Introduction

Scientists have uncovered hundreds of genetic variants that purportedly cause familial forms of various neurodegenerative diseases, including Alzheimer’s, Parkinson’s, frontotemporal dementias, and amyotrophic lateral sclerosis. But are all of these associations causative? Which of the mutations are truly pathogenic? Non-geneticists often spend precious time and research funds studying the biology downstream of such polymorphisms. Can the field self-correct by pointing out, with all due collegial respect, which associations to set aside?

On 31 July 2012, Alzforum held a Webinar led by Rita Guerreiro, University College, London, UK, who was joined for a panel discussion by Marc Cruts, VIB, University of Antwerp; Bart De Strooper, VIB, University of Leuven, both in Belgium; and John Hardy, also from University College, London. Guerreiro presented algorithms that test the likelihood that inherited mutations are pathogenic, and the panel discussed associations that may or may not stand up to scrutiny.

Background

Background Text
By Tom Fagan

Familial forms of neurodegenerative diseases such as Alzheimer’s, frontotemporal dementias, and amyotrophic lateral sclerosis are explained by Mendelian genetics, which posits that these inherited disorders are traits caused by mutations in a single genetic locus. That led to the discovery of mutations in the genes for amyloid precursor protein (APP) and presenilins 1 and 2, three genes responsible for familial forms of AD. This, in turn, inspired mechanistic work that led to the amyloid cascade hypothesis. Genetic studies continue to report novel mutations in APP and the presenilins, as well as in genes such as progranulin and TDP-43 that underlie other neurodegenerative diseases. But are these genetic variants all pathogenic? The answer is important, because scientists use mutation studies as triggers to explore cell biology in hopes of uncovering underlying pathobiology. While AlzGene and PDGene are good guides for genetic risk associations, weak or incorrect Mendelian variants that are not subsequently corrected in print could send biologists astray.

In light of new evidence, geneticists charge that some previous genetic associations don’t seem to hold up. Newer data question not only specific mutations in disease genes, but also whether some genes associate with disease at all. For example, 197 presenilins and 33 APP variants have been discovered, but just because these genes can cause AD does not mean that every variant that exists will be pathogenic. Like many other polymorphisms in the human genome, some presenilin and APP mutations may be innocuous, others may be just risk factors for sporadic forms of the disease, and some may even be protective.

For instance, the E318G variant in presenilin 1 was deemed a familial AD (FAD) gene at first, but in fact pops up in numerous studies in late-onset AD (LOAD) and in healthy controls. It appears at best a risk factor for LOAD (Aldudo et al., 1998; Zekanowski et al., 2004; Helisalmi et al., 2000). Other presenilin 1 mutations reported to associate with AD but which are not likely pathogenic include R35G and V191A.

In presenilin 2, the R62H (Cruts et al., 1998), M174V, and R71W (Guerreiro et al., 2010) mutations that were classified as pathogenic now seem to be benign since they do not always segregate with disease and they turn up in healthy, non-demented individuals (Cruchaga et al., 2012). Likewise, L143H and A252T now seem innocent in the light of further studies.

Mutations in APP that are most likely benign include E665D, A673T, and H677R (see APP Mutations Table). For the tau gene, recent work suggests that the R5H mutation (see Table), thought to be pathogenic for frontotemporal dementia, is more likely benign as well (Cruchaga et al., 2012).

Growing data also call whole gene associations into question. Independent of its role in Aβ production, presenilin 1 was originally linked to a form of FTD characterized by ubiquitin inclusions and an absence of tau pathology (Amtul et al., 2002). If true, mutant presenilin would cause amyloid-free forms of dementia, i.e., act through a different pathway. The presenilin variant, an R352 insertion, acts as a dominant negative, suppressing γ-secretase activity. The purported association of presenilin 1 InsR352 with a form of FTD supported the notion that loss, rather than gain, of presenilin function drives certain pathologies (Amtul et al., 2002). However, further analysis revealed that progranulin, not presenilin 1 mutations, explains those FTD cases (see Pickering-Brown et al., 2006). Likewise, scientists now question the presenilin 1 G183V mutation reported to cause FTD (Dermaut et al., 2004) and which seems to lower presenilin 1 gene expression (Watanabe et al., 2012). And according to De Strooper, FTD cases attributed to the L113P presenilin 1 mutation (see Raux et al., 2000) turned out to have a frontal variant of Alzheimer’s rather than FTD. Are scientists who might still be studying these mutations barking up the wrong family tree? Sixty-nine mutations have been reported for the progranulin gene and 23 for FUS. Do they all cause disease?

