Challenges and disparities in the application of personalized genomic medicine to populations with African ancestry

Michael D. Kessler, Laura M. Yerges-Armstrong, Margaret A. Taub, Amol C. Shetty, Kristin A. Maloney, Linda Jo Bone Jeng, Ingo Ruczinski, Albert M. Levin, L. Keoki Williams, Terri H. Beaty, Rasika A. Mathias, Kathleen C. Barnes, Meher Preethi Boorgula, Monica Campbell, Sameer Chavan, Jean G. Ford, Cassandra Foster, Li Gao, Nadia N. Hansel, Edward A. Horowitz, Lili Huang, Romina Ortiz, Joseph Potee, Nicholas Rafaels, Alan L. Scott, Candelaria Vergara, Jingjing Gao, Yi‐Juan Hu, H. Richard Johnston, Zhaohui Qin, Badri Padhukasahasram, Georgia M. Dunston, Mezbah U. Faruque, Eimear E. Kenny, Kimberly Gietzen, Mark Hansen, Rob Genuario, Dave Bullis, Cindy Lawley, Aniket Deshpande, Wendy E. Grus, Devin P. Locke, Marilyn G. Foreman, Pedro C. Avila, Leslie C. Grammer, Kwang-YounA Kim, Rajesh Kumar, Robert P. Schleimer, Carlos D. Bustamante, Francisco M. De La Vega, Chris Gignoux, Suyash Shringarpure, Shaila Musharoff, Genevieve L. Wojcik, Esteban Burchard, Celeste Eng, Pierre‐Antoine Gourraud, Ryan D. Hernandez, Antoine Lizée, Maria Pino‐Yanes, Dara G. Torgerson, Zachary A. Szpiech, Raúl Torres, Dan L. Nicolae, Carole Ober, Christopher O. Olopade, Olufunmilayo I. Olopade, Oluwafemi Oluwole, Ganiyu Arinola, Wei Song, Gonçalo R. Abecasis, Adolfo Correa, Solomon K. Musani, James G. Wilson, Leslie A. Lange, Joshua M. Akey, Michael J. Bamshad, Jessica X. Chong, Wenqing Fu, Deborah A. Nickerson, Alexander P. Reiner, Tina V. Hartert, Lorraine B. Ware, Eugene R. Bleecker, Deborah Meyers, Victor E. Ortega, Maul R. N. Pissamai, Maul R. N. Trevor, Harold Watson, Maria Ilma Araújo, Ricardo Riccio Oliveira, Luis Caraballo, Javier Marrugo, B. Martínez, Catherine Meza, Gerardo Ayestas, Edwin Francisco Herrera-Paz, Pamela Landaverde-Torres, Said Omar Leiva Erazo, Rosella MartinezNature Communications 2016 · DOI

TL;DR

This paper investigates how biases in genomic databases and filtering methods disproportionately affect the interpretation of genetic variants in individuals of African ancestry. The study reveals that common precision medicine tools exhibit ancestry-related biases, leading to challenges in accurate diagnosis and increased costs for minority populations.

What Problem Does This Paper Address?

Precision genomic medicine aims to tailor medical treatments based on an individual's genetic makeup. A critical step in this process is identifying disease-causing genetic variants within a vast sea of genetic variations – often metaphorically described as finding a "needle in a haystack." However, the "haystack" itself, meaning the databases and analytical tools used, is not neutral. Historically, these resources have been heavily populated with data from individuals of European ancestry. This imbalance raises a significant concern: can these tools accurately and efficiently identify disease-causing variants in populations with different ancestral backgrounds, particularly those of African ancestry, who have distinct patterns of genetic variation? The paper directly addresses this by quantifying the extent of ancestry-related biases in these tools and exploring how these biases impact diagnostic accuracy and cost-effectiveness for minority populations.

