Content uploaded by Emil O. W. Kirkegaard
Author content
All content in this area was uploaded by Emil O. W. Kirkegaard on Jan 06, 2020
Content may be subject to copyright.
READ REVIEWS
WRITE A REVIEW
CORRESPONDENCE:
pifferdavide@gmail.com
DATE RECEIVED:
June 10, 2015
KEYWORDS:
intelligence, adhd, polygenic
selection
,
© Piffer et al. This article is
distributed under the terms of
the Creative Commons
Attribution 4.0 International
License, which permits
unrestricted use, distribution,
and redistribution in any
medium, provided that the
original author and source are
INTRODUCTION
Attention deficit hyperactivity disorder (ADHD) is a psychiatric disorder characterized by having trouble
concentrating and impulsive behavior. Prevalence depends on specific criteria used (e.g. self-report
vs. clinical diagnosis) but is around 5-7% (Willcutt, 2012). Genetically informative designs find high
heritabilities of ADHD. One review of 20 twin studies found a mean heritability of 76% (Faraone et al,
2005). ADHD is related to many other psychiatric problems such as bipolar disorder and drug
dependency (Kessler et al, 2006). More importantly, ADHD is negatively related to general intelligence
(GI; Kuntsi et al, 2003) and learning ability (Mayes et al, 2000). The correlation with GI is about -.3.
This may seem small, but in practice it means that the ADHD group has an IQ about 9 points below
the non-ADHD group (see Frazier 2004 for a meta-analysis of ADHD x IQ correlations). Moreover,
Kunstsi et al. (2003) found that the genetic overlap between ADHD and GI was 100%; the same genes
have an influence on both traits.
Martin et al. (2014) found that a genetic composite risk score, based on a case-control genome-wide
association study (GWAS) for clinical ADHD (Stergiakouli et al., 2012), was independently associated
with lower IQ. However, the association was rather weak (beta = -.05). On the other hand, the
discovery sample was quite small for a GWAS (700 with ADHD, 5100 controls), so the association will
BIOLOGICAL SCIENCES
Strong negative relationship between population-level
general intelligence and ADHD genetic factors inferred
from allele frequencies
DAVIDE PIFFER , EMIL KIRKEGAARD
ABSTRACT
ADHD and general intelligence are negatively correlated (within populations) and this correlation is driven by common genetic
variants shared between the two phenotypes. This paper analyzes the population frequency patterns of alleles associated with
ADHD and intelligence in two samples of 26 and 50 populations (1000 genomes and ALFRED). Factor analysis of allele frequencies
was used to estimate the strength of natural selection on the two traits. The two factors, indicating selection for general intelligence
and ADHD, show strong negative correlations in both 1000 Genomes and ALFRED samples (r= -0.93 and -0.90, respectively).
Alleles with lower p-values would be less likely to be false positives, so the more significant ADHD GWAS hits are expected to be
more strongly negatively correlated with the general intelligence SNP and the ADHD SNP factors, which were also found (r=-0.26
and 0.37, respectively).
The ADHD factor predicted national IQs also after accounting for a measure of population structure (Fst).
Results are interpreted in a framework based on evolutionary convergent selection pressure for higher general intelligence and lower
ADHD.
✎
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 1
credited. likely be larger in a better powered GWAS.
Molecular genetic studies of height, general intelligence and educational attainment
Over the last few years, researchers have started moving away from the study of genetic evolution
using a single-gene, Mendelian approach towards models that examine many genes together
(polygenic). The more genes are involved in a given phenotype, the more the signal of natural
selection will be “diluted” across different genomic regions (because each gene accounts for a tiny
effect) making it difficult to detect it using approaches focused on a single gene (Pritchard et al., 2010;
Piffer, 2014; Davies et al, 2011; Groen-Blokhuis et al, 2014). A first attempt at empirically identifying
polygenic selection was made by Turchin et al. (2012) on two populations (Northern and Southern
Europeans), providing evidence for higher frequency of height-increasing alleles (obtained from
GWAS studies) among Northern Europeans. A drawback of that study was the reliance on populations
from a single continent and that crude pairwise comparisons (e.g. French vs. Italian) were used without
correlating frequency differences to average population height. Moreover, the strength of selection was
not determined.
Rietveld et al. (2013)’s meta-analysis found ten SNPs that increased educational attainment,
comprising three with nominal genome-wide significance and seven with suggestive significance. A
recent study has replicated the positive effect of these top three SNPs, rs9320913, rs11584700 and
rs4851266, on mathematics and reading performance in an independent sample of school children
(Ward et al., 2014).