Genetic associations for other neurodegenerative diseases have fared no better. Angiogenin polymorphisms (Greenway et al., 2006) now seem an unlikely cause of familial ALS since they poorly segregate with the disease and have been found in normal healthy controls (Corrado et al., 2007). Vascular endothelial growth factor and chromatin-modifying protein 2b also seem less likely candidates for ALS genes (see Schymick et al., 2007). And UCHL-1, Nurr1/NR4A2, and Htra2/Omi rest on weak genetic evidence, at best, as Parkinson's disease genes (Hardy et al., 2009).

“As two general rules, the more ‘biological’ data are used to convince the reader of the case for the genetic involvement of the gene, the more likely they are to be wrong. And the more complex the analytical protocol to ‘prove’ a gene involvement, the more suspicious the reader should be,” Hardy said.

How can molecular and cell biologists choose fruitful topics for investigation from the field of genetics? A start exists in the Alzheimer Disease & Frontotemporal Dementia Mutation Database curated by Marc Cruts at the Flanders Institute for Biotechnology in Antwerp, Belgium. This database indicates whether a given mutation is likely to be pathogenic. For example, the presenilin 1 page), as of 11 July 2012, listed four of 197 mutations as “not pathogenic” and eight as “pathogenic nature unclear.” One approach would be to establish strict standards for linking genes with causation. Guerreiro will present algorithms that test the robustness of genetic associations for presenilin, tau, and progranulin mutations. These weigh how strongly variants segregate with disease, whether they are absent in normal healthy controls, and how likely they are to interfere with protein function. Such algorithms could help researchers decide which genetic variants are worth pursuing in the lab.

Main Conclusions

  • Not all mutations in disease-causing genes are pathogenic, and not all mutations listed in databases should be considered as pathogenic without further scrutiny of the literature.
  • Mutations in genes that cause disease in Mendelian fashion may not be simple Mendelian mutations. Some may lack full penetrance in other families, while others may be risk or protective factors.
  • Better criteria are needed for determining whether specific mutations cause disease and if the use of decision trees like those presented by Rita Guerreiro should be implemented broadly.
  • Some criteria, for example, that the mutation should segregate with disease and be absent in controls, can be broadly applied. This is still the gold standard to decide whether a mutation is disease causing. However, criteria that depend on amino acid conservation and structure/function relationships may only be applicable to well-characterized genes.
  • As researchers avail themselves of newer technologies for deep sequencing, the number of sequence data will grow exponentially, only serving to compound the difficulties in assigning pathogenicity to specific mutations.
  • Sharing of data will be increasingly important as sequence databases grow.
  • More genetic data are needed on healthy, elderly controls to help rule out pathogenicity and to discriminate between Mendelian mutations and risk and protective variants.
  • African datasets may prove particularly useful because of their wide genetic variation, but researchers should be aware that phenotypic data is often poor, especially for controls.
  • The field needs a means to self-correct for dubious genetic associations when newer and more parsimonious explanations emerge in the literature.
  • Geneticists and cell biologists should cooperate early in the discovery process.
  • Researchers should be cautious when cell biology data are used to support genetic conclusions, and vice versa.
  • Researchers in the field should cooperate to devise newer, more stringent criteria for determining pathogenicity of genetic variants in disease genes.

Questions With Panelists' Answers

Q: This is a question for John. Are there some kind of guidelines to publish or deposit genetics data so researchers like us know a high-quality dataset has been generated? Or who will control the quality of the database for the field?

John Hardy: Marc Cruts does a good job. His database is pretty much definitive. But you should always be careful, and if you’re really dependent on something—like a cell biologist working on a particular gene and mutation—then you really should (as Bart said) read critically the original papers and understand enough to know whether you can trust it.

Marc Cruts: I believe that it is extremely useful to get as many data on genetic variability of the disease genes available to the community as possible. Current publication standards have a discouraging effect on those trying to publish novel variations in known disease genes if they don’t bring essential novel insights into the biology. It is likely that there are many interesting variations in disease genes hidden in private laboratory databases of which we have no knowledge at all. This information would, for example, help identify neutral/protective variants or help develop criteria to evaluate pathogenic variations. Introducing strict publishing criteria for genetic data will have an even more negative effect, by which only the most obvious pathogenic variations get into the public domain.

What we do need are improved annotations of pathogenicity. We are working hard to improve the labels in the Alzheimer Disease & Frontotemporal Dementia Mutation Database, but as was obvious from the Webinar, this is a non-trivial issue. The decision trees presented by Rita Guerreiro are a good start toward that direction. However, for some variations, conclusive evidence will never be reached based on available evidence. For those variations, the original literature is the best resource to form a personal opinion.

Bart De Strooper: I agree that more data should become available in the public domain. It is generally true that the hurdles to publish repetitions of experiments and negative data or observations are quite high. Alternative ways of getting this knowledge on the Web would be very helpful. Could the database of Marc not be updated or linked to an online repository for short reports on such variants that include brief descriptions of clinical data and of the functional data obtained?

