Archival Report Biological

Psychiatry

Polygenic Risk Scores, School Achievement,

and Risk for Schizophrenia: A Danish
Population-Based Study
Holger J. Sørensen, Jean-Christophe Debost, Esben Agerbo, Michael E. Benros,
John J. McGrath, Preben Bo Mortensen, Anne Ranning, Carsten Hjorthøj, Ole Mors,
Merete Nordentoft, and Liselotte Petersen

ABSTRACT
BACKGROUND: Studies have suggested that poor school achievement is associated with increased risk of
schizophrenia; however, the possible genetic contribution to this association is unknown. We investigated the
possible effect of the polygenic risk score (PRS) for schizophrenia (PRSSCZ) and for educational attainment (PRSEDU)
on the association between school performance and later schizophrenia.
METHODS: We conducted a case-cohort study on a Danish population-based sample born from 1987 to 1995
comprising 1470 individuals with schizophrenia and 7318 subcohort noncases. Genome-wide data, school
performance, and family psychiatric and socioeconomic background information were obtained from national
registers and neonatal biobanks. PRSSCZ and PRSEDU were calculated using discovery effect size estimates from
a meta-analysis of 34,600 cases and 45,968 controls and 293,723 individuals.
RESULTS: Higher PRSSCZ increased the risk (incidence rate ratio [IRR]: 1.28; 95% confidence interval [CI], 1.19–1.36),
whereas higher PRSEDU decreased the risk of schizophrenia (IRR, 0.87; 95% CI, 0.82–0.92) per standard deviation.
Not completing primary school and receiving low school marks were associated with increased risk of
schizophrenia (IRR, 2.92; 95% CI, 2.37–3.60; and IRR, 1.58; 95% CI, 1.27–1.97, respectively), which was not
confounded by PRSSCZ or PRSEDU. Adjusting for social factors and parental psychiatric history, effects of not
completing primary school and receiving low school marks were attenuated by up to 25% (IRR, 2.19; 95% CI,
1.75–2.73; and IRR, 1.39; 95% CI, 1.11–1.75, respectively). Increasing PRSEDU correlated with better school
performance (p , .01; R2 = 7.6%). PRSSCZ and PRSEDU was significantly negatively correlated (r = 2.31, p , .01).
CONCLUSIONS: The current PRS did not account for the observed association between primary school performance
and risk of schizophrenia.
Keywords: Epidemiology, Genetics, Polygenic risk score, Register-based, Schizophrenia, School achievement
https://doi.org/10.1016/j.biopsych.2018.04.012

Cognitive impairments in schizophrenia often manifest pre-
morbidly during childhood and adolescence (1), with the
largest effect observed for IQ (2). School achievement, which is
correlated with IQ (3), has less consistently been linked with
the risk of later development of schizophrenia (4,5), but recent
evidence from large-scale epidemiological studies does sug-
gest an association between poorer school performance and
higher risk of schizophrenia later in life (6,7). There is still limited
understanding of the possible genetic underpinnings of the
relationship between IQ (or school achievement) in the general
population and schizophrenia (8). For the genetic relationship
between IQ and schizophrenia, a recent meta-analysis of
genome-wide association studies (GWASs) in 78,308 in-
dividuals suggests a moderate negative genetic correlation
between schizophrenia and intelligence (3). For educational
attainment, recent studies using GWAS results have shown a
small but positive genetic correlation between schizophrenia
and educational attainment (9,10).
A link between poorer school performance and higher risk of
developing schizophrenia (6,7) supports the possibility that
impaired neurodevelopment increases risk of schizophrenia, as
evidenced by several longitudinal studies showing premorbid IQ
deficit in individuals who subsequently developed schizophrenia
(11–15). The association between lower school performance and
risk of developing schizophrenia is not specific to schizophrenia,
as it has been observed for mood disorder as well (16,17). It is
currently not known in any great detail whether the association
between poorer school performance and risk of schizophrenia
could be partly due to genetic factors or is predominantly influ-
enced by environmental factors.
From the large GWASs of schizophrenia (18) and educa-
tional attainment (9), polygenic risk scores (PRSs) can be