Two different approaches to identify selection based on the correlation of allele frequencies across
different populations have been recently developed by Piffer (2013) and Berg & Coop (2014).
Piffer (2013) obtained two samples comprising 14 and 50 populations (1000 Genomes and ALFRED
databases, respectively) and applied principal components analysis to the frequencies of the ten
alleles reported in Rietveld et al. (2013). The alleles loaded highly and in the expected direction
(positively) on a single factor accounting for most of the variance. The factor scores were correlated to
indexes of country educational achievement (PISA) and IQ, producing high correlations (r’s around
0.9). This factor was interpreted as indicating the strength of polygenic selection. This was the first time
that genetic frequencies had been used from a cross-racial sample and an estimate of selection
strength was provided, thus correlating it with measured average phenotypic scores.
The genetic correlation between ADHD and GI within a population makes it plausible that the same
pattern will be found across populations (Jensen, 1998, “default hypothesis”). Based on the above, we
formed the following hypotheses:
1. If the ADHD SNPs are true positives and have been selected for or against, then there should be a
general ADHD SNP selection factor across populations.
2. There should be a negative correlation between the ADHD SNP general factor and measures of
national GI, as well as with the previously reported GI SNP selection factor.
3. There should be a negative correlation between the p-values of ADHD SNPs and their relationships
with the GI/educational attainment SNP genetic factor and the ADHD SNP general factor because
SNPs with higher p-values are more likely to be false positives.
METHODS
IQ/educational attainment-increasing alleles: Piffer’s factor, extracted from 4 SNPs affecting general
intelligence, was used (Piffer, 2015a). The top three SNPs (not in LD), were obtained from the most
recent GWAS of cognitive function (Davies et al., 2015).This study focused on fluid intelligence.
ADHD risk alleles were obtained from the largest GWAS to this date, including 5,621 clinical patients
and 13,589 controls (Grohen-Blokuis et al., 2014).
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 2
National/Ethnic IQs were obtained from Lynn and Vanhanen (2012). The IQ for Tuscany was obtained
from Piffer & Lynn, 2014, as the IQ of Central Italy. There were 3 missing cases (Chinese Dai, Gujarati
Indian, Indian Telegu). See suppl. material (“gadhdfactors” spreadsheet).
As previously done, we conducted the analysis on two independent databases of population genetic
data: 1000 Genomes (http://browser.1000genomes.org/index.html), comprising 26 populations, and
ALFRED(.http://alfred.med.yale.edu/), comprising 50 populations after list-wise deletion of missing
cases.
RESULTS (STUDY 1 AND 2)
Since the analyses were generally identical in both databases, we go through the first in detail and
mostly summarize the results for the second.
STUDY 1: 1000 Genomes (phase 3)
Factor analysis of GI-increasing alleles including the latest three GWAS hits (Davies et al.,
2015).
Factor analysis was on the three new hits and the four SNPs (Piffer, 2015a) minus one SNP
(rs9320913) which was in strong LD with one of the three new hits (rs10457441).
Table 1. Factor loadings of 3 “new” SNPs
Chr. # Gen. region SNP Factor loading
6 98678841 rs10457441.C 0.83
14 32372633 rs17522122.G -0.76
19 50098513 rs10119.G 0.39
Table 2. Factor loadings of 4 SNPs minus overlapping one.
Chr. # Gen. region SNP Factor loading
1 204576983 rs11584700,G 0.83
2 100818479 rs4851266.T 0.91
1 94059554 rs236330.C 0.83
The two factors were strongly correlated with each other (r=0.79, N=26, p<0.0001).
A factor analysis was then carried out on the 6 SNPs together. 5 of 6 had strongly positive loadings,
but one SNP (rs17522122) loaded in the wrong direction (-0.55).
Table 3. Factor loadings of 6 GI SNPs.
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 3
Chr. # Gen. region SNP Factor loading
6 98678841 rs10457441.C 0.69
14 32372633 rs17522122.G -0.55
19 50098513 rs10119.G 0.76
1 204576983 rs11584700.G 0.82
2 100818479 rs4851266.T 0.92
1 94059554 rs236330.C 0.85
The correlation between the 6 SNPs GI factor and national IQs was r=0.87 (N=23, p<0.0001). Factor
scores are reported in Table 4.
Table 4. Factor scores for two GI SNP sets and total set of 6 SNPs.
