Forensic Biology and DNA | Session 4 from the 2026 NIJ Forensic Science Research and Development Webinar Series

Tiffany Layne: Good afternoon, and welcome to the 2026 NIJ Forensic Science Research and Development webinar series. We're excited for you to join us. My name is Tiffany Layne. I'm a physical scientist in the Office of Investigative and Forensic Sciences, and I will be moderating today's session.

The National Institute of Justice, or NIJ, is the research, evaluation and technology agency of the US Department of Justice, and is dedicated to improving knowledge and understanding of crime and justice issues through science. Today's webinar is on forensic biology and DNA, highlighting transformative research and forensic DNA analysis and biological evidence examination.

Presentations will feature advances in DNA technologies, mixture interpretation, nontraditional biological evidence, genealogy applications, and emerging molecular methods that expand investigative capabilities.

Before we begin, we'll review some important information. To adjust your audio settings, click on the down arrow button next to the greyed out mic icon to adjust as needed. Questions can be submitted at any time and will be answered after each presentation during a Q&A session. To submit your questions, click the Q&A icon at the top of your screen. Please note we may not be able to address all questions. Closed captioning is available. To access, click the more button with three dots, select a language and speech, and then select Show Live Captions. A recording of the webinar will be posted to the NIJ website in the coming days, and certificates are not being issued for this webinar series.

So today's presenters are in the following order. First, Professor Bruce McCord of Analytical and Forensic Chemistry at Florida International University. Second is the chair and associate Professor Daniele Podini in the forensic molecular biology at George Washington University. Third is biochemist and professor Glendon Parker from the Department of Environmental Toxicology at the University of California Davis. Fourth will be Sarah Cavanaugh, who's a senior research scientist at Bode Technology. And fifth will be Dr. August Woerner, an associate professor of the Department of Microbiology, Immunology and Genetics at the University of North Texas Health Science Center.  

So with that, let's hear from Bruce.

Bruce McCord: Good morning everyone. Today I'm going to be discussing our work over the past few years, investigating the potential of microbial transfer for the detection of sexual assault using digital PCR and deep shotgun sequencing.

We are extremely grateful to the NIJ for funding our work and highlighting our project in the seminar series. As always, the comments and findings in this presentation are our own and do not necessarily reflect the official views of the US Department of Justice. I also want to thank Nur A. Hasan and Bassam El-Fahmawi of Mawi DNA technologies for their collaborative work and continuing support of this project. As most of you know, sexual assault is defined as a non-consensual sexual contact, often violent, that affects 1 in 3 women globally. In the United States, it actually estimated that over 81% of women have faced some sort of sexual assault.  

The development of forensic DNA typing over the past 30 years or so, utilizing PCR and automated methods for extraction, amplification, and analysis, has greatly assisted forensic investigation and has brought closure to many victims of crime. Legislation providing funding through the NIJ and other agencies has also brought about major advances in the field. Particularly important has been mandating that sexual assault kit analysis be expedited. Unfortunately, the results of many sexual assault kits can be compromised by poor recovery of male DNA due to overwhelming female DNA. These problems can often be overcome by Y-STR analysis.  

However, a second and even more important issue is the problem of perspective. Sexual assault is a crime of violence, and often male perpetrators do not ejaculate. In fact, as many as 60% of sexual assault kits do not show male DNA. An alternate method is needed to detect the assault and provide closure to victims who must submit to this invasive procedure. So our research question is — can we develop an assay that will not depend on Y-chromosome or autosomal DNA and still be able to detect sexual contact? And we believe that the forensic microbiome can provide this by basically giving us invisible traces that we leave behind. In this project, we've been looking at this other way, utilizing the microbiome. Microbial DNA, which contains bacterial, viral and fungal DNA, are present throughout our bodies.  

Recent estimates indicate there are at least as many bacterial cells in the human body as there are human cells. The types and variety of these bacteria depend on the location in the body and the physiological, cultural, and environmental aspects of where an individual lives. There have been many studies on microbial forensics with regards to poisoning, post-mortem analysis and human individualization, but at the time we started this work, there was very little on aspects of the genital microbiome, with most work involving aspects of disease and vaginosis. The germ of our idea began with this project, with our long-term collaboration of Doctor George Duncan, who was previously the DNA manager of the Broward Sheriff's Office Forensic Laboratory and is currently an adjunct professor at Nova Southeastern University and FIU.  

In an amazing project, Doctor Duncan assisted Jack Gilbert of Argonne National Laboratory in examining the potential of microbial DNA as a method for human identification and for tracking suspects. With the assistance of the Broward Sheriff's Office and Time magazine in 2015, they organized a fake burglary with multiple suspects in an abandoned home. The idea was to see if the actions of the burglars could be tracked throughout the home, using microbial signatures on the items they had touched.  

The results were amazing. Not only could they track the path of the individuals through the house, but they could also correlate the individual's diet and medical history using next generation sequencing and thus providing additional potential investigative leads. The team also produced an amazing video describing the results for which I wish I had time to show. The work of Jack Gilbert and others who had investigated the role of the microbiome on human individualization and the changes in the microbiome that occur as someone becomes sick or changes diet, provide us with much to think about.  

Thus, we were curious, if different individuals have unique microbiomes, could there also be traceable differences between the penile and the vaginal microbiome? Information in the literature suggested this could be true. Could these possibly be detected and used as a tool to track sexual assault? Our initial team included Doctor Duncan, Mirna Ghemrawi, formerly a member of my research group and now a professor at FIU, and Andrea Ramirez, a former graduate student at FIU. We decided to investigate this hypothesis. Okay, so here are our original results. To check to see if there were significant differences between male and female genital DNA, we undertook a short pilot study investigating ten individuals, five male and five female, using deep shotgun sequencing. Color in this chart indicates a different species of bacteria. The results were very interesting. Female vaginal bacterial DNA was dominated by two types of bacterial—Lactobacillus and Gardnerella.

The role of Lactobacillus is commensal. It lowers pH and reduces infection, while high levels of Gardnerella can indicate vaginosis. Male penile bacteria is more diverse—less is recovered, and a wide range of bacteria is seen. Interestingly, one of our co-authors, Rita Colwell, former head of the NSF and founder of CosmosID, who assisted in this study, mentioned that the profile of the male penile microbiome resembles what she had seen an analysis of DNA on the feet of houseflies. And you can take this as you see it. Okay, so a fundamental principle of forensic science is Locard’s Principle of Exchange, which he postulates that physical contact between two individuals or items results in a transfer of material. Would there not be a similar transfer following an instance of sexual assault?  

One might expect that a co-mingling could occur between the two microbiomes that would be detected on both victim and suspect. Thus, with assistance of funds provided by Mawi DNA technologies and later some additional funding, through the NIJ, once we had provided and developed the preliminary results you saw before, we embarked on a study of the transfer of microbial strains following sexual contact. This work was initiated by a contact between doctors El-Fahmawi and Ghemrawi at a scientific poster session. Interestingly, the seed funding directly from this interaction directly led to our first paper, which you saw, and to the current NIJ grant. My only comment about this is that there's lots of science that can get done in a poster session, and our deep gratitude to all involved in pushing this project further, including Mawi DNA technologies, EzBiome, and the NIJ.  

The results of our initial study on microbial transfer following sexual contact appears here. Seven different couples are examined with sampling conducted by the test subjects, by themselves, using sterile cotton swabs and gloves. The samples include pre and post coital samples of vaginal, labial and penile microbiome profiles. Immediately obvious to you, and I don't know if you can see my cursor, but basically, if you look at the next to last, profile, that's the penile microbiome before sexual contact and the last one is post sexual contact. And what you see is that the penile microbiome tends to disappear and begins to appear like the vaginal one. And to make it easier and more clear, let's go to the next slide where I’ve blown this up.

Okay. Here in this slide is a blow up showing four of the couples. Again, each color represents a different species of bacteria and the order is pre-coital, post-coital labial samples, pre-coital and post-coital vaginal samples, and pre-coital and post-coital penile samples. As before, the pre-coital penile samples, which is the fifth results in each of the four sets of of data, stand out very clearly. But we also see minor differences in pre and post coital vaginal and labial samples. Equally striking is the way the penile samples appear similar to the vaginal samples following intercourse. Thus, we can conclude that post-intercourse there is a significant transfer of the vaginal microbiome and somewhat less transfer because of the amount of the microbes on the penile samples, to the female samples as well. Questions also arrive. How significant are these results and how are they affected by time and by the activities of participants in terms of sampling? Also, can we individualize these results in some way, as is done with STR SNPs and mitochondrial DNA? This was the subject of the current NIJ grant.

To illustrate this a little bit further and show some statistical results using principal components of the data you just saw, the question of significance can be answered, by this slide, which demonstrates that prior to intercourse, there's a clear and compelling difference between the penile and the labial and vaginal microbiome. However, following sexual content in the penile microbiome cannot be distinguished from the labial and vaginal microbiome.  

Furthermore, we can also use SNP data to check how closely the post-coital penile samples relate to vaginal and labial samples, with slight differences between the couples. In the upper couple, E, the vaginal and post-penile samples are most similar, while the second couple, B, the labial, and penile samples are more similar. Perhaps these results suggest differences in behavior, not something that we actually checked in our data, but it's interesting. But perhaps most impressive of all is the fact that we can use the SNP data to check how closely, how we can individualize it across different couples. Here we examine the differences in single nucleotide polymorphisms of the bacteria across each of the different sample types post-coital.  

