Prenatal Nutrition and IQ:

A causal analysis using a Mendelian

randomization approach

Sarah Lewis

A few earlier studies

Common problems in observational studies

• Measurement error• Reporting/

interviewer bias• Reverse causation• CONFOUNDING

4

Mendelian randomization

5

 Comparison of groups of individuals defined by genotype should only differ with respect to the locus under study. With control for population structure, there should be little confounding by any behavioural, socioeconomic or physiological factors.

Mendel’s second law – the law of independent assortment

6

“The behaviour of each pair of differentiating characteristics in hybrid union is independent of the other differences between the two original plants, and, further, the hybrid produces just so many kinds of egg and pollen cells as there are possible constant combination forms” 

Gregor Mendel, 1865.

Problems of observational epidemiology can be overcome using genetic variants as proxies

• Confounding

• Reverse causation• Biological• Due to reporting bias

• Measurement error

7

Independent assortment

Fixed at conception

Can be measured (genotyped)

Clustered environments and randomised genes (93 phenotypes, 23 SNPs)

Phenotype / phenotype4278 pairwise associations

Phenotype / genotype2139 pairwise combinations

Genotype / genotype253 pairwise combinations

43% significant at p<0.01

20 significant at p<0.01 vs 21 expected

4 / 253 significant at p<0.01 vs 3 expected

Davey Smith et al. PLoS Medicine 2008

MR Vs RCT

9

Objective To determine whether exposure to specific dietary

factors in utero and infancy influences cognition

Hypotheses Suboptimum levels of nutrients in utero lead to

impaired neurodevelopment and cognitive ability Variation in genes related to nutrient metabolism is

associated with neurodevelopment in infancy Association of maternal genotype with measurements

of cognitive ability, independent of child’s genotype, will indicate a role of the prenatal environment

Project design

Maternal genetic variants affecting nutrient levels

Nutrient exposureIQ at age 8

confounders

Offspring genetic variants

Association between diet and cognition can be affected by confounding by lifestyle factors, reverse causality and measurement error

Polymorphisms in genes that metabolise nutrients can be used as proxies for differences in dietary intake and therefore to infer causal relationships between nutrients and cognition without the above problems

ALSPAC (Avon Longitudinal Study of Parents and Children)

Population-based prospective study conducted in Bristol, England to evaluate factors that affect health and development of children

~ 14,000 pregnant women enrolled between April 1991 and December 1992

Information on mother and child collected at regular intervals and ongoing

DNA samples available for mothers and children (~10000 each, ~7000 duos)

Population Characteristicstotal IQ verbal IQ performance IQ

mean 104.1 107.0 99.5sd 16.5 16.7 17.1

min 45 46 46max 151 155 151

N 4111 4124 4127

mean sd N p-value* maternal age (years) 28.5 4.7 6590 <0.0001child's age (months) 103.5 3.1 4183 <0.0001gestation (weeks) 39.5 1.8 7373 0.525

birthweight (grams) 3422.4 536.8 7294 <0.0001

*p-value for association with total IQ

gene associated SNP nutrient/diseasePTGS2 rs20417 omega-3 fatty acidsGPX4 rs713041 selenium

ADAMTS9 rs4607103 T2DCDC123/CAMK1D rs12779790 T2D

CDKAL1 rs10946398 T2DHHEX-IDE rs1111875 T2D

HNF1B rs757210 T2DIGFBP2 rs4402960 T2DKCNJ11 rs5219 T2DNOTCH2 rs10923931 T2DTCF7L2 rs7903146 T2DTHADA rs7578597 T2D

TSPAN8-LGR5 rs7961581 T2DWFS1 rs10010131 T2DPPARG rs1801282 T2D/obesityGCKR rs780094 triglycerides

CDKN2A/B rs10811661 triglycerides/T2DJAZF1 rs864745 triglycerides/T2D

SLC2A9 rs16890979 uric acidFUT2 rs492602 vitamin B12TCN2 rs1801198 vitamin B12

ALPL/NBPF3 rs4654748 vitamin B6SLC23A1 rs33972313 vitamin C

VDBP rs7041 vitamin DVDR rs2228570 vitamin DVDR rs1544410 vitamin DVDR rs731236 vitamin DVDR rs7975232 vitamin D

