What this is
- This research investigates the causal relationship between maternal coffee consumption and various pregnancy outcomes, including miscarriage, stillbirth, birthweight, gestational age, and pre-term birth.
- Using Mendelian randomization (MR), the study analyzes genetic variants linked to coffee consumption to mitigate confounding factors present in traditional observational studies.
- The findings indicate that maternal coffee consumption does not significantly impact adverse pregnancy outcomes, although some associations with birthweight were noted.
Essence
- Maternal coffee consumption does not appear to causally influence miscarriage, stillbirth, or pre-term birth. While increased coffee intake is associated with higher birthweight, the effect size varies.
Key takeaways
- No significant causal relationship was found between maternal coffee consumption and miscarriage, stillbirth, or pre-term birth. This was consistent across analyses using both individual-level and summary-level data.
- Increased coffee consumption was associated with higher birthweight; however, the effect size was inconsistent across different analyses, suggesting variability in the impact of coffee on birthweight.
Caveats
- The study's findings may lack power to detect small causal effects due to the limited variance explained by the genetic risk scores for coffee consumption.
- The genetic instruments used were identified in non-pregnant populations, which may not fully capture the effects of coffee consumption during pregnancy.
AI simplified
Introduction
Coffee is the most consumed beverage worldwide, with an average daily intake of >400 mg of caffeine, equivalent to approximately four cups,per capita in many European countries, such as Norway, Sweden and Netherlands.Coffee is the primary source of caffeine in most populations, but other beverages such as tea and soft drinks, or foods such as chocolate, can also contribute to an individual’s total caffeine intake. 121 121 121 121 ,
It has been estimated that ∼70% of pregnant women in the USA consume caffeine during pregnancy, with coffee being the main source of caffeine.Several physiological changes occur during pregnancy that decrease the rate at which caffeine is metabolized—especially in the third trimester—due to decreased activity of the liver enzyme CYP1A.Therefore, caffeine can accumulate in the body throughout the pregnancy, with its half-life increasing from an average of 3 h for non-pregnant women to ≤18 h for pregnant women at the end of pregnancy.Since caffeine can freely cross the placenta and the fetus is unable to metabolize the molecule, the fetus is exposed to caffeine and its metabolites in proportion to levels consumed by the mother. 121 121 121 121 121 ,
The current World Health Organization guidelines recommend a caffeine intake of <300 mg/day during pregnancy,whereas the American College of Obstetricians and Gynecologists recommends a maximum caffeine intake of 200 mg/day,with one cup of coffee typically containing between 70 and 140 mg of caffeine depending on the type of bean and coffee, the degree of roasting and the serving size.A Scandinavian expert committee concluded that caffeine exposure during pregnancy should be limited due to a potential increased risk of spontaneous miscarriage and fetal growth restriction.However, they emphasized the lack of clear evidence of an association between coffee consumption and adverse outcomes, such as low birthweight,stillbirthsand miscarriages,as many of the prior observational studies failed to address the potential confounding effects of smoking and alcohol intake during pregnancy. 121 121 121 121 121 121 121 121 121 ,
Given it would be infeasible and/or unethical to randomize women to consume different levels of coffee during pregnancy, alternative methods for addressing the possible causal effect of maternal coffee consumption on pregnancy outcomes are needed. Mendelian randomization (MR), which uses genetic variants related to an exposure of interest as instruments for that exposure, is less prone to confounding by social, environmental and behavioural factors that affect many traditional observational epidemiological studies.In this paper, we use MR to investigate whether the observational relationships between coffee consumption and adverse pregnancy outcomes are causal. Specifically, we investigate the causal relationship between maternal coffee consumption and miscarriages, stillbirths, pre-term birth, gestational age and birthweight using two-sample summary data MR analysis.Additionally, we perform a one-sample genetic risk score (GRS) analysis using individual participant data from two large UK studies: UK Biobank (UKBB)and the Avon Longitudinal Study of Parents and Children (ALSPAC)(). 121 121 121 121 121 Figure 1 ,
Overview of data sources and phenotypes used in the two-sample MR, individual-level MR and traditional observational analysis. GRS, genetic risk score; MR, Mendelian randomization; UKBB, UK Biobank; ALSPAC, Avon Longitudinal Study of Parents and Children; EEG, Early Growth Genetics.