ª 2018 Society of Biological Psychiatry. 1

ISSN: 0006-3223 Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
Biological
Psychiatry Polygenic Risk, School Achievement, and Risk for Schizophrenia
constructed from risk alleles at thousands of genetic loci. For
instance, a higher PRS for schizophrenia (PRSSCZ) predicts
schizophrenia in the general population (19,20), and a higher
PRS has also been associated with certain nonpsychotic
symptoms (21) as well as with lower performance IQ (8). It is
uncertain if the PRSSCZ is associated with primary school
performance and if this association is explained by PRSSCZ or
by other factors (e.g., genetic factors not assessed by PRSSCZ,
early life environmental factors that impact on cognitive
reserve, socioeconomic factors) (22). Differences in educa-
tional length are under substantial genetic influence, with an
estimated 15% single nucleotide polymorphism (SNP) herita-
bility in unrelated individuals (9). From GWAS of educational
attainment (9), a higher PRS for educational attainment
(PRSEDU) has been shown to correlate positively with school
performance (23) and longevity (24) but to be unrelated to risk
of depression (17). It may be important to examine the influ-
ence of PRSEDU on school outcomes and risk of developing
schizophrenia in the general population because PRSEDU may
pleiotropically influence the association between school per-
formance and later schizophrenia.
The aims of our study were to investigate the possible effect
of the PRSSCZ and PRSEDU on the association between school
performance (in ninth grade) and later schizophrenia. We
wanted to answer the following questions: Are PRSSCZ and
PRSEDU associated with the risk of schizophrenia? Is there a
correlation between PRSSCZ and school performance in cases
with schizophrenia and in noncases? Is there a correlation
between PRSEDU and school performance in cases with
schizophrenia and in noncases? Do PRSSCZ and PRSEDU
confound the association between adolescent school
achievement and risk of schizophrenia?
METHODS AND MATERIALS
Data Sources
This was an observational case-cohort study linking Danish
population-based registers using a unique personal identifi-
cation number. This identification number has been assigned
to all live-born children and new residents in Denmark since
1968 across all registration systems (26). The Danish Psychi-
atric Central Research Register has covered all psychiatric
inpatient facilities since 1969 and outpatient contacts since
1995 (26). Diagnostic information was based on the ICD-10
from 1994 (27). The Danish Civil Registration System (25) con-
tains information on dates of birth, death, immigration, emigra-
tions, and links to family members. The Integrated Database for
Longitudinal Labor Market Research covers the entire population
and contains yearly information from 1980 including income,
marital status, education, and birthplace (28). The Danish
Neonatal Screening Biobank stores dried blood spots taken after
birth from nearly all infants born in Denmark after 1981 (29).
Study Population
All singleton births in Denmark, born in the period from 1987 to
July 1, 1994, who were registered in the Danish Neonatal
Screening Biobank and who had a first diagnosis of schizo-
phrenia (ICD-10 code F20.x) after July 1 registered in the Danish
Psychiatric Central Register, the year they turned 18 years of age
2 Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
(between 2005 and 2012), were included as cases. We con-
ducted a case-cohort study in which subcohort (SC) members
were randomly selected among singletons born in Denmark by
Danish-born parents between 1987 and 1995. The case-cohort
sample used here was drawn from the larger iPSYCH sample,
which has been described in detail elsewhere (30). A proportion
of the schizophrenia cases included had been selected from two
prior Danish subsamples (Genomic Medicine for Schizophrenia
[GEMS] 1 and GEMS2), used in an earlier GWAS (30), but there
was no overlap between the former and latter case selection, as
we only used incident schizophrenia cases. Parents were
identified using the Civil Registration System (25).
Genetic Analysis
For all three samples (GEMS1, GEMS2, and iPSYCH), DNA
was extracted from the Danish Newborn Screening Biobank,
and whole-genome amplified (in triplicate using the Qiagen
REPLI-g mini kit [Qiagen, Hilden, Germany] and the three
separate reactions were pooled). The GEMS1 and GEMS2
samples were genotyped with Illumina Human 610-Quad
BeadChip array and Illumina Infinium CoreExome beadchip
(Illumina, San Diego, CA), respectively (details have been
published previously) (30). The iPSYCH samples were pro-
cessed at the Broad Institute (Boston, MA) using the Infinium
PsychChip v1.0 array (Illumina). For the iPSYCH sample gen-
otyping was conducted in 23 waves, and samples with overall
call rates ,95% failed sample quality control (further details
have been described elsewhere) (29).
PRS Estimation
PRSs were calculated for schizophrenia (PRSSCZ) and educa-
tional attainment (PRSEDU). To calculate PRSSCZ, association
results from the latest Psychiatric Genomics Consortium (PGC)
schizophrenia sample (18), except the Danish results were
included in a meta-analysis, which resulted in a discovery
sample of 34,600 cases and 45,968 controls. The PRS was
calculated using the summary statistics from the discovery
dataset (effect allele, size, and direction), after excluding un-
common SNPs (minor allele frequency ,2%), low-quality
variants (info score ,90%), and SNPs missing in .50% of
subsamples from the PGC, and only keeping the most signif-
icant marker in the extended major histocompatibility complex
region (chromosome 6: 25–34 Mbp). Data were clumped in
windows of 500 kbp, removing SNPs in linkage disequibrilium
(LD) (R2 ..1) with another more significant marker.
For educational attainment, the PRSEDU was based on the
largest GWAS with summary statistics on 293,723 individuals
(9). GWAS summary statistics for educational attainment were
obtained from the Social Science Genetic Association Con-
sortium (https://www.thessgac.org/data). We calculated
PRSEDU based on summary statistics from this discovery
dataset excluding rare SNPs (minor allele frequency ,5%), low-
quality imputed variants (info score ,90%), indels, ambiguous
markers (A/T and C/G), and SNPs in the extended major histo-
compatibility complex region (chromosome 6: 25–34 Mbp).
Data were clumped in windows of 500 kbp, discarding variants
in LD (R2 ..1) with another more significant marker.
PRSSCZ and PRSEDU were standardized in our study sample
separately for GEMS1, GEMS2, and iPSYCH using arithmetic

