Retrieving Y chromosomal haplogroup trees using GWAS data

Abstract

Phylogenetically informative Y chromosomal single-nucleotide polymorphisms (Y-SNPs) integrated in DNA chips have not been sufficiently explored in most genome-wide association studies (GWAS). Herein, we introduce a pipeline to retrieve Y-SNP data. We introduce the software YTool (http://mitotool.org/ytool/) to handle conversion, filtering, and annotation of the data. Genome-wide SNP data from populations in Myanmar are used to construct a haplogroup tree for 117 Y chromosomes based on 369 high-confidence Y-SNPs. Parallel genotyping and published resequencing data of Y chromosomes confirm the validity of our pipeline. We apply this strategy to the CEU HapMap data set and construct a haplogroup tree with 107 Y-SNPs from 39 individuals. The retrieved Y-SNPs can discern the parental genetic structure of populations. Given the massive quantity of data from GWAS, this method facilitates future investigations of Y chromosome diversity.

Introduction

The non-recombining portion of the Y chromosome is the most genealogically informative haploid marker in the human nuclear genome.¹ In 2002, the Y Chromosome Consortium constructed a Y chromosomal haplogroup tree based on 245 biallelic single-nucleotide polymorphisms (Y-SNPs) including indels that were generally treated as binary SNPs. Analyses of these data defined 153 human Y chromosomal haplogroups.² Subsequently, the Y chromosome tree was modified^{1, 3} and updated.⁴ The updated tree provides an important foundation for studies of evolutionary anthropology,⁵ genealogical reconstruction,⁶ molecular forensics,⁷ and medical genetics.⁸

Current commercial DNA chips (microarrays) for massive genome-wide association studies (GWAS)⁹ are designed to contain many Y-SNPs. In 2007, Underhill and Kivisild retrieved 295 Y-SNPs from customized Perlegen arrays in a previous study¹⁰ and used them to reconstruct the Y chromosome tree for 33 males.³ Their work suggested that Y-SNP data from DNA chips—usually the byproduct of GWAS—provide opportunities to explore Y chromosome diversity within and among populations. Although Y-SNP data from DNA chips helped improve assessments of population stratification,¹¹ many GWAS do not consider or evaluate Y-SNPs. It is difficult to tease out Y-SNPs from bulk GWAS data and then to assign the Y chromosome to a haplogroup.

Herein, we develop a de novo pipeline for retrieving Y-SNPs in DNA chips. We test our approach using data of DNA chips from populations in Myanmar (also known as Burma). Only sporadic sampling exists for Myanmar¹² and, thus, a systematic investigation of the Y chromosome diversity is wanting. In addition to providing a practical protocol of analyzing Y-SNPs from GWAS data, our work adds an essential piece to the genetic puzzle for Southeast Asians.

Materials and methods

Sample collection

We collected blood samples from 106 unrelated male individuals living in Myanmar (Table 1). Four different ethnic groups were surveyed: Bamar (also Burman, n=59), Chin (n=19), Naga (n=15), and Rakhine (n=13). Blood samples of eight unrelated individuals of the Jingpo ethnic group from Yunnan, China, were collected to represent the Kachin people of northern Myanmar because the two names appear as synonyms for the same ethnic group.¹³ Blood samples of three Nigerian males were used as the outgroup reference. All 117 subjects were interviewed to obtain informed consent before sample collection. The Institutional Review Board of the Kunming Institute of Zoology approved the protocols and study.

Table 1. General information for populations genotyped in this study.

Code	Group	Location	Region	Size
BA	Bamar	Ayeyarwady region	Lower Myanmar	11
BB	Bamar	Bago region	Lower Myanmar	14
RR	Rakhine	Rakhine state	Lower Myanmar	13
BM	Bamar	Magway region	Upper Myanmar	21
BS	Bamar	Sagaing region	Upper Myanmar	13
CC	Chin	Chin state	Upper Myanmar	19
NS	Naga	Sagaing region	Upper Myanmar	15
JY	Jingpo	Yunnan, China	Upper Myanmar	8
AN	African	Nigeria	—	3
Total				117

DNA extraction and genome-wide SNP genotyping

Genomic DNA was extracted and purified by using QIAmp DNA Blood Mini Kit (Qiagen, Hilden, Germany). Genome-wide SNPs for each sample were genotyped with HumanOmniZhongHua-8 BeadChip (http://support.illumina.com/array/array_kits/humanomnizhonghua-8_beadchip_kit.ilmn; Illumina, San Diego, CA, USA) according to the manufacturer's protocols. The 900015 markers on this chip captured 81% variation (r²>0.8) with minor allele frequency >5% in East Asians (CHB+JPT). Given the close relationship between East and Southeast Asians,^{12, 14} we used the HumanOmniZhongHua-8 BeadChip to investigate Myanmarese populations. Compared with the contemporary Illumina HumanOmni1-Quad BeadChip (1140419 markers in total, including 2322 Y-SNPs), the HumanOmniZhongHua-8 BeadChip provided a considerable quantity of markers (900015, including 2041 Y-SNPs) at a much lower cost.

