More than a year ago, researchers identified variants in phospholipase D3 (PLD3) as risk factors for Alzheimer’s disease. First found in two families and then confirmed in several larger cohorts, the variants reportedly doubled the risk of AD. Earlier this month, three letters to Nature reported otherwise. The writers had failed to turn up any association between AD risk and variations in the gene. A fourth letter reported a marginally significant risk associated with one of the variants.

The search for rare variants, such as those in PLD3, will always be fraught with contrasting results derived from different cohorts, said Carlos Cruchaga of Washington University in St. Louis, who was the first author of the original study. In two responses to the letters that appeared April 1 in Nature online, Cruchaga and senior author Alison Goate, now at the Icahn School of Medicine at Mount Sinai Hospital in New York, claimed that one likely reason for the conflicting data was differing genetic backgrounds in subsequent studies. Other researchers insisted that the preponderance of evidence in the letters—one of which included a meta-analysis of data from all of the letters and Cruchaga’s data combined—strongly dispute that PLD3 is a significant risk factor. In the end, everyone who spoke with Alzforum agreed that larger sequencing studies should be analyzed before putting a final nail in the PLD3 coffin.  

Cruchaga and colleagues had used exome sequencing to identify the PLD3 coding variant Val232Met. It segregated with AD in two families in the National Institute on Aging Late-Onset Alzheimer’s Disease (NIA-LOAD) study. Then the geneticists screened seven independent data sets from the United States, which together included nearly 5,000 sporadic AD cases and more than 6,000 controls. Val232Met had a higher frequency in cases than controls in all of the data sets, and on average doubled AD risk. Much like ApoE4, the frequency of Val232Met decreased in healthy people with increasing age, suggesting that those who live longer without dementia are less likely to carry the variant. Further sequencing of PLD3 in more than 2,000 additional cases and controls of European descent revealed 20 variants in PLD3, 14 of which occurred more frequently in cases than in controls and nine of which only occurred in cases. Two variants in addition to Val232Met—Met6Arg and Ala442Ala, the result of a synonymous nucleotide substitution that the researchers found affected alternative splicing—were significantly associated with AD risk. The Val232Met and Ala442Ala variants also lined up with AD in a small African-American cohort of 302 cases and controls (see Dec 2013 news).

Researchers led by Cornelia van Duijn at Erasmus Medical Center in Rotterdam, Netherlands, partially confirmed Cruchaga’s findings. They looked for the Val232Met variant in three population-based and three case-control studies conducted in the United States and Europe. First author Sven van der Lee and colleagues found that the frequency of the methionine version varied widely across cohorts. Between 0.34 and 1.42 percent of controls carried it—but it was always present at a higher frequency in cases. The variant did not significantly predict AD risk in any one cohort, but combining the odds ratios from all of them yielded a significant, 53 percent increase in AD risk. After adjusting for age and sex, Val232Met nearly doubled AD risk, putting these results in line with those of Cruchaga’s. However, some authors of the other letters pointed out, and van Duijn agreed, that the frequencies of the allele differed so much that the controls in one population sometimes had a higher carrier frequency than the cases of another, which made the association a bit dubious. Also, each cohort was relatively small, most containing five or fewer carriers of the variant. “When this happens, it is a warning that you should be careful in your interpretation,” van Duijn told Alzforum.

Van Duijn and colleagues next sifted through whole-genome sequencing data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and a Dutch AD cohort, and found several of the other variants Cruchaga and colleagues identified. However, none, including Ala442Ala, significantly associated with AD risk. The researchers next conducted a burden analysis, which pools together all of the coding variants in a gene rather than just looking at one at a time. They concluded that even when combined, the variants did not confer increased AD risk at the gene level. Van der Lee and colleagues concluded that PLD3 did not meet the criteria for a bona fide AD risk gene.

Researchers led by Philippe Amouyel of the Pasteur Institute in Lille, France, looked for AD risk association in more than 2,000 French cases and controls. They found that no one in the study carried the Met6Arg variant, deeming it monomorphic. Neither Val232Met nor Ala442Ala associated with disease risk. Also, in contrast to Cruchaga's findings, first author Jean-Charles Lambert and colleagues found that the frequency of the Val232Met variant in healthy controls was similar across age groups (see Lambert et al., 2015).

A third study, led by Alfredo Ramirez at the University of Bonn in Germany, looked for AD risk associated with four of the PLD3 variants—Val232Met, Ala442Ala, Met6Arg, and Pro76Ala—in two Spanish and two German cohorts, which together comprised more than 3,000 non-familial AD cases and controls. First author Stefanie Heilmann and colleagues found no association between any of the variants and AD risk. Again, a burden analysis indicated that PLD3 was not a risk factor.

In response to these three letters, Cruchaga and Goate wrote that population stratification—which occurs when people within a cohort have differing genetic backgrounds—may have hindered efforts to replicate their findings. Population stratification could produce spurious results, for example, if two genetically distant populations with differing frequencies of a certain variant exist within a cohort. If the population with higher carrier frequency predominates in the controls, the variant might not seem to associate with disease, producing a false negative. To control for this possibility, researchers often incorporate data from genome-wide association studies conducted on their cohorts, which gives them an idea of genetic heterogeneity. They also match cases and controls with similar ethnicities. Cruchaga, van der Lee, and Lambert all performed such adjustments, but Ramirez did not. Goate and Cruchaga suggested that this lack of control could have produced a false negative result. As evidence that variant frequencies differ even among people from the same country, Cruchaga and Goate pointed out that the frequency of the Val232Met was 0.36 percent in one of the Spanish cohorts of nearly 3,000 controls, and zero in a much smaller cohort of 180 controls, again Spanish.  