This is only possible because we are doing deep shotgun sequencing in a large format, Illumina NovaSec. This process allows us to detect SNP differences between the same couple, which is the white data, that is vastly different between that found among different couples, which is the blue data.  

So we compare two different couples and the data—the microbial information that we detect—and what you see is the number of SNP differences when comparing across the same couples are almost, are very small in the tens of differences where there's thousands, tens of thousands of differences if we compare across couples. What this is telling you is that the bacterial strains in the town found in the two couples are quite different, and the bacterial strains in the same couple can be individualized based on the SNP data, giving us a pathway towards not just individualization, but towards identification. In our current project, we are expanding these results by adding longitudinal studies for both female and male, adding more couples, and investigating the possibility of utilizing clothing as another way to check, DNA transfer and bacterial transfer.  

Again, we are having our subjects collect samples and submit them to us for deep sequencing and ultimate publication. And of course, all these results are collected using appropriate IRB protocols. The details of the sample collection studies are shown here. 50 groups of samples are collected with the total number of samples at 300. The samples include longitudinal male and female coital samples and undergarments. We're nearly finished with all our samples, including extraction and quantitation, and the first half of these samples are currently undergoing shotgun sequencing. The results of the quantitation are shown here.

As mentioned previously, the vaginal samples provide a much higher input of total DNA. And I might add that we use Qubit only for rough quantitation as it contains both human and bacterial DNA. While the qPCR uses cyber green for quantitation checked against a communal bacterial standard. The total amount collected as determined by qPCR is in the fourth column. And basically what you see is there's a significant, the larger amount of, bacterial DNA in the vaginal samples compared to the penile, as we showed earlier. That being said, we anticipate that the bacterial signatures transferred from the general microbes between sexual partners can also be screened using quantitative PCR assays as a presumptive test.  

Due to its sensitivity, we are utilizing nano plate digital PCR and the work that you previously saw on quantitation and this new study on, using a presumptive test, has been performed by Dipanwita Raha who is a new graduate student in our laboratory. The methodology for collecting the samples for our presumptive bacterial detection includes taking a portion of the extracted pre and post coital samples in the previous study, and developing a digital PCR assay for them. A number of bacteria have been previously identified, both by, Dipa and by Andrea. Basically, what we've noticed is that, among the most, commonly transferred bacteria include Lactobacillus crispatus, Lactobacillus iners, Gardnerella and Bifidobacterium breve. On the penile samples is Corynebacterium jeikeium, Staphylococcus haemolyticus, and two others.  

We ultimately plan to create a digital PCR multiplex assay to detect all of these bacteria as a presumptive test, for the detection of sexual assault. The primer and probe design is used by these bioinformatic tools. We're using Taqman hydrolysis probes. Our digital PCR system allows up to 12 multiplex amps. The digital PCR workflow is interesting. We're using nanoplate digital PCR, which allows us to directly count the number of positive and negative wells in a 96 well plate that contains thousands of individual net wells. We count the number of bacteria and correct it using a Poisson distribution.  

Because I'm running out of time, I can't really go over that. But basically, the fact is that within these nano wells, you could have 1 or 2 or many, bacterial genomes. And the idea is that we can correct for the distribution using a Poisson distribution. Real time or digital PCR can be used for this.

And the initial results, from Dipa are shown here. What you see is prior to intercourse, the Lactobacillus that’s detected on the penile is very slight and can be, and is greatly increased, following and post-intercourse. We also see a slight increase in the vaginal bacterial post-intercourse as well. But of course these are vaginal markers, so that would be what you'd expect. Showing the same in a different couple.  

Conclusions. There's a significant difference between the penile and vaginal microbiome. Sexual contact can be detected based on the easily detected transfer of the vaginal microbiome to the penile microbiome of the sexual partner. There's also a similar transfer that occurs male to female, but the quantity of bacteria transferred is much lower. Additional studies involving more couples, longitudinal transfer and transfer to clothing are underway, with the sequencing data currently being generated. We’ve also been developing presumptive test using real time and digital PCR.  

Our preliminary data shows that Lactobacillus and also Gardnerella can be promising targets to trace microbial transfer after sexual intercourse, additional bacteria being selected. I'd like to thank all of the co-workers and co-authors in this study. Major support for this research was provided by the National Institute of Justice. Additional funding through Mawi DNA Technologies. And we also thank QIAGEN for their support of this study, through the lending us of a digital PCR. Thank you very much.

Tiffany Layne: Great. Thank you. Bruce. We're now moving into the Q&A session. Please submit your questions in the Q&A box. We already have a couple in, so first one up for you, Bruce, is...you mentioned male assailants may not ejaculate, but won’t pubic hairs be found which can help in collecting data? If not pubic hairs, can't traces of sweat be used?

Bruce McCord: When I talked about this, I was mainly talking about the results of the initial processing of sexual assault kits. There was a study that we're referencing that I believe that took place in San Diego. And that's where this number, 60% comes from, from a large number of kids that were tracked and traced. Yes, I'm sure there are other alternative processes that may or may not be available. But in the cases that we had the data from that were reviewed that number stands.

Tiffany Layne: Another one is, it's an issue to get victim females to agree to a report or, claim, rather than agree with testing. Will this collection technique be considered invasive? Especially for the females with possibly religious concerns? Could there be another way, like, some other body fluid or something like that?

Bruce McCord: We are testing, of course, just vaginal and labial samples, first of all. Second of all, this test would be no more invasive than a standard rape kit given, probably less so. But of course, I would expect the rape kit would be done first. And this would be an alternate procedure when, no male DNA is found.

Tiffany Layne: Another one is, how do other body fluids, such as blood, impact these results? Have you looked at that?

Bruce McCord: Well, you would expect that the level or the amount of bacteria in fresh blood is going to be relatively low compared to the bacteria that you would find in the vaginal tract or on the skin. But we have not, that would be a subject and an interesting subject for future study. Of course.

Tiffany Layne: Okay. Another question is, are the donor couples collecting the samples themselves or by a health care practitioner?

Bruce McCord: They collect them themselves. We give them a kit and, some fairly complete instructions. As well as, some, minimal amount of data about their behavior that helps us, interpret the results.

Tiffany Layne: Another one is, how long after intercourse does the vaginal microbiome signature persist in the penile samples? Any time series component to this study?

Bruce McCord: I expect to have results on that in the next 2 or 3 months. When the sequencing data from our current work is being completed. And I'm very excited to be able to let you know soon. That is that is the subject of our current work.

Tiffany Layne: Okay. Another question is, have you studied if condom usage affects your results?

Bruce McCord: We have not. But that is another thing that we've been thinking about and of course, there's other aspects that could be looked at in terms of including digital penetration and other forms of assault that could be looked at. And, yeah, I think we're just scratching the surface of the things that could be done. Absolutely. Thank you for that question.

Tiffany Layne: Okay. So lots of questions. Another one is in a real case, would you be able to process sexual assault kit swabs that were already sampled for STR or Y?

Bruce McCord: Yes. We haven't looked at that, but I think that could be done. I'm looking at Dipa, who's with me. She agrees. Yeah. It would require, of course, validation.

Tiffany Layne: Okay. Another question is, has the part of the menstrual cycle been considered and how the transfer may vary?

Bruce McCord: Yes. That's involved in the way the samples are collected. Dipa do you have any comment on that?

Dipanwita Raha: Yeah. Hi We are currently looking at that because, along with the transfer study, we are also doing a longitudinal study where we are collecting the samples from females on their different days of their menstrual cycle. So we are also going to look at how that's, that are affecting the gender microbiome in general. So as I said, we want to have the results in next two months. But like we have sent out our samples for shotgun sequencing, we are waiting for that.

Tiffany Layne: Great, okay. Fantastic. Well, thank you so much for your presentation today, Bruce.

Bruce McCord: Thank you so much.

Tiffany Layne: We will now hear from Daniele.

Daniele Podini: Good afternoon. So thank you very much for, allowing me to present, our research here on capturing nuclear DNA from hair shafts. So, hair is a common type of evidence that can be found at crime scenes, but it generally doesn't get tested for DNA because of its challenges. As we know that nuclear DNA is very degraded in the hair shaft. And the mitochondrial DNA, is not as informative as nuclear DNA. So these types of evidence generally don't get processed. And we know the issues with, morphological comparisons. So we're not going to go down that route right now.  

So the goal of our project was to develop and optimize a method for extraction of DNA, and then subsequent enrichment for specific SNPs that could allow us for, identification of an individual, from the hair shaft. So, we started with obtaining an IRB, approval for collecting, samples, hair shafts, and or hair, I should say and buccal swabs, and some information from donors. We initially started with 50 donors, but we actually were able to expand it to 130 donors.

So we optimized an extraction method, using a kit from BioChain, a company that uses, that has developed this cell-free DNA extraction method that we adapted to hair shafts. And the process is, pretty standard. So we lyse the hair in a lysis buffer. Then we centrifuge to pellet the insoluble materials. So some non-digested melanin can then inhibit the following steps. And so we try to get rid of that. And then we use a mix of magnetic beads that are enveloped in silica. With the special binding buffer., we allow the DNA to bind to the beads, and then we purify the DNA, kind of conventional method. But it was originally developed for cell-free DNA.

