Research Article | Open Access
Corinne E. Sexton, Mark T. W. Ebbert, Ryan H. Miller, Meganne Ferrel, Jo Ann T. Tschanz, Christopher D. Corcoran, Alzheimer’s Disease Neuroimaging Initiative, Perry G. Ridge, John S. K. Kauwe, "Common DNA Variants Accurately Rank an Individual of Extreme Height", International Journal of Genomics, vol. 2018, Article ID 5121540, 7 pages, 2018. https://doi.org/10.1155/2018/5121540
Common DNA Variants Accurately Rank an Individual of Extreme Height
Polygenic scores (or genetic risk scores) quantify the aggregate of small effects from many common genetic loci that have been associated with a trait through genome-wide association. Polygenic scores were first used successfully in schizophrenia and have since been applied to multiple phenotypes including multiple sclerosis, rheumatoid arthritis, and height. Because human height is an easily-measured and complex polygenic trait, polygenic height scores provide exciting insights into the predictability of aggregate common variant effect on the phenotype. Shawn Bradley is an extremely tall former professional basketball player from Brigham Young University and the National Basketball Association (NBA), measuring 2.29 meters (76, 99.99999th percentile for height) tall, with no known medical conditions. Here, we present a case where a rare combination of common SNPs in one individual results in an extremely high polygenic height score that is correlated with an extreme phenotype. While polygenic scores are not clinically significant in the average case, our findings suggest that for extreme phenotypes, polygenic scores may be more successful for the prediction of individuals.
Polygenic, or genetic risk, scores are aggregate measurements of the effects of multiple common genetic loci that are associated with a trait. First used in schizophrenia , they have been applied to many complex traits such as multiple sclerosis , rheumatoid arthritis , and cardiovascular risk . However, polygenic scores are not generally expected to be clinical predictors of an individual’s phenotype. For example, Machiela et al. observed that the calculated AUC for the prediction of breast cancer from the polygenic score did not exceed 53%, which suggests that more validated variants (increased sample size) are necessary for a better prediction or that other factors besides common variants account for a large part of the disease phenotype . Similarly, Evans et al. found that while adding genome-wide variant information can slightly improve prediction accuracy, it is unlikely to be used for the prediction of individual phenotypes until larger datasets can improve the number of validated associated variants .
Most phenotypes (e.g., height, Alzheimer’s disease, Parkinson’s disease, etc.) are complex and polygenic, and our understanding of the underlying biology is limited because of high data dimensionality and small sample sizes. Approximately 80% of adult height variation has been attributed to genetic factors [7–10], and common SNPs are believed to account for approximately 50% of that variation [11, 12]. The Genetic Investigation of ANthropometric Traits (GIANT) consortium recently identified 697 SNPs across 423 loci that explain 20% of adult height heritability and further demonstrated that the 2000, 3700, and 9500 most significantly associated SNPs explained 21%, 24%, and 29% of height variation , respectively. Using 160 of these SNPs, which explain 10% of variation in height as reported by the GIANT consortium, Chan et al. observed that weighted polygenic allele scores were as predictive as expected in the extreme height phenotypes . This conclusion was also validated by Liu et al., who reported an AUC of 0.75 for a weighted allele score prediction for 180 SNPs on tall stature .
Shawn Bradley is an extremely tall former professional basketball player from Brigham Young University and the National Basketball Association (NBA), measuring 2.29 m (7 6) tall (Figure 1) and has no known medical conditions. Mr. Bradley’s height is 8.6 standard deviations (standard deviation = 6.05 cm) above the average height for US males (176.8 cm), putting him in the 99.99999th percentile . While height is known to be polygenic, exceptional outliers for height and other phenotypes remain intriguing because their rarity may present exciting genetic insights. Possible explanations for their rare height may include a combination of rare genetic variants, environmental factors (e.g., diet) and an extremely rare combination of common SNPs. Here, we present evidence of a relationship between common SNPs and an extreme polygenic phenotype and demonstrate that in Mr. Bradley’s specific case, the polygenic score predicts his height ranking as expected.
