Introduction
In 2010, a high-profile Genome-Wide Association Study (GWAS) identified several genetic variants (SNPs) purportedly linked to an increased risk of esophageal squamous cell carcinoma (ESCC) in Chinese populations. This corrigendum provides a critical re-evaluation of those findings, revealing that several major genetic associations previously linked to the disease were likely "statistical ghosts." The authors demonstrate that these false positives were the result of inadequate control for population stratification—a classic and persistent pitfall in genomic research.
The Problem: Population Stratification
The validity of any GWAS hinges on the fundamental assumption that the case group (individuals with the disease) and the control group (healthy individuals) are genetically similar in every respect except for the trait being studied. If the cases are inadvertently drawn from a different ancestral sub-group than the controls, any genetic variation between those sub-groups will appear to be a disease marker, even if it is merely a marker of regional ancestry. This phenomenon, known as population stratification, led to the identification of susceptibility loci that did not hold up under more rigorous ancestral matching.
Methodology and Re-analysis
To address these potential errors, the authors performed a comprehensive re-analysis of their original data while incorporating several new datasets to validate the previous findings. The primary methodological shift involved the use of genetically matched subsets.
Instead of analyzing the broad original pool of subjects, the researchers employed principal component analysis (PCA) and other ancestry-matching techniques to ensure the cases and controls were truly comparable. This refined group consisted of 1,789 subjects (1,076 cases and 713 controls) selected for ancestral homogeneity. These results were then cross-referenced against:
- A New Henan Cohort: 669 new subjects (229 cases, 370 controls).
- NCI GWAS Data: An independent dataset from the National Cancer Institute.
- Beijing GWAS Data: A dataset representing a "low-risk" population used for comparative replication.
Key Insights: The Vanishing Signal
The most significant finding of this corrigendum is the dramatic evaporation of statistical significance once population matching was applied.
- The C20orf54 Locus: In the original study, the SNP rs13042395 at the C20orf54 locus showed a highly significant p-value of $1.51 \times 10^{-11}$, well beyond the standard threshold for discovery. However, in the combined analysis of matched and new subjects, the Odds Ratio (OR) shifted from 0.66 to 0.88, and the p-value plummeted to a non-significant $0.09$ ($9.12 \times 10^{-2}$).
- Other Retracted Loci: The authors found no evidence to support the association for three other previously highlighted SNPs: rs17761864 (17p13), rs6503659 (17q21), and rs2847281 (18p11).
These results serve as a stark reminder that large sample sizes (N) cannot compensate for poor population control. The failure of the rs13042395 variant to replicate in either high-risk (NCI) or low-risk (Beijing) populations confirms that the original signal was a statistical artifact rather than biological reality.
Conclusion and Future Directions
This corrigendum updates the scientific record by providing corrected estimates for these SNPs, effectively removing them from the list of confirmed ESCC susceptibility factors. The study highlights a fundamental limitation in early GWAS: the reliance on geographic proximity as a proxy for genetic similarity.
For the bioinformatics and genomics communities, this work underscores the necessity of using robust ancestral informative markers (AIMs) and stricter population stratification controls. Furthermore, it demonstrates the vital role of the "post-publication" phase of science, where independent replication and rigorous re-analysis serve as essential self-correcting mechanisms for the field.
Summary updated