Then we prepare libraries, using we prepared libraries using the Takara ThruPlex DNA sequencing kit. And, used, indices from the Takara kit. And then we did a couple of HiFi Hot Starts. So we generated a library for each one of our samples. And then we, proceeded to capture hybridization. I'm going to talk about the how we chose the SNPs in the next slide. But the process is basically developing, a panel of RNA probes; we use the Arbor Bioscience’s products. So these are 60 nucleotide probes that are bound with biotin and they hybridize to complementary sequences. And then we pull them out using magnetic beads that have a Streptavidin attached to them, and kind of similarly to you know conventional binding, I mean magnetic rack process.  

And so here is a bit of information on the targeted SNPs. So we targeted 813 SNPs overall. And the focus here was identity. Not necessarily kinship, but identity, developing a way that we could identify an individual. Obviously, because we're typing SNPs, there's no database. So we would need a reference sample. But that was the target. And we started with, kind of piggied off the force panel. Which targeted over 5000 SNPs. And so, our idea here was a smaller panel, initially thinking that the smaller would be more efficient. We realized afterwards that that really doesn't change, even a million SNPs can be captured kind of with the same efficiency. But at least with this smaller panel, we can better optimize coverage and real estate on a flow cell so we could do more samples compared to other greater coverage, greater number of SNP assays.

So there is an overlap here with the force panel. As you can see, 583 of these SNPs are part of that panel as well. So, what we did, we extracted the DNA. Then we took samples that yielded more than 50 picograms, quantified with Quantifiler.

We prepared the libraries. Captured and then sequenced on the NextSeq 2000. And then analyzed everything on the CLC Genomics workbench from QIAGEN. So this was, the this is a plot of an example of DNA.

The size of the DNA fragments that were obtained from a hair shaft. And you see this particular pattern of up and down, which is, around ten nucleotides. And so, oh, and this is instead the pattern that you get from the buccal swabs. From the buccal swabs, we extracted the DNA fragmented and then sequenced it. And this is the pattern that you see from a buccal swab. Obviously these are artificially fragmented because the quality of the DNA from the buccal swab is high. But just to show you the difference compared to the previous hair DNA. And so, we were thinking about this around 10 nucleotides and we know that the DNA is bound around the nucleosome. And other scientists have seen this from both from cell free DNA and from hair shaft and the distance between the length of a loop of the double helix is about 10.4 nucleotides. So when you take, two of these peaks, for example, starting from 48 to 90, you see that it’s actually equal to 10.4 times 5. So if you take a fragment that's 38 nucleotides, you add 10.4 times 5, it equals exactly 90, which is consistent with this 10.4. So what we think is that there's a part of the double helix, it's more susceptible to breakage. And that's why we see this pattern.

This is showing us the proportion of number of SNPs that we were able to type with the coverage greater than ten reads, plotted against the input DNA. So as we expect, the greater the input DNA, the greater are the amount of SNPs that we're able to genotype.

And this is the comparison between the buccal swab, and the hair DNA, plotted in increasing number of SNPs genotyped. So the black line is essentially the input DNA, and in green are the percentage of exact concordance between the buccal swab and the hair. In yellow is a partial concordance which is consistent with potential dropout from the hair. So the hair was homozygous and the buccal swab was heterozygous. And then, discordant. So this is in red. So we saw a little bit of discordance which is not unexpected in this type of low input DNA samples when doing this type of sequencing.

Then we took a software, FamLink2, that was designed to treat low coverage DNA for comparison purposes, for, well, it was designed obviously for, as the word says, “FamLink,” so to look at linkage and relationship testing from low coverage DNA. And this was done in collaboration with the armed forces DNA identification laboratory and Daniel Kling and Andreas Tillmar from Sweden. They had developed a pipeline, tailored to the force panel. So we took that and then we adapted it to our SNPs, to our set of SNPs, and we analyzed, we compared essentially the hair, the DNA obtained from the hair and the DNA from the buccal swab. So this is an example of the output. So each dot represents one of these analyses. In green are when we compared each. So we took 20 samples, 20 hair shafts and we compared them to 20 buccal swabs.

One of these was the correct donor. The other 19 were not. And we did this in, 5 different populations. So what you see there is the likelihood ratio, Log10 LR of the likelihood ratio, where H1 is when we compared the buccal swab to the actual donor. And H2-TRUE is when we compared the buccal swab, sorry, the hair shaft to a non-donor. And plotted is the obtained LR. So in this sense the probability of observing these genotypes if they are from the same individual versus from two unrelated individuals. We can zoom in here. And you see this is the ones that yielded the lowest LRs. And the x-axis represents the SNPs with the greater coverage than 10 nucleotides, than 10 reads. And if you see we're still in very high LRs for H1-TRUE.  

And the one that yielded the lowest is actually 3.28 times 10 to the 16. So it's a log LR of 16.5, which essentially around 10 quadrillions. So and this is from 5 centimeters of hair shaft. So if we have a reference we can actually identify the individuals. And this is the one that yielded the lowest, log LR. So the ones that had a greater coverage and yielded obviously more starting and greater starting amount of DNA were in log10 LR of 200s. So, you know, way over even a conventional STR kit. We also tested and this was using, something that was developed for the Force kit.

And we had Kim Andreaggi from AFDIL run this for us because they had a pipeline specific for this, to look at the ancestry. So we haven't, done that yet on all of our samples, but this is just to show that it actually is able to predict the ancestry with, you know, with decent accuracy. And this is one of the samples, so with principal component analysis.

And obviously, I couldn't leave out the microhaplotypes, which are a very informative markers, a lot more informative than an individual SNP. And this is an example of a region that we were targeting on IGV. So in our probes we targeted around 70 microhaps, very small microhaps, less than 70 nucleotides. And one of the sources of these was this paper from de la Puente. So when you zoom in to IGV, you can actually see that by looking at the sequences, you're able to identify the phase of, the two alleles from the two chromosomes. And so in this case one has the four polymorphisms and the other one matches the reference.  

So this microhaplotype, these particular haplotypes, are a lot more informative than the individual SNPs. And so this demonstrates that it's actually possible to do that. We haven't done the LR based on the microhaps, but we that's our next step. We're actually working on that right now. One of my student is looking specifically at that. And this shows three different individuals and how just that one locus they can be differentiated. So like I said earlier, it's a lot more informative than an individual SNP.  

One of the big things here is the bioinformatic aspect of this project. So, this slide shows you all the steps, but I want to focus now briefly on the bioinformatic aspect.  

So we've been using CLC but CLC is very expensive. And so as part of this project, we are collaborating with Chantal Roth who's a bioinformatician. And we've worked with her in the past on developing Java scripts for analyzing microhaps. And so we asked her to get involved in the project and develop a pipeline for analyzing data from capture hybridization. So starting from the FASTQ file, where she developed this pipeline that we can also batch multiple samples that does the conventional analysis. I won't go through every step, but there's the, you know, the alignment, the genotyping, and you can set a bunch of parameters and it's kind of very user friendly compared to, CLC.  

And our plan is once it's optimized is to give it to the community or whoever is interested in using it so it will be free. And this shows you how user friendly it is. Essentially you select the R1, the FASTQ files, the R1 and R2. And then you have a bunch of parameters that you can select depending on the type of analysis that you want to do. And then you just kind of click “go” and it does the analysis for you. So right now, we are testing another 30 samples as part of the process. And we're also looking at the microhaplotypes. And an interesting aspect of this project that we, are investigating now is that we're actually seeing that we're capturing a lot more than what we were targeting.  

So that, the number of SNPs, we're targeting 800 SNPs, but we're not specifically capturing, 99% more DNA. And so we're trying to see now if that is useful, what kind of information, if it's consistent across samples and so there's a lot more data than we initially anticipated. With that, I obviously want to thank this can’t be done without NIJ support and without the support of a lot of the members of the lab, particularly Doctor Elaine Lewis who is a research professor here at GW. The help of Andreas and Daniel from Sweden and also Kim from AFDIL. And obviously my students, Grace, Quinn, Mia, and Chantal has been an incredible resource.

That’s it, thank you. I'm happy to answer questions.

Tiffany Layne: Fantastic. Thank you so much, Daniele. Okay. We are now moving into the Q&A. First one up—is five centimeters of hair, is that a typical size for hair evidence?

Daniele Podini: There's a lot of variation.

So I don't have specific data on what's the average size, hair fragment size, that is obtained at crime scenes.

So we said five centimeters just because it was, seemed like a reasonable size, a hair. We cut off. Well, if it was longer, we cut off the two ends to make sure that we were not including the follicle of the hair.

Tiffany Layne: Okay, great.

Another question is, with the low input of the DNA from your hair, were your concordance results expected, or did you expect to see something different?

Daniele Podini: Well, we kind of went in this without knowing exactly, what we're going to find. So, I don't know. I would say I was very pleased for what we’ve seen, that there was a concordance between what, we were getting from the hair shaft versus the versus the buccal swab, even at the low level. Of course, as you saw in the plot, the lower the input, the greater the amount of partial concordance between the two. FamLink2 accounts for that, so when we input the genotypes, we input everything, even what is a mismatch. And obviously if it's a mismatch, it lowers the likelihood there. And if it's a full, you know, concordance then it's going to, you know, increase the LR of that particular SNP. And then also takes into account the coverage, meaning how much that particular allele is represented in terms of number of sequences. So the greater the number of SNPs covered, the greater the weight it will give to whether it's concordant or not.