3 “new” snps 3 old snps (4 minus 1 in LD) 6 SNPs IQ
ACB -1.20 -1.22 -1.28 83
ASW -1.10 -1.23 -1.24 85
ESN -1.21 -1.43 -1.47 71
GWD -1.03 -1.38 -1.41 62
LWK -1.71 -1.47 -1.77 74
MSL -0.38 -1.29 -1.16 64
YRI -1.00 -1.46 -1.47 71
CLM -0.08 -0.25 -0.26 83.5
MXL -0.30 0.05 0.06 88
PEL -0.86 0.08 0.12 85
PUR 0.73 -0.17 0.04 83.5
CDX 0.94 1.06 1.14 N/A
CHB 0.18 1.31 1.22 105
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 4
CHS 0.27 1.24 1.22 105
JPT 0.01 1.22 1.07 105
KHV 0.63 1.45 1.40 99.4
CEU 1.12 0.52 0.51 99
FIN 1.27 0.47 0.56 101
GBR 0.88 0.60 0.55 100
IBS 1.39 0.33 0.46 97
TSI 1.41 0.29 0.52 99
BEB 0.24 -0.09 0.08 81
GIH 0.25 0.56 0.51 N/A
ITU -0.20 0.31 0.29 N/A
PJL -0.18 0.34 0.21 84
STU -0.07 0.14 0.08 79
Factor analysis of ADHD risk alleles (1000 Genomes)
Analysis including linked loci
Since the probability that an SNP is a true positive depends on its p-value, only alleles with a p-
value<5*e-5 were included. This is already a very liberal threshold, because the conventional threshold
to be considered significant is 5*10-8 after correction for multiple testing issues (Clarke et al., 2011).
However, the SNPs with the lowest p-value in this study were only <2*e-6, so we chose as a-priori
convention to include all SNPs with a p-value of an order of magnitude lower. This resulted in a set of
42 SNPs. For allele frequencies, see the supplementary material. The set of SNPs (N=42) included
many hits within the same genomic region (500Kb). To avoid redundancy, linked loci were not counted
more than once and, in order to reduce noise, if two or more SNPs were in linkage, only the one with
the best p-value was included. Since having a case-to-variable ratio of at least 2:1 is recommended for
factor analysis (Zhao, 2009) and there were a total of 17 SNPs, we created two sets of 9 and 8 SNPs,
ordered according to their p-value (i.e. the first set comprising the half with the lower p-values).
A factor analysis using minimum residuals was carried out and factor scores saved with the Thurstone
method. Other extraction methods were not used as they produced nearly identical solutions. The
same factor analytic procedure was used throughout this paper.
Most of the alleles’ loadings (11/17) were in the right direction (positive), p=0.166. These are reported
in Tables 5 and 6.
The two factors (obtained from the two separate and unlinked SNP sets) were highly correlated with
each other (r=0.96) and with the genetic population GI SNP factor (from the 4 alleles) (r=-0.90 and -
0.89, respectively).
Table 5. First set of ADHD risk SNPs.
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 5
Chr. # Gen. region SNP Factor loading
18 64443239 rs17232800.C -0.76
5 78146288 rs6453417.T 0.93
19 8675892 rs301407.T 0.45
11 120025839 rs4245040.G 0.74
2 232026998 rs16828074.G 0.89
14 96126480 rs7160641.T 0.95
4 97191397 rs7674977.G -0.59
20 6624355 rs6038589.C 0.97
8 24338666 rs7012077. -0.26
Table 6. Second set of ADHD risk SNPs.
Chr. # Gen. region SNP Factor loading
9 133903199 rs10512416.A 0.7
1 203260558 rs2802837.C 1
7 85478161 rs17160554.C 0.52
6 40761488 rs9471400.C 0.67
3 62876746 rs4635724.G -0.9
17 85123530 rs17712565.G -0.73
12 33865802 rs12582885.T 0.33
22 36781026 rs8142185.C -0.21
Factor scores for GI and ADHD alleles are reported in Table 7.
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 6
Table 7. Factor scores: GI-enhancing and ADHD risk alleles in 1000 Genomes.