APOA5 rs12272004 vitamin ESLC30A8 rs13266634 zinc

gene associated SNP nutrient/diseaseBCMO1 rs6564851 beta carotene

TAS2R38 rs1726866 bitter taste receptorTAS2R38 rs713598 bitter taste receptor

CASR rs1801725 calciumTRPV5/C7orf34 rs11979330 calcium

TRPV6 rs4987667 calciumSLC8A1 rs2110922 calciumCELSR2 rs646776 cholesterol

LDLR rs2228671 cholesterolSLC5A7 rs1013940 cholineSLC5A7 rs333229 cholineMTHFR rs1801133 folic acid

GCK rs1799884 glucoseDIO1 rs2235544 iodineHFE rs1799945 ironHFE rs1800562 ironTF rs3811647 iron

TMPRSS6 rs4820268 ironLCT rs4988235 lactose

APOB rs693 lipidsAPOE rs2075650 lipidsLIPC rs1532085 lipids

TRPM6 rs2274924 magnesiumTRPM6 rs3750425 magnesiumTRPM7 rs8042919 magnesiumMAT1A rs1143693 methionineMAT2A rs17026419 methionine

MTR rs1805087 methionineMTRR rs1801394 methionineFTO rs9939609 obesity

Potential ConfoundersMother: Age Education Social class Marital status Parity Inter-pregnancy interval Any infection during pregnancy Housing tenure Ever smoked Alcohol consumption Iron, zinc, calcium, folic acid, vitamins, other

supplements during pregnancy

Potential ConfoundersChild: Sex Age Gestation Birth weight Breastfeeding duration

total IQ mean sd N p-valuematernal factors I

education <0.0001less than O level 95.9 15.4 855

O level 102.1 15.0 1411more than O level 110.4 15.8 1730

social class <0.0001manual 98.5 15.3 551

non-manual 106.2 16.2 2941

marital status <0.0001never married 101.4 16.8 493

married 104.8 16.3 3347separated/divorced/widowed 101.1 15.8 176

parity <0.0001no children 105.7 16.4 1869

1 child 104.4 16.0 14112 children 100.7 16.9 534

3 or more children 98.3 15.9 169

interpregnancy interval 0.016< 3 months 105.2 16.8 1653-5 months 103.0 16.6 120

6-11 months 104.8 16.3 31312-17 months 103.7 16.7 38818-35 months 103.4 16.3 77836-59 months 102.4 16.6 275

60+ months 99.7 14.8 192

any infection during preg 0.038no 104.6 16.5 3101yes 103.3 16.2 827

housing tenure <0.0001mortgaged 105.3 16.2 3407

rented/other 102.2 16.3 294council rented 93.6 15.3 309

total IQ mean sd N p-valuematernal factors II

ever smoked <0.0001no 105.2 16.5 2273yes 103.0 16.3 1735

alcohol consumptionbefore this pregnancy <0.0001

never 98.3 17.0 215< 1 glass per week 103.1 16.0 1482

1+ glasses per week 105.3 16.5 18361+ glasses per day 106.4 16.6 483

during this pregnancy 0.009never 103.8 16.3 1731

< 1 glass per week 104.4 16.7 16461+ glasses per week 105.8 15.8 5621+ glasses per day 99.9 18.8 72

iron supplements 0.595no 104.2 16.4 3256yes 104.5 16.8 755

zinc supplements 0.001no 104.1 16.4 3948yes 111.0 18.2 64

calcium supplements 0.039no 104.2 16.5 3875yes 107.1 16.1 134

folate supplements 0.004no 104.0 16.4 3651yes 106.7 16.8 359

vitamin supplements <0.0001no 103.7 16.4 3320yes 106.8 16.4 680

other supplements 0.029no 104.3 16.3 3753yes 107.7 17.0 115

Confounders continuedtotal IQ mean sd N p-value

child-related factorssex 0.831

male 104.0 17.1 2053female 104.1 15.9 2058

duration of breastfeeding <0.0001never 98.1 15.2 782

< 3 months 102.5 15.8 8853-5 months 104.5 16.0 6746 months+ 108.8 16.5 1487

confounder dbSNP p-valuemother's age rs2075650 0.012

education rs1799945 0.023interpreg. Interval rs7041 0.020

any infection rs492602 0.053rs1726866 0.003rs713598 0.006

housing tenure rs10811661 0.015gestation rs10811661 0.005

rs1544410 0.013rs731236 0.008

breastfeeding rs864745 0.058iron supps rs731236 0.028zinc supps rs1799945 0.057

folate supps rs13266634 0.024rs1544410 0.017rs731236 0.014

vitamin supps rs10811661 0.040other supps rs1726866 0.045child's age rs1544410 0.015

