August E. Woerner

Forensic human identification using skin microbiomes.

ABSTRACT The human microbiome contributes significantly to the genetic content of the human body. Genetic and environmental factors help shape the microbiome, and as such, the microbiome can be unique to an individual. Previous studies have demonstrated the potential to use microbiome profiling for forensic applications; however, a method has yet to identify stable features of skin microbiomes that produce high classification accuracies for samples collected over reasonably long time intervals. A novel approach is described here to classify skin microbiomes to their donors by comparing two feature types: Propionibacterium acnes pangenome presence/absence features and nucleotide diversities of stable clade-specific markers. Supervised learning was used to attribute skin microbiomes from 14 skin body sites from 12 healthy individuals sampled at three time points over a >2.5-year period with accuracies of up to 100% for three body sites. Feature selection identified a reduced subset of markers from each body site that are highly individualizing, identifying 187 markers from 12 clades. Classification accuracies were compared in a formal model testing framework, and the results of this analysis indicate that learners trained on nucleotide diversity perform significantly better than those trained on presence/absence encodings. This study used supervised learning to identify individuals with high accuracy and associated stable features from skin microbiomes over a period of up to almost 3 years. These selected features provide a preliminary marker panel for future development of a robust and reproducible method for skin microbiome profiling for forensic human identification. IMPORTANCE A novel approach is described to attribute skin microbiomes, collected over a period of >2.5 years, to their individual hosts with a high degree of accuracy. Nucleotide diversities of stable clade-specific markers with supervised learning were used to classify skin microbiomes from a particular individual with up to 100% classification accuracy for three body sites. Attribute selection was used to identify 187 genetic markers from 12 clades which provide the greatest differentiation of individual skin microbiomes from 14 skin sites. This study performs skin microbiome profiling from a supervised learning approach and obtains high classification accuracy for samples collected from individuals over a relatively long time period for potential application to forensic human identification.

Full-gene haplotypes refine CYP2D6 metabolizer phenotype inferences

CYP2D6 is a critical pharmacogenetic target, and polymorphisms in the gene region are commonly used to infer enzyme activity score and predict resulting metabolizer phenotype: poor, intermediate, extensive/normal, or ultrarapid which can be useful in determining cause and/or manner of death in some autopsies. Current genotyping approaches are incapable of identifying novel and/or rare variants, so CYP2D6 star allele definitions are limited to polymorphisms known a priori. While useful for most predictions, recent studies using massively parallel sequencing data have identified additional polymorphisms in CYP2D6 that are predicted to alter enzyme function but are not considered in current star allele nomenclature. The 1000 Genomes Project data were used to produce full-gene haplotypes, describe their distribution in super-populations, and predict enzyme activity scores. Full-gene haplotypes generated lower activity scores than current approaches due to inclusion of additional damaging polymorphisms in the star allele. These findings are critical for clinical implementation of metabolizer phenotype prediction because a fraction of the population may be incorrectly considered normal metabolizers but actually may be poor or intermediate metabolizers.

Targeted sequencing of clade-specific markers from skin microbiomes for forensic human identification

The human skin microbiome is comprised of diverse communities of bacterial, eukaryotic, and viral taxa and contributes millions of additional genes to the repertoire of human genes, affecting human metabolism and immune response. Numerous genetic and environmental factors influence the microbiome composition and as such contribute to individual-specific microbial signatures which may be exploited for forensic applications. Previous studies have demonstrated the potential to associate skin microbial profiles collected from touched items to their individual owner, mainly using unsupervised methods from samples collected over short time intervals. Those studies utilize either targeted 16S rRNA or shotgun metagenomic sequencing to characterize skin microbiomes; however, these approaches have limited species and strain resolution and susceptibility to stochastic effects, respectively. Clade-specific markers from the skin microbiome, using supervised learning, can predict individual identity using skin microbiomes from their respective donors with high accuracy. In this study the hidSkinPlex is presented, a novel targeted sequencing method using skin microbiome markers developed for human identification. The hidSkinPlex (comprised of 286 bacterial (and phage) family-, genus-, species-, and subspecies-level markers), initially was evaluated on three bacterial control samples represented in the panel (i.e., Propionibacterium acnes, Propionibacterium granulosum, and Rothia dentocariosa) to assess the performance of the multiplex. The hidSkinPlex was further evaluated for prediction purposes. The hidSkinPlex markers were used to attribute skin microbiomes collected from eight individuals from three body sites (i.e., foot (Fb), hand (Hp) and manubrium (Mb)) to their host donor. Supervised learning, specifically regularized multinomial logistic regression and 1-nearest-neighbor classification were used to classify skin microbiomes to their hosts with up to 92% (Fb), 96% (Mb), and 100% (Hp) accuracy. All samples (n=72) regardless of body site origin were correctly classified with up to 94% accuracy, and body site origin could be predicted with up to 86% accuracy. Finally, human short tandem repeat and single-nucleotide polymorphism profiles were generated from skin swab extracts from a single subject to highlight the potential to use microbiome profiling in conjunction with low-biomass samples. The hidSkinPlex is a novel targeted enrichment approach to profile skin microbiomes for human forensic identification purposes and provides a method to further characterize the utility of skin microflora for human identification in future studies, such as the stability and diversity of the personal skin microbiome.