Data sets

Genome-wide SNPs in the HumanOmniZhongHua-8 BeadChip were scanned by iScan (Illumina). The chip data with PLINK format (.MAP and.PED files) were exported through GenomeStudio (Illumina). The alleles were mapped to the forward strand. Excluding pseudoautosomal region loci coded by chromosome 25, we extracted data for 2041 Y-SNPs in the male-specific region (coded by chromosome 24) of 117 males using PLINK 1.07 (http://pngu.mgh.harvard.edu/∼purcell/plink/).¹⁵

Data filtering

Owing to the unusual nature of Y chromosomal characters, genotyping of human Y-SNPs using DNA chips has been problematic.¹⁰ The high level of similarity between certain regions of Y and X chromosomes/autosomes^{16, 17} can blur signals of genotyping Y-SNPs, as can the very high proportion of ampliconic repeats that are often highly similar to each other (usually >99.9% identity).^{16, 17} Consequently, we employed specific quality controls and excluded the following data in 117 male samples: (1) Y-SNPs with heterozygous alleles; (2) Y-SNPs genotyped with a missing rate more than 5% and (3) known Y-SNPs with identical genotypes (ie, invariants).

Phylogenetic analyses

We constructed median-joining network¹⁸ with the filtered Y-SNPs by using NETWORK 4.611 (http://www.fluxus-engineering.com/sharenet.htm) to infer the genealogy for each of Y chromosomes. Each Y-SNP was conveniently mapped on branches of the genealogy during network construction. The Y chromosome haplogroup tree was constructed from the genealogy. To reconcile our results with the widely accepted Y chromosomal phylogeny and haplogroup definitions,⁴ we consulted Y-DNA Haplogroup Tree 2013 of ISOGG (http://www.isogg.org/tree/),¹⁹ which was based on Karafet et al⁴ but constantly updated. We followed the previously used nomenclature (ie, lineage-marker) for each haplogroup proposed by Karafet et al⁴ To automatically address the data conversion, filtering, and annotation, we also developed C++ stand-alone software with GUI: YTool (http://mitotool.org/ytool/; Supplementary Methods). YTool was also designed to export Y-SNP data in the FASTA format for use in alternative software.

Validation

Two parallel strategies were adopted to validate the DNA chip data. First, in terms of the known Y chromosome tree,⁴ 28 diagnostic Y-SNPs defining the common Y chromosomal haplogroups (M168-CT, M145-DE, M174-D, M55-D2, M69-H, M410-J2a, M241-J2b2, M11-L, P188-NO, P191-O, P203.1-O1a, M50-O1a2, P31-O2, M95-O2a, M88-O2a1, M122-O3, P199-O3a, P201-O3a3, M134-O3a2c1, M117-O3a2c1a, M162-O3a2c1a1, M120-Q1a1, M207-R, P231-R1, M198-R1a1a, M87-R1a1a1c, M269-R1b1b, and M124-R2a) in Southeast Asia^{20, 21} and Northeast India²² were selected. These Y-SNPs were amplified in three panels of multiplex PCR reactions (Supplementary Methods) and then were genotyped with GenomeLab SNPstream (Beckman Coulter, Fullerton, CA, USA). Based on our previous studies,^{20, 22} primers for multiplex PCR and single base extension reactions were designed by Autoprimer software (Beckman Coulter). Further sequencing with an ABI 3730 DNA Analyzer (Applied Biosystems by Life Technologies, Foster City, CA, USA) was used to confirm specific Y-SNPs. The PCR amplification and sequencing primers were previously reported.⁴ In terms of the diagnostic Y-SNPs, we classified each of the 117 males into their known haplogroup, and then compared the results with those based on DNA chips. Second, we consulted the 6662 Y-SNPs reported in a recent study of Y chromosome using next-generation sequencing (ie, NGS data).²³ We compared our results with the NGS data, especially by checking the annotation of phylogenetic status for each of the Y-SNPs.

Application

We downloaded Y-SNPs of CEU (ie, genotypes_chrY_CEU_r28_nr.b36_fwd.txt.gz) from HapMap (ftp://ftp.ncbi.nlm.nih.gov/hapmap/genotypes/2010-08_phaseII+III/forward/). Because the samples were genotyped using different platforms, we only considered 44 individual males investigated by all chips. The data were transformed into PLINK format. Five individuals with high proportions of missing genotypes (>61%) were excluded. Finally, we tested the data of 943 Y-SNPs in 39 CEU males with our method implemented in YTool. All sample data as well as results for analyses in this work were made available at http://mitotool.org/ytool/ytool.zip.