Polygenic Risk, School Achievement, and Risk for Schizophrenia
mean and standard deviation in controls only to account for
the differences in genotyping procedures (e.g., type of chip
used). Both scores were calculated based on permissive p
value thresholds ranging from p , 5 3 10–8 to p , 1. In this
study, we used a pT of .05, as this has been deemed optimal
for the discovery of signals in the training sample (31).
Genetic Correlation Between Schizophrenia and
Educational Attainment
The genetic correlation between educational attainment and
schizophrenia was calculated using LD score regression (32),
based on summary statistics of our sample and the one for
educational attainment (9). The schizophrenia sample used to
calculate correlation with educational attainment was slightly
different from the sample used in the main analyses in the
article. The schizophrenia sample only included waves with
more than 40 cases, and the number of controls was down-
sampled to four times the number of cases (to avoid domi-
nating genetic structure from the controls). The sample was
corrected for population structure by adjusting for the first two
principal components derived from principal component
analysis. Furthermore, variants in the sample were filtered so
they only included intersecting variants across all three sam-
ples (GEMS1, GEMS2, and iPSYCH) with an info score .0.9
and minor allele frequency .0.01.
Adolescence School Variables
Data on grade point averages and the failure to complete
primary education were retrieved from the Primary Education
Register. Primary education in Denmark lasts 10 years and is
comparable to elementary/middle school in North America and
primary/lower-secondary school in the United Kingdom. Stu-
dents are typically 16 years of age at graduation.
Exams are mandatory for students. Grades are based on
agreement on the child’s performance between the child’s
teacher and an external examiner. Children were graded on a
7-point scale according to absolute criteria, rather than being
compared with their peers. Information about grade point
average was obtained from Statistics Denmark and data on
primary education are available from June 2002 onward. A
priori, we defined individuals who had not completed ninth-
grade exams in both math and Danish by the 17 years of
age as noncompleters.
Here, we used the grade point average for each child for
written and oral math and Danish (excluding grades for order).
These two subjects were chosen because they were available
during the school years included in the study, and because
these tests are the most standardized ones. For analytic pur-
poses, we included two measures: completion of primary ed-
ucation as a binary measure (yes/no), and grade point average
(only available for those having completed).
Assessment of Socioeconomic Status
We generated a parental socioeconomic status score (0–6) for
the year before the proband’s birth as a sum of six previously
described risk factors in which 6 is lowest socioeconomic
status (10), consisting of father’s or mother’s gross income in
the lowest quintile, father or mother being unemployed or
Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
Biological
Psychiatry
otherwise outside of the labor market, and father’s or mother’s
highest educational level less than high school completion.
Assessment of Schizophrenia and Mental Disorders
in Probands and in Their Parents
The psychiatric register (26) was used to identify individuals
with schizophrenia (ICD-10 code F20.x) and extract explana-
tory variables indicating whether the patient’s parents had
been given a diagnosis according to the hierarchy: schizo-
phrenia or related (schizophrenia-like) psychosis (ICD-8 codes
295, 297, 298.39, and 301.83 [301.83 was included as
schizophrenia-like psychosis as in previous publications
(33,34)] and ICD-10 codes F20–F29), or any other psychiatric
disorder before the proband turned 10 years (assuming that
parental psychiatric history is most relevant before 10 years of
age) (35).
Statistical Analysis
Individuals were followed from their 18th birthday until first
diagnosis of schizophrenia, death, emigration, or end of study
(December 31, 2012), whichever came first. As some in-
dividuals may finish ninth grade 1 or 2 years later than the
average, we chose to start follow-up at 18 years of age instead
of 16 years of age. For analyses using primary education as an
exposure (and schizophrenia as outcome), we examined the
effect of not having completed primary education and of
school grades using the median quintile as the reference. For
the case-cohort sample, incidence rate ratios (IRRs) for
schizophrenia were estimated in a weighted Cox model. We
used the Lawless and Kalbfleisch estimator with robust stan-
dard errors, as this estimator has been deemed optimal (36).
The basic model was adjusted for sex, age, and year of birth
(1987–1988, 1989–1991, 1992–1994). We then adjusted for
PRS or for psychiatric family history and socioeconomic sta-
tus, ending up with the final model taking all effects into ac-
count. To control for population stratification, we restricted the
sample to offspring of parents born in Denmark and adjusted
for the first two principal components.
Mean PRSSCZ and PRSEDU were compared using analyses
of variance for each of the six categories of adolescence
school performance.
RESULTS
A total of 1470 cases with schizophrenia and 7318 subcohort
noncases were available for analysis. Details as to the repre-
sentativeness of the case-cohort sample to the general pop-
ulation has been described elsewhere (29).
Table 1 contains characteristics of cases with schizophrenia
and noncases in the cohort. More cases than SC noncases
were noncompleters of primary school (of the 1470 cases and
7318 SC noncases, a total of 455 [30.9%] of all individuals with
schizophrenia and 962 [13.1%] among the SC noncases were
noncompleters). For the overall average, and Danish and math
grade scores, lower grade point averages were consistently
found in the group with subsequent schizophrenia diagnosis (p
, .0001). Cases had higher proportions of parental schizo-
phrenia, other parental psychiatric diagnosis, and lower
average lower socioeconomic status than controls (all p ,
.0001).
3
Biological
Psychiatry Polygenic Risk, School Achievement, and Risk for Schizophrenia

Table 1. Characteristics of the Population-Based Cohort 0.5

With SZ and Noncases
p Value for 0.4
SZ Cases Noncases Difference
Total 1470 7318

Mean PRS schizophrenia

Passed Primary Schoola 1015 (69.0) 6356 (86.9) , .0001 0.3

Grade Point Average 5.7 6 2.0 6.5 6 2.2 , .0001

Danish Grade Point Average 5.9 6 2.2 6.4 6 2.3 , .0001 0.2 non-cases
Math Grade Point Average 5.4 6 2.7 6.7 6 2.8 , .0001 SZ cases
Parental Schizophrenia 69 (4.7) 92 (1.3) , .0001
0.1
Parental Any Other 342 (23.3) 829 (11.3) , .0001
Psychiatric Diagnosis
Low Socioeconomic Statusb 124 (8.4) 244 (3.3) , .0001 0
High PRSSCZc 263 (17.9) 759 (10.4) , .0001 NC 1 2 3 4 5