Flanking Variation Influences Rates of Stutter in Simple Repeats

It has been posited that the longest uninterrupted stretch (LUS) of tandem repeats, as defined by the number of exactly matching repeating motif units, is a better predictor of rates of stutter than the parental allele length (PAL). While there are cases where this hypothesis is likely correct, such as the 9.3 allele in the TH01 locus, there can be situations where it may not apply as well. For example, the PAL may capture flanking indel variations while remaining insensitive to polymorphisms in the repeat, and these haplotypic changes may impact the stutter rate. To address this, rates of stutter were contrasted against the LUS as well as the PAL on different flanking haplotypic backgrounds. This study shows that rates of stutter can vary substantially depending on the flanking haplotype, and while there are cases where the LUS is a better predictor of stutter than the PAL, examples to the contrary are apparent in commonly assayed forensic markers. Further, flanking variation that is 7 bp from the repeat region can impact rates of stutter. These findings suggest that non-proximal effects, such as DNA secondary structure, may be impacting the rates of stutter in common forensic short tandem repeat markers.

Modeling SNP array ascertainment with Approximate Bayesian Computation for demographic inference

Single nucleotide polymorphisms (SNPs) in commercial arrays have often been discovered in a small number of samples from selected populations. This ascertainment skews patterns of nucleotide diversity and affects population genetic inferences. We propose a demographic inference pipeline that explicitly models the SNP discovery protocol in an Approximate Bayesian Computation (ABC) framework. We simulated genomic regions according to a demographic model incorporating parameters for the divergence of three well-characterized HapMap populations and recreated the SNP distribution of a commercial array by varying the number of haploid samples and the allele frequency cut-off in the given regions. We then calculated summary statistics obtained from both the ascertained and genomic data and inferred ascertainment and demographic parameters. We implemented our pipeline to study the admixture process that gave rise to the present-day Mexican population. Our estimate of the time of admixture is closer to the historical dates than those in previous works which did not consider ascertainment bias. Although the use of whole genome sequences for demographic inference is becoming the norm, there are still underrepresented areas of the world from where only SNP array data are available. Our inference framework is applicable to those cases and will help with the demographic inference.

Exploring the 1000 Genomes Project haplotype reporting for the CYP2D6 pharmacogene

The Gaedigk et al. article BA perspective by PharmVar: Are the hundreds of CYP2D6 haplotypes predicted by Wendt and colleagues real?^ describes shortcomings of the 2017 Wendt et al. article BFull-gene haplotypes refine CYP2D6 metabolizer phenotype inferences^ [1]. To summarize, they discuss (1) the lack of submission of novel variants to www. PharmVar.org; (2) inaccurate activity score reporting, namely for those haplotypes containing the 843T>G SNP; (3) use of 1000 Genomes Project (1kGP) data from the inaccessible regions of the database; and (4) lack of sequence and structural validation for any of the described haplotypes. We thank Gaedigk and colleagues for their review of the Wendt et al. 2017 findings and in many ways share their concerns. In general, the authors’ letter raises valid concerns for the data presented in the original Wendt et al. study and many pharmacogenomics studies utilizing publically available data. However, the authors’ appear to overstate our reported findings and seem to ignore where we already transparently discuss the major limitations of using such a database for this type of data exploration. We summarize our responses to their concerns below. In general, we urge PharmVar to actively update its nomenclature table as to reflect most recent submitted findings. Additionally, we encourage PharmVar, its affiliates, and other pharmacogenomics researchers to release full-gene information as it becomes available, rather than only those sites relevant to the PharmVar nomenclature table(s) or the repository of knowledge for their respective gene(s) of interest. In doing so, the initiative described by Gaedigk and colleagues will continue to thrive. The Wendt et al. paper was intended to explore the use of full-gene CYP2D6 haplotype diversity in publically available data. A number of single-nucleotide polymorphisms from the 1kGP and the Wendt et al. study are not found on www. PharmVar.com. Gaedigk and colleagues note that PharmVar accepts submission of high-quality haplotype data. The Pilot Criteria of the 1kGP Phase3 Paired-end Accessible Regions are quite stringent, requiring Ba depth of coverage between 8,960 and 35,840 inclusive (between one-half and twice the average depth) and that no more than 20% of covering reads have mapping quality zero^ [2]. Indeed, read depth is a limiting factor for using data such as those of the 1kGP; however, Wendt et al. never recommended or even suggested that the data were high quality and be considered for submission to PharmVar. Such a recommendation would have been inappropriate and misleading to the community. Indeed, we stress in our paper that Bempirical data are required to confirm their enzyme activity [of the resulting haplotypes]^ and thus share similar concerns. Wendt et al. indicated that the relatively low sequencing read depth of the 1kGP is a major limitation of their findings. However, low read depth of pharmacoand immunogenes does not warrant ignoring the public availability of 1kGP short-read data for exploratory purposes. It is a great resource used by many scientists for developing hypotheses and addressing probing questions. There appears to be some confusion by Gaedigk et al. regarding the methods ofWendt et al. in which 1kGP haplotypes were characterized first using only those sites recognized and published on the PharmVar website (Human Cytochrome p450 Allele Nomenclature Database at the time of Wendt et al. analyses). Here, the consortium defines CYP2D6 * Frank R. Wendt frw5010@gmail.com