Tiffany Layne: Okay, great. Another question is, can you speak to how you anticipate the microhaplotype data being stored, searched and matched?

Daniele Podini: So that's, a broader question and it's something that I am discussing with the microhaplotype working group, which is an international effort. We're trying to kind of address some of those questions, right. First of all, I think there needs to be consensus on the loci that the community wants to use kind of a bit like CODIS or the expanded CODIS, right? We have to all agree on which loci to use, and the parameters for selection. Should we focus on their ancestry information, should we focus on their size, or should we look at smaller versus larger? Larger loci are better for mixtures; smaller loci are better for degraded DNA. Should it be a combination of all these? So all these things are being discussed. So I don't have an answer for it. I wish I had, but I am trying to get one and I'm not alone in this effort.

Tiffany Layne: Awesome. Another question is, some hair is specific to, different parts of your body. And so, head hair was used in here, but have you looked at pubic hair or hair from other parts of the body to see if you get the same result?

Daniele Podini: That's a great question. No, we haven't looked at other parts. One, it's a lot easier to get an IRB just for hair than for pubic hair. So, and also I again, I don't have specific data but, I assume that in general crime scenes hair from our head is more likely than any other type of hair. Of course, in sexual assault, evidence that may be may not be the case. But here we were focusing on, you know, more general finding the hair on somebody's clothing that's probably hair from the head.

Tiffany Layne: Okay, perfect. Another question is, how diverse were the hair donors? Did the extraction process work equally well with hair from say Asian donors versus European donors? And did the age of the hair donor have any effect?

Daniele Podini: So we collected the age of the donor. But we haven't done a correlation with that. We've seen a lot of variation across the board.

And we've seen that darker hair tends to yield more. And we've seen that there's consistency between donors, a bit like a good shedder or bad shedder. There is, individuals whose hair tends to yield more DNA versus others that doesn’t. But there is variation even within the same individuals. So there's more data to look at to be able to answer that well.

Tiffany Layne: Perfect. Awesome. Well that’s time. So thank you so much, Daniele. It was great presentation.

Daniele Podini: Thank you. Thank you for the opportunity.

Tiffany Layne: Absolutely. So now we will hear from Glendon.

Glendon Parker: Thank you, Tiffany, and thank you to the NIJ for this opportunity to share our research, with you and with the audience today. So, just before jumping in, we have some disclaimers. These are my opinions, and not those of the NIJ or the US Department of Justice. We mention commercial products, but we are not endorsing any. For the purposes of this study, we will be talking about species and species-specific markers, and we define species here based on the algorithm and the taxonomic characterization by the National Library of Bioinformatic Information, the NCBI.

We acknowledge that there's a controversy as to what a species is, where those boundaries occur and, we defer to that. We will, we'll be using the word “species-specific” during this talk, and we're referring to that situation at the NCBI. We do intend to offer some of the technologies developed today as a commercial service for the forensic science community and as part of that, we also will be developing intellectual property.

So today we'll be sort of talking a bit about the statement of the problem. We'll be talking about what’s the current best practices within the wildlife forensics community. We’ll be introducing the concepts and some of the underlying technologies behind phylogenetic proteomics. And then we'll be jumping into the third project that was funded by the NIJ. At the end, we'll sort of have some conclusions. And also sort of point out some of the potential and limitations of the technology.

So wildlife forensic crime is actually fairly pernicious and growing threat, both to the wildlife but also to us as well. We live in an ecologically, we depend on the environment around us and that also, part of that is species that are sort of keystone species as part of that. And so if these species are lost, then we have environmental degradation and that threatens food sources. It threatens our well-being and part of that is a growing threat, both in the trafficking of some of these species. There's some perverse incentives—the rarer something is, the more it's going to be trafficked, the more valuable it is.  

And as part of that, people have recognized now for over 100 years, that we need to protect the environment that we live in. And as part of that, there's a range of legal standards and precedents to do that. The problem is, is that a lot of the traffic occurs through pelts and other sort of preserved taxidermied samples. So these are actually very poor sources of DNA. There's, as we just very beautifully illustrated by the previous talk by Daniele, there is not a lot of DNA in here and as part of the process of creating a pelt, you have to cook it in acid. There's a lot of different rather harsh chemical processes that are involved and these will destroy any residual DNA that's going to be in there.  

And so how do we approach this? How can we actually, bring in some new, ideas and technologies to try and determine the species of a given hair? As a result of that could then help in developing legal cases and enforcing some of the wildlife laws and treaty obligations that we have. Often we can also do morphology, but again, morphology doesn't have a lot of species resolution. And it's very difficult to scale. You need a lot of expertise and a lot of training in order to do that.

So currently if we have a good, sample that has DNA, we would collect the sample, we would extract out the DNA and we would target typically amplicons in mitochondrial DNA. A lot of these proteins that are part of the electron transport chain. These are able to be very sensitive. There's a lot of copies of mitochondrial DNA in tissues. And also there's a lot of resolution. A lot of these you can actually get in and speciate. There's a lot of, different, DNA, a lot of evolutionary processes at play here. So you can actually resolve closely related species. One of the things that I would just point out, however, is that all of the SNPs that are occurring, about half of them, actually change in amino acid assignment. And I have a little panel here in the middle of cytochrome c oxidase subunit one. This is within the primate taxa here. And we are in the middle there with humans. And this is actually a human specific, sequence of protein that relates to human specificity that is in a DNA barcode. So protein is also a carrier of phylogenetic information.

So what does that look like. So obviously over time and over sort of phylogenetic and evolutionary distance, what starts out as a mutation can become a common minor allele then a major allele. And eventually a small subset of these are going to lock in as being represented in each member of that population. At that point, it can then become a species marker and then accumulates as long as the descendants of that species survive. And this is one example. Here, we're looking at amelogenin in that figure below it. This is from European bison and from mammoths.  

And you can actually see that there's been an accumulation of different amino acids that are occurring. And then and it's very sort of, it's preserved very well. This is an example, the bison data is from a 400,000 year old individual. So the question is, can we actually, are we going to get enough of these variants to actually do and obtain some of these specific markers? Are there actually markers in for that can be used to identify species unambiguously?

So next slide. So what we have here may also be illustrated here on the right figure that we have here. So you can sort of see over evolutionary distance, you can accumulate some of these markers. And so if there's less evolutionary distance it's going to be less sequence or less difference between them. And so you have less to start with. But once you have these sort of differences between different species, there's a series of filters that occur both biological and technical. The biological ones, well, they have to be expressed in hair. Now, there's about 800 proteins that occur in hair.

So there's going to that's a fairly large window on the genome, but it's not the whole genome. Also, proteins occur at different levels. And so some might be below the level of detection. Either way you can end up getting very large numbers of peptides detected in a hair sample, many thousands. And so that gives you a starting point where you can then go in and start to filter out those and actually try to identify, develop workflows for identifying the small subset of peptide sequences that only occur within that given species. At the end of that, after we have a data set, we can then do additional filters to then say these are ones that are going to be better for developing as biomarkers for use in forensics and in other wildlife studies.

So this is the process that we went through. We did a process called peptide spectral matching where we would have mass spectrometry. We would go and actually screen, we would go through a workflow, we would process the samples, we would then screen those, and then we would go ahead and obtain those sequences of amino acids within that given sample. We would generate a peptide list, and then we would consolidate those sequences, we would break it down and we would say, well, we'll get rid of repetition and we'll bioinformatically remove different mass shifts or different modifications on these peptides. We would then select out those that would be more likely to have both taxonomic information, but also would perform well in mass spectrometry.  

From that, we then had a very large data set that we used. And so we would do an internal search initially only to eliminate—actually the vast bulk of those that would occur in different species from our own data. And then we ended up with something like about 30,000 sequences that we then, submitted to the NCBI and then get the answers back as to how many species in the NCBI actually had that given sequence. And so from that, we were able to pull out those that only showed up with one individual or one individual species.  

We then developed an additional test ourselves, where we actually went ahead and downloaded all of the mammalian proteomes on our own computers, and then we actually screened them a second time. And this time, we were able to add different sort of additional stringency to that as well.  

So this is the overall results that we had. So we went ahead with the third project here with the NCBI. We had 45 species that we worked with. We processed 175 individuals. We did basically three individuals from each of these species, and then we went ahead and did each of those in biological triplicates. So we ended up with something, 465 mass spectrometry data sets. We then conducted, we didn't have, we weren't able to get high quality proteomes for some of these species. Unfortunately, they're not yet available, but we're able to do it on 40 of the species.

We then conducted bioinformatic analysis of a mass spectrometry data. We ended up with a total of 600,000 peptide identifications. And then we were able to screen it down for those different steps that we were talking about before. In the end, we ended up with something like, 1426 peptides. And from that, that at least passed the NCBI initial tests that we did. So this is on average about 36 per species. After doing a more stringent test, it we're looking at, different patterns, different mass patterns that we're looking at. We were able to bring it down to 1183. So each one of these peptides only occurs within one species. And, we were able to get that in 21 plus or minus 28 individuals. So there's a very large error there. And that's because when you're looking at sister species, there's going to be very low numbers. When you're looking at species that are isolated phylogenetically like pangolins or, rabbits and things like that, you're going to get, you're going to have, a lot of different, a lot of unique, sequences.

