Figures
Abstract
Background
As availability of primary cells can be limited for genetic studies of human disease, lymphoblastoid cell lines (LCL) are common sources of genomic DNA. LCL are created in a transformation process that entails in vitro infection of human B-lymphocytes with the Epstein-Barr Virus (EBV).
Methodology/Principal Findings
To test for genotypic errors potentially induced by the Epstein-Barr Virus transformation process, we compared single nucleotide polymorphism (SNP) genotype calls in peripheral blood mononuclear cells (PBMC) and LCL from the same individuals. The average mismatch rate across 19 comparisons was 0.12% for SNPs with a population call rate of at least 95%, and 0.03% at SNPs with a call rate of at least 99%. Mismatch rates were not correlated across genotype subarrays run on all sample pairs.
Conclusions/Significance
Genotypic discrepancies found in PBMC and LCL pairs were not significantly different than control pairs, and were not correlated across subarrays. These results suggest that mismatch rates are minimal with stringent quality control, and that most genotypic discrepancies are due to technical artifacts rather than the EBV transformation process. Thus, LCL likely constitute a reliable DNA source for host genotype analysis.
Citation: Herbeck JT, Gottlieb GS, Wong K, Detels R, Phair JP, Rinaldo CR, et al. (2009) Fidelity of SNP Array Genotyping Using Epstein Barr Virus-Transformed B-Lymphocyte Cell Lines: Implications for Genome-Wide Association Studies. PLoS ONE 4(9): e6915. https://doi.org/10.1371/journal.pone.0006915
Editor: Philip Awadalla, University of Montreal, Canada
Received: April 7, 2009; Accepted: August 7, 2009; Published: September 4, 2009
Copyright: © 2009 Herbeck et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding for this study was provided to J.I.M. by R37 AI47734, to J.T.H. and J.I.M. by the University of Washington Center for AIDS Research (P30 AI27757), to J.T.H. by NIH T32 AI07140. The MACS is funded by the National Institute of Allergy and Infectious Diseases, with additional supplemental funding from the National Cancer Institute and the National Heart, Lung and Blood Institute. UO1 AI35042, 5MO1 RR00722 (GCRC), UO1 AI35043, UO1 AI37984, UO1 AI35039, UO1 AI35040, UO1 AI37613, UO1 AI35041. Website located at http://www.statepi.jhsph.edu/macs/macs.html. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Advances in microarray technology have allowed high-throughput rapid genotyping of hundreds of thousands of single nucleotide polymorphisms (SNPs) across the human genome. These large sets of individual genotypes can be used for genome-wide association studies (GWAS), in which SNP allelic variation across a study population is tested for statistical associations with a particular disease phenotype. This method for studying the genetics of human disease has become widespread, and the genomic DNA necessary is generally provided by archived primary cells or tissue samples collected in prospective or longitudinal cohorts. As these samples are used for a wide range of studies and will become limited as more studies related to human disease are performed, the establishment of cell lines as permanent resources of genomic DNA is considered a potential solution. This process entails in vitro infection of human B-lymphocytes with the Epstein-Barr Virus, resulting in “immortalized” cell lines termed lymphoblastoid cell lines (LCL).
The feasibility of LCL for use in genetic studies has been evaluated primarily with regard to genomic copy number variation. Redon et al. compared copy number variations found in 268 HapMap (www.hapmap.org) LCL to copy number variations seen in each individual's blood cell-derived DNA [1]. They estimated that putative LCL-specific genomic errors accounted for less than 0.5% of observed deletions and considered LCL to be robust sources of genomic DNA for studies of copy number variation. Subsequent studies have supported the conclusion that LCL are likely to have little or minor effects on genomic structural variation (e.g. [2], [3], [4]).
With regard to gene expression and LCL, Choy et al. examined the utility of using HapMap LCL to identify expression quantitative trait loci (eQTL) that contribute to drug response phenotypes [5]. They found that non-genetic factors, such as the genomic EBV copy number, in vitro growth rate, and cellular ATP levels of individual LCL were more strongly associated with drug response and mRNA expression level phenotypes than any genotypic variation (genetic factors). The substantial noise from non-genetic factors impaired the ability to detect significant associations between genotypic variation and drug response or mRNA expression phenotypes. They noted that the non-genetic factors may be due to the EBV transformation process. In addition, Gimelbrant et al. and Plagnol et al. have documented that LCL exhibit extensive random, monoallelic expression [6], [7]. Plagnol et al. (2008) have suggested that LCL mRNA expression data may not be suitable for eQTL association analyses, given that LCL expression data can represent the random sampling of LCL clones with monoallelic expression not representative of the (source) cellular population as a whole.