High PRSEDUc 108 (7.3) 733 (10.0) , .001

Values are mean 6 SD or n (%).
PRSEDU, polygenic risk score for educational attainment; PRSSCZ,
polygenic risk score for schizophrenia; SZ, schizophrenia.
a
Defined as having at least a grade point average of 2 in Danish and
math before July 1 the year they turn 18 years of age.
b
Measured on a scale of 0 to 6 in which 6 is lowest.
c
Highest decile.
Associations of PRSSCZ and PRSEDU With
Schizophrenia
We found that a higher PRSSCZ predicted schizophrenia, with
an IRR of 1.28 (95% confidence interval [CI], 1.19–1.36) per
standard deviation of PRSSCZ in this sample. The highest (vs.
lowest) decile of PRSSCZ predicted schizophrenia, with an IRR
of 2.26 (95% CI, 1.74–2.94).
We also found that PRSEDU was inversely associated with
schizophrenia, with an IRR of 0.87 (95% CI, 0.82–0.92) per
standard deviation of the score. The highest decile (vs. lowest)
PRSEDU predicted a decreased risk of schizophrenia (IRR,
0.68; 95% CI, 0.51–0.90) compared with the lowest decile.
PRSSCZ and PRSEDU were significantly negatively correlated
(r = 2.31, p , .01).
-0.1
School achievement groups
Figure 1. Mean (95% confidence interval) polygenic risk score (PRS) for
schizophrenia (SZ) according to category of primary school performance (1
depicts the lowest and 5 the highest quintile) in cases with SZ and in non-
cases. NC, noncompleters.
association between PRSEDU was different according to whether
the child/adolescent develop schizophrenia or not. As shown in
Figure 2, PRSEDU was associated with positive school perfor-
mance (category of adolescent school performance) in a dose-
response association. (p , .001; test for trend) in cases with
schizophrenia and noncases. There was no evidence that the
effect of PRSEDU was different according to whether the child/
adolescent developed schizophrenia or not (p = .98). In the entire
sample of cases and noncases, PRSEDU accounted for 7.6% of
variance in school performance.
Primary School Performance and Associations With
Schizophrenia
Table 2 shows that poorer school performance was associated
with an increased risk of schizophrenia in both basic and

PRSSCZ and Associations With School Performance 0.6

We wanted to investigate whether there is a correlation be-

tween PRSSCZ and school performance (ninth grade) in cases 0.4

with schizophrenia and noncases. Figure 1 shows the mean

PRSSCZ (and 95% CIs) for six categories of school perfor- 0.2
Mean PRS educaon

mance in cases with schizophrenia and SC noncases. Among

the cases, we found no clear correlation as there was no clear
0
pattern that poorer school performance was associated with non-cases
higher mean PRSSCZ. Among the SC noncases, status as SZ cases
noncompleter of primary school was associated with a -0.2

significantly higher mean PRSSCZ (from regression of PRSSCZ

on completers vs. noncompleters). We also tested for effect -0.4
modification by examining whether the association between
PRSSCZ and school performance was different according to
-0.6
whether the child/adolescent developed schizophrenia or not, NC 1 2 3 4 5
but found no significant difference (p = .99). School achievement groups

PRSEDU and Associations With School Performance Figure 2. Mean (95% confidence interval) polygenic risk score (PRS) for
educational attainment according to category of primary school perfor-
We also investigated the correlation between PRSEDU and mance (1 depicts the lowest and 5 the highest quintile) in cases with
adolescence school performance and tested whether the schizophrenia (SZ) and in noncases. NC, noncompleters.