So then we did a final selection basically trying to screen them based off characteristics that actually would be favorable to developing these as biomarkers. So, we selected for those which had very high signals, that they showed up regularly in our data that they had good performance in that they came off at the same time on the instrument and that they're affordable. And for that you needed small peptides and also peptides that don't have special chemistries on them. So we developed this algorithm at the end of that. And we were aiming for the best five per species. Some species we had less than five. And so we had to deal with what we had. But in the end we had 186 peptides now suitable for development as a test where we could actually screen these in the as an open resource.  

So I just thought it would be good to just sort of take you through one of the steps or the steps for one of these, peptides. So when we are actually going through and we're doing our mass spectrometry, we're looking for masses. And when you get that, you get two pieces of information, you get the actual mass, but you also get how much of it is present there. And what you can see here on this sort of figure is that we’re at a given point in the run. So this is something, that occurs every two seconds. You actually get a screen and this is something that's occurring over a, this is something that's occurring over like an hour run. And so you can imagine that there's a lot of these scans that are happening, many thousands. And you can actually see each one of these as a peptide. You can see that if we take one of these tiny ones here, that has a mass to charge ratio of 692 that it's occurring at a given point in this run. Then you could actually fragment it. And so I take that one and do additional chemistries on it or physics on it. Next slide.

When you do that, that tiny little peptide breaks up. And then you can measure the pieces. What are the size of those pieces? It's called a fragmentation spectrum. And while it is difficult to sort of look at and interpret, it actually follows a lot of fairly strict rules as to how it breaks up. And so we could actually take that and compare it to a library of possible sequences. And then we can actually resolve what a lot of these different peaks are.

So when that happens with this particular sequence here, we're able to identify most of these fragments. Some of these larger ones can also be, assigned as well. And there's some different features here. You'll notice that it is very precise. So that's the error rate occurring here is within 0.04 Daltons. So one hydrogen atom is one Dalton. And so we're talking about very, very accurate, very precise measurements here. The other thing is we have a lot of these signals. We're able to measure the current of each of these. So one molecule is basically one electron. And so, or one proton in this case. And so you could actually take these and measure all of them. And then we see up to a million or more of these different peptides in a given run. And we have them here, we do a logarithmic transformation, and you can see here that it occurs in all nine of these individuals. The feature of this particular peptide, of course, is that the only way to make to get this particular pattern is that you have to have that particular combination of amino acids. And the only way to have that is if you were a bobcat. The bobcat is the only animal currently in the databases where this actually has that particular combination of nucleotides that result in that amino acid sequence. So we have something that is unique to a given animal.

The amount that we're actually applying onto the mass spec is half of a microgram, which is, I think, the equivalent of probably 0.1mm of a hair shaft. So we're talking about exquisite levels of sensitivity, and we're able to detect many orders of magnitude. So some of the molecules we're seeing at ten to the 12 and things like that, we see six orders of dynamic range. In other words, we could handle up to a range that covers something like 1 million fold differences. So it's very sensitive, exquisitely precise. And these proteins have survived. This is from a taxidermied sample.

We could do additional validation on it. And so we could actually predict, using AI and other sort of machine learning tools, what the what we should see, what sort of markers should show up and the degree with which we should see these fragmentations. And so the bottom plot is the predicted and the top plot is what we see in the data. So this acts as a validation of what we're looking at. So there are limitations that we talked about before. So we have a lot of pluses. But it's also you don't always see peptides. You rely on peptides being at a certain mass. There's some that are very long, some that are too short. We rely on the proteins being present. So there's a whole range of filters. So, if we don't have a huge amount of genetic differences at the front end, then sometimes you get to the end of this process and you end up with very little. That being said, what is the effective species resolution? Now DNA barcoding, I think at this point is slightly better.

You are able to resolve sister species quite well. But, with this methodology, you are able to get, you're able to have something that is more robust and is going to be present in highly degraded samples where there's no DNA being present.

There is some phylogenetic erosion that you would see. That's true of DNA as well as with proteins. And phylogenetic erosion, basically is a fancy way of saying that if you get a new species that's closely related, what was quote unquote “species-specific,” may not be once you actually get additional information. One way to protect against that is to get a lot of different potential markers at the front end, and you will lose some as you get more information. But ultimately, you should have enough at the end.

Just to conclude, this is a third fundamentally different way of doing analytical, of actually doing species identification.

Manufactured fur is a very poor, almost impossible source of DNA barcoding. And now we have a way to go ahead and do it and to do that. We're able to go ahead and we could discover these specific markers in 39 to 40 species that we had. So, we believe that this is a way, a third fundamental way, of actually doing species identification.

And just to thank everyone.

I'd like to thank my team at the University of California Davis. I'd like to thank my colleagues, at the CDFW, California Department for Fish and Wildlife, particularly Ashley Spicer, who is the Co-PI on this. Barry Baker at the US Fish and Wildlife Service. And of course, the Department of Justice. Thank you.

Tiffany Layen: Fantastic. Thanks so much.

Glendon Parker: I'm sorry, I thought I was getting carried away.

Tiffany Layne: No, you're totally fine.

Glendon Parker: Okay.

Tiffany Layne: So now we're going to move into the question and answer session. Please submit your questions into the Q&A box.

We have a couple in. So how much hair do you need to use for this species identification?

Glendon Parker: We do, we apply 0.5 micrograms. That's, or even less, frankly. We basically do a single shaft that's 20mm long, but we're only applying a very small amount into the mass spectrometer. So in theory, we could probably go down to 1 or 2 mm or even less. But we did 20 for this purpose, and we just did a single hair shaft, 20mm. And we only applied less than 1% of this onto the instrument.

Tiffany Layne: Awesome, great. Is there any other data you can gain from looking at the single fur hairs? Are you able to see anything related to environmental or dietary conditions for the species?

Glendon Parker: Not with this particular analysis. This is fairly focused and specialized, but there is a lot of other information in the hair shaft. You could, there are people who are able to do isotopic analysis, which is a different sort of workflow, but you can get that from it. So you can work out where hair came from based off the pattern of strontium and oxygen isotopes that you can extract out of the protein in the hair.

That's certainly possible. It's just it's not with without particular analysis. I should also point out that with humans, this is part of my other work, we can look for not just species-specific variants, but we can look at any variant. So we could also do non synonymous SNP genotyping, which you can do in humans. But you can't do that in animals because there just isn't enough representation to develop the databases to get a low frequencies.

Tiffany Layne: Awesome. Well thank you so much. You have more questions that we'll be sending to you later. So thank you so much Glendon. Great job.

Glendon Parker: No problem. Thank you for the opportunity.

Tiffany Layne: And now we will hear from Sarah.

Sarah Cavanaugh: All right. Thank you to the NIJ for the opportunity to present our current work on evaluating target enrichment to improve genotyping by sequencing of skeletal remains.

For my requisite disclosure, this project was funded by the National Institute of Justice. Any opinions or conclusions are my own. And any mention of commercial products does not imply any sort of endorsement. All right. Take a look at the NamUs June 2026 report. Over 16,000 unidentified unclaimed persons cases reported to NamUs have been solved in the last ten years. Forensic investigative genetic genealogy has become crucial to naming and returning hundreds of these solves. To define FIGG, it's a technique that combines the scientific process of developing a SNP profile with the specific intent to identify different distant relatives, such as by uploading to and searching a genetic genealogy database. However, as I have highlighted in orange, thousands await resolution. The problem is that many of these cases present unique challenges.

And these unique challenges and potential limitations continue to emphasize the need for technological improvements to the overall process. Here I show a rather simplified process map of steps necessary to resolve unidentified remains cases from discovery to DNA extraction and processing and identification. Identification can entail several different methods, but for the purpose of this presentation, we're going to focus on the FIGG route, utilizing genealogy databases. Foremost, these challenges are derived from the state in which the human remains are recovered. Location of recovery, environment exposure, conditions of the remains, age of the remains, ultimately how they're stored, all impact the quality of recoverable DNA. Presence of exogenous non-human DNA content—environmental, bacterial, fungal species—can overwhelm and damage the low human content.

Previous BODE-University of Tennessee collaborative grant demonstrated such bones can be unsuccessful with the standard method of genealogy SNP typing via microarray. So we come to the need to continue improving laboratory processes of DNA extraction and processing to overcome the limitation imposed by the conditions of the remains. There has been a push to apply techniques pioneered by the ancient DNA community to improve steps two and three.

To produce the most useful information for FIGG based identification. This study primarily focuses on method improvement to DNA processing. Next please. Where we perform high density SNP genotyping via next generation sequencing, and look at optimizing the library preparation by comparing library preparation from double stranded DNA to library preparation from single stranded DNA, as it has been demonstrated that highly degraded age remains may contain significant proportions of highly nicked or denatured and those single stranded DNA, and incorporate methods to isolate human content via hybridization capture. Hybridization capture can either target and capture the whole human genome from a background of non-human content, or we can limit the target regions to only those of interest for our downstream SNP genotyping purposes.  

