributed positively to victim identification. The addition of more DNA technologies increases discrimination and certainty in assigning a victim to a pedigree tree. 3.5. Importance of profile type Additionally, it is recommended to have complete CODIS 13 core loci profiles for UHRs. The partial profiles contained a minimum of 8 complete loci and 2 partial loci where an allele had dropped out. There was no significant difference between the complete 13 CODIS core loci profiles, partial profiles based from the complete profiles and AmpF‘STR1 Profiler PlusTM profiles, however it is always beneficial to have more loci available in victim identification (Table 7). Alternative methods may be used to further identify UHRs from the same family. Direct reference samples can be used to clarify Pedigree Search results. For example when two siblings of the same gender are both returned as high ranks to a pedigree tree, they can be distinguished by their exemplar DNA profiles [2]. It is also recommended that more than one exemplar sample be submitted for each victim to counter the possibility of shared personal effects and partial profiles [4]. Metadata entered in CODIS 6.0 can be referenced for further individual identification.
Table 6 Results from a Kolmogorov–Smirnov statistical analysis showing the mean LRs for victims who had a JPLR value with no additional DNA information compared to victims who had a CLR with mtDNA and/or Y-STR DNA information available for their pedigree searches. Group
Mean LR value
JPLR with no additional information
2.59E+12 N = 183 1.21E+19 N = 164
CLR (JPLR plus mtDNA and/or Y-STR information)
Total
N = 347
296
L. Bradford et al. / Forensic Science International: Genetics 5 (2011) 291–296
Table 7 Results from a Kruskal–Wallis statistical analysis comparing mean ranks from victims with complete 13 CODIS loci profiles, victims with partial profiles and victims with only Profiler PlusTM DNA profiles used in their identifications. Group
N value
Mean rank
1 Complete profile 2 Partial profile 3 AmpF‘STR1 Profiler PlusTMprofile
143 22 12
90.33 89.68 71.83
Total
177
Table 8 Breakdown of the average JPLR and CLR values for all profile types in the airplane and Tsunami mass disasters scenarios.
JPLR
CLR
Complete profiles
Partial profiles
Profiler PlusTM profiles
Mixed profiles
1.58E+14 N = 127 9.55E+17 N = 173 4.53E+16 N = 127 1.18E+21 N = 174
4.16E+10 N = 20 2.12E+07 N=2 2.64E+13 N = 28 7.30E+10 N=2
1.97E+09 N = 12
1.59E+05 N=3
3.91E+12 N = 12
1.20E+05 N=4
processes; Dr. John Butler (NIST) for Y-STR databases and references; Insp. Kevin Miller (RCMP) for providing invaluable Disaster Victim Identification information; the NDDB CODIS administrators and Dave Morissette (Acting OIC NDDB) for their assistance and support; Centre of Forensic Science (Toronto, ON, CA); Laboratoire de sciences judiciaires et de me´decine le´gale (Montre´al, QC, CA); RCMP National Forensic Services CODIS administrators; Dr. Darrell Williams (RCMP) for his analysis of our data; Dr. George Carmody for his review of our database and statistical interpretation; and Tim Zolandz, Dr. Douglas Hares, Scott Carey and Jennifer Luttman for their assistance in reviewing a draft of this paper. We appreciate the strong support and keen interest from all involved in making our contribution to the FBI UAT testing of CODIS 6.0 a success. Appendix A. Supplementary data
N.B. The Tsunami mean values are shown in bold.
The creation of a victim list and use of a manifest when available is a valuable option as has been recommended by International Society for Forensic Genetics (ISFG) [11]. The victims can be crossreferenced with the list as they are identified creating a record of which victims remain to be identified. Even in the case of an open mass disaster a list can be made of reported missing individuals from the relative and investigator interviews. 4. Conclusions CODIS 6.0 offers many new features useful for the identification of mass disaster victims and missing persons. The new user interface allows for multiple tabs to be open simultaneously, allowing for smoother navigation while searching and ease of associating DNA information to pedigree trees in Pedigree Tree Designer. In addition CODIS 6.0 includes the option of using other DNA technologies to assist identifications. Identity searching and the new feature Pedigree Searching, are both available as options for investigating missing persons and mass disaster identification processes. The new features increase the efficiency of identifying UHRs from mass disasters and reduce the occurrence of fortuitous matches that cause delays in mass disaster investigations. The UHRs are identified rapidly and with confidence, provided in the calculation of a JPLR or combined likelihood ratio (CLR) value (Table 8). Acknowledgements The authors wish to thank Dr. John Planz, Steve Gammon and Melody Josserand from the University of North Texas for their generous review of our Y-STR profiles and for supplying our anonymous mtDNA sequences. We thank Dr. Benoıˆt Leclair (Myriad Genetics Inc., Salt Lake City) for the extensive database of DNA profiles; Dr. Chantal Fre´geau and Dr. Ron Fourney (RCMP) for consultation on SwissAir Flight 111 victim identification
Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.fsigen.2010.05.005.