2. Materials and Methods
2.1. Sample Collection and Sequencing
The Cache County Study on Memory Health and Aging was initiated in 1994  and consists of 5092 participants representing approximately 90% of the Cache County population aged 65 and older in 1994. Specific details about data collection, obtaining consent, and phenotyping individuals in the Cache County population were reported previously , and other additional information on this dataset can be found in previous reports [16, 17].
Whole genome sequences (WGS) from 809 individuals (432 males, 354 females, and 23 unknown) were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). ADNI is a large collaboration from several academic and private institutions, and subjects have been recruited from over 50 sites across the US and Canada. Currently, over 1500 adults (ages 55 to 90) participate, consisting of cognitively normal older individuals, people with early or late MCI, and people with early stage Alzheimer’s disease. For up-to-date information, see http://www.adni-info.org.
We combined WGS from ADNI with WGS for 211 individuals (82 males and 129 females) from the Cache County study. All samples were sequenced using the Illumina HiSeq technology at an average of 30x coverage. We sequenced Mr. Bradley’s exome using the Ion Torrent and the Ion Ampliseq Exome Kit at an average coverage of 30x. Sequence data from all studies were mapped to the human reference genome, version GrCh37 with BWA (Burrows-Wheeler Aligner) . We further genotyped Mr. Bradley using the Illumina HumanOmniExpress chip and imputed additional SNPs using Impute2  and the 1000G reference panel . Subsequently, we filtered imputed SNPs with low information (info <0.4). Mr. Bradley and all individuals in the ADNI and Cache County cohorts are of Northern European ancestry .
SNP data from the Alzheimer’s Disease Genetics Consortium (ADGC) were used to examine patterns of linkage disequilibrium. The ADGC consists of 32 studies collected over two phases that include 16,000 cases and 17,000 controls. All subjects were self-reported as being of European American ancestry. More information about this dataset can be found in the study of Naj et al.  and the ADGC data preparation description .
The GIANT Consortium reported 22,539 genome-wide significant SNPs associated with human height. We extracted these SNPs from the ADGC data and identified unique tag SNPs within each LD block to (1) estimate the number of unique signals in the GIANT data and (2) prevent counting the same signal more than once. We identified tag SNPs using default settings in Haploview  for each chromosome individually (). We then extracted as many of the remaining SNPs as possible from Mr. Bradley’s data, the ADNI samples, and Cache County samples. We calculated an additive polygenic height score  for each individual and their respective ranks in the distribution of height scores. We also calculated the maximum possible score across the selected SNPs.
To estimate the number of SNPs needed to elevate Mr. Bradley’s height score to the highest in the distribution, we performed a random selection of SNPs (bootstrap) at various SNP-set sizes ranging from 100 to 2000 SNPs, recalculating Mr. Bradley’s height score and rank each time. We performed 1 million replicates for each SNP-set size and measured the range (minimum and maximum), first and third quartiles (25th and 75th percentiles), and the median for each SNP-set size.
We also explored the difference in height scores between the observed distribution of height scores amongst the 1020 individuals from ADNI and Cache County compared to the null distribution, assuming no evolutionary constraints. We simulated genotypes and height scores across the extracted common SNPs for 20 billion individuals. Specifically, for each SNP, we randomly chose one of three possible genotypes and calculated the simulated individual’s height score.
Understanding whether Mr. Bradley’s height is attributed to an increased proportion of heterozygous or homozygous genotypes associated with increased height could shed additional light on whether the SNP effects are additive or nonadditive (i.e., being homozygous has a greater effect than the sum). We tested for a difference between Mr. Bradley’s genotype distribution and the average ADNI and Cache County genotype distribution using a goodness-of-fit test. Alleles with a positive effect size are associated with increased height, while alleles with a negative effect size are associated with decreased height. A significant difference that could indicate the effects on height are nonadditive, though more data from extremely tall individuals would be necessary to provide definitive evidence.