Evaluation of the precision ID mtDNA whole genome panel on two massively parallel sequencing systems

Sequencing whole mitochondrial genomes by capillary electrophoresis is a costly and time/labor-intensive endeavor. Many of the previous Sanger sequencing-based approaches generated amplicons that were several kilobases in length; lengths that are likely not amenable for most forensic applications. However, with the advent of massively parallel sequencing (MPS) short-amplicon multiplexes covering the entire mitochondrial genome can be sequenced relatively easily and rapidly. Recently, the Precision ID mtDNA Whole Genome Panel (Thermo Fisher Scientific by Applied Biosystems™) has been introduced. This panel is composed of 162 amplicons (in two multiplexes) that are considerably smaller in length (∼163bp) and thus are more amenable to analyzing challenged samples. This panel was evaluated on both the Ion S5™ System (Thermo Fisher Scientific) and the MiSeq™ FGx Desktop Sequencer (Illumina). A script was developed to extract phased haplotypes associated with these amplicons. Levels of read-depth were compared across sequencing pools and between sequencing technologies and haplotype concordances were assessed. Given modest thresholds on read depth, the haplotypes identified by either technology were consistent. Nuclear mitochondrial sequences (Numts) were also inferred, and the effect of different mapping strategies commonly used to filter out Numts were contrasted. Some Numts are co-amplified with this amplification kit, and while the choice of reference sequence can mitigate some of these effects, some data from the mitochondrial genome were lost in the process in this study. This study demonstrates that the Ion and MiSeq platforms provide consistent haplotype estimation of the whole mitochondrial genome, thus providing further support for the reliability and validity of the Precision ID mtDNA Whole Genome Panel.

Forensic human identification with targeted microbiome markers using nearest neighbor classification

From the perspective of forensics genetics, the human microbiome is a rich, relatively untapped resource for human identity testing. Since it varies within and among people, and perhaps temporally, the potential forensic applications of the use of the microbiome can exceed that of human identification. However, the same inherent variability in microbial distributions may pose a substantial barrier to forming predictions on an individual as the source of the microbial sample unless stable signatures of the microbiome are identified and targeted. One of the more commonly adopted strategies for microbial human identification relies on quantifying which taxa are present and their respective abundance levels. It remains an open question if such microbial signatures are more individualizing than estimates of the degree of genetic relatedness between microbial samples. This study attempts to address this question by contrasting two prediction strategies. The first approach uses phylogenetic distance to predict the host individual; thus it operates under the premise that microbes within individuals are more closely related than microbes between/among individuals. The second approach uses population genetic measures of diversity at clade-specific markers, serving as a fine-grained assessment of microbial composition and quantification. Both assessments were performed using targeted sequencing of 286 markers from 22 microbial taxa sampled in 51 individuals across three body sites measured in triplicate. Nearest neighbor and reverse nearest neighbor classifiers were constructed based on the pooled data and yielded 71% and 78% accuracy, respectively, when diversity was considered, and performed significantly worse when a phylogenetic distance was used (54% and 63% accuracy, respectively). However, empirical estimates of classification accuracy were 100% when conditioned on a maximum nearest neighbor distance when diversity was used, while identification based on a phylogenetic distance failed to reach saturation. These findings suggest that microbial strain composition is more individualizing than that of a phylogeny, perhaps indicating that microbial composition may be more individualizing than recent common ancestry. One inference that may be drawn from these findings is that host-environment interactions may maintain the targeted microbial profile and that this maintenance may not necessarily be repopulated by intra-individual microbial strains.

Potential highly polymorphic short tandem repeat markers for enhanced forensic identity testing

Due to their polymorphic nature, short tandem repeats (STRs) are well-studied and routinely used genetic markers for forensic DNA typing. However, even the largest STR multiplexes are limited in their ability to parse out individuals in a DNA mixture sample, due to alleles shared by size detected by capillary electrophoresis and challenges in resolving minor alleles from stutter, and inherent heterozygote imbalance. In this study, STRs were explored in public datasets that displayed sequence variation and may have limited allele length spread. STRs were first selected using fundamental criteria of high heterozygosity, tetra-, penta-, or hexanucleotide repeat length, and overall relative narrow allele spread (based on length). All candidates were further scrutinized for chemistry compatibility. The resulting STRs were multiplexed and sequenced by massively parallel sequencing in a limited sample population set. Each candidate STR was evaluated for analytical performance and desired biological properties. The findings presented describe a refined set of 53 potential highly polymorphic STR markers (high sequence diversity and heterozygosity; reduced allele spread) that may be suitable to supplement the current core marker set(s) for possible enhanced characterization of complex DNA profiles.