Results

Data filtering

Call rates for each sample were above 98% for the HumanOmniZhongHua-8 BeadChips. For the 2041 Y-SNPs (Supplementary Data 1 and 2), the proportions of missing genotypes varied from 10.3– to 20.4% (11.3%, on average). First, 241 Y-SNPs (∼11.8%, 241/2041) genotyped with heterozygous alleles in the 117 male samples were disregarded. Second, 79 Y-SNPs (∼3.9%, 79/2041) with missing genotypes in more than 5% of the male samples were excluded. Third, 1346 Y-SNPs (∼65.9%, 1346/2041) identified as invariant were not considered. As a result of this filtering, 375 variants were remained (Table 2). Y-SNPs filtered in each step were deposited in downloadable files (http://mitotool.org/ytool/ytool.zip).

Table 2. Comparison of Y-SNPs from HumanOmniZhongHua-8 BeadChip (117 males) and CEU (Utah residents with Northern and Western European ancestry from the CEPH collection) in HapMap (39 males) with NGS data.

			ISOGG	NGS dataa
Chip data	Y-SNPs	No. SNPs	annotated	In YSURsa	Annotated
OmniZhongHua	Heterozygous alleles	241	13	51	22
	Missing rate >5%	79	8	30	13
	Invariants	1346	308	745	96
	Potential recurrent	6	2	4	2
	Qualified variants	369	174	294	244
	Total	2041	505	1124	377
HapMap-CEU	Heterozygous alleles	601	1	0	0
	Missing rate>5%	16	9	16	9
	Invariants	218	100	133	74
	Potential recurrent	1	0	1	1
	Qualified variants	107	52	78	71
	Total	943	162	228	155

Abbreviations: ISOGG, International Society of Genetic Genealogy; NGS, next-generation sequencing; SNP, single-nucleotide polymorphism; Y-SNP, Y chromosomal single-nucleotide polymorphism; YSUR, Y-specific unique region.

^aNGS data and YSURs (Y-specific unique regions) were depicted in ref 23.

Y chromosomal haplogroup tree

Of the 375 Y-SNPs for 117 males, six Y-SNPs occurred multiple times on different branches of the median-joining network. These likely represented multiple mutational events. Because recurrent mutations are rare^{2, 4, 23} and should be treated with caution,^{24, 25} we took a conservative approach and excluded these six Y-SNPs from subsequent analyses. Thus, we reconstructed the Y chromosomal haplogroup tree based on 369 phylogenetically informative Y-SNPs (Figure 1). According to this tree, all samples were assigned into 22 (sub-)haplogroups (paragroups) within macrohaplogroup CT (Supplementary Table 1).

Figure 1.

Y chromosomal haplogroup tree of 117 male samples. Reference sequence numbers (rs#) of the Y-SNPs defining the tree structure are shown along the branches. Population codes are given in Table 1 followed by sample sizes for each of the haplogroups. Y-SNPs without rs# are noted in italics: 610147 (200610-147), 610155 (200610-155), 610158 (200610-158), 610191 (200610-191), 610195 (200610-195), 610214 (200610-214), 610226 (200610-226), 610245 (200610-245), 610281 (200610-281), 610363 (200610-363), 610366 (200610-366), 8Y2897 (2010-08-Y-2897), and 8Y3175 (2010-08-Y-3175). Y-SNPs are annotated with their corresponding markers deposited in ISOGG (http://www.isogg.org/tree/ISOGG_YDNA_SNP_Index.html). Y-SNPs identified in NGS data are in red, whereas those out of nine YSURs are in blue (Table 2).²³ Those distributing in YSURs but not found in NGS data are in black. Y-SNPs genotyped by SNPstream or direct sequencing are underlined. Problematic Y-SNPs are given in parentheses.

Validation

The phylogeny of Myanmarese Y chromosomal lineages was largely supported by genotyping and direct sequencing the candidate diagnostic Y-SNPs (Figure 1; Supplementary Table 2). There was one conflict because of variant rs34893929 (ie, Page23) detected in the branch of haplogroup O3a2c-8Y2897 (Figure 1). Page23 is equivalent to M117 defining haplogroup O3a2c1a-M117, which was a downstream diagnostic variant of M134 defining O3a2c1-M134.¹⁹ However, both M117 and M134 were absent in our samples (Supplementary Table 2). Thus, genotyping of Page23 was a false positive signal. Similarly, variant rs34441307 (ie, L137), which defined haplogroup I2a2a1a1-L137¹⁹ and occurred on branch O3*-M122, was likely a phantom.