We also tested whether height scores were correlated with actual height in 407 individuals from the ADNI and Cache County datasets for each individual with both height and genetic data available. We tested for a correlation between the two using Pearson’s product moment correlation coefficient, which is calculated using the R statistical package .
3. Results and Discussion
We tested whether a simple polygenic height score, calculated using SNPs that were statistically associated with human height in the GIANT consortium data , could accurately predict Mr. Bradley’s height rank amongst 1020 individuals of Northern European descent. We used Haploview to identify tag SNPs for each LD block across the 22,539 GIANT SNPs to avoid counting a single signal multiple times and to estimate how many independent signals exist in the GIANT SNPs. Using the Alzheimer’s Disease Genetics Consortium (ADGC) [22, 23] data with over 30,000 individuals, we identified 3428 unique signals, suggesting that most of the GIANT SNPs tag redundant effects. This is consistent with the GIANT result that most of the adult height variability explained by their SNPs is captured in the top 697 SNPs identified. After extracting genome-wide significant GIANT SNPs from Mr. Bradley’s exome and SNP data and using only a single tag SNP within each linkage disequilibrium (LD) block, 2910 SNPs (2491 genotyped, 419 imputed, Supplementary Table 1) remained and were included in the analysis. These represent 2910 of the 3428 LD blocks identified across the 22,539 significant GIANT SNPs using the ADGC dataset. Each allele included in this study is estimated by the GIANT consortium to affect an individual’s height by −0.14 to 0.19 millimeters.
We calculated height scores weighted by effect size (see Supplementary Table 1 for effect betas) for Mr. Bradley and 1020 individuals from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Cache County Study on Memory Health and Aging. Because Mr. Bradley’s height is 8.6 standard deviations above the average height of a male in the US, it is expected that his height score would be much higher than the average of the 1020 individuals for whom height scores were calculated. Mr. Bradley’s height score (10.32), calculated using the 2910 SNPs, was ranked highest, while the next highest was 7.43 (Figure 2). The mean height score within the ADNI and Cache County data was 0.98 with a standard deviation of 2.22, making Mr. Bradley’s height score 4.2 standard deviations above the mean, as expected.
In order to determine how few SNPs could be used for Mr. Bradley’s height score to rank highest when compared to the ADNI and Cache County population data, we created subsets of SNPs randomly from the 2910 available SNPs and then calculated height scores for all 1020 individuals as well as Mr. Bradley. We then ranked the resulting height scores and recorded Mr. Bradley’s percentile (Table 1). This procedure was replicated 1 million times for each SNP subset size. Choosing a subset of 100 SNPs randomly 1 million times, Mr. Bradley’s height scores calculated from the SNP subsets range from the lowest to the highest when compared to the ADNI and Cache County SNP subset height scores. His median height percentile settles at 96.9. Using a subset of 250 SNPs across 1 million iterations, Mr. Bradley’s median height percentile rises to 99.6 with his minimum height percentile at 20.4 and his maximum ranking highest. By using 750 SNPs, Mr. Bradley’s Q1 height rank is the top of the distribution, meaning that at least 75% of the time, his height score was ranked highest in the distribution. His lowest percentile using 750 SNPs was 78.8. Randomly selecting 1500 of the 2910 SNPs, Mr. Bradley’s lowest rank was in the 99.2 percentile (1017 of 1021).
Table 1 Shawn Bradley’s height score quickly stabilizes at the highest rank as SNP-set size increases. Data are represented in percentiles. The “” indicates that his score was the highest.
We also explored the difference in height scores between the observed distribution of height scores amongst the 1020 individuals from ADNI and Cache County when compared to the null distribution, based on 20 billion simulated individuals created from ADNI and Cache County genotypes, assuming no evolutionary constraints. The mean simulated height score (−0.30) was 1.28 mm lower than the observed height score mean (0.98). The maximum simulated height score (8.37) was 1.95 mm lower than Mr. Bradley’s (10.32).