LCL are commonly used in SNP analyses and genome-wide association studies, although the relationship between LCL genomic structural variation and genotype (SNP) fidelity is unclear. Redon et al. (2006) suggested that genomic duplications can result in Mendelian inconsistencies, and that genomic deletions can result in departures from Hardy-Weinberg equilibrium due to a lack of heterozygous genotypes. In this study, we attempted to quantify errors in SNP fidelity (genotypic discrepancies) that are potentially induced by the EBV transformation processes. To do so we compared the fidelity of SNP genotype calls in DNA obtained from paired samples of peripheral blood mononuclear cells (PBMC) and LCL from the same donor.
Materials and Methods
Samples
We compared the fidelity of SNP genotype calls in DNA obtained from paired samples of PBMC and LCL from 16 individuals of European American ancestry (mean age = 32.1 years) from the Multicenter AIDS Cohort Study (MACS). The MACS is an ongoing prospective study of the natural and treated histories of HIV-1 infection in homosexual and bisexual men conducted by study sites located in Baltimore, Chicago, Pittsburgh, and Los Angeles [8]. A total of 6,973 HIV-1 infected and uninfected men have been enrolled since 1984. The MACS collects plasma and PBMC, among other laboratory specimens and clinical data, from study participants at six-month intervals. In order to meet expected research demands on specimen availability, LCL were created from the PBMC from many participants.
The procedure for establishing LCL was adapted by the MACS from several previously described methods [9], [10], [11]. Briefly, B lymphocytes were separated from PBMC and incubated with Epstein-Barr Virus until immortalized. Immortalization was confirmed by continual increase in cell number, observation of cell blastogenesis and morphology, and detection of EBV antigens expressed in immortalized cell lines. After immortalization, transformed cells were grown in culture until they reached a concentration of 1×107 to 1×108 cells/ml. This stock culture served as the source for cell pellets of 5×106 cells/vial stored at −80°C until use. These cells were not passaged after immortalization.
The participants in this analysis were selected from an ongoing study of associations between host genetic factors and prognostic phenotypes of HIV-1 infection. This study includes 210 MACS individuals, of which 118 were genotyped from PBMC samples and 71 were genotyped from LCL samples (due to limited PBMC availability). The selection of the individuals for the LCL genotypic fidelity study described here was random, and is not expected to affect the estimates of LCL fidelity. We genotyped both PBMC and LCL source DNA from 16 individuals. Four individuals were also genotyped in duplicate from the same DNA source (two from PBMC, two from LCL); these served as control analyses of genotypic fidelity, as well as adding three replicates of PBMC vs. LCL tests, bringing the total number of PBMC versus LCL mismatch comparisons to 19.
Genotyping
DNA was extracted from PBMC and LCL cell pellets using the Qiagen QiaAmp blood mini-kit (Qiagen, Valencia, CA). We genotyped PBMC and LCL samples using the Affymetrix GeneChip Human Mapping 500 k array set (Affymetrix, Santa Clara, CA), which includes a combined 500,568 SNPs on two arrays, the 250 k Nsp I restriction enzyme assay (∼262,000 SNPs) and the 250 k Sty I restriction enzyme assay (∼238,000 SNPs). All samples were genotyped according to the Affymetrix protocol. In this approach, 250 ng of genomic DNA is first digested with a restriction enzyme (either Nsp I or Sty I), ligated to an adaptor, and amplified by PCR. The resulting amplicons are fragmented, labeled with biotinylated dideoxy ATP using terminal deoxynucleotidyl transferase, and hybridized to the array. Hybridization is detected by incubation with streptavidin-phycoerythrin conjugates, followed by scanning the array for phycoerythrin fluorescence and quantitation.