While a whole genome method can recover FIGG relevant SNPs, here I demonstrate that targeting the whole genome actually results in limited capture of the SNPs we’re most interested in for FIGG purposes, but utilizing a target capture method, we increase the human specific proportion and increase the sequence read coverage depth of these FIGG relevant loci ultimate, resulting in more valuable genotype calls for genealogy database comparisons. Therefore, the goals of this project were as follows. Expand the current body of FIGG protocols to enhance the ability to identify unknown human remains. Develop a protocol to enrich human DNA from environmentally contaminated compromised bones. Develop a protocol to generate more than 600,000 SNP genotypes for FIGG applications that can be performed in forensic laboratory under forensic QA practices, and ultimately reduce current backlog of missing persons. I'll start by discussing the experimental design to compare DNA processing methods.

The processing workflow was first evaluated with control samples to assess input sensitivity and parameter optimization. These samples included single source blood extracts and controlled mixtures of human blood extract, combined with a microbial DNA mixture consisting of three common human associated bacterial strains. These single source extracts and human microbial DNA mixtures were also subjected to controlled degradation through mechanical shearing and UV irradiation. Once optimized, the workflow was applied to skeletal samples. These samples consisted of ten bones obtained from the University of Tennessee Knoxville's Forensic Anthropology Center. The originating bodies were allowed to skeletonize while exposed to the Tennessee environment.  

Some were left exposed on the surface. Some were buried in shallow graves. Seven of the ten bone selected for sampling are shown here. Nine of the ten bones selected were right calcaneus, or heel bones. The 10th bone was a left first metatarsal. Bones were selected based on our previous work, it was attempting to genotype aged bones using microarray technology. These particular bones performed poorly with the microarrays, and also demonstrated high concentrations of exogenous DNA from a variety of environmental, bacterial and fungal species. Up to a microgram of bone powder was sampled from each bone, and we were also supplied with paired blood samples, from each bone donor to generate reference profiles. Here is the general overview of the processing workflows tested with all of our samples. For the bone specifically, they were extracted with BODE bone prep, BODE’s proprietary bone extraction protocol that has been validated for forensic analysis of human remains. Up to 500 micrograms of bone powder were utilized per extraction. Thus, all the bone samples were divided into at least two extractions.

Extracts were quantified with Quantifiler Trio and Qubit Double Stranded Broad Range assay. The extracts were then tested with two library prep methods, which were subsequently directed into two methods of targeted hybridization capture, which I'll describe in detail in just a bit. For comparisons and, downstream concordance, one nanogram of bone extract was also genotyped with the Qiagen Forensic Intelligence Kit. For this effort, these are the two commercial kits tested. The TWIST Bioscience’s library preparation kit or library preparation EF Kit 2.0 and the IDT xGen single stranded DNA and low input DNA library prep kit. The TWIST method begins with enzymatic fragmentation to fragment DNA molecules to about 200-250 base pairs. Then the method is a standard double stranded DNA library preparation, with end repair and a-tailing to polish the fragment ends and allow ligation of universal adapters. Following AB-based cleanup, the adapter ligated fragments are amplified to add unique dual indexes and Illumina P5 P7 adapters, producing the ready to sequence libraries.The IDT xGen method begins with mechanical fragmentation using a Covaris focused ultrasonicator to again fragment DNA molecules to about 200-250 base pairs. Then, the samples denatured to convert all DNA molecules to single stranded DNA. Next, the xGen chemistry uses a proprietary enzyme called adaptase to end repair and amend tails and a stubby R2 adapter. The complementary strand is extended from this R2 adapter. Then an R1 stubby adapter is ligated to the original DNA strand.  

Finally, libraries are amplified to add unique dual indexes and P5 P7 adapters. Ultimately producing directional libraries based on the original template strand. Protocols were followed according to the manufacturer's recommendations, with the exception that 15 cycles of library amplification performed with both kits. All amplified libraries were purified  via bead based cleanup and library QC consisted of Qubit double strand DNA fluorometry and fragment sizing on an Agilent tape station.  

The generated libraries were then subjected to target enrichment, and the two methods tested for this project were the Daicel Arbor Biosciences MyBaits Compass 1.2 million SNP kit, and a custom panel purchased from TWIST Biosciences. First, the custom assay was constructed with assistance by TWIST to target just over a million FIGG relevant SNPs across all autosomes X and Y. The TWIST custom assay is actually built on top of TWIST’S commercially available diversity panel, which captures the SNPs found on Illumina’s GSAB chip that we added in some Ancestry.com specific SNPs, all of the Kintelligence SNPs, and the FORCE panel SNPs.

The TWIST assay utilizes 120 base pair double stranded DNA biotinylated probes to capture both forward and reverse strands of the target SNP. Input for this assay is up to 1500 nanograms of library, with hybridization at 70°C for 16 hours. The myBaits compass assay is now commercially available and targets 1.2 million FIGG relevant SNPs across all autosomes XY and also full mito, with single stranded RNA probes. Input for this assay is up to a thousand nanograms of library with hybridization at 62°C for 48 hours. Approximately 700,000 SNPs overlapped the two panels.

The schematic in the center was taken from G et al. and nicely depicts the comparison between the two capture methods tested. So the dark and light gray strands represent on target and off target sequences, respectively. The blue strands represent biotinylated probes, with the biotin attached to the three prime ends. Once hybridization is complete, libraries are incubated with streptavidin beads to capture the biotin molecules isolated by a magnet, washed, removed, unbound off target molecules, and then finally subjected to a limited cycle library amplification and a final bead based purification.

Our captured libraries were once again QC’d with Qubit fluorometry or Kappa QPCR library quantification and Agilent tape station fragment sizing. All libraries were sequenced on the Element Biosciences AVITI using cloud break free cell chemistry in 2 x 150 cycle runs. For QC purposes, our pre captured libraries were low pass sequence first, generating about three gigs of data per library. Then enriched libraries were sequenced to generate up to 20 gigs of data per library. And our bone donor references were shotgun sequenced to generate about 100GB of data per library. All libraries were then analyzed through the pipeline shown, which utilizes open source methods commonly applied as best practice for the analysis of human genome sequence data.  

Pre-processing includes fastQC evaluation, adapter removal and reemerging; reads are then mapped using BWA-MEM and deduped with MarkDuplicates. The pipeline enables several methods of data quality review, such as damage assessment and endogenous DNA estimation. Generates a summary report with multi-QC. BCF tools is then used to perform genotype calling and filtering, and a custom script converts the BCF tool's output into a GEDmatch compatible format. Alright take a look at some of the trends we've observed. So comparing just the two library preparation methods, the TWIST double strand method generates statistically higher library concentrations compared to the xGen method, but the xGen single strand DNA kit on average the human content proportion converted to library was higher.  

So, I'm not seeing a statistical difference, but that additionally, the mechanical fragmentation method utilized with xGen resulted in longer median insert lengths. Though overall, I did not observe any impact to capture performance due to variation in fragment length. And overall, no significant differences have been observed for key genotype client metrics such as match reads, coverage, call rates, concordance, that are driven specifically by the library prep method utilized. Here you see the comparison of the two methods of hybridization capture from the controlled human microbial mixtures based on four key metrics: on target percentage, or the proportion of sequenced reads mapped to the target SNP regions, post enrichment endogenous content, for the percentage of generated reads mapped to the human genome, mean coverage of the target regions, or how many times loci of interest are the loci of interest are covered by unique reads, and finally, post enrichment call rate, or the percentage of target SNPs that are genotyped. For this evaluation, a locus must have a minimum depth of 5X to be called.  

The scatter plots compare the TWIST enrichment metrics on the X axis to the Compass metrics on the Y axis. For these controlled mixtures, enrichment with the TWIST custom assay yielded higher on target rates, higher coverage, and ultimately higher call rates than enrichment with the Compass assay, including from our control mixtures with less than 1% human content. And now here we compare the efficiencies of the two methods of enrichment from the degraded controlled human microbial mixtures.  

Again, based on the previously described four metrics. And again, TWIST enrichment metrics are on the X axis with the Compass metrics on the Y axis. For these degraded mixtures, enrichment with the TWIST custom assay again yielded higher on target rates, higher coverage, and ultimately higher call rates than enrichment with the Compass assay, and this is also from mixtures exposed to UV irradiation with DI values ranging from 16 to 45. I highlight the data points with the lowest coverage and consequently lowest call rates to point out that these correspond to low input, low human content mixtures of less than half a nanogram human DNA content in the background of more than 90% exogenous DNA and they demonstrated DI values of about 34, over sample with a DNA input of at least two and a half nanograms based on Trio quant and a UV induced DI over 100 resulted in more than 74% capture, SNPs captured with the TWIST enrichment. This overall input amount, starting at library prep appears to have a larger impact than the ultimate DI value.  

Okay, we'll move on to the FAC bone samples. Just to give a brief summary of the bone extract quality, Human DNA quantification ranged from about 100 picograms to just under five nanograms per microliter. Total DNA ranged from 12 nanograms per microliter to over 200 nanograms per microliter. These bones had been in archival storage, so they do not exhibit high levels of DNA degradation as determined by Trio DI values, and the estimated human content ranged from 0.06% to right around 10%.  

Bone samples were first analyzed with the ForenSeq Kintelligence Kit as an initial gauge of sample quality, and to be used for later concordance evaluations of the Kintelligence SNPs that are captured by the TWIST assay. Eight of the ten bones yielded call rates greater than 90%, with 95 greater than 95% concordance of the amplified SNP loci to the high coverage whole genome reference data. For these eight bones, estimated human content was greater than 0.6% of the total DNA.  