Population ADHD SNP
second order
factor
ADHD unlinked
first set, SNP
factor
ADHD unlinked
second set, SNP
factor
ADHD
Composite
GI 4
SNP
factor
GI 6
SNP
factor
Afr.Car.Barbados 1.53 1.17 1.43 1.30 -1.26 -1.28
US Blacks 1.21 1.12 1.05 1.09 -1.21 -1.24
Esan Nigeria 1.93 1.69 1.65 1.67 -1.45 -1.47
Gambian 1.47 1.48 1.53 1.50 -1.45 -1.41
Luhya Kenya 1.50 1.47 1.63 1.55 -1.54 -1.77
Mende Sierra
Leo
1.71 1.64 1.65 1.65 -1.24 -1.16
Yoruba 1.66 1.66 1.70 1.68 -1.46 -1.47
Colombian -0.32 -0.78 -0.46 -0.62 -0.12 -0.26
Mexican LA -0.42 -0.58 -0.57 -0.58 0.02 0.06
Peruvian -0.46 -0.69 -1.07 -0.88 -0.30 0.12
Puerto Rican -0.32 -0.41 -0.17 -0.29 0.01 0.04
Chinese Dai -0.82 -1.05 -0.96 -1.00 1.18 1.14
Han Chin. Bejing -0.65 -1.06 -0.80 -0.93 1.40 1.22
Han Chin. South -0.83 -0.80 -0.97 -0.89 1.30 1.22
Japanese -0.25 -1.32 -0.90 -1.11 1.23 1.07
Vietnam -0.77 -0.91 -0.83 -0.87 1.60 1.40
UtahWhites -0.57 -0.75 -0.77 -0.76 0.76 0.51
Finns -0.63 -0.30 -0.98 -0.64 0.71 0.56
British -0.66 -0.46 -0.58 -0.52 0.85 0.55
Spanish -0.55 -0.75 -0.68 -0.71 0.60 0.46
Tuscan (Italy) -0.66 -0.70 -0.44 -0.57 0.57 0.52
Bengali
Bangladesh
-0.68 -0.06 -0.22 -0.14 -0.26 0.08
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 7
Gujarati Ind. Tx -0.51 0.22 -0.24 -0.01 0.47 0.51
Indian Telegu UK -0.71 0.18 -0.06 0.06 0.02 0.29
Punjabi Pakistan -0.54 -0.11 0.23 0.06 0.19 0.21
Sri LankanUK -0.67 0.11 -0.18 -0.03 -0.61 0.08
Given the high correlation between the factors extracted from the two sets of unlinked alleles, a
composite factor was obtained as the mean of the two vectors, to make the analysis more
parsimonious and less erroneous.
Figures 1(a,b) shows a scatterplot with the factor scores for the ADHD (composite) and GI SNP
factors. The correlations between the two variables were r= -0.91 and -0.93 for the 4 and 6 SNPs
factors, respectively.
Figure 1a. Regression of ADHD SNP factor on the 4 SNPs GI SNP factor.
Figure 1b. Regression of ADHD SNP factor on the 6 SNPs GI SNP factor.
Visual inspection of the plot revealed that the African groups were outliers. After removal of the African
groups, the correlations between the ADHD SNP composite factor and the 4 or 6 SNPs GI factors were
still negative and significant (respectively, r=-0.64 and r=0.61, N=19, p<0.05). This indicates that the
relationship also persists across the other continents.
Method of correlated vectors
The method of correlated vectors consists of correlating the factor loadings of indicator variables with
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 8
the correlations between each indicator variable and the criteria variable (see Kirkegaard, 2014). The
method was employed to test the hypothesis that SNPs with higher p-values would be more likely to be
false positives, hence if there has been natural selection, less selection signal would be detected. The
composite ADHD factor (Table 10, 5th col.) was a criterion variable and the unlinked SNPs were
correlated with it. Spearman rank-order correlation was slightly negative (in line with predictions) but
nonsignificant: r= -0.26 (N=17, p=0.31). Using all the 42 SNPs, the Spearman rank-order correlation
was r= 0.37 and significant (p= .015). This relationship is plotted in Figure 2.
Figure 2. Regression of ADHD risk alleles’ correlation with ADHD composite genetic factor on
alleles’ p-values.
Correlation of population genetic factors with aggregate phenotypic measures
A systematic review and meta-analysis concluded that “the large variability of ADHD/HD prevalence
rates worldwide resulted mainly from methodological differences across studies” (Polanczyk et al.,
2007). Since prevalence rates do not reflect an underlying phenotype, correlating them with genetic
scores would not be useful. However, since there is a genetic correlation between GI and ADHD, and
natural selection has stronger effects upon higher-order constructs with pervasive effects on life
outcomes and survival, such as GI (Jensen, 1998; Gottfredson, 1997; Gordon, 1997; Gottfredson,
2004) we expect to find a correlation between population-level ADHD genetic scores and aggregate
measures of cognitive capacity.