Methods

The researchers employed a multi-faceted approach to dissect ancestry-related biases:

  • Variant Classification Pipeline: They developed a pipeline to categorize genetic variants into four groups based on their potential clinical relevance:
  • Deleterious Pathogenic Annotated Variants (PAVs)

    • Non-deleterious PAVs
    • Deleterious Non-Annotated Variants (NAVs)
    • Non-deleterious NAVs This classification involved using established databases (OMIM, HGMD, ClinVar), allele frequency filters (e.g., minor allele frequency < 5%), and in silico prediction tools to assess variant deleteriousness.
  • Ancestry Estimation: They used the ADMIXTURE software to estimate the proportion of African ancestry in each individual.

  • Correlation Analysis: Statistical methods were used to quantify the correlation between the estimated African ancestry proportion and the number of variants in each of the four classified groups. This was done both with and without applying various filtering criteria.
  • Temporal Analysis of ClinVar: To understand how biases evolve, they analyzed archived versions of the ClinVar database over time, correlating variant counts with African ancestry at different monthly updates.

Datasets

  • Consortium on Asthma among African-ancestry Populations in the Americas (CAAPA) Project: This study utilized whole-genome sequencing data from 642 individuals within this consortium. The CAAPA project is notable for its diverse representation of individuals with African ancestry from North America, South America, the Caribbean, and continental Africa.
  • Public Databases:
    • ClinVar: A public archive of relationships among sequence variation and human phenotype.
  • Human Gene Mutation Database (HGMD): A database of germline mutations associated with human inherited diseases.
  • Online Mendelian Inheritance in Man (OMIM): A catalog of human genes and genetic disorders.
  • Allele Frequency Databases:
    • 1000 Genomes Project (Phase 3): A comprehensive catalog of human genetic variation.
  • Exome Sequencing Project (ESP): Provides exome sequencing data from diverse populations.
  • Exome Aggregation Consortium (ExAC): A large-scale aggregation of exome sequencing data.

Key Insights and Ideas

  • The "Haystack" is Biased: The study powerfully demonstrates that the tools and databases used in precision medicine are not neutral and carry significant ancestry-related biases. This means the "haystack" is not uniform for all individuals.
  • Filtering Amplifies Bias (Sometimes): While filtering aims to refine variant calls, the specific filters applied can interact with ancestry in complex ways, sometimes reducing and sometimes even reversing correlations with African ancestry.
  • ClinVar is Dynamic and Ancestry-Sensitive: The ClinVar database, a critical resource for clinical interpretation, exhibits significant fluctuations in its ancestry-related bias over time, highlighting the dynamic nature of curated genetic information and its susceptibility to population-specific data contributions.
  • NAVs are the "Hidden Needles": Non-Annotated Variants (NAVs), particularly deleterious ones, are more likely to be correlated with African ancestry. This suggests that true causal variants for individuals of African ancestry might be disproportionately classified as NAVs, making them harder to find within the larger "haystack" of unannotated variants.
  • Cost and Implementation Hurdles: The identified biases directly translate into increased diagnostic effort, time, and cost for individuals of African ancestry, posing a significant barrier to equitable healthcare.

Key Results

  • Strong Correlations: Significant correlations were found between estimated African ancestry and the number of variants per individual across most variant classification sets, with the exception of deleterious PAVs.
  • HGMD vs. ClinVar: Variants identified in HGMD showed a strong positive correlation with African ancestry, while those prioritized by ClinVar were more abundant in individuals of European ancestry.
  • ClinVar's Temporal Instability: The correlation between African ancestry and ClinVar-based pathogenic variants showed a dramatic shift from positive (r=0.733) to negative (r=-0.683) between March and April 2014, driven by 68 specific SNPs.
  • Estimated Cost Increase: For Sanger confirmation of variants, an individual of African ancestry was estimated to have approximately 4.5 prioritized variants needing validation, compared to 2.8 for an individual of European ancestry, leading to an estimated cost difference of over $800 per patient.

Limitations and Future Work

  • Limitations:
    • The study acknowledges that the estimates for confirmation costs are simplified and conservative, not fully accounting for the clinician's review time.
  • While the paper identifies SNPs driving changes in ClinVar bias, the exact biological or curation reasons for these shifts are not fully elucidated.
  • The study focuses on African ancestry; similar biases might exist for other underrepresented populations.
  • Future Work:
    • The authors emphasize the need for further investigation into the specific mechanisms driving ancestry-related biases in predictors and databases.
  • They suggest that a more inclusive approach to variant prioritization, casting a wider net for variant identification and confirmation, is necessary.
  • Addressing these biases is crucial for improving the implementation, interpretation, and cost-effectiveness of genomic medicine for minority populations.