We tested whether Mr. Bradley’s extreme height may be caused by an increased proportion of heterozygous or homozygous genotypes using a goodness-of-fit test (). Mr. Bradley has an increased proportion of homozygous genotypes for alleles with a positive effect (Table 2). He has nearly identical numbers of heterozygous genotypes for positive (associated with increased height) and negative (associated with decreased height) effect sizes with 621 and 634, respectively. The additive effects on his score for the positive and negative heterozygous genotypes are approximately equal and opposite at 15.12 and −15.27, respectively, summing to −0.17. There is a large difference, however, when comparing the homozygous genotypes for alleles with a positive and negative effect. Mr. Bradley has 465 genotypes where he is homozygous for GIANT alleles with a positive effect and only 267 genotypes where he is homozygous for GIANT alleles with a negative effect. The additive effects where Mr. Bradley is homozygous for positive and negative alleles are 25.89 and −15.42, respectively. The sum of all four scores equates to his height score of 10.32. Based on these data, Mr. Bradley’s height score rank is largely attributed to an excess of 198 positive-effect homozygous genotypes.
Using available height data from the ADNI and Cache County data, we tested whether the height scores calculated using the 2910 SNPs were correlated with the self-reported heights (at age 18) for the 407 individuals for which we have both height and genetic data. We failed to detect significant correlation between the two (correlation coefficient = 0.06, ; Figure 3). This is consistent with the findings of the GIANT consortium. With a population of 1914 individuals, Wood et al. found a predictive for 697 SNPs (20% variation explained) . It is expected that this r2 should be stronger than the correlation coefficient in our findings because of our smaller population size of 407 individuals of the ADNI and Cache County individuals as well as the fact that the GIANT consortium identified the 697 SNPs used for prediction directly from their population of 1914 individuals.
While research has shown that height is a polygenic trait heavily influenced by common SNPs [7–12], a polygenic score that quantifies common SNP effect is generally insufficient for successful individual phenotype prediction. We demonstrate that in the case of Mr. Bradley, a rare combination of common SNPs corresponds to an extremely high polygenic score that predicts an extreme phenotype. Because Mr. Bradley is an outlier, studying his genetic makeup provides a unique context to understand the complex nature of human height. Using a simple polygenic model across approximately 2000 SNPs, we accurately predicted Mr. Bradley’s height rank amongst a population of 1020 individuals.
The accurate prediction of tall individuals based on polygenic score has been found by both Chan et al.  and Liu et al. , confirming that in the case of an extremely tall phenotype, such as Mr. Bradley’s, polygenic scores can predict height rank. While these studies used a population of tall individuals to confirm their findings, we provide a validation of one individual polygenic height score rather than a distribution.
Mr. Bradley’s height score—like his actual height—was an extreme outlier (4.2 standard deviations above the mean). This appears to be driven by an increased proportion of homozygous genotypes for SNPs associated with increased height when compared to the average ADNI and Cache County genotype values. Despite this, his height score only predicted him to be 10.32 mm taller than average. This suggests that while Mr. Bradley’s extreme polygenic score could accurately rank his height amongst 1020 individuals, it does not accurately predict his actual height measurement, demonstrating that there are significant factors unaccounted for. Similarly, and as expected, this model was not able to accurately predict actual heights among the 407 ADNI and Cache County individuals for which we had both height and genetic data. These results as well as Mr. Bradley’s predicted height (10.32 mm taller than average) suggest that other factors such as environmental factors , nonadditive individual loci , and both epistasis (gene by gene interactions) and gene by environment interactions  play a significant role in determining actual height measurement. Recent studies of heritability in height and other complex traits suggest significant contributions of nonadditive factors [30, 31].
Height is a complex trait that may serve as an effective phenotype model for other complex traits and diseases because it is a noninvasive and easily-measured phenotype to study. By developing new models and studies to better understand all genetic contributors to an individual’s height, researchers will be able to apply the methods to other complex data.