4 Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal


Polygenic Risk, School Achievement, and Risk for Schizophrenia
Table 2. IRR of Schizophrenia According to School Achievement at Primary School (Not Passing and Quintiles of Average
Grades)
Basic Model Adjusted for PRSSCZ
Not Passing 2.92 (2.37–3.60) 2.87 (2.34–3.55)
Lowest Quintile 1.58 (1.27–1.97) 1.60 (1.28–1.99)
Second Lowest 1.44 (1.15–1.79) 1.44 (1.15–1.80)
Median Quintile 1 (reference) 1 (reference)
Second Highest 0.84 (0.68–1.06) 0.85 (0.68–1.07)
Highest Quintile 0.55 (0.42–0.72) 0.55 (0.42–0.72)
Values are presented as IRR (95% CI).
CI, confidence interval; IRR, incidence rate ratio; PRSEDU, polygenic risk score for educational attainment; PRSSCZ, polygenic risk score for
schizophrenia; SES, socioeconomic status.
a
According to hierarchy of parental diagnoses (see text).
b
Mutually adjusted for all variables.
adjusted analyses. In basic analyses, noncompleters of pri-
mary school had an increased risk of schizophrenia, with an
IRR of 2.92 (95% CI, 2.37–3.60) compared with completers.
Adjusting for PRSSCZ and PRSEDU (without adjusting for psy-
chiatric family history and socioeconomic status) did not
substantially alter the effect of being noncompleter on ninth
grade, which indicates that there was no evidence of con-
founding by PRS. In the fully adjusted model that additionally
included psychiatric family history and socioeconomic status,
noncompleters of primary school had an increased risk of
schizophrenia, with an IRR of 2.19 (95% CI, 1.75–2.73). The
fully adjusted model thus took into account both genetic lia-
bility and parental psychiatric history.
Among completers of primary school with grades in the
lowest quintile, the IRR of schizophrenia was increased with
1.58 (95% CI, 1.27–1.97) compared with the median quintile,
whereas probands with grades in the highest quintile had an
IRR of 0.55 (95% CI, 0.42–0.72). We also found that adjusting
for PRSSCZ and PRS EDU (without adjusting for psychiatric
family history and socioeconomic status) did not substantially
alter these estimates.
DISCUSSION
Main Findings
The present study investigated the possible effect of the
polygenic risk score (PRSSCZ) for schizophrenia and for
educational attainment (PRSEDU) on the association between
school performance and later schizophrenia. We found that
increasing PRSSCZ predicted schizophrenia and that PRSEDU
also predicted schizophrenia. PRSSCZ did not convincingly
correlate with ninth-grade school performance in cases with
schizophrenia, but PRSEDU correlated with adolescence school
performance in a dose-response association, with 7.6% of the
variance explained by PRSEDU.. Finally, we replicated a robust
association between poor school performance and risk of
schizophrenia, which was not explained by either PRSEDU or
PRSSCZ.
PRSEDU Analyses
Consistent with the findings of others (9,10), we found that the
genes identified for educational attainment showed a genetic
Biological
Psychiatry
Adjusted for SES and
Adjusted for PRSEDU Parental Psycha Fully Adjustedb
2.93 (2.37–3.61) 2.19 (1.75–2.73) 2.19 (1.75–2.73)
1.59 (1.28–1.98) 1.36 (1.08–1.70) 1.39 (1.11–1.75)
1.44 (1.15–1.80) 1.31 (1.05–1.64) 1.33 (1.06–1.67)
1 (reference) 1 (reference) 1 (reference)
0.85 (0.68–1.06) 0.89 (0.71–1.11) 0.89 (0.70–1.12)
0.57 (0.43–0.74) 0.61 (0.47–0.80) 0.61 (0.46–0.81)
overlap with schizophrenia. Each 1 SD higher PRSEDU pre-
dicted a significantly 13% reduced risk. This result contrasts
with the findings of Okbay et al. (9) and Krapohl et al. (37), who
found a positive genetic correlation between educational
attainment and schizophrenia (r = .08) (9). Low cognitive per-
formance and low school performance have been associated
with schizophrenia (7,15,38). Evidence from a population-
based Swedish Multi-Generation Register suggested a nega-
tive correlation (20.11) between IQ and psychosis, in line with
our results (39). Also, it has been demonstrated that copy
number variants conferring risk for schizophrenia predict lower
cognitive performance in healthy individuals (40). Hence, a
priori we would expect variants associated with more years of
schooling to occur less frequently in schizophrenia cases. On
the other hand, the PRS for schizophrenia has also been re-
ported to predict creativity in independent samples (41), and a
high prevalence of psychosis in individuals performing highly in
certain creative and cognitive tasks has been reported (42),
and it is evident that the relationship between schizophrenia
and educational attainment is complex. Lead SNPs for both
educational attainment and schizophrenia show enrichment in
brain areas important for cognition (9), and it is possible that
SNPs associated with schizophrenia are in LD with SNPs
associated with educational attainment because they occur in
the same brain regions and hence would create genetic cor-
relations, as suggested in Okbay et al. (9). Furthermore, in
Okbay et al. (9), they used schizophrenia GWAS summary
statistics from the full PGC2 meta-analysis, whereas we used
the Danish subsample included in the PGC2 (GEMS1 and
GEMS2) and a sample from the iPSYCH cohort. It is plausible
that the heterogeneity or genetic drift is responsible for allele
frequency differences between the PGC2 samples and the
Danish sample, and that this would be reflected in different
estimates of genetic correlation between educational attain-
ment and schizophrenia.
We found a rather high correlation between PRSEDU and
school performance in cases with schizophrenia as well as in
noncases. Findings from a recent UK-based study using
teacher-obtained information about school performance (En-
glish and mathematics) suggest that educational PRS
explained 9% of the variance at around 16 years of age (23). A
generally poorer primary school performance in the schizo-
phrenia group compared with controls could be due to effects
Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal 5
Biological
Psychiatry Polygenic Risk, School Achievement, and Risk for Schizophrenia
of other factors (such as early morbidity) expressed more
strongly in the schizophrenia group. Of the genome-wide sig-
nificant loci associated with educational length (9,43), the
relevant SNPs are disproportionately found in genomic regions