Methods
Genetic instruments were identified in a genome-wide association study (GWAS) of coffee consumptionfrom the Coffee and Caffeine Genetics Consortium. We used a two-sample MR approach to assess the effect of these genetic instruments of coffee consumption on pregnancy outcomes using GWAS of birthweight,miscarriages,stillbirths (; 10 December 2020, date last accessed) and gestational age and pre-term birth from 23andMe, Inc.Additionally, we performed an individual-level MR analysis using GRS for offspring birthweight, gestational age and pre-term birth in ALSPAC and for offspring birthweight, number of miscarriages and number of stillbirths in UKBB using linear regression. Finally, we assessed the association between coffee consumption in pregnancy and offspring birthweight, gestational age and risk of pre-term birth in ALSPAC using multivariable regression (for an overview, seeand, available asatonline). Detailed descriptions of these analyses are listed below in the sections ‘Two-sample MR analyses’, ‘Individual-level data MR analysis using GRS (ALSPAC and UKBB)’ and ‘Associations between self-reported coffee consumption and pregnancy outcomes in the ALSPAC study’. 121 121 121 121 Figure 1 Supplementary Table S1 Supplementary data http://www.nealelab.is/uk-biobank↗ IJE
Two-sample MR analyses
Data sources
Coffee consumption
Eight single-nucleotide polymorphisms (SNPs) that were significantly associated with coffee consumption were extracted from summary results data from a GWAS of coffee consumption published by the Coffee and Caffeine Genetics Consortium (= 91 462 individuals of European ancestry).The study reports a GWAS meta-analysis of coffee consumption (cups/day) across 28 population-based studies. The SNPs included in the MR analyses in this paper are listed in(available asatonline). N IJE 121 Supplementary Table S2 Supplementary data
Birthweight
Summary results statistics from the GWAS of birthweight were obtained from the Early Growth Genetics (EGG) consortium and UKBB meta-analysis (maternal-specific genetic effects on offspring birthweight) ().The birthweight GWAS included 297 356 individuals reporting their own birthweight and 210 248 females reporting the birthweight of their first offspring. A structural equation model was used to partition the genetic effects into maternal-specific and fetal-specific genetic effects on birthweight. Only the maternal-specific genetic effects on birthweight were used in the present study because the aim was to assess whether offspring birthweight was causally affected by maternal coffee consumption. https://egg-consortium.org/birth-weight-2019.html↗ 121
Miscarriage
Summary results statistics for sporadic miscarriages were obtained from Laisk,who performed GWAS on 49 996 females of European ancestry who self-reported miscarriages and 174 109 female controls. Sporadic miscarriage was defined as one or two (self-reported) miscarriages. et al. 121
Stillbirth
Summary results statistics from the GWAS of the number of self-reported stillbirths were obtained from publicly available summary statistics published by the Neale lab (; 10 December 2020, date last accessed) that involved 60 453 women from UKBB who reported the number of stillbirths they experienced. http://www.nealelab.is/uk-biobank↗
Gestational age and pre-term birth
GWAS summary results statistics for gestational age and pre-term birth (defined as a birth at <37 weeks of gestation) were obtained from 23andMe. These results contributed to the discovery data set for a previously published GWAS by ZhangThe original GWAS analysis included 43 568 unrelated women of European ancestry from 23andMe who self-reported length of gestation (in weeks) from their first live singleton birth. The full GWAS summary statistics for the 23andMe discovery data set were made available through 23andMe, under an agreement that protects the privacy of the 23andMe participants. Please visitfor more information and to apply to access the data. et al. 121 https://research.23andme.com/collaborate/#dataset-access/↗
Genetic instruments
We selected three sets of SNPs to instrument coffee intake (, available asatonline). First, we selected eight independent SNPs that reached genome-wide significance in the GWAS of coffee consumption mentioned above (rs1481012, rs4410790, rs9902453, rs7800944, rs17685, rs2472297, rs1260326 and rs6265).Second, we kept six out of the eight SNPs for which there was no strong evidence of horizontal pleiotropy based on the results of a phenome-wide scan using PhenoScannerat a-value threshold of 5 × 10. Thevariant (rs1260326) was associated with alcohol intake frequency, type 2 diabetes, fasting glucose levels, triglycerides levels, cardiovascular disease risk factors and chronic kidney disease, and the SNP rs6265 in thegene had previously been associated with several traits, including body mass index, weight, body fat and smoking (, available asatonline). Supplementary Table S2 Supplementary data 121 121 121 Supplementary Table S2 Supplementary data IJE P GCKR BDNF IJE , − 8