Ramirez noted that while the lack of variant carriers in the smaller cohort was likely due to its size, the 0.36 percent frequency in the larger cohort was similar to the German control cohort (0.43 percent) in his study, and to the French control cohort in Lambert’s study (0.41percent). While he could not rule out population stratification as a confounding factor in his study, Ramirez said he felt it was unlikely to explain the results.

Lambert told Alzforum that of all cohorts analyzed so far, he considers the American cohorts in the Cruchaga study to be most susceptible to this confound, simply due to their higher diversity. “The American cohort comprises people from all over Europe,” Lambert said. “Even though Cruchaga controlled for population stratification and genetic substructure, in my point of view it may not have been enough.” Lambert and Ramirez agreed that genetic heterogeneity, combined with low variant frequency and small sample sizes, could have yielded false positive results in the case of Cruchaga’s study, or even false negative results in their own. Only larger studies will help sort this out.

In support of their argument that Ramirez’s study could have false negatives, Cruchaga and Goate cited a recent paper reporting several PLD3 coding variants enriched in German people with late-onset AD (see Schulte et al., 2015). However, as other researchers pointed out, Schulte et al. found no significant association between Val232Met or Ala442Ala variants and AD risk. Rather, the Val232Met variant occurred at greater frequency in controls than in AD cases. The study instead identified several other variants that associated with AD risk. 

Rounding out the quartet of letters in Nature, researchers led by Lars Bertram of the University of Lübeck in Germany and Rudolph Tanzi of Massachusetts General Hospital in Boston used whole-genome sequencing data to look for PLD3 variants in 439 families comprising 1,440 people in the National Institute of Mental Health (NIMH) Alzheimer’s Disease Genetics Initiative study sample. First authors Basavaraj Hooli (from Mass. General) and Christina Lill (from Lübeck) and colleagues found that the frequency of the Val232Met variant was actually higher in unaffected family members than in those with AD. “When you see this sort of fluctuation between studies, where some people are saying risk and others protection, usually it indicates that this is just a chance finding,” Bertram said.

In their published reply to Hooli and colleagues, Cruchaga and Goate wrote that they were surprised by the high number of variants the researchers identified in some families, as well as the number of families that harbored at least one variant. For example, Bertram and colleagues found 10 non-synonymous coding variants in just five families, suggesting that those families had more than one variant each. They also found that nearly half of the 439 families harbored at least one of 30 coding variants identified in the sequencing. Cruchaga and Goate wrote that this contrasted with their own findings, which identified variants in only 8 percent of cases and 3 percent of controls across the NIA-LOAD, NIA-UK, and Knight Alzheimer’s Disease Research Center data sets. Bertram did not think this discrepancy called his data into question.  Since each of the variants is indeed so rare, a few extra carriers have enough power to bump up the frequency in small cohorts, he said.

In the grand finale experiment, Bertram and colleagues conducted a meta-analysis that included the original data from Cruchaga as well as the data from all four letters to Nature. They found no significant association between either the Val232Met or the Ala442Ala mutant and AD risk. “When you put the data from all of the validation studies together, none of the variants are significantly associated with AD risk,” Bertram said. “Especially with rare variants, what you see in the first study sometimes turns out to not be correct. It’s the nature of this research, and you need really large numbers to be sure.” Geneticists refer to this as the winner’s curse.

Cruchaga and all the other researchers agreed that emerging data from large-scale sequencing efforts, such as the Alzheimer’s Disease Sequencing Project and the Alzheimer’s Disease Genetics Consortium, will likely clarify the role of PLD3 variants in AD. Cruchaga believes that while PLD3-associated risk may vary between populations analyzed so far, the protein likely plays a role in AD biology. The lipase is expressed at lower levels in the brains of AD patients, and seems to reduce Aβ production in cell culture studies, he reported in the original study. Another recent study reported accumulation of PLD3 in neuritic plaques in AD brains (see Satoh et al., 2015).

Regardless of whether PLD3 ends up in the AD risk hall of fame, Ramirez said these conflicting studies are a crucial part of the process of understanding how to uncover rare variants. “These rare variants must be analyzed carefully, and we have to figure out how to do this,” he said. “Having this discussion is the only way that we will find the truth.”—Jessica Shugart


No Available Comments

Make a Comment

To make a comment you must login or register.


News Citations

  1. Phospholipase D3 Variants Double Sporadic AD Risk

Paper Citations

  1. . PLD3 and sporadic Alzheimer's disease risk. Nature. 2015 Apr 2;520(7545):E1. PubMed.
  2. . PLD3 is accumulated on neuritic plaques in Alzheimer's disease brains. Alzheimers Res Ther. 2014;6(9):70. Epub 2014 Nov 2 PubMed.

External Citations

  1. Schulte et al., 2015
  2. Alzheimer’s Disease Sequencing Project
  3. Alzheimer’s Disease Genetics Consortium

Further Reading


  1. . PLD3 in Alzheimer's Disease. Mol Neurobiol. 2014 Jun 17; PubMed.

Primary Papers

  1. . PLD3 variants in population studies. Nature. 2015 Apr 2;520(7545):E2-3. PubMed.
  2. . PLD3 and sporadic Alzheimer's disease risk. Nature. 2015 Apr 2;520(7545):E1. PubMed.
  3. . PLD3 in non-familial Alzheimer's disease. Nature. 2015 Apr 2;520(7545):E3-5. PubMed.
  4. . PLD3 gene variants and Alzheimer's disease. Nature. 2015 Apr 2;520(7545):E7-8. PubMed.