Genotype Fidelity
SNP genotypes were called using the Affymetrix BRLMM algorithm [12], and the mean SNP array call rate for the combined 500 k array across all individuals was 98.5%. We define genotype fidelity as the SNP genotype concordances in LCL and PBMC from the same individual. We analyzed genotype fidelity using the genetic association software PLINK [13], with which we calculated identity-by-state pairwise distances for the 500 k combined array set and for each 250 k array separately. Prior to analysis, we subjected SNPs to multiple levels of quality control based on individual SNP call rates across the larger population data set of 210 individuals (including both the LCL and PBMC genotypes from the 16 individuals described here). Four sets of comparisons were done: one with no SNPs filtered; and three in which SNPs were filtered if they were not called (had no call or had an ambiguous genotype call) in at least 90%, 95%, and 99% or more of the population data set, respectively. Estimates of genomic copy number variation were inconsistent for this data set, as batch effects from different array processing dates were confounding.
It has been suggested [1] that genomic deletions will result in an increased number of SNPs that depart from Hardy-Weinberg equilibrium (HWE) due to a lack of heterozygous genotypes. While we could not directly test for genomic deletions in our study, we compared the number of SNPs deviating from HWE in LCL to the number deviating from HWE in PBMC. We tabulated SNPs that deviated from HWE at three levels of statistical significance, P<0.05, 0.01, and 0.001, in PBMC and LCL populations independently (in the larger population of 210 individuals, with no replicate samples included, after filtering by population call rate).
Care should be taken to distinguish the population call rate from the array call rate: the population call rate is the frequency of successful genotype calls for a given SNP across a sampled population; the array call rate is the frequency of successful genotype calls for a particular SNP array.
Results
Genotypic fidelity and SNP population call rate
We estimated genotypic fidelity of LCL genomic DNA by comparing SNP genotypes inferred from LCL and from the original PBMC, from the same donor, for 16 individuals. We additionally genotyped three individuals in replicate (for 19 total LCL and PBMC comparisons), and genotyped four individuals in duplicate from identical source DNA (two LCL, two PBMC).
Genotypic fidelity increased with more stringent SNP calling quality control (Table 1, Figure 1). With SNP filtering at a 95% population call rate across the larger population study of 210 individuals, mean pairwise distance between PBMC and LCL genotypes was 0.12% for the 500 k combined array. This genotypic mismatch rate falls within the 0.10% to 0.30% mismatch rate reported by Affymetrix for replicate genotyping assays of the same purified DNA sample (after equivalent quality control). Further, mismatch rates at the 99% population call rate were ∼0.03%, equivalent to less than 100 mismatched SNPs in more than 300,000 genotype calls (Table 1). More stringent levels of quality control (increasing from no filtering upward to 99% population call rate) yielded significantly improved genotypic fidelity between PBMC and LCL genotypes (P = 3.82×10−6, Wilcoxon signed rank test, for the comparison of mismatch rates for no filtering and rates for filtering at the 99% population call rate level).
Genotypic fidelity is shown as mean pairwise distances among 19 paired comparisons of LCL and PBMC genotypes, for increasingly stringent SNP filtering by population call rate. A) Genotypic fidelity between LCL and PBMC source DNA from the same individual, estimated using the Affymetrix GeneChip Human Mapping 500 k Array set. B) SNP numbers remaining after filtering, shown for the combined 500 k array. C) Genotypic fidelity between LCL and PBMC source DNA from the same individual shown for the Nsp 250 k array. D) Genotypic fidelity between LCL and PBMC for the Sty 250 k array.
Comparisons of mismatch rates between LCL and PBMC pairs and duplicate pairs (duplicate genotyping of identical source DNA) revealed equivalent mean pairwise distances (P = 0.42, at the 99% population call rate level; Mann-Whitney U test, one sided, unequal sample sizes; Table 2). This test has a power of 0.62 to detect a difference of one standard deviation at a significance level of P = 0.05 (Figure 2). Considering LCL and PBMC samples from the same individual genotyped with 500 k array and filtered for 99% population call rate, one standard deviation is equivalent to ∼90 mismatched SNPs out of >300,000 compared.
Relationship between effect size, in units of (pooled) standard deviation, and statistical power, for a Mann-Whitney U test (one-sided, unequal sample sizes) with group samples sizes of 19 and 4. There is power of 0.62 to detect a difference in means of one standard deviation between LCL and PBMC pairs and duplicate pairs, equal to ∼90 SNPs out of >300,000 compared.