rs1799945 0.044rs731236 0.022

Do low levels of vitamin B in utero lead to a lower IQ at age 8?

Vitamin B12 facts Only synthesised by microorganisms Main sources: fish, shellfish, eggs, meat, dairy products Recommended Daily Amount: 2-3 ug/day Dietary deficiency rare (vegans at risk) B12 deficiency: <150 pmol/l Main functions: red blood cell formation, DNA synthesis,

maintenance of healthy nervous system Transport: 80% bound to transcobalamin I (HC)

20% bound to transcobalamin II (holoTC). holoTC delivers B12 to cells.

Stored in the liver

indicators of B12 deficiency

modified from Nexo and Hoffmann-Lucke (2011)

• Birth defects• Spontaneous abortion• Pre-eclampsia• Prematurity• Low birth weight• Cardiovascular disease• Cognitive deficit• Dementia

Vitamin B12 status during pregnancy

and cognition in children Lower cognition tests scores among offspring of mothers

with deficient intake of B12 (Mexico; del Rio Garcia et al., 2009)

Children of mothers with low B12 levels performed worse in sustained-attention and working memory tests (India; Bhate et al., 2008)

No association of maternal B12 levels with cognitive performance in children. Although verbal ability scores were higher in children of mothers with low B12 (India; Veena et al., 2010)

Problem: residual confounding?

Observational Study

Model 1: Adjusted for offspring sex and age at time of IQ assessment, and maternal energy intake.Model 2: Model 1 + maternal education, social class, age at delivery, parity, any infection in pregnancy, ever smoked, alcohol consumption before and during pregnancy, folate supplementation.Model 3: Model 2 + gestational length and birth weight.

N = 5004

mean difference in child IQ

per doubling of maternal vitamin B12

intake

95% CI p-value

Model 1 2.03 1.30, 2.76 < 0.001Model 2 0.74 0.04, 1.44 0.04Model 3 0.70 0.002, 1.39 0.05

Instrumental variables

Maternal SNPs vs offspring IQSNP genotype N IQ

mean (SD)

FUT2rs492602

TT 1009 103.3 (16.8)TC 1940 104.2 (16.1)CC 1012 105.0 (16.7)

mean difference in child IQ per C allele (95% CI)

3961 0.86 (0.14, 1.58)

p-value 0.02

TCN2rs1801198

GG 784 103.9 (16.6)CG 1978 104.1 (16.0)CC 1208 104.6 (17.0)

mean difference in child IQ per C allele (95% CI)

3970 0.35(-0.38, 1.08)

p-value 0.35

TCN2rs9606756

AA 3673 104.5 (16.0)AG 1038 104.7 (17.0)GG 75 108.3 (14.5)

mean difference in child IQ per G allele (95% CI)

4786 0.60 (-0.39, 1.58)

p-value 0.24

Conclusions• Genotypes associated with high vitamin B12

levels are associated with higher IQ.

• However, the effect of genotype on exposure is small, therefore the power of the study to confidently detect an effect is low.

• Replication is needed.

Do low prenatal iron levels affect IQ?