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that there is no conflict of interest regarding the publication of this paper.
Corinne E. Sexton and Mark T. W. Ebbert contributed equally to this work.
The authors thank Mr. Bradley and the participants and staff of the centers that were involved in the data collection for ADNI and the Cache County study for their important contributions to this work. The data used in the preparation of this article were obtained in part from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in the analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf. Whole-genome data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI), National Institutes of Health Grant (U01 AG024904), and DOD ADNI, Department of Defense award (W81XWH-12-2-001). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd. and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Synarc Inc.; and Takeda Pharmaceutical Company. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (http://www.fnih.org). The grantee organization is the Northern Rev December 5, 2013 California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by the Laboratory for Neuroimaging at the University of Southern California.
Supplementary Table 1: provides the 2910 SNPs used to create the height scores as well as the beta scores from the GIANT consortium for each SNP. (Supplementary Materials)
- S. M. Purcell, N. R. Wray, J. L. Stone et al., “Common polygenic variation contributes to risk of schizophrenia and bipolar disorder,” Nature, vol. 460, pp. 748–752, 2009.
- The International Multiple Sclerosis Genetics Consortium (IMSGC), “Evidence for polygenic susceptibility to multiple sclerosis—the shape of things to come,” The American Journal of Human Genetics, vol. 86, no. 4, pp. 621–625, 2010.
- E. A. Stahl, D. Wegmann, G. Trynka et al., “Bayesian inference analyses of the polygenic architecture of rheumatoid arthritis,” Nature Genetics, vol. 44, no. 5, pp. 483–489, 2012.
- M. A. Simonson, A. G. Wills, M. C. Keller, and M. B. McQueen, “Recent methods for polygenic analysis of genome-wide data implicate an important effect of common variants on cardiovascular disease risk,” BMC Medical Genetics, vol. 12, no. 1, p. 146, 2011.
- M. J. Machiela, C. Y. Chen, C. Chen, S. J. Chanock, D. J. Hunter, and P. Kraft, “Evaluation of polygenic risk scores for predicting breast and prostate cancer risk,” Genetic Epidemiology, vol. 35, no. 6, pp. 506–514, 2011.
- D. M. Evans, P. M. Visscher, and N. R. Wray, “Harnessing the information contained within genome-wide association studies to improve individual prediction of complex disease risk,” Human Molecular Genetics, vol. 18, no. 18, pp. 3525–3531, 2009.
- R. A. Fisher, “XV.—The correlation between relatives on the supposition of Mendelian inheritance,” Earth and Environmental Science Transactions of the Royal Society of Edinburgh, vol. 52, no. 2, pp. 399–433, 1919.
- K. Silventoinen, S. Sammalisto, M. Perola et al., “Heritability of adult body height: a comparative study of twin cohorts in eight countries,” Twin Research and Human Genetics, vol. 6, no. 5, pp. 399–408, 2003.
- P. M. Visscher, S. E. Medland, M. A. R. Ferreira et al., “Assumption-free estimation of heritability from genome-wide identity-by-descent sharing between full siblings,” PLoS Genetics, vol. 2, no. 3, article e41, 2006.
- A. R. Wood, T. Esko, J. Yang et al., “Defining the role of common variation in the genomic and biological architecture of adult human height,” Nature Genetics, vol. 46, no. 11, pp. 1173–1186, 2014.
- J. Yang, B. Benyamin, B. P. McEvoy et al., “Common SNPs explain a large proportion of the heritability for human height,” Nature Genetics, vol. 42, no. 7, pp. 565–569, 2010.
- J. Yang, T. A. Manolio, L. R. Pasquale et al., “Genome-partitioning of genetic variation for complex traits using common SNPs,” Nature Genetics, vol. 43, no. 6, pp. 519–525, 2011.