Q: What percentage of familial AD cases is not explained by mutations in APP, PS1, and PS2, and what is the frequency of ApoE4 in FAD cases, compared to sporadic AD cases and controls?

John Hardy: In my view, autosomal dominant disease (three generations, fully penetrant with reasonable onset age) is fully explained by APP, PSEN1, and PSEN2. On ApoE4, the percentage in “sporadic” European AD is about 36 percent, and in familial it is about 43 percent. I published on this a very long time ago. I would bet that others have, too.

Marc Cruts: We have still a few genetically unexplained autosomal dominant AD families. Unexplained "familial" AD patients are not exceptional.

Q: What do you make of the report in The New York Times quoting the main author of the recent Nature DECODE paper as saying that the APP variant protected all E4/E4 carriers in their study from developing AD?

John Hardy: I think it is probably correct (about 90 percent confident) that it is protective. I would hesitate to suggest it was 100 percent protective because the (n) is very small.

Comment: If two variants are found in the same patient, for example, one in PS1 and one in GRN, and that patient has a TDP-43 pathology, then I think using Occam's razor is a sensible approach; that is, the most simple explanation is the most likely. In this scenario, the GRN is clearly the pathogenic mutation, and a new function for PS1 causing TDP-43 pathology does not need to be invented.

John Hardy: I agree.

Marc Cruts: in general terms, carrying a pathogenic mutation in one gene does not protect one from having a second pathogenic mutation in another gene. Consequently, the presence of a probable pathogenic mutation in the one gene does not affect the probability of pathogenicity of the other mutation. The pathological and clinical picture can be modulated by both variations.

Comments

Make a Comment

To make a comment you must login or register.

Comments on this content

No Available Comments

References

Paper Citations

  1. . Missense mutation E318G of the presenilin-1 gene appears to be a nonpathogenic polymorphism. Ann Neurol. 1998 Dec;44(6):985-6. PubMed.
  2. . The E318G substitution in PSEN1 gene is not connected with Alzheimer's disease in a large Polish cohort. Neurosci Lett. 2004 Mar 11;357(3):167-70. PubMed.
  3. . Is the presenilin-1 E318G missense mutation a risk factor for Alzheimer's disease?. Neurosci Lett. 2000 Jan 7;278(1-2):65-8. PubMed.
  4. . Estimation of the genetic contribution of presenilin-1 and -2 mutations in a population-based study of presenile Alzheimer disease. Hum Mol Genet. 1998 Jan;7(1):43-51. PubMed.
  5. . Genetic screening of Alzheimer's disease genes in Iberian and African samples yields novel mutations in presenilins and APP. Neurobiol Aging. 2010 May;31(5):725-31. PubMed.
  6. . Rare variants in APP, PSEN1 and PSEN2 increase risk for AD in late-onset Alzheimer's disease families. PLoS One. 2012;7(2):e31039. PubMed.
  7. . A presenilin 1 mutation associated with familial frontotemporal dementia inhibits gamma-secretase cleavage of APP and notch. Neurobiol Dis. 2002 Mar;9(2):269-73. PubMed.
  8. . Mutations in progranulin explain atypical phenotypes with variants in MAPT. Brain. 2006 Nov;129(Pt 11):3124-6. PubMed.
  9. . A novel presenilin 1 mutation associated with Pick's disease but not beta-amyloid plaques. Ann Neurol. 2004 May;55(5):617-26. PubMed.
  10. . Familial frontotemporal dementia-associated presenilin-1 c.548G>T mutation causes decreased mRNA expression and reduced presenilin function in knock-in mice. J Neurosci. 2012 Apr 11;32(15):5085-96. PubMed.
  11. . Dementia with prominent frontotemporal features associated with L113P presenilin 1 mutation. Neurology. 2000 Nov 28;55(10):1577-8. PubMed.
  12. . ANG mutations segregate with familial and 'sporadic' amyotrophic lateral sclerosis. Nat Genet. 2006 Apr;38(4):411-3. PubMed.
  13. . Variations in the coding and regulatory sequences of the angiogenin (ANG) gene are not associated to ALS (amyotrophic lateral sclerosis) in the Italian population. J Neurol Sci. 2007 Jul 15;258(1-2):123-7. PubMed.
  14. . Genetics of sporadic amyotrophic lateral sclerosis. Hum Mol Genet. 2007 Oct 15;16 Spec No. 2:R233-42. PubMed.
  15. . The genetics of Parkinson's syndromes: a critical review. Curr Opin Genet Dev. 2009 Jun;19(3):254-65. PubMed.

Other Citations

  1. APP Mutations Table

External Citations

  1. AlzGene
  2. PDGene
  3. Alzheimer Disease & Frontotemporal Dementia Mutation Database
  4. presenilin 1 page

Further Reading

Papers

  1. . Signal transduction through prion protein. Science. 2000 Sep 15;289(5486):1925-8. PubMed.