The correlation between country/ethnic phenotypic IQ and the ADHD genetic factor was highly
negative (r=-0.91, N= 23, p= <.001).
STUDY 2 (ALFRED dataset)
The 4 SNPs influencing general intelligence used by Piffer (2015a) were searched on ALFRED. When
a SNP was not found in ALFRED, the SNP in close linkage disequilibrium (r>0.8) was used. Linkage
calculator (http://www.broadinstitute.org/mpg/snap/) based on 1000 Genomes phase 1, CEU data.
Corresponding SNPs are in brackets in the spreadsheet file.
None of the 3 new SNPs was found in ALFRED with this method, apart from the one loading in the
wrong direction (rs17522122) in the 1000 Genomes analysis. Thus, analysis of ALFRED was limited to
the 4 SNPs.
Table 8 shows the factor loadings of the 4 GI-related SNPs in ALFRED.
Table 8. Factor loadings of 4 GI-related SNPs in ALFRED.
SNP Factor loading
rs9320913 A 0.43
rs11584700 G 0.70
rs4851266 T 0.84
rs236330 C 0.72
Factor analysis of ADHD risk alleles
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 9
In total, 13 (out of the 42) SNPs were found in ALFRED (including those in close LD). Results of factor
analysis are reported in Table 9.
Table 9. Factor loadings of 13 ADHD SNPs in ALFRED.
SNP Factor Loading
rs507533.G -0.22
rs6453417.T 0.89
rs7160641.T 0.81
rs4936536.G -0.31
rs10026084.A 0.63
rs4673145.A -0.24
rs2206922.T 0.72
rs1982863.T -0.48
rs2841633.T 0.72
rs12668989.T 0.32
rs4635724.G -0.79
rs4580847.G 0
rs3026685.T -0.40
The factor was negatively correlated to the GI SNP factor (r=-0.81, N=50).
Since the method of correlated vectors showed an abundance of noise in the SNPs with lower
significance, another factor analysis was carried out including only the SNPs with the p-value in the top
half of the total 13 (N=6). Results are reported in Table 10.
Table 10. Top 6 ADHD SNPs loadings.
SNP Factor Loading
rs507533.G -0.28
rs6453417.T 0.90
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 10
rs7160641.T 0.75
rs4936536.G -0.30
rs10026084.A 0.65
rs4673145.A -0.28
In accord with the hypothesis of stronger selection signal among the genuine hits, the correlation
between the extracted factor and the GI SNP factor slightly improved over the all-inclusive factor
analysis (r= -0.9 vs -0.81). Factor scores are reported in Table 11.
Visual inspection of the regression plot suggests that the distribution of scores is not driven by outliers
as in the 1000 Genomes database but is fairly uniform across continental clusters and populations
(Figure 3).
Figure 3. Relationship between 4 GI SNP factor and ADHD (top 6) SNP risk factor, by ethnic
group.
Table 11. Factor scores for ADHD risk and GI-related alleles in ALFRED.
Continent Population ADHD factor ADHD top 6 4 GI factor
Africa Bantu 1.88 1.66 -2.13
Africa San 2.52 2.58 -1.85
Africa Biaka 1.91 1.83 -1.48
Africa Mbuti 1.90 1.76 -1.68
Africa Yoruba 1.86 1.35 -2.07
Africa Mandenka 1.66 1.14 -1.70
Africa Mozabite 0.49 0.51 -0.67
Middle East Bedouin 0.15 -0.12 -0.43
Middle East Druze 0.01 0.31 0.10
Middle East Palestinian 0.47 0.36 -0.14
Europe Adygey -0.01 0.08 0.37
Europe Basque -0.06 0.16 -0.27
Europe French 0.02 0.08 0.06
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 11
Europe Italians -0.17 -0.14 0.23
Europe Orcadian 0.06 0.00 0.63
Europe Russian 0.32 0.43 -0.32
Europe Sardinian 0.20 0.64 -0.36
C Asia Burusho -0.05 -0.24 0.06
C Asia Kalash 0.36 0.50 0.33
C Asia Pashtun 0.49 0.55 -0.36
C Asia Balochi 0.37 0.30 0.29
C Asia Brahui 0.26 0.26 -0.17
C Asia Hazara -0.42 -0.15 0.64
C Asia Sindhi 0.80 0.79 -0.35
E Asia Mongolia -0.90 -0.81 1.27
E Asia Dai -0.29 -0.30 0.65
E Asia Daur -1.21 -1.21 1.15
E Asia Han -0.48 -0.24 0.87
E Asia Hezhe -1.29 -1.80 1.18
E Asia Japanese -1.01 -0.96 0.59
E Asia Koreans -0.89 -0.80 0.93
E Asia Lahu -0.67 -0.81 0.71
E Asia Miao -0.73 -1.19 0.62
E Asia Naxi -1.07 -0.92 0.05
E Asia Oroquen -0.68 -0.68 0.75
E Asia She -1.07 -1.27 0.62
E Asia Tu -0.50 -0.14 0.61
E Asia Tujia -0.64 -0.61 1.42
E Asia Uyghur -0.38 -0.33 0.66
E Asia Xibe -1.04 -1.13 1.04
E Asia Yi -0.76 -0.45 0.98
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 12