Iron deficiency anemia in pregnant women

Source: Worldwide Prevalence of Anemia - WHO report 2008

Effects of iron deficiency in early life

• developing brain structures (striatum, hippocampus)

• neurotransmitter systems (dopamine, serotonin)

• Synaptogenesis

• Myelination

• different gene and protein profiles in the ID brain

Consequences for the child

• recognition memory function• temperament (irritability, lower alertness and soothability)• hand-eye movement• locomotion• comprehension of language• fine motor skills• school performance• long lasting abnormalities, even after iron repletion

Instruments: TF, TMPRSS6, HFE SNPs

rs3811647 G/Aserum transferrin

serum ferritintransferrin saturation

A = iron lowering

Benyamin et al. (2009)

rs4820268 A/G D521Dserum iron

haemoglobintransferrin saturation

G = iron lowering

Benyamin et al. (2009)

rs1799945 C/G H63Drs1800562 G/A C282Y

serum ironserum transferrin

serum ferritinhaemoglobin

transferrin saturationC & G= iron lowering

Kullo et al. (2010)

Maternal genotype & Hb levels

*p-value adjusted by gestational age at the time of measurement

SNP gene genotype haemoglobin (g/dl) SD N unadjusted

p-value adjusted p-value

H63D rs1799945 HFE

GG 12.5 0.9 144 5.5x10-4 4.1x10-5 CG 12.4 1.0 1532 CC 12.3 1.0 4731

C282Y rs1800562 HFE

AA 12.6 1.1 29 3.0x10-6 4.6x10-8 GA 12.5 1.0 828 GG 12.3 1.0 5541

rs3811647 TF

GG 12.3 1.0 2767 0.81 0.76 GA 12.3 1.0 2885 AA 12.3 0.9 720

D521D rs4820268 TMPRSS6

AA 12.4 1.0 1725 2.8x10-9 2.9x10-10 AG 12.3 1.0 3316 GG 12.2 1.0 1356

rs1799945/rs1800562

HFE no. low iron

alleles

2 12.6 1.0 297 2.3x10-9 4.9x10-12 3 12.4 1.0 2080 4 12.3 1.0 3931

genotypic score

HFE/TMPRSS6 no. low iron

alleles

2 or 3 12.5 1.0 826 1.1x10-15 1.9x10-18 4 12.4 1.0 2094 5 12.3 1.0 2464 6 12.2 1.0 846

Maternal genotypes and iron supplementation

P = 0.001

P = 4x10-5

Maternal genetic score and child’s IQ

2 or 3 4 5 6

9899

100101102103104105106

iron at 32 weeksiron at 18 weeksNo iron supp

No of low iron alleles

IQ score

P=0.90

P=0.43

P=0.14

dbSNP gene genotype full scale IQ SD N

unadjusteda

effect (95% CI)p-value

N

adjusteda,b

effect (95% CI)p-value

N

rs1799945 HFEGG 105.9 17.0 79 0.50 (-0.59, 1.58)

0.373543

1.09 (-0.27, 2.45)0.392402

CG 102.9 16.3 850CC 104.1 16.4 2614

rs1800562 HFEAA 107.1 12.8 15 -0.52 (-2.05, 1.00)

0.503535

0.38 (-1.57, 2.32)0.712352

GA 104.2 16.6 455GG 103.8 16.5 3065

rs3811647 TFGG 104.0 16.3 1511 -0.13 (-0.95, 0.69)

0.763514

-0.36 (-1.40, 0.68)0.502378

GA 103.9 16.3 1613AA 103.6 17.3 390

rs4820268 TMPRSS6AA 104.3 16.0 962 -0.74 (-1.52, 0.05)

0.073527

-0.23 (-1.23, 0.77)0.662389

AG 104.1 16.7 1837GG 102.7 16.3 728

rs1799945/rs1800562

2 risk alleles 0 106.1 16.3 1560.11 (-0.83, 1.05)

0.823490

0.99 (-0.19, 2.18)0.1023123 risk alleles 1 103.1 16.5 1163

4 risk alleles 2 104.0 16.4 2171

genotypic score

≤3 risk alleles 0 103.8 16.1 464-0.32 (-0.94, 0.30)

0.313444

0.22 (-0.54, 0.98)0.572254

4 risk alleles 1 104.4 16.7 1143

5 risk alleles 2 103.5 16.1 1392

6 risk alleles 3 103.3 16.8 445

Table 5. Maternal genotypes at SNPs in iron-related genes and full scale IQ of their children at 8 years of age.