- Y. Chan, O. L. Holmen, A. Dauber et al., “Common variants show predicted polygenic effects on height in the tails of the distribution, except in extremely short individuals,” PLoS Genetics, vol. 7, no. 12, article e1002439, 2011.
- F. Liu, A. E. J. Hendriks, A. Ralf et al., “Common DNA variants predict tall stature in Europeans,” Human Genetics, vol. 133, no. 5, pp. 587–597, 2014.
- C. D. Fryar, Q. Gu, C. L. Ogden, and K. M. Flegal, “Anthropometric reference data for children and adults; United States, 2011-2014,” Vital and Health Statistics, vol. 3, no. 392016, 2016.
- J. C. S. Breitner, B. W. Wyse, J. C. Anthony et al., “APOE-ε4 count predicts age when prevalence of AD increases, then declines the Cache County study,” Neurology, vol. 53, no. 2, pp. 321–331, 1999.
- M. T. W. Ebbert, P. G. Ridge, A. R. Wilson et al., “Population-based analysis of Alzheimer’s disease risk alleles implicates genetic interactions,” Biological Psychiatry, vol. 75, no. 9, pp. 732–737, 2014.
- H. Li, “Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM,” 2013, http://arxiv.org/abs/1303.3997.
- B. N. Howie, P. Donnelly, and J. Marchini, “A flexible and accurate genotype imputation method for the next generation of genome-wide association studies,” PLoS Genetics, vol. 5, no. 6, article e1000529, 2009.
- The 1000 Genomes Project Consortium, “A global reference for human genetic variation,” Nature, vol. 526, pp. 68–74, 2015.
- A. R. Sharp, P. G. Ridge, M. H. Bailey et al., “Population substructure in Cache County, Utah: the Cache County study,” BMC Bioinformatics, vol. 15, Supplement 7, pp. S8–S8, 2014.
- A. C. Naj, G. Jun, G. W. Beecham et al., “Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer's disease,” Nature Genetics, vol. 43, no. 5, pp. 436–441, 2011.
- K. L. Boehme, S. Mukherjee, P. K. Crane, and J. S. Kauwe, “ADGC 1000 Genomes combined data workflow,” October 2015, http://kauwelab.byu.edu/Portals/22/adgc_combined_1000G_09192014.pdf.
- J. C. Barrett, B. Fry, J. Maller, and M. J. Daly, “Haploview: analysis and visualization of LD and haplotype maps,” Bioinformatics, vol. 21, no. 2, pp. 263–265, 2005.
- F. Dudbridge, “Power and predictive accuracy of polygenic risk scores,” PLoS Genetics, vol. 9, no. 3, article e1003348, 2013.
- R Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, 2018, http://www.R-project.org/.
- B. Bogin and L. Rios, “Rapid morphological change in living humans: implications for modern human origins,” Comparative Biochemistry and Physiology Part A: Molecular & Integrative Physiology, vol. 136, no. 1, pp. 71–84, 2003.
- G. Su, O. F. Christensen, T. Ostersen, M. Henryon, and M. S. Lund, “Estimating additive and non-additive genetic variances and predicting genetic merits using genome-wide dense single nucleotide polymorphism markers,” PLoS One, vol. 7, no. 9, article e45293, 2012.
- P. M. Visscher, J. Yang, and M. E. Goddard, “A commentary on ‘Common SNPs explain a large proportion of the heritability for human height’ by Yang et al. (2010),” Twin Research and Human Genetics, vol. 13, no. 06, pp. 517–524, 2010.
- T. A. Manolio, F. S. Collins, N. J. Cox et al., “Finding the missing heritability of complex diseases,” Nature, vol. 461, no. 7265, pp. 747–753, 2009.
- J. Yang, J. Zeng, M. E. Goddard, N. R. Wray, and P. M. Visscher, “Concepts, estimation and interpretation of SNP-based heritability,” Nature Genetics, vol. 49, no. 9, pp. 1304–1310, 2017.
Copyright © 2018 Corinne E. Sexton et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.