Summary updated

Abstract

To characterize the extent and impact of ancestry-related biases in precision genomic medicine, we use 642 whole-genome sequences from the Consortium on Asthma among African-ancestry Populations in the Americas (CAAPA) project to evaluate typical filters and databases. We find significant correlations between estimated African ancestry proportions and the number of variants per individual in all variant classification sets but one. The source of these correlations is highlighted in more detail by looking at the interaction between filtering criteria and the ClinVar and Human Gene Mutation databases. ClinVar's correlation, representing African ancestry-related bias, has changed over time amidst monthly updates, with the most extreme switch happening between March and April of 2014 (r=0.733 to r=-0.683). We identify 68 SNPs as the major drivers of this change in correlation. As long as ancestry-related bias when using these clinical databases is minimally recognized, the genetics community will face challenges with implementation, interpretation and cost-effectiveness when treating minority populations.
BibTeX
@inproceedings{kessler2016challenges,
  author = {Michael D. Kessler and Laura M. Yerges-Armstrong and Margaret A. Taub and Amol C. Shetty and Kristin A. Maloney and Linda Jo Bone Jeng and Ingo Ruczinski and Albert M. Levin and L. Keoki Williams and Terri H. Beaty and Rasika A. Mathias and Kathleen C. Barnes and Meher Preethi Boorgula and Monica Campbell and Sameer Chavan and Jean G. Ford and Cassandra Foster and Li Gao and Nadia N. Hansel and Edward A. Horowitz and Lili Huang and Romina Ortiz and Joseph Potee and Nicholas Rafaels and Alan L. Scott and Candelaria Vergara and Jingjing Gao and Yi‐Juan Hu and H. Richard Johnston and Zhaohui Qin and Badri Padhukasahasram and Georgia M. Dunston and Mezbah U. Faruque and Eimear E. Kenny and Kimberly Gietzen and Mark Hansen and Rob Genuario and Dave Bullis and Cindy Lawley and Aniket Deshpande and Wendy E. Grus and Devin P. Locke and Marilyn G. Foreman and Pedro C. Avila and Leslie C. Grammer and Kwang-YounA Kim and Rajesh Kumar and Robert P. Schleimer and Carlos D. Bustamante and Francisco M. De La Vega and Chris Gignoux and Suyash Shringarpure and Shaila Musharoff and Genevieve L. Wojcik and Esteban Burchard and Celeste Eng and Pierre‐Antoine Gourraud and Ryan D. Hernandez and Antoine Lizée and Maria Pino‐Yanes and Dara G. Torgerson and Zachary A. Szpiech and Raúl Torres and Dan L. Nicolae and Carole Ober and Christopher O. Olopade and Olufunmilayo I. Olopade and Oluwafemi Oluwole and Ganiyu Arinola and Wei Song and Gonçalo R. Abecasis and Adolfo Correa and Solomon K. Musani and James G. Wilson and Leslie A. Lange and Joshua M. Akey and Michael J. Bamshad and Jessica X. Chong and Wenqing Fu and Deborah A. Nickerson and Alexander P. Reiner and Tina V. Hartert and Lorraine B. Ware and Eugene R. Bleecker and Deborah Meyers and Victor E. Ortega and Maul R. N. Pissamai and Maul R. N. Trevor and Harold Watson and Maria Ilma Araújo and Ricardo Riccio Oliveira and Luis Caraballo and Javier Marrugo and B. Martínez and Catherine Meza and Gerardo Ayestas and Edwin Francisco Herrera-Paz and Pamela Landaverde-Torres and Said Omar Leiva Erazo and Rosella Martinez},
  title = {{Challenges and disparities in the application of personalized genomic medicine to populations with African ancestry}},
  booktitle = {Nature Communications},
  year = {2016},
  doi = {10.1038/ncomms12521},
}
Figure