Limitations• instrument is associated with confounder• supplementation blurs association of genotype

and Hb• small numbers for stratified analysis• HFE variants are rare

Conclusions• HFE and TMPRSS6 variants were strongly associated with Hb

levels. Not so TF rs3811647.• Genotypes associated with low Hb levels were also associated with

higher risk of iron supplementation.• Mothers with rare HFE homozygote genotypes were more likely to

have an educational level greater than O-level.• The association between iron-related genotypes and child’s IQ was

strongest for women who took iron supplements during pregnancy. Adjustment by child’s genotype, maternal education, confounders and ancestry informative markers did not change this result.

• Results suggest that exposure to low levels of iron in fetal life adversely affects brain development and therefore IQ in childhood.

• Replication is needed.

Lipids and cognition• lipids are vital for membrane biogenesis during cellular growth

processes

• On the other hand, elevated serum cholesterol is a well-known risk factor for cognitive decline, dementia and Alzheimer’s disease.

• And higher serum LDL has been recently associated with decreased white matter integrity among healthy older adults.

Genetic scores for Lipid levels

• TG score = 27 SNPs• LDL-C score = 35 SNPs• HDL-C score = 44 SNPs• TC score = 46 SNPs

Weighted by effect sizes reported by Teslovich et al. (2010), Aulchenko et al. (2009) and Kettunen et al. (2012)

Additive model Risk alleles increase TG, LDL-C and TC, and decrease HDL-

C Scores are not associated with confounders

Association between offspring lipid genetic scores and plasma lipids, adjusted for age and sex.

triglycerides (mg/dl) total cholesterol (mg/dl) effect 95% CI p-value R2 (%) F effect 95% CI p-value R2 (%) F

TG score 1.03 1.02, 1.03 1.0x10-18 4.3 175.0 0.85 0.63, 1.08 1.1x10-13 1.4 56.0 LDL-C score 1.01 1.00, 1.01 2.7x10-5 0.5 17.4 1.40 1.20, 1.60 1.0x10-18 4.8 193.4 HDL-C score 1.01 1.00, 1.01 1.9x10-8 0.8 30.8 -0.29 -0.45, -0.12 7.3x10-4 0.3 11.8

TC score 1.01 1.00, 1.01 3.6x10-9 0.9 34.8 1.22 1.04, 1.40 1.0x10-18 4.3 174.2

LDL-C (mg/dl) HDL (mg/dl) effect 95% CI p-value R2 (%) F effect 95% CI p-value R2 (%) F

TG score 0.86 0.68, 1.05 1.0x10-18 2.1 82.5 -0.38 -0.49, -0.28 1.7x10-13 1.4 55.0 LDL-C score 1.31 1.15, 1.47 1.0x10-18 6.1 250.5 -0.13 -0.22, -0.04 0.01 0.2 7.9 HDL-C score 0.33 0.20, 0.47 1.9x10-6 5.5 21.7 -0.74 -0.81, -0.67 1.0x10-18 9.5 408.7

TC score 1.11 0.97, 1.26 1.0x10-18 5.2 211.0 -0.11 -0.20, -0.03 0.01 0.2 7.0

Association between offspring and maternal genotype scores and offspring IQ

Offspring IQ Offspring scores effect 95% CI p-value

TG score -0.05 -0.18, 0.08 0.46 LDL-C score -0.12 -0.24, -0.01 0.03 HDL-C score -0.04 -0.13, 0.06 0.44

TC score -0.12 -0.23, -0.02 0.02

Maternal scores effect 95% CI p-value TG score 0.07 -0.10, 0.24 0.42

LDL-C score -0.10 -0.25, 0.05 0.20 HDL-C score -0.04 -0.17, 0.09 0.52

TC score -0.10 -0.24, 0.04 0.16

Quartiles of LDL-C score, LDL-C plasma levels and IQ in ALSPAC children.

Conclusions• Scores have been developed which are strongly

related to lipid levels.• Maternal lipid scores are not associated with

offspring IQ• Offspring LDL score is associated with offspring IQ

such that an increase in score increases IQ

AcknowledgementsBristolLuisa Zuccolo OxfordCarolina Bonilla David SmithJean Golding Helga RefsumAndy NessYoav Ben-Shlomo AustraliaDavid Gunnell Marie-Jo BrionGeorge Davey Smith Craig PennellDebbie Lawlor Raine team and

participantsAmy TaylorPauline EmmettNic TimpsonBeate St. PourcainKate NorthstonePhil Lobb & Sue RingALSPAC team and participants