The two bones performing the worst, bones O and X, are the two samples with the lowest estimated human content at less than 0.5% of the total DNA. Now look at the enrichment of the more than 1 million FIGG relevant SNPs from these bone samples, using both the TWIST assay and the Compass assay. And the trend holds that the TWIST captured again demonstrates higher on target rates shown in the left scatter plot, which corresponds to higher proportions of the sequence data mapping as endogenous human and leading to higher mean coverage of the target regions just shown on the right. Higher coverage then translates to generally higher call rates.  

Here you'll see the bone genotyping metrics in the bottom panels juxtaposed against the original human content estimate based on DNA quantification values in the top panel. 1% human content is annotated by the dashed line across that top panel.

Do want to note that for the higher human content bones like E, G, J, and N, though the compass call rate percentage is slightly lower, the denominator being higher at 1.2 million, than versus TWIST, the raw SNP counts are very similar between TWIST and Compass, if not just a tad higher when enriched with Compass. So essentially both assays are quite efficient when the estimated human content is between 2 and 10%, though with an additional caveat. While the TWIST enrichment overall resulted in higher call rates, with no real difference between the two library prep methods, enrichment with the Compass assay did indicate an improvement based on which library prep method was utilized.  

When the bone extract was prepared using the xGen single stranded DNA kit, Compass call rates were consistently higher over Compass call rates from the double stranded library prep method, which is most obvious with bone N and S. But for those bones with human proportion estimated to be less than 1%, TWIST enrichment shows the greatest efficiency at capturing target SNPs as seen in bones F, K, O, and X, Regardless of method employed, concordance to the whole genome reference data was high, greater than 96% for most of the captured libraries.

So data analysis is ongoing but conclusions thus far are that genotyping by sequencing is an efficient protocol to develop SNP genotype profiles for FIGG. High coverage SNP genotypes can be recovered from skeletal remains with human content estimates less than 1%. Both the double stranded DNA library preparation method and a single stranded library prep method are effective for constructing genotyping by sequencing libraries from said remains. And for this particular set of samples, TWIST hybridization capture with our custom panel is demonstrating improved performance over the myBaits hybridization capture chemistry.  

Like I said data review is ongoing. Some of the additional analysis we are looking into include matching capabilities of both panels within GedMatch PRO for enriched libraries with low coverage, either low in general or by, say, running on an instrument with lower overall data output. How does the use of imputation impact or improve genotype recovery? And then what other information might be gleaned from the set of SNPs evaluated, such as incorporating Y-leaf software from Erasmus University into the pipeline for Y haplogroup terminations or phenotype and ancestry predictions. And then how do they compare to known phenotypes of the original donors.  

And with that, thank you to NIJ for the opportunity and I can take questions.

Tiffany Layne: Great job. Thank you, Sarah. We're now moving into the Q&A session. Sarah, if a body is severely burned, how much data can you collect from the remains?

Sarah Cavanaugh: I am not the one to ask that question. We are currently not looking at burn samples. I know there is some very good literature out there that has, that is specifically looking at burns, burned remains. And if they are interested in and don't know the publications, I am happy to send those to them.

Tiffany Layne: Awesome. Another one is, real quick. Can you repeat how much bone powder was used for your extraction?

Sarah Cavanaugh: We received up to or just under a microgram of bone powder from each of our samples.

Tiffany Layne: Great. And then last real quick, one. Why were calcaneus bones chosen for this study? Do you think any other types of bones would give you different results?

Sarah Cavanaugh: So the calcaneus bones, the heel bones, feet bones in particular, have been identified in previous work by the University of Tennessee's Amy Mundorff and in collaboration with our, John Davern from body as being one of the best sources of DNA for skeletal remains. So given that they have previously identified that as the highest source, that's why those bones were selected.

Tiffany Layne: All right, so we have more questions we'll send to you that you can answer offline. So thank you. Great presentation today, Sarah.

Sarah Cavanaugh: Thank you.

Tiffany Layne: And now we will hear from August.

August Woerner: Hi, everyone. All right. I'm August Woerner. I'm here to talk to you today about forensic low pass whole genome sequencing.

This is just standard slide because I've already seen many times. In essence, the words I say are my own. I'm not endorsing anything.

I would also like to give a big shout out to the research and development unit at the CHI. So much of the work that you're seeing is really because of them. So thank you. As well as, the staff’s facility at Sam Houston State University. Okay, so let's talk a little bit about the economics of genomics. So this is a plot that many of you have seen before, which shows you the price of the first human genome, rolling in at about $100 million to the price that you see today, which is about $500. However, was not stated is that this is done using million dollar instruments. One such instrument comes from Alto Genomics, and it rolls in at about 1.5 million. However, I'm interested in these smaller, benchtop sequencers like the Illumina MiSeq i100. So if you look at the promotional material for this instrument, they say it's good for targeted gene sequencing and that if you want to do large whole genome sequencing, that's human genome sequencing, they want you to purchase a bigger machine. But remember these are people that are selling you this product. Oh that background is not great. Oh well we'll just roll with it. Okay. So if we think about targeted, gene sequencing, what's happening is you pick a couple loci and then you do PCR on it, so you make lots of copies.

And then you sequence it, and then the sequencing that you do is priced per read. So imagine a little dollar sign associated with every single one of those reads. In contrast, if you do shotgun sequencing, which is a de facto way to do whole genome sequencing, you pick everything and then you do less PCR. And then you remove that non-independent information, which of course broaches a natural question. What happens if you sequence aka pay less? What you get, what you should be seeing, if the background weren’t messed up, is less information true, but also less redundancy. So the current approaches that we have for genotyping are redundant. So you take some allele and you measure it some large number of times.  

To consider what patterns of polymorphism looked like in human populations, there already is redundancy. So SNPs are incredibly redundant with each other. So if you're wondering if maybe you can leverage linkages equilibrium to call SNPs, you absolutely can. And one excellent tool to do so, as mentioned earlier is Glimpse. And really with Glimpse, what we've seen is this explosion and low pass whole genome sequencing, which is a thought experiment we did earlier, right. Where we sequence something but we sequence less of it. So low passes largely replaced genome wide association studies. You see it a lot in clinical genomics. You also see it in consumer genetics. So if you submit a DNA swab to MyHeritage, they're probably not using a microarray. They probably are using low pass. So the open question of course is does it have a place in forensics? I would actually say anytime you see genomics in forensics, it probably is low pass.

And it's worth noting that really the use case, at least right now in, forensics is forensic genetic genealogy. We try to identify individuals based off of matchings to cousins and these large databases like GEDmatch. All right. So we have all sorts of issues that we have to worry about if we're going to try to rule out low pass in forensics, right. So we have a template, microbial content, DNA may be damaged or fragmented, and of course you also have the perennial issue of mixtures.

First of all, all these things could be possible, right? But are any of these techniques actually feasible right? Do we need to rely on these multi-million dollar instruments or can we get away with something cheaper? And then even if something can theoretically happen, you need to have those bioinformatic pipelines cemented in place. So that can be easily used by the community. Okay. So spoiler for you. The answer to both of these questions is yes. So for the younger people in the audience, Yes was a progressive rock band of the 1970s. And for the older people in the audience, this is what that rock band looks like today.  

Okay, so we need to do some pilot studies to be able to demonstrate the feasibility of low pass. So we picked two cell line DNAs and we picked two cousins, two first cousins. And we put them each on to a MiSeq. And you got about a 1X genome when you do that. And then we extended those pedigrees all the way down to the third cousins. And we estimated genotypes. And we tested our ability to recover relatives just like you might with GEDmatch. And what you find is that you get either the correct degree classification or you're off by one at about the same rate as you would get had you had perfect genotyping.

Okay, so now it's time for us to be a little bit greedy. We know that 1X genomes might be sufficient. So what happens if we try going lower. So to leverage or to test this particular hypothesis we took advantage of the Gambian Genome Variation project which has both low pass data and microarray data. Further is doing it in these really challenge populations. So these are populations that are most distinct from our imputation panel.

And we do essentially the same procedures as before—we estimate genotypes, we simulate relatives, and then we try to see how little data do we actually need. So down sampling started at a 1X genome and went all the way down to 0.001X. And then we looked at the things that the genotype provides. And that's a genotype call and an estimate of the posterior probability.

And what you see is that if you have a mild threshold on that probability, genotype accuracy starts high and slowly starts to wane. If you put a more stringent, filter on that probability, your accuracy stays high and your call rate instead drops.

So in other words, if you remove the genotypes that are likely to be wrong, those that remain are correct, which is obvious. Okay. What's less obvious is that the IBD-segment recovery can also be really poor. So to demonstrate this you can look at the IBD-segment recovery within the array on the X-axis or between the array and imputed sample on the Y-axis. And what you want to see is all your samples showing up on that main diagonal. With a 1X genome that's largely true. With the 0.1X genome, you do well if you filter to have a posterior probability of at least 0.99.

If you go lower, things get worse and kind of interesting. So what you see is this essentially false segment sharing, where people that show no DNA matching, based off the array data, but based off the imputed data, you see lots of matching where is even worse than the samples with that high posterior probability cutoff. So those are genotypes that are largely right that are giving a very wrong inference.  