SE Asia Cambodians -0.84 -0.24 0.13
Oceania Papuan N.G 1.12 1.06 -1.16
Melanesian Melanesian 1.74 1.77 0.26
Siberia Yakut -0.84 -0.95 0.54
America Pima,Mx -0.21 -0.21 -0.25
America Maya,Yuc -0.79 -0.71 -0.70
America Amerindians -0.29 -0.18 -1.33
America Karitiana -0.70 -0.88 0.07
America Surui -0.60 -0.67 -0.40
Controlling for migration and drift
Piffer (2015c) developed a method to assess the impact of population structure on estimates of
selection inferred from allele frequency patterns.
The correlations between IQ distances and Fst, ADHD factor distances were, respectively: r= 0.588,
0.63.
Accordingly, Fst distances published by Piffer (2015c) were used in a multiple linear regression with
ADHD factor distances to predict IQ distances between populations. After list-wise deletion there were
253 caes (NA=72). Standardized Betas were 0.168 and 0.483 for Fst and ADHD factor, respectively.
4 and 6 SNPs g factor distances were also employed as dependent variables to assess the predictive
power of ADHD distances net of population structure. Fst and ADHD had similar Betas (0.409,0.443),
(0.5,0.429) for the 4 and 6 SNPs factors, respectively.
DISCUSSION
Factor analysis of allele frequencies obtained from GWAS hits by independent studies on different but
partly overlapping phenotypes (ADHD and GI) shows that they follow an inverse spatial distribution.
Robustness (reliability) of the findings was provided by the internal consistency of the measures: factor
analysis of two sets of unliked ADHD risk alleles (using 1000 Genomes) located on different
chromosomes yielded two very similar factors (r= 0.96). Moreover, most of the factor loadings (11/17)
were positive (in the right direction), (p= 0.17). Rietveld’s and Davies’ three top hits were related to two
distinct phenotypes (educational attainment and fluid intelligence, respectively), yet the two sets of
alleles produced similar frequency patterns (r=0.79). Moreover, the effect on educational attainment of
a genomic region on chromosome 6 was replicated by Davies for fluid intelligence.
Validity was confirmed by the strong negative correlation of the factor of ADHD SNPs with a factor
extracted from an independent set of 6 GI SNPs produced by several studies, with frequencies
obtained from the ALFRED and 1000 Genomes datasets, comprising 50 and 26 populations
respectively. GWAS hits with a lower p-value had a stronger relationship with the ADHD SNP factor,
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 13
confirming the hypothesis that a stronger signal of selection would be found in alleles that are likely to
be true positives. Moreover, the factor was negatively related to average phenotypic country/ethnic GI.
The 6 GI-related SNPs were not in linkage disequilibrium (LD) with the 17 unlinked ADHD alleles, so
the correlation between the two factors is not confounded by LD. However, two of them (GI-related
rs11584700 and ADHD risk rs2802837) were in nearby genomic regions (around 1.2 Mbp apart), on
chromosome 1.
Another noteworthy finding is that groups which are the most distant genetically (Africans and
Oceanians) appear to have the most similar genetic scores for cognitive traits, whereas genetically
similar groups (East Asians and Native Americans) differ by almost a standard deviation (Tables 15
and 16). This fact is hard to square with an explanation of the results in terms of genetic drift and
instead favours a model of differential selection pressures faced by isolated populations living in
different environments.
We controlled for the effect of population structure using the method outlined by Piffer(2015c). This
showed that although the ADHD factor was a stronger predictor of IQ distances between populations
than Fst, as indicated by Fst, it lost part of its predictive power (Beta= 0.483). Moreover, it did not have
more predictive power of GI factor distances than Fst. These results together are weaker than what
found by Piffer for the 4 SNPs g factor(2015c), which retained all its predictive power on IQ distances
when regressed with Fst.