regulating gene expression in the fetal brain (9), possibly
affecting school performance in the preschizophrenia group
adding to the hypothetical effects of other risk genes
expressed in the fetus and involved in synaptogenesis (44,45).
PRSSCZ Analyses
PRSSCZ had power to predict schizophrenia in line with pre-
vious findings (46) despite a low degree of chronicity of
schizophrenia in our sample. The association with schizo-
phrenia identified in the latest PGC GWAS paper (46) is
stronger in more chronic schizophrenia samples (47). We found
no evidence that PRSSCZ explained poorer primary school
performance in schizophrenia. Limited statistical power may
underlie our negative findings, but it is also possible that
environmental risk factors could have impacted school per-
formance more in cases with schizophrenia. However, within
the control group, a higher mean PRSSCZ was found in pro-
bands that had not completed primary school, but no asso-
ciation was found among completers. Interestingly, higher
PRSSCZ has previously been shown to be associated with
nonpsychotic phenotypes including anxiety disorder in young
people (21).
Social Factors and Family-Level Psychiatric History
Influences
As in previous studies (19,48), positive family-level psychiatric
history and low parental socioeconomic status were associated
with the risk of schizophrenia in offspring as well as with school
performance (5). These factors clearly confounded the asso-
ciation between school performance and risk of schizophrenia.
Our multivariate approach disentangled effects of common
genetic variants from some of the presumed environmental
effects of social factors and parental psychiatric history. After
adjustment for all factors, we still found a 2.19-fold higher risk
for schizophrenia to be associated with not having completed
primary school, and an elevated risk was also associated with
low school marks. This suggests that other factors (than our
currently measured risk factors such as early morbidity) could
have played a role. In the study of 5825 UK individuals, their
polygenic score for education was associated with family so-
cioeconomic status (w7%), and no interaction between their
educational attainment PRS and family socioeconomic status
on school performance was found (23).
Strengths and Limitations
Strengths include the uniformly organized Danish health care
system, access to the Primary Education Register allowing a
population-based design, the ability to link different data
sources with prospectively collected data, and a virtually
complete follow-up. Access to the Primary Education Register
enabled us to examine earlier markers of educational success
in the general population, at a time before the typical age of
onset of schizophrenia. Moreover, we used publicly available
GWAS summary statistics for schizophrenia and educational
attainment to calculate PRSSCZ (19) and PRSEDU (9).
6 Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
Several limitations should be considered. Schizophrenia is a
heterogeneous disorder, but a diagnosis of schizophrenia in
the Danish Psychiatric Central Register has high diagnostic
validity (49). Despite the PRSs being based on big samples,
they may not be sufficiently powerful to capture signs of dis-
rupted neurodevelopment. PRSs capture only a proportion of
the variation attributable to underlying genetic risk (50), as they
do not include, for example deleterious exonic mutations (51)
or rare copy number variations (52), which may be important
for the behavioral trait. Also, our measure of socioeconomic
status might only have captured a part of the actual circum-
stances during an individual’s upbringing.
Conclusions
We found that poor school performance before the typical age of
onset of schizophrenia predicted increased risk of schizophrenia
later in life. PRSEDU predicted primary school performance and
explained 7.6% of its variance. However, neither current PRSSCZ
nor PRSEDU explained the school performance–schizophrenia
association.
Poor school performance may be an antecedent or may be
part of the extended prodrome for schizophrenia. Collectively,
these findings suggest that other factors (e.g., genetic factors
not assessed by PRS environmental factors that impact on
cognitive reserve and neurodevelopmental factors) are impor-
tant in explaining the school performance–schizophrenia
association.
ACKNOWLEDGMENTS AND DISCLOSURES
The work was supported by an unrestricted grant from the Lundbeck
Foundation (iPSYCH), a National Health and Medical Research Council John
Cade Fellowship (Grant No. APP1056929) (to JM) and a Niels Bohr Pro-
fessorship from the Danish National Research Foundation (to JM). The au-
thors report no biomedical financial interests or potential conflicts of
interest.
ARTICLE INFORMATION
From the Mental Health Centre Copenhagen (HJS, MEB, AR, CH, MN),
Copenhagen University Hospital; and i-PSYCH initiative for Integrative
Psychiatric Research (HJS, J-CD, EA, MEB, PBM, AR, CH, OM, MN, LP),
Lundbeck Foundation, Copenhagen; and National Centre for Register-
Based Research (J-CD, EA, JJM, PBM, LP) and Centre for Integrated
Register-Based Research and National Centre for Register-Based Research
(EA, OM, LP), Aarhus University, Aarhus; and Department of Psychosis
(J-CD, OM), Aarhus University Hospital, Risskov, Denmark; and the Queens-
land Brain Institute (JJM), University of Queensland, St Lucia, Australia.
MN and LP contributed equally to this work as joint senor authors.
Address correspondence to Holger J. Sørensen, M.D., Ph.D., Mental
Health Centre Copenhagen & Copenhagen University, Kildegaardsvej 28, 4,
2900 Hellerup, Denmark; E-mail: holger.jelling.soerensen@regionh.dk.
Received Dec 15, 2017; revised and accepted Apr 17, 2018.
REFERENCES
1. Reichenberg A, Caspi A, Harrington H, Houts R, Keefe RSE,
Murray RM, et al. (2010): Static and dynamic cognitive deficits in
childhood preceding adult schizophrenia: A 30-year study. Am J
Psychiatry 167:160–169.
2. Woodberry KA, Giuliano AJ, Seidman LJ (2008): Premorbid IQ in
schizophrenia: A meta-analytic review. Am J Psychiatry 165:579–
587.
3. Sniekers S, Stringer S, Watanabe K, Jansen PR, Coleman JRI,
Krapohl E, et al. (2017): Genome-wide association meta-analysis of