Genotypic fidelity comparing Nsp and Sty 250 k arrays
The Affymetrix 500 k genotype array is composed of two separate arrays, each containing ∼250 k SNPs. In our study, all individuals were genotyped with both arrays. Overall, the Sty 250 k array had lower genotype call rates and lower genotypic fidelity than the Nsp 250 k array (Figure 1), although at 95% and 99% population call rate levels the genotypic fidelity approached equal levels.
In order to best investigate the association of genotypic errors with LCL, it is heuristic to compare the LCL and PBMC mismatch rates seen for Nsp and Sty arrays across multiple individuals. In effect, our experimental design included duplicate tests for every individual LCL and PBMC comparison, as each pair was genotyped with two arrays (Nsp and Sty). If mismatch rates seen with Nsp and Sty arrays are correlated, we can infer that genotypic discrepancies between LCL and PBMC pairs reflect underlying genomic errors potentially associated with LCL (because identical source DNA was genotyped on different arrays). If mismatch rates between Nsp and Sty arrays are unrelated, we can infer that observed genotypic discrepancies reflect technical artifacts related to the genotyping process. We found no correlation between LCL and PBMC mismatch rates seen for Nsp and Sty arrays (P = 0.919, Spearman's rank correlation coefficient, rho = −0.33).
Genotypic fidelity and genotype array SNP call rate
Genotypic fidelity (SNP genotype concordance) between LCL and PBMC varied across sample pairs, with mismatch rates ranging from 0.175% to 0.664% (with no filtering). To investigate the source of this variation, we compared the array call rate to the mismatch rate. For each LCL versus PBMC pair, we compared the LCL versus PBMC mismatch rate to the lesser of the two array call rates (we assumed the lesser array call rate was more likely to be the source of potential genotypic mismatch). There was a strong relationship between array call rate and mismatch rate, and the relationship is seen for both Nsp and Sty arrays (Figure 3). The correlation was strongest with no filtering by population call rate (Table 3; Nsp array: slope = −0.167, r2 = 0.70, P = 6.42×10−7), and is alleviated with increasingly stringent filtering by population call rate, as seen at the 99% population call rate (Table 3; Nsp array: slope = −0.044, r2 = 0.60, P = 1.43×10−5).
Relationship between genotypic fidelity of LCL and PBMC pairs (mismatch rate, estimated using pairwise distance) and the array call rate. For each LCL versus PBMC genotypic pair, we compared the LCL versus PBMC mismatch rate to the lesser of the two array call rates. Mismatch rates are estimated at different levels of SNP filtering by population call rate: no filtering, 90%, 95%, and 99% percent population call rate.
Departures from Hardy-Weinberg equilibrium
We compared the numbers of SNPs that depart from HWE in PBMC and LCL sample populations after filtering by population call rate and removing replicate samples. While a trend existed toward increased SNPs deviating from HWE in LCL samples, the differences were not statistically significant (e.g., comparing the number of SNPs removed from LCL and PBMC sample populations, with HWE filtering at P<0.05, from populations with no call rate filtering, showed Χ2 = 0.12, P = 0.73), and more stringent population call rate filtering alleviated the discrepancy (Table 4).
Discussion
Here we report a study of SNP genotype fidelity between PBMC and EBV-transformed lymphoblastoid cell lines. We genotyped 16 PBMC and LCL pairs with the Affymetrix 500 k array (with three replicates genotyped, for 19 total comparisons). We estimated mean genotypic mismatch rates (estimated with pairwise genetic distances) and compared these rates to genotyping error rates reported by Affymetrix and also to mismatch rates seen in control pairs of duplicate genotyping of identical DNA samples. We also evaluated the effect of various levels of SNP quality control based on population call rate, and estimated mismatch rates seen with no SNP filtering, and with filtering those SNPs not called in 90%, 95%, and 99% or greater of individuals in a larger (N = 210) population. We found PBMC and LCL mismatch rates to be within ranges reported by Affymetrix for duplicate genotyping, when filtered at 95% or greater population call rate. PBMC and LCL mismatch rates are also not significantly greater than rates seen in control pairs.
We next compared mismatch rates seen for the Nsp and Sty 250 k subarrays, and found that rates for individual pairs were not correlated across subarrays, suggesting that the genotypic discrepancies observed are most likely due to technical artifacts of the genotyping process rather than the EBV transformation process. If the LCL genomic DNA contained gentoypic errors, mismatch rates would be consistent in Nsp and Sty arrays across the individuals examined. Additionally, the genotypic mismatch rate is highly correlated with the array call rate, and this relationship is seen for both Nsp and Sty arrays.