We see at least two possible explanations for the general results: 1) The SNPs found by the ADHD
GWAS are in reality GI-increasing alleles and have no specific effect on ADHD. 2) ADHD and GI have
undergone opposite selection pressure, that is ancestral environments that selected for higher GI also
selected against behavioral predispositions to ADHD, such as inattention and impulsivity. In such a
hypothetical scenario, harsher environments could have placed more demands on focused attention
and problem solving skills that resulted in survival and fitness differentials among carriers of different
alleles.
Obviously, a third possible explanation is that the ADHD hits were not under selection and represent
population structure just as any other random set of SNPs. The not entirely satisfying results of
regressing these with Fst distances does not allow us to rule out this possibility.
Another limitation of this study is that ADHD prevalence rates are too affected by different diagnostic
criteria to provide a universal measure that can be used for a cross-cultural comparison.
Future research
Future studies should replicate the present analysis when new GI and ADHD SNPs are reported.
REFERENCES
Benyamin, B., Pourcain, B.St., Davis, O.S., Davies, G., Hansell, N.K., Brion, M.-J.A. et al. (2013).
Childhood intelligence is heritable, highly polygenic and associated with FNBP1L. Molecular
Psychiatry, 19: 253-258.
Berg, J.J. & Coop, G. (2014). A Population Genetic Signal of Polygenic Adaptation. PLOS Genetics,
Doi: 10.137/journal.pgen.1004412
Clarke G.M., Anderson .CA., Pettersson F.H., Cardon L.R., Morris A.P., Zondervan K.T. (2011). Basic
statistical analysis in genetic case-control studies. Nat Protoc 6: 121–33. doi:10.1038/nprot.2010.182
Davies, G., Tenesa, A., Payton, A., Yang, J., Harris, S.E., Liewald, D., Xiayi, K., Le Hellard, S. et al.
(2011). Genome-wide association studies establish that human intelligence is highly heritable and
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 14
polygenic. Molecular Psychiatry, 16: 996-1005.
Davies, G., Armstrong, N.,... Deary, I.J. (2015). Genetic contributions to variation in general cognitive
function: a meta-analysis of genome-wide association studies in the CHARGE consortium (N=53949).
Molecular Psychiatry, 1-10.doi:10.1038/mp.2014.188
Faraone, S. V., Perlis, R. H., Doyle, A. E., Smoller, J. W., Goralnick, J. J., Holmgren, M. A., & Sklar, P.
(2005). Molecular genetics of attention-deficit/hyperactivity disorder. Biological psychiatry, 57(11),
1313-1323.
Frazier, T. W., Demaree, H. A., & Youngstrom, E. A. (2004). Meta-analysis of intellectual and
neuropsychological test performance in attention-deficit/hyperactivity disorder. Neuropsychology,
18(3), 543.
Gordon, R. A. (1997). Everyday life as an intelligence test: Effects of intelligence and intelligence
context. Intelligence, 24(1), 203-320.
Gottfredson, L. S. (2004). Life, death, and intelligence. Journal of Cognitive Education and Psychology,
4(1), 23-46.
Gottfredson, L. S. (1997). Why g matters: The complexity of everyday life. Intelligence, 24(1), 79-132.
Groen-Blokhuis, M. M., Middeldorp, C. M., Kan, K. J., Abdellaoui, A., van Beijsterveldt, C. E., Ehli, E.
A., ... & Psychiatric Genomics Consortium ADHD Working Group. (2014). Attention-
Deficit/Hyperactivity Disorder Polygenic Risk Scores Predict Attention Problems in a Population-Based
Sample of Children. Journal of the American Academy of Child & Adolescent Psychiatry, 53(10), 1123-
1129.
Jensen, A. R. (1998). The g factor: The science of mental ability. Westport, CT: Praeger.
Kessler, R., Adler, L., Barkley, R., Biederman, J., Conners, C., Demler, O., ... & Zaslavsky, A. (2006).
The prevalence and correlates of adult ADHD in the United States: results from the National
Comorbidity Survey Replication. American Journal of Psychiatry, 163(4), 716-723.
Kirkegaard, E. O. W. (2014). Crime, income, educational attainment and employment among
immigrant groups in Norway and Finland. Open Differential Psychology.
Kuntsi, J., Eley, T. C., Taylor, A., Hughes, C., Asherson, P., Caspi, A., & Moffitt, T. E. (2004).