To validate the remaining Y-SNPs in the chip, we referenced the 6662 Y-SNPs identified in a recent resequencing study of nine Y-specific unique regions (YSURs) spanning 8.97 Mb in which pseudoautosomal, heterochromatic, X-transposed, and ampliconic segments were excluded.²³ Comparison of two data sets was shown in Table 2. About 50.4% (842/1672) of the Y-SNPs removed in our data filtering occurred out of the YSURs. In contrast, only ∼20.3% (75/369) of the Y-SNPs passing through our filtering distributed out of the YSURs. The difference was statistically significant (χ²test; p<0.0001; Table 3). Further, no conflicts were found upon checking the phylogenetic status for each of 244 variants annotated in NGS data (Figure 1). For the branches investigated in both studies (eg, CF-F-IJK-P-R-R1), almost all variants distributed within the YSURs were confirmed by NGS data, except for rs2049642 in the branch directing to haplogroup IJK (Figure 1). Because the NGS data were from 36 males only representing haplogroups A, D, E, G, I, N, Q, and R, Y-SNPs for other haplogroups (eg, C, H, J, L, and O) were not evaluated.²³

Table 3. Statistic tests for Y-SNPs distributing in or out of YSURsa.

Chip data	Quality control	In YSURs	Out of YSURs	χ2 test
OmniZhongHua Beadchip	Filtered	830	842	P<0.0001
2041 Y-SNPs, 117 males	Passed	294	75
HapMap-CEU	Filtered	150	686	P<0.0001
943 Y-SNPs, 39 males	Passed	78	29

Abbreviations: Y-SNP, Y chromosomal single-nucleotide polymorphism; YSUR, Y-specific unique region.

^aYSURs were depicted in ref 23.

Application: CEU in HapMap

The data filtering for 943 Y-SNPs in 39 males was shown in Table 2. The Y chromosomal haplogroup tree based on 107 qualified Y-SNPs was constructed (Figure 2). The phylogeny was generally consistent with the annotation of ISOGG and NGS data. All 39 males were assigned into two major European Y chromosome haplogroups I and R1, suggesting a European ancestry. Three samples (NA06994, NA07357, and NA12891) that were both genotyped by chips and sequenced by NGS were concordant (Figure 2).

Figure 2.

Y chromosomal haplogroup tree of 39 CEU males in HapMap. The phylogenetic status of seven Y-SNPs cannot be determined, and these Y-SNPs are shown in box with dashes. For the information of the labels, see Figure 1 and its legend.

Discussion

We provide a practical strategy to retrieve informative Y-SNPs from GWAS data using YTool software. The software also exports retrieved Y-SNPs into FASTA format for further analysis by many phylogenetic tools. Both traditional genotyping and published NGS data generally validate our method. Our pipeline can be used to discern data quality, mutational stabilities, and phylogenetic status of multiple populations involved massive GWAS. The Y chromosomal haplogroup tree can be constructed (eg, Figure 1) and the haplogroup profiles of populations can be depicted (eg, Supplementary Table 1). Further analyses, such as principal component analysis²⁶ and analysis of molecular variance,²⁷ are able to clarify the parental genetic structures.

The Y chromosomal haplogroup tree retrieved from GWAS data is valuable for exploring population genetic diversity. Most studies rely on genotyping 20–30 candidate diagnostic Y-SNPs in Southeast Asians (eg, He et al²⁰). In contrast, resolution of the Y chromosomal haplogroup tree for Myanmar is improved by analyzing hundreds of high-confidence variants (Figure 1). In terms of this tree, the future employment of Y-STRs²⁸ promises to reveal more details about parental demographic history in Myanmar.

One should consider the quality of Y-SNP data integrated in DNA chips. Of the 2041 and 943 Y-SNPs retrieved from the HumanOmniZhongHua-8 BeadChip and multiple chips used in CEU, 81.9% (1672/2041) and 88.7% (836/943) Y-SNPs were removed in data filtering, respectively (Table 2). Although we could not exclude inherent genotyping or calling errors in DNA chips, the high proportion of filtered data likely reflect the design of the probes. Around 50.4% (832/1672) and 82.1% (686/836) of filtered Y-SNPs for two respective data sets occur outside of the nine YSURs (Table 3). These filtered Y-SNPs may distribute in the heterochromatic, X-transposed, and ampliconic segments of the Y chromosome. As a result, the specificity of probes in the DNA chips needs to be improved. This can be achieved, at least partially, by replacing the problematic markers with those identified in Y chromosome resequencing (eg, see Wei et al,²³ Cruciani et al,²⁹ Xue et al³⁰) and then redesigning the probes. We also suggest that DNA chip manufacturers use our method to test their new products before release.

The authors declare no conflict of interest.