Genomic deletions in LCL have been hypothesized to result in an increased number of SNPs that depart from Hardy-Weinberg equilibrium, owing to a decreased number of heterozygotes. We tested for increased numbers of SNPs that depart from HWE in comparisons of PBMC and LCL samples, and while a trend existed in this direction, it was not statistically supported. We conclude that LCL are likely to have no apparent gross genotypic errors due to the process of EBV-transformation, and that SNP genotypes assayed from LCL may, with stringent quality control, be considered robust.
Acknowledgments
We thank Rob Hall, Trevor King, and Roger Bumgarner at the Center for Array Technologies, University of Washington, and Janet Schollenberger at the Center for the Analysis of MACS Data (CAMACS), Johns Hopkins University. The Multicenter AIDS Cohort Study (MACS) includes centers (Principal Investigators) at The Johns Hopkins University Bloomberg School of Public Health (Joseph B. Margolick, Lisa Jacobson), Howard Brown Health Center and Northwestern University Medical School (Steven Wolinsky), University of California, Los Angeles (Roger Detels), and University of Pittsburgh (Charles Rinaldo). Website at http://www.statepi.jhsph.edu/macs/macs.html.
Author Contributions
Conceived and designed the experiments: JTH GSG JIM. Performed the experiments: JTH. Analyzed the data: JTH. Contributed reagents/materials/analysis tools: JIM. Wrote the paper: JTH. Purified the DNA from archived samples: KGW. Collected data and samples, and served as principal investigators for the MACS: RD JP CRR LJ JBM.
References
- 1. Redon R, Ishikawa S, Fitch KR, Feuk L, Perry GH, et al. (2006) Global variation in copy number in the human genome. Nature 444: 444–454.
- 2. Jeon JP, Shim SM, Nam HY, Baik SY, Kim JW, et al. (2007) Copy number increase of 1p36.33 and mitochondrial genome amplification in Epstein-Barr virus-transformed lymphoblastoid cell lines. Cancer Genet Cytogenet 173: 122–130.
- 3. Simon-Sanchez J, Scholz S, Fung HC, Matarin M, Hernandez D, et al. (2007) Genome-wide SNP assay reveals structural genomic variation, extended homozygosity and cell-line induced alterations in normal individuals. Hum Mol Genet 16: 1–14.
- 4. McElroy JP, Nelson MR, Caillier SJ, Oksenberg JR (2009) Copy number variation in African Americans. BMC Genet 10: 15.
- 5. Choy E, Yelensky R, Bonakdar S, Plenge RM, Saxena R, et al. (2008) Genetic analysis of human traits in vitro: drug response and gene expression in lymphoblastoid cell lines. PLoS Genet 4: e1000287.
- 6. Gimelbrant A, Hutchinson JN, Thompson BR, Chess A (2007) Widespread monoallelic expression on human autosomes. Science 318: 1136–1140.
- 7. Plagnol V, Uz E, Wallace C, Stevens H, Clayton D, et al. (2008) Extreme clonality in lymphoblastoid cell lines with implications for allele specific expression analyses. PLoS ONE 3: e2966.
- 8. Kaslow RA, Ostrow DG, Detels R, Phair JP, Polk BF, et al. (1987) The Multicenter AIDS Cohort Study: rationale, organization, and selected characteristics of the participants. Am J Epidemiol 126: 310–318.
- 9. Louie LG, King MC (1991) A novel approach to establishing permanent lymphoblastoid cell lines: Epstein-Barr virus transformation of cryopreserved lymphocytes. Am J Hum Genet 48: 637–638.
- 10. Pelloquin F, Lamelin JP, Lenoir GM (1986) Human B lymphocytes immortalization by Epstein-Barr virus in the presence of cyclosporin A. In Vitro Cell Dev Biol 22: 689–694.
- 11. Caputo JL (1991) An effective method for establishing Human B lymphoblastic cell lines using Epstein-Barr Virus. J Tissue Culture Methods 13: 39–44.
- 12.
Affymetrix (2007) BRLMM: an improved genotype calling method for the GeneChip® human mapping 500 K array set. http://wwwaffymetrixcom/support/technical/whitepapers/brlmm_whitepaperpdf.
- 13. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, et al. (2007) PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 81: 559–575.