Polygenic Risk, School Achievement, and Risk for Schizophrenia
78,308 individuals identifies new loci and genes influencing human
intelligence. Nat Genet 49:1107–1112.
4. Dickson H, Laurens KR, Cullen AE, Hodgins S (2012): Meta-analyses
of cognitive and motor function in youth aged 16 years and
younger who subsequently develop schizophrenia. Psychol Med
42:743–755.
5. Lin A, Di Prinzio P, Young D, Jacoby P, Whitehouse A, Waters F, et al.
(2017): Academic performance in children of mothers with schizo-
phrenia and other severe mental illness, and risk for subsequent
development of psychosis: A population-based study. Schizophr Bull
43:205–213.
6. Kendler KS, Ohlsson H, Mezuk B, Sundquist K, Sundquist J (2016):
A Swedish National prospective and co-relative study of school
achievement at age 16, and risk for schizophrenia, other nonaffective
psychosis, and bipolar illness. Schizophr Bull 42:77–86.
7. MacCabe JH, Lambe MP, Cnattingius S, Torrång A, Björk C,
Sham PC, et al. (2008): Scholastic achievement at age 16 and risk of
schizophrenia and other psychoses: A national cohort study. Psychol
Med 38:1133–1140.
8. Hubbard L, Tansey KE, Rai D, Jones P, Ripke S, Chambert KD, et al.
(2016): Evidence of common genetic overlap between schizophrenia
and cognition. Schizophr Bull 42:832–842.
9. Okbay A, Beauchamp JP, Fontana MA, Lee JJ, Pers TH, Rietveld CA,
et al. (2016): Genome-wide association study identifies 74 loci asso-
ciated with educational attainment. Nature 533:539–542.
10. Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh PR,
et al. (2015): An atlas of genetic correlations across human diseases
and traits. Nat Genet 47:1236–1241.
11. Khandaker GM, Barnett JH, White IR, Jones PB (2011): A quantitative
meta-analysis of population-based studies of premorbid intelligence
and schizophrenia. Schizophr Res 132:220–227.
12. Zammit S, Allebeck P, David AS, Dalman C, Hemmingsson T,
Lundberg I, et al. (2004): A longitudinal study of premorbid IQ score
and risk of developing schizophrenia, bipolar disorder, severe
depression, and other nonaffective psychoses. Arch Gen Psychiatry
61:354–360.
13. Reichenberg A, Weiser M, Rapp MA, Rabinowitz J, Caspi A,
Schmeidler J, et al. (2005): Elaboration on premorbid intellectual per-
formance in schizophrenia: Premorbid intellectual decline and risk for
schizophrenia. Arch Gen Psychiatry 62:1297–1304.
14. Mortensen EL, Sørensen HJ, Jensen HH, Reinisch JM, Mednick SA
(2005): IQ and mental disorder in young men. Br J Psychiatry 187:407–
415.
15. Jones P, Murray R, Jones P, Rodgers B, Marmot M (1994): Child
developmental risk factors for adult schizophrenia in the British 1946
birth cohort. Lancet 344:1398–1402.
16. Sørensen HJ, Sæbye D, Urfer-Parnas A, Mortensen EL, Parnas J
(2012): Premorbid intelligence and educational level in bipolar and
unipolar disorders: A Danish draft board study. J Affect Disord
136:1188–1191.
17. Peyrot WJ, Lee SH, Milaneschi Y, Abdellaoui A, Byrne EM, Esko T,
et al. (2015): The association between lower educational attainment
and depression owing to shared genetic effects? Results in w25,000
subjects. Mol Psychiatry 20:735–743.
18. Ripke S, Neale BM, Corvin A, Walters JTR, Farh K-H, Holmans PA,
et al. (2014): Biological insights from 108 schizophrenia-associated
genetic loci. Nature 511:421–427.
19. Agerbo E, Sullivan PF, Vilhjálmsson BJ, Pedersen CB, Mors O,
Børglum AD, et al. (2015): Polygenic risk score, parental socioeco-
nomic status, family history of psychiatric disorders, and the risk for
schizophrenia. JAMA Psychiatry 72:635–641.
20. Benros ME, Trabjerg BB, Meier S, Mattheisen M, Mortensen PB,
Mors O, et al. (2016): Influence of polygenic risk scores on the asso-
ciation between infections and schizophrenia. Biol Psychiatry 80:609–
616.
21. Jones HJ, Stergiakouli E, Tansey KE, Hubbard L, Heron J, Cannon M,
et al. (2016): Phenotypic manifestation of genetic risk for schizophrenia
during adolescence in the general population. JAMA Psychiatry
73:221–228.
Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
Biological
Psychiatry
22. McGrath JJ, Wray NR, Pedersen CB, Mortensen PB, Greve AN,
Petersen L (2014): The association between family history of mental
disorders and general cognitive ability. Transl Psychiatry 4:e412.
23. Selzam S, Krapohl E, von Stumm S, O’Reilly PF, Rimfeld K, Kovas Y,
et al. (2017): Predicting educational achievement from DNA. Mol
Psychiatry 22:267–272.
24. Marioni RE, Ritchie SJ, Joshi PK, Hagenaars SP, Okbay A, Fischer K,
et al. (2016): Genetic variants linked to education predict longevity.
Proc Natl Acad Sci U S A 113:13366–13371.
25. Pedersen CB, Gøtzsche H, Møller JO, Mortensen PB (2006): The
Danish Civil Registration System. A cohort of eight million persons.
Dan Med Bull 53:441–449.
26. Mors O, Perto GP, Mortensen PB (2011): The Danish Psychiatric
Central Research Register. Scand J Public Health 39(7 suppl):54–57.
27. World Health Organization (1993): ICD-10 Classification of Mental and
Behavioral Disorders. Diagnostic Criteria for Research. Geneva,