All right. So why, why are we seeing this. So if we remember how that posterior is calculated right. The posterior is proportional to your likelihood times your prior, right. Your prior is essentially fixed, comes from the population. The likelihood depends on your data. Right. So as your data start to become reduced, that likelihood becomes flat. And your posterior and your prior really becomes the same thing. So maybe we are, only including these common genotypes, and we're excluding these genotypes that might differentiate individuals.

So there's a prescription for this, right. So you can design a filter such that you include only sites that are driven by data. And you can do that with a Bayes factor. We'll consider a different data set and apply Bayes factor filter all these issues essentially go away. So as coverage is reduced, your variance in IBD segment estimation goes up a little bit. But you fail, you stop seeing these false IBD segment matches.  

Okay, so let's recap for a second. So the MiSeq i100 was argued to be able to do targeted gene sequencing. It can. And maybe it can do large whole genome sequencing right. And why do I say maybe? Let's try walking through that. So we know that if you put a single sample onto a MiSeq you get a 1X genome. Which should mean that if we put ten samples on you get a 0.1X genome, right. And 0.1X was generally good enough to be able to recover those IBD segments. But that's all assuming like five bar data. That's not really what those DNA samples are going to look like right. But maybe we're looking at effectively one bar data right.  

Which could mean that you could put two samples on to a MiSeq and generate a 0.1X genome. So you know, you're talking about $800 a sample, which is a pretty good deal. All right. So to assess this hypothesis we just raised, we need to have two things. You need really good data. And then from the same individuals you need really bad data. So we reached out to Doctor Sheree Hughes at Sam Houston State University, and she was able to furnish us with both fresh blood, age accelerated blood, as well as decomposed samples. And this is all within subject. So you know exactly what those original ground truth genotypes are.

All right. So we started first with the heat study where we had eight subjects, 100 picograms worth of DNA, DIs that are off the charts, and we gave ourselves a budget of eight MiSeq flow cells. And what we did is we put all eight samples onto that flow cell and then sequenced that thing eight times.

As it happens, one of those runs failed. I would note that even though the run failed, we didn't actually lose any data, which is pretty cool. And then we considered all the 2 to the 7 minus 1, or 127 ways of combining all those results. And this is what you see. So each pane are the results from an individual.

And the one eighths corresponds to those individual MiSeq runs, the two eights corresponds to all the ways of taking pairs of runs for an individual and combining them at the looking at the resulting coverage on the Y-axis. And what you see is that everything is very sample specific, right? So that the sample on the top left has generally poor performance. You're not going to give get above 0.1X until you get about maybe half a MiSeq flow cell dedicated to them, where the last sample probably would go, you probably get that 0.1X genome on that first try. What you also notice in that last sample is that light shading of blue. So when you do whole genome sequencing you're going to estimate a molecular complexity. So you know how big that original library was that you sequenced which is super cool. And that's the reason why you get this bend in that last plot. So it's sub-additive and you can measure and estimate the extent to that sub-additivity. So in general, performance is variable, sample specific, and depends heavily on fragmentation. I would note the numbers that are boxed shows you the average template length associated with each one of these samples.  

Okay. So if we know that our in vitro study, which is controlled, had considerable variance, your in vivo study has got to be worse, right? So we adopted a pre sequencing strategy, where you sequence it first at high plexity. In this case we put 11 to 12 samples onto a MiSeq flow cell. And you use that information for triage. And you don't just have how much coverage but you also have the estimate of molecular complexity which is shown in color, right.  

And so that we broke our samples into three classes, either we already have a 0.1X genome, in which case we're done. It's salvageable, which means if we sequenced it again and maybe play with that weight, how much it goes into that next run, we'll get enough data. And then those that are just cost prohibitive, you would need to put it onto an S4 flow cell by itself and be able to get what you need. So we gave ourselves a budget of an additional five MiSeq flow cells, which let us take 32 out of our 87 original samples, and generate, roughly a 0.1X genome, sometimes a little bit less. We can also play with that pool weighting. So you can over represent those samples that are, poor performing. And you got an overall accuracy in genotyping of more than 99.5%.  

If you think that maybe this only applies to MiSeq, it doesn't. We also dedicated about 20% of an S2 flow cell, equivalent of maybe two samples, onto a MiSeq. And we did the same thing. We broke our samples into those that needed to be triaged and thrown away sequenced again or the ones that were sufficient. And in this case, all the samples that redeemed sufficient had genotype accuracies north of 99%, and all the coverages were above 0.08.  

Okay. So I'm really interested in making sure our work translates. So we try to take our findings and generalize them a little bit to suggest maybe a forensic workflow that might work well in practice. So you might start with a DNA sample. You do a library prep on it, and then you do low pass whole genome sequencing, like ultra low. And then you can think of three classes that that might arise. Either when you sequence that, you usually get no DNA or is cost prohibitive, which case you do go back to your extract to do something different.  

Case two is the library looks promising, but you didn't happen to sequence it enough. So then maybe you play with those pooling weights, over represent sample a little bit on the next run, and then maybe you get a good enough profile. And then in the last case, case number three, your data are sufficient. You have enough data to go ahead and move forward to genotyping.  

All right. So let's take a pause for a second. I gave you a spoiler, gave you all some potential problems. But oh I guess I, I would hopefully I convinced you that the answer was yes. All these problems are solvable.  

But if you're really paying attention, We have this issue of mixtures. Right. So what happens when your sample is a mixture. You might not know it, especially with that little DNA. So we fine-tuned our mixture algorithm to be able to detect DNA mixtures all the way down to 0.1X. And that same framework is being extended to do convolve DNA mixtures as well. So super cool. If you're wondering if you can actually roll any of these technologies out in your own lab, you can. We have free open source software that does just that.

And if you're wondering that the whole point of doing this is just forensic genetic genealogy, it's not. You can take these same outputs and feed them into DBLR, or Familias, or wgsLR, to be able to infer identity. You can also estimate surnames and come to find that this this data can also be really useful for doing phenotyping. So methylation will tend to associate with where those break points happen. Super cool. And you can also use heteroplasmy, nucleosome occupancy and mitochondrial copy number to be able to estimate what sort of sample that you're working with. So is this a blood sample or is it semen? Sometimes it's not so clear. And that last bit of research is being conducted by Jennifer Cihlar and Elizabeth Kowalcyzk shown below. And with that I would like to thank everyone, starting with staffs and Dr.Sheree Hughes, as well as the CHI and their Research and Development Unit. And last but certainly not least, the NIJ for funding this research. Thank you. And with that, I'll take questions.

Tiffany Layne: Great. Awesome. Thank you so much, August. Now we're going to move into the final Q&A session. Okay. So August, if a body is found months or years later, how much data can you collect from the scene for this type of DNA sampling? Is it similar to the kinds of samples that you guys tested?

August Woerner: Right. So the skeletal samples and the decomposed samples were, anywhere from 150 to 300 years..ugh...days aged. [laughter] So they were, some, some time had elapsed. Not as much as you just said when I said years. So sorry about that. And I would stress that I wouldn't do any of this. I'm an informatician, so you don't you don't want me anywhere near anything on the wet side of things.

Tiffany Layne: Okay, cool. Does DNA like saliva or body tissue, fade away over time from the scene? And so would that, do you think just not do as well? Or would you still be able to get that with this low pass strategy?

August Woerner: I mean, like, DNA will degrade. I think that's unavoidable right? And the time frames that we're, that we've evaluated so far, it seems like you get a modest amount of DNA recovery. And if the conditions are right, like we've recovered DNA from, you know, from neanderthals right. But those are, like, really perfect conditions, essentially samples that are put in the equivalent of a freezer. I would love to know more about how the long term preservation of DNA, actually behaves. And historical DNA analysis is like the bread and butter of many whole genome, researchers. So they go out and take archeological specimens that are hundreds, if not thousands of years old and attract DNA out of them.

Tiffany Layne: Cool. Another question, what type of throughput would a lab need in order to take advantage of a low pass strategy?

August Woerner: So the line we drew in the sand is about 0.1X as being suitable. Right. And so the giant mystery there for translation is how much coverage do your samples tend to generate? So if you tend to work with skeletonized remains, where you have a large mass of sample, then I can see, like the previous doc, look at some enrichment strategies that could work well. And by the way, it could also work well on a MiSec. And by the way, I've already tested glimpse out on those markers and you've got about 1X. So it's not, those strategies are not mutually exclusive.  

If you're working with samples that are like, more contemporary and unlikely to be mixtures and unlikely to have high bacterial load, I think the MiSeq is perfectly suitable. And if you want, you can try playing with how much these instruments produce and what their costs are, you can estimate how many runs you might be doing. And it's pretty easy. The back-of-the-envelope to figure out which instrument is right for you. To me, it doesn't really matter if it's a MiSeq or an AVITI or whatever. You should just figure out which is most cost efficient and run with it.

Tiffany Layne: Awesome. So, that's the time we have for today. So any other questions we'll send to you, August, to answer offline.

And so that is a wrap for the Forensic Biology and DNA session today. Thank you for dedicating your afternoon to today's webinar. We hope the information provided was helpful and informative. For any follow up questions, please contact [email protected]. A recording of this webinar will be posted to the NIJ website within the coming days. And if you enjoyed this week's webinars and would like to hear more NIJ supported researchers, we have one more session tomorrow on our graduate student research. We look forward to your participation in future events. Have a great day.