Friday, August 13, 2010

A Concise Critique of the Methods Used in the Personality Genetics Paper

The previous post, Bad News for the Genetics of Personality, discussed a genome-wide association study that searched for common genetic variants associated with personality ratings from Cloninger's temperament scales. None were found:
Participants' scores on Harm Avoidance, Novelty Seeking, Reward Dependence, and Persistence were tested for association with 1,252,387 genetic markers. We also performed gene-based association tests and biological pathway analyses. No genetic variants that significantly contribute to personality variation were identified, while our sample provides over 90% power to detect variants that explain only 1% of the trait variance. This indicates that individual common genetic variants of this size or greater do not contribute to personality trait variation, which has important implications regarding the genetic architecture of personality and the evolutionary mechanisms by which heritable variation is maintained.
The post discussed possible problems with the Tridimensional Personality Questionnaire but remained agnostic on the GWA aspects of the paper:
[Those technical and statistical methods are beyond the scope of this blog, so I will leave it to someone else to describe and critique the genotyping aspects of the paper.]
A comment at Gene Expression on Heritability, personality, and genomics by Princeton Professor Lee M. Silver criticized the methods used by Verweij et al. (2010):
20. LeeMSilver Says:
August 10th, 2010 at 8:04 am

If you read the original article describing the research, you’ll find numerous serious problems with the way the study was conducted. The GWAS approach is designed to work with a sample of unrelated individuals. But the sample set used by Wray and colleagues consisted of 5117 subjects, who only came from 2567 families. The sample set also included 1702 monozygotic twins (797 pairs), and it also included an entire cohort of people diagnosed with bipolar disorder. All of this substructure can muddy the waters, which seems to be perfectly fine to Wray and colleagues whose language gives away their bias against GWAS.

However, the really fundamental problem is the subjective complexity of all the traditional scales for measuring personality. For thousands of years, breeders tried to find patterns in heredity, with no success. Mendel succeeded not by stuffing multiple measurements into individual traits but by eliminating all the variables except one or two in each experiment. This won’t get you the genes for “harm avoidance” (which has no objective meaning) but it could uncover loci that influence the response to one important question in the harm avoidance panel, before investigating the next.


ADDENDUM: Authors Karin Verweij and Brendan Zietsch have replied to Lee Silver in the comments. Briefly, they pointed out that their analysis did take into account relatedness between family members, including monozygotic twins. They also noted that the small subsample with high vs. low depression scores was treated appropriately in the analysis. Furthermore, these high depression individuals were not diagnosed with bipolar disorder.

Besides his critique that the population was not suitable for GWAS, Professor Silver also raised a good point on reducing traits or temperaments to their constituent elements. For starters, Harm Avoidance has four subscales:
  1. Anticipatory worry (HA1)
  2. Fear of uncertainty (HA2)
  3. Shyness/Shyness with strangers (HA3)
  4. Fatigability/Fatigability and asthenia (HA4)
The same is true for Novelty Seeking, except the subscales seem even more diverse:
  1. Exploratory excitability (NS1)
  2. Impulsiveness (NS2)
  3. Extravagance (NS3)
  4. Disorderliness (NS4)
LMK made similar points in a comment on The Neurocritic's previous post. There are a number of other interesting observations, including a link to Kevin Mitchell on Nature, Nurture, and Noise. And we also have Genomes Unzipped Setting the record straight on Genetic Heterogeneity of Human Disease. So head over there and keep the conversation going. As for me, I'm still 8 comments behind...

Reference

Verweij KJ, Zietsch BP, Medland SE, Gordon SD, Benyamin B, Nyholt DR, McEvoy BP, Sullivan PF, Heath AC, Madden PA, Henders AK, Montgomery GW, Martin NG, Wray NR. (2010). A genome-wide association study of Cloninger's Temperament scales: Implications for the evolutionary genetics of personality. Biol Psychol. Aug 3. [Epub ahead of print].

Subscribe to Post Comments [Atom]

2 Comments:

At August 16, 2010 7:44 PM, Anonymous Anonymous said...

The paper does not suggest that personality variation between individuals is not influenced by their genetic makeup. That personality variation is substantially heritable has been demonstrated in countless studies using twin, family, and adoption designs. What the paper says is that the genetic influence on personality must consist of the cumulative effect of many genes or SNPs of individually very small effect - this is because our study, as well as other large-scale genome-wide association and linkage studies on other personality traits (e.g. Big Five, Eysenck’s), have failed to find any individual genetic variants with large or moderate effects on personality. Lessons from GWAS of, for example, human height (eg Visscher 2008 Sizing up human height, Nature Genetics 40: 489-490) and lipids (Teslovich et al. 2010 Biological, clinical and population relevance of 95 loci for blood lipids, Nature 466: 707-13) suggest that we will start to find associated variants as sample sizes increase.

To address a couple of erroneous criticisms from Prof Lee M. Silver, highlighted by Neurocritic:

- As we clearly describe, our analysis takes into account relatedness between family members, including monozygotic twins. Family-based genome wide association analysis is very standard; see for example the review of Benyamin et al. (2009, Family-based genome-wide association studies, Pharmacogenomics, 10 (2): 181).

- The sample includes a relatively small subsample that was based on their high vs low scores on depression symptoms (they are not diagnosed with bipolar disorder, as claimed). We took extra care before including these individuals, given the potential for problematic stratification. Firstly, we tested for mean differences on the personality scales between that selected subsample and the rest of the sample - there were no differences. Secondly, we compared the GWAS results with and without inclusion of these individuals, and there were no substantive differences. Therefore we included them in the analyses, increasing the sample size and statistical power to find any individual SNP effects.

Kind regards, Karin Verweij and Brendan Zietsch

 
At August 17, 2010 1:03 AM, Blogger The Neurocritic said...

Thanks very much for stopping by to comment. I amended the main text of the post to provide a summary and link. I'd also like to commend you for your paper, which has sparked a lot of interest and discussion.

 

Post a Comment

<< Home

eXTReMe Tracker