Switzerland: World Health Organization.
28. Petersson F, Baadsgaard M, Thygesen LC (2011): Danish registers on
personal labour market affiliation. Scand J Public Health 39(7 suppl):95–98.
29. Pedersen CB, Bybjerg-Grauholm J, Pedersen MG, Grove J, Agerbo E,
Bækvad-Hansen M, et al. (2018): The iPSYCH2012 case-cohort
sample: New directions for unravelling genetic and environmental ar-
chitectures of severe mental disorders. Mol Psychiatry 23:6–14.
30. Børglum AD, Demontis D, Grove J, Pallesen J, Hollegaard MV,
Pedersen CB, et al. (2013): Genome-wide study of association and
interaction with maternal cytomegalovirus infection suggests new
schizophrenia loci. Mol Psychiatry 19:325–333.
31. Wray NR, Lee SH, Mehta D, Vinkhuyzen AAE, Dudbridge F,
Middeldorp CM (2014): Research review: Polygenic methods and their
application to psychiatric traits. J Child Psychol Psychiatry 55:1068–1087.
32. Bulik-Sullivan BK, Loh P-R, Finucane HK, Ripke S, Yang J,
Patterson N, et al. (2015): LD score regression distinguishes con-
founding from polygenicity in genome-wide association studies. Nat
Genet 47:291–295.
33. Pedersen CB, Mortensen PB (2001): Family history, place and season
of birth as risk factors for schizophrenia in Denmark: A replication and
reanalysis. Br J Psychiatry 178:46–52.
34. Sørensen HJ, Nielsen PR, Benros ME, Pedersen CB, Mortensen PB
(2015): Somatic diseases and conditions before the first diagnosis of
schizophrenia: A nationwide population-based cohort study in more
than 900 000 individuals. Schizophr Bull 41:513–521.
35. Mortensen PB, Pedersen MG, Pedersen CB (2003): Psychiatric family
history and schizophrenia risk in Denmark: which mental disorders are
relevant? Psychol Med 40:201–210.
36. Petersen L, Sørensen TIA, Andersen PK (2003): Comparison of case-
cohort estimators based on data on premature death of adult
adoptees. Stat Med 22:3795–3803.
37. Krapohl E, Euesden J, Zabaneh D, Pingault J-B, Rimfeld K, von
Stumm S, et al. (2016): Phenome-wide analysis of genome-wide
polygenic scores. Mol Psychiatry 21:1188–1193.
38. Osler M, Lawlor DA, Nordentoft M (2007): Cognitive function in
childhood and early adulthood and hospital admission for schizo-
phrenia and bipolar disorders in Danish men born in 1953. Schizophr
Res 92:132–141.
39. Zammit S (2012): A population-based study of shared genetic variation
between premorbid IQ and psychosis among male twin pairs and
sibling pairs from Sweden. Arch Gen Psychiatry 69:460–466.
40. Stefansson H, Meyer-Lindenberg A, Steinberg S, Magnusdottir B,
Morgen K, Arnarsdottir S, et al. (2013): CNVs conferring risk of autism
or schizophrenia affect cognition in controls. Nature 505:361–366.
41. Power RA, Steinberg S, Bjornsdottir G, Rietveld CA, Abdellaoui A,
Nivard MM, et al. (2015): Polygenic risk scores for schizophrenia and
bipolar disorder predict creativity. Nat Neurosci 18:953–955.
42. Kyaga S, Landén M, Boman M, Hultman CM, Långström N,
Lichtenstein P (2013): Mental illness, suicide and creativity: 40-Year
prospective total population study. J Psychiatr Res 47:83–90.
43. Rietveld CA, Medland SE, Derringer J, Yang J, Esko T, Martin NW,
et al. (2013): GWAS of 126,559 individuals identifies genetic variants
associated with educational attainment. Science 340:1467–1471.
7
Biological
Psychiatry Polygenic Risk, School Achievement, and Risk for Schizophrenia
44. Jaaro-Peled H, Hayashi-Takagi A, Seshadri S, Kamiya A, Brandon NJ,
Sawa A (2009): Neurodevelopmental mechanisms of schizophrenia:
Understanding disturbed postnatal brain maturation through
neuregulin-1-ErbB4 and DISC1. Trends Neurosci 32:485–495.
45. Ross RG, Hunter SK, McCarthy L, Beuler J, Hutchison AK,
Wagner BD, et al. (2013): Perinatal choline effects on neonatal path-
ophysiology related to later schizophrenia risk. Am J Psychiatry
170:290–298.
46. Ripke S, O’Dushlaine C, Chambert K, Moran JL, Kähler AK, Akterin S,
et al. (2013): Genome-wide association analysis identifies 13 new risk
loci for schizophrenia. Nat Genet 45:1150–1159.
47. Meier SM, Agerbo E, Maier R, Pedersen CB, Lang M, Grove J, et al.
(2016): High loading of polygenic risk in cases with chronic schizo-
phrenia. Mol Psychiatry 21:969–974.
8 Biological Psychiatry - -, 2018; -:-–- www.sobp.org/journal
48. Sørensen HJ, Nielsen PR, Pedersen CB, Benros ME, Nordentoft M,
Mortensen PB (2014): Population impact of familial and environmental
risk factors for schizophrenia: A nationwide study. Schizophr Res
153:214–219.
49. Uggerby P, Østergaard SD, Røge R, Correll CU, Nielsen J (2013): The
validity of the schizophrenia diagnosis in the Danish Psychiatric
Central Research Register is good. Dan Med J 60:A4578.
50. Dudbridge F (2013): Power and predictive accuracy of polygenic risk
scores. PLoS Genet 9:e1003348.
51. Purcell SM, Moran JL, Fromer M, Ruderfer D, Solovieff N, Roussos P,
et al. (2014): A polygenic burden of rare disruptive mutations in
schizophrenia. Nature 506:185–190.
52. Malhotra D, Sebat J (2012): CNVs: Harbingers of a rare variant revo-
lution in psychiatric genetics. Cell 148:1223–1241.