References [1] A. Alonso, P. Martin, C. Albarra´n, P. Garcia, L. Ferna´ndez de Simon, M. Iturralde, A. Ferna´ndez-Rodriguez, I. Atienza, J. Capilla, J. Garcia-Hirschfeld, P. Martinez, G. Vallejo, O. Garcia, E. Garcia, P. Real, D. A´lvarez, A. Leo´n, M. Sancho, Challenges of DNA profiling in mass disaster investigations, Croatian Medical Journal 46 (4) (2005) 540–548. [2] B. Leclair, R. Shaler, G. Carmody, K. Eliason, B. Hendrickson, T. Judkins, M. Norton, C. Sears, T. Scholl, Bioinformatics and human identification in mass fatality ncidents: the World Trade Center disaster, Journal of Forensic Sciences 52 (4) (2007) 806–819. [3] B. Leclair, Large-scale comparative genotyping and kinship analysis: evolution in its use for human identification in mass fatality incidents and missing persons databasing, International Congress Series 261 (2004) 42–44. [4] B. Leclair, C. Fre´geau, K. Bowen, R. Fourney, Enhanced Kinship Analysis and STRbased DNA typing for human identification in mass fatality incidents: the Swissair Flight 111 disaster, Journal of Forensic Sciences 49 (5) (2004) 939–953. [5] L. Biesecker, J. Bailey-Wilson, J. Ballantyne, H. Baum, F. Bieber, C. Brenner, B. Budowle, J. Butler, G. Carmody, P. Michael Conneally, B. Duceman, A. Eisenberg, L. Forman, K. Kidd, B. Leclair, S. Niezgoda, T. Parsons, E. Pugh, R. Shaler, S. Sherry, A. Sozer, A. Walsh, DNA Identifications after the 9/11 world trade center attack, Science 310 (2005) 1122–1123. [6] C. Brenner, B. Weir, Issues and strategies in the DNA identification of World Trade Center victims, Theoretical Population Biology 63 (2003) 173–178. [7] B. Budowle, F. Bieber, A. Eisenberg, Forensic aspects of mass disasters: strategic considerations for DNA-based human identification, Legal Medicine 7 (2005) 230–243. [8] B. Leclair, S. Niezgoda, G. Carmody, R. Shaler, Kinship analysis and human identification in mass disasters: the use of MASS DISASTERKAP for the world trade center tragedy, in: 13th International Symposium on Human Identification, 2002. [9] Lessons Learned from 9/11: DNA Identification in Mass Fatality Incidents. U.S. Department of Justice, Washington, DC, 2006. [10] J. Butler, B. McCord, Y-STRs and mtDNA, in: 58th Annual Meeting of the American Academy of Forensic Sciences, Seattle, Washington, 2006. [11] M. Prinz, A. Carracedo, W.R. Mayr, N. Morling, T.J. Parsons, A. Sajantila, R. Scheithauer, H. Schmitter, P.M. Schneider, DNA Commission of the International Society for Forensic Genetics (ISFG): recommendations regarding the role of forensic genetics for disaster victim identification (DVI), Forensic Science International Genetics 1 (2007) 3–12. [12] KLM Website, www.klm.com/travel/gb_en/travel_information/on_board/sea ting_plans/737-900.htm. [13] http://www.interpol.int/Public/DisasterVictim/. [14] Interpol Disaster Guidelines. CBRN. Emergency Management. [15] NIST Website, http://www.cstl.nist.gov/div831/strbase/NISTpop.htmm. [16] Yhrd Website, www.yhrd.org/rcms/navigation/1000010.html. [17] http://www.mapquest.com/maps?city=Tofte&state=MN. [18] SWGDAM Subcommittee on Missing Persons and Mass disasters (2008) (Dr. John Planz, UNT, Committee Chair). [19] N. Ritter, Identifying remains: lessons learned from 9/11, NIJ 256 (2007).