Co‐occurrence of ADHD and low IQ has genetic origins. American Journal of Medical Genetics Part B:
Neuropsychiatric Genetics, 124(1), 41-47.
Martin, J., Hamshere, M.L., Stergiakouli, E., O’Donovan, M.C., Thapar, A. (2014). Neurocognitive
abilities in the general population and composite genetic risk scores for attention-deficit hyperactivity
disorder. Journal of Child Psychology and Psychiatry, doi:10.1111/jcpp.12336
Mayes, S. D., Calhoun, S. L., & Crowell, E. W. (2000). Learning disabilities and ADHD overlapping
spectrum disorders. Journal of learning disabilities, 33(5), 417-424.
Lynn, R., & Vanhanen, T. (2012). Intelligence: A unifying construct for the social sciences. London:
Ulster Institute for Social Research.
Piffer, D. (2013). Factor Analysis of Population Allele Frequencies as a Simple, Novel Method of
Detecting Signals of Recent Polygenic Selection: The Example of Educational Attainment and IQ.
Mankind Quarterly, 54: 168200.
Piffer, D. (2014). Simple statistical tools to detect signals of recent polygenic selection. IBC 2014;6:1,
1-6 • DOI: 10.4051 / ibc.2014.6.1.0001
Piffer, D. & Lynn, R. (2014). New evidence for differences in fluid intelligence between North and South
Italy and against school resources as an explanation for the north-south IQ differential. Intelligence, 46:
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 15
246-249.
Piffer, D. (2015a). Estimating the genotypic intelligence of populations and assessing the impact of
socioeconomic factors and migrations., The Winnower 2:e142299.93508 (2015).
DOI:10.15200/winn.142299.93508
Piffer, D. (2015b). Evidence of polygenic selection on human stature inferred from spatial distribution of
allele frequencies [v1; ref status: awaiting peer review, http://f1000r.es/4ye] F1000Research 2015, 4:15
(doi:10.12688/f1000research.6002.1)
Piffer, D. (2015c). Intelligence GWAS hits: Selection signal or population structure? A test of the null
hypothesis, The Winnower2:e142593.30710 (2015). DOI:10.15200/winn.142593.30710
Polanczyk, G., Silva de Lima, M., Horta, B.L., Biederman, J., Rohde, L.A. (2007). The Worldwide
Prevalence of ADHD: A Systematic Review and Metaregression Analysis. American Journal of
Psychiatry, 164: 942-948.
Rietveld, C. A., Medland, S. E., Derringer, J., Yang, J., Esko, T., Martin, N. W., Westra, H. J.,
Shakhbazov, K., Abdellaoui, A., Agrawal, A., et al. (2013). GWAS of 126,559 individuals identifies
genetic variants associated with educational attainment. Science 340, 1467-1471.
Stergiakouli, E., Hamshere, M., Holmans, P., Langley, K.,Zaharieva, I., Hawi, Z., ... & Thapar, A.
(2012). Investigatingn the contribution of common genetic variants to the risk and pathogenesis of
ADHD. American Journal of Psychiatry, 169,186–194.
Turchin, M. C., Chiang, C. W., Palmer, C. D., Sankararaman, S., Reich, D., Genetic Investigation of, A.
T. C., and Hirschhorn, J. N. (2012). Evidence of widespread selection on standing variation in Europe
at height-associated SNPs. Nat Genet 44, 1015-1019.
Ward, M.E., McMahon, G., St Pourcain, B., Evans, D.M., Rietveld, C.A., et al. (2014) Genetic Variation
Associated with Differential Educational Attainment in Adults Has Anticipated Associations with School
Performance in Children. PLoS ONE 9(7): e100248. doi:10.1371/journal.pone.0100248
Willcutt, E. G. (2012). The prevalence of DSM-IV attention-deficit/hyperactivity disorder: a meta-
analytic review. Neurotherapeutics, 9(3), 490-499.
Zhao, N. (2009). The minimum sample size in factor analysis. URL
https://www.encorewiki.org/display/~nzhao/The+Minimum+Sample+Size+in+Factor+Analysis
accessed 19th Dec. 2014.
STRONG NEGATIVE RELATIONSHIP BETWEEN POPULATION-LEVEL GENERAL INTELLIGENCE AND
ADHD GENETIC FACTORS INFERRED FROM ALLELE FREQUENCIES : BIOLOGICAL SCIENCES
PIFFER & KIRKEGAARD The Winnower MARCH 11 2015 16
