Intronic haplotypes in GBA modify age at diagnosis of Parkinson's: replication in a subgroup

In Schierding et al we identified noncoding variants within GBA that were associated with age at PD onset and diagnosis. Toffoli et al (this issue) failed to replicate our findings using data from the RAPSODI study and AMP-PD cohort. Here we provide evidence that supports our original findings and discuss the hypothesis that differing diagnostic criteria and/or data conglomeration is a potential basis for the replication failure of Toffoli et al.


Intronic Haplotypes in GBA Modify Age at Diagnosis of Parkinson's: Replication in a Subgroup
In Schierding et al we identified noncoding variants within GBA that were associated with age at PD onset and diagnosis. 1Toffoli et al (this issue) failed to replicate our findings using data from the RAPSODI study and AMP-PD cohort.
Here we provide evidence that supports our original findings and discuss the hypothesis that differing diagnostic criteria and/or data conglomeration is a potential basis for the replication failure of Toffoli et al.

Methods
The cohort and methods for polymerase chain reaction amplification and sequencing the GBA gene, and not the the pseudogene GBAP1, were previously described. 2For this analysis, patients were classified according to referring neurologist (Fig. 1).

Results
Haplotyping analysis of the Netherlands cohort of 1242 patients lacking GBA exonic variants did not replicate our findings (Fig. 1A, All).However, stratification by referral source identified a significant association (P = 0.0022) between the GBA1 intronic haplotype and age at diagnosis (AAD) in individuals who were referred to the study by tertiary center-based neurologists (Fig. 1A, Tertiary).The difference between the median ageat diagnosis for the AA and BB GBA1 intronic haplotypes was 10 years with weak evidence for a dosage effect (Fig. 1B).This finding was consistent with our original observation of a dosage effect and 3.4-year median difference in age at diagnosis observed between the 208 deeply phenotyped PD patients (AA vs BB) in the NZBRI cohort, who were diagnosed by a single clinician at a movement disorders clinic. 1,3e observed the identification of a significant haplotype-AAD relationship within the tertiary-diagnosed patients and not those from the other categories (peripheral, mix, and other(mix); Fig. 1).This observation may suggest that populations of patients who are at tertiary clinics are distinct from other populations.There are at least 2 nonexclusive explanations for this.First, it could reflect a scenario in which the diagnostic process for PD, and consequently AAD, varies between cohorts.If so, amalgamating patients diagnosed using differing diagnostic processes into a cohort is likely to obscure potential haplotype-AAD associations.As such, the observation that the RAPSODI study and the multiple cohorts that make up AMP-PD use differing diagnostic criteria is a concern (Supplementary Table 1).Data conglomeration issues like these are a recognized confounder for genomic studies because of variability in the phenotyping. 4Alternatively, it could be argued that some subtypes (eg, early onset or high familial burden) of PD patients are preferentially referred to and examined by tertiary neurologists.This could lead to the tertiary cohort having specific characteristics that are associated with the observed genetic trend.
Alternative explanations for our observations also include: (1) the sample sizes of the NZBRI and Netherlands PD (tertiary) are not sufficiently large, and the association is a false positive; (2) founder effects are present in both the NZBRI and Netherlands PD cohorts.
Finally, it is possible that the haplotype-AAD association was not detected in the AMP-PD because the accurate mapping of short-sequencing reads to GBA in AMP-PD is confounded by reads from the highly similar GBAP1 pseudogene.By contrast, the NZBRI and Netherlands cohorts underwent targeting sequencing of GBA.
Large cohorts with harmonized clinical, genomic, and transcriptomic datasets are critical resources for the breakthrough discoveries required to substantially advance our understanding of disease, its different trajectories, and the identification of potential therapeutic targets.However, as this study has indicated, potential variation in phenotyping, either within a cohort or between cohorts, has the capacity to diminish evidence of possibly important findings.