Further, we tested the association between each of the eight SNPs and alcohol consumption and smoking behaviour in 194 196 UKBB women who reported either offspring birthweight (= 173 259), having a stillbirth (= 78 867) or having a spontaneous miscarriage (= 78 681). Five out of eight SNPs were associated with either alcohol consumption or smoking behaviour, after Bonferroni correction (<0.05/8 = 0.00625) (, available asatonline). Therefore, for the third subset of SNPs, we kept three out of the eight SNPs (rs1481012, rs4410790 and rs9902453), which were not associated with smoking or alcohol intake in UKBB and had no strong evidence for pleiotropy in the above search in PhenoScanner(, available asatonline). This analysis was performed because the GRS for coffee consumption had been found to be associated with smoking behaviour and alcohol intake frequency in UKBB previously—two phenotypes strongly associated with adverse pregnancy outcomes. N N N P IJE IJE Supplementary Table S3 Supplementary data 121 121 Supplementary Table S2 Supplementary data 121 ,
Statistical analysis
A two-sample inverse-variance weighted (IVW) MR analysis was conducted using either eight, six or three genetic variants as explained above to estimate the causal relationship of maternal coffee consumption on sporadic miscarriages, stillbirths, gestational age, pre-term birth and offspring birthweight. All SNPs and their effect sizes for these phenotypes are listed in(available asatonline). A heterogeneity test of causal effect estimates was conducted across each of the variants for coffee consumption using Cochran’s Q. Directional pleiotropy was tested by assessing whether the MR–Egger intercept was different from zero. Sensitivity analyses were performed using MR–Egger regression, weighted median, weighted and simple mode estimation approaches.was calculated to assess the potential for weak instrument bias in the MR–Egger regression analyses. However, due to the small number of variants used in the MR analyses, the ability of these methods to control for latent horizontal pleiotropy is likely to be limited. Bonferroni correction of-value for the two-sample MR analysis was 0.05/5 phenotypes tested = 0.01. Supplementary Table S4 Supplementary data IJE I P 2 GX
A flowchart showing the number of individuals included in each analysis is presented in(available asatonline). The TwoSampleMR packagein R version 3.4.3 was used to perform all the analyses. Supplementary Figure S1 Supplementary data 121 IJE
Individual-level data MR analysis using GRS (ALSPAC and UKBB)
Data sources
UKBB
UKBB is a large prospective population-based cohort containing ∼500 000 individuals (aged 37–73 years; 54% female), representing 5.5% of those who were invited to participate.UKBB has ethical approval from the North West Multi-Centre Research Ethics Committee, which covers the UK, and all participants provided written informed consent. Up to 251 058 European women were available for analysis; a detailed description of the selection of these women is provided in(available asatonline). 121 Supplementary Note S1 Supplementary data IJE
We removed UKBB women of European ancestry who did not report any information on their offspring's birthweight (= 20 937), stillbirth (= 115 329) or spontaneous miscarriage (= 115 329), leaving a total of 194 196 women who reported any of the birth outcomes along with their coffee consumption habit. The exact number of women who reported the number of stillbirths and miscarriages was 78 867 and 78 681, respectively. Out of the 194 196 women, only a total of 155 242 reported both their offspring birthweight and maternal age at birth. A detailed description of the phenotypes is presented in(available asatonline). N N N IJE Supplementary Note S2 Supplementary data
ALSPAC
ALSPAC is a longitudinal birth cohort study established to understand how genetic and environmental factors influence health and development in parents and children.A detailed description of the cohort is presented in(available asatonline). 121 121 Supplementary Note S3 Supplementary data , IJE
Birthweight data for each child in the study were ascertained in a variety of ways, including from obstetric data records, recorded by the ALSPAC research team and from the birth notification service in the UK. If the birthweight measurement for an individual differed across the different sources (defined as >100 grams difference), the individual was excluded from the analysis. When the differences were ≤100 grams, the lower birthweight measurement was used. In the study, we excluded offspring with birthweight of <2.5 or >4.5 kg to match the inclusion criterion of the GWAS of birthweight. Gestational age was defined as the weeks of gestation based on the final clinically estimated date of delivery and pre-term birth was defined as <37 weeks of gestation.
Maternal coffee intake was defined as cups of coffee per day at Week 32 of pregnancy (= 6845). Maternal alcohol drinking was derived from units of alcohol drunk per week at Week 32 of pregnancy, with those who did not drink defined as non-drinkers (= 2826) and the rest defined as drinkers (= 1525). Maternal smoking was derived from the smoking behaviour in the first trimester of pregnancy, with those smoking any types of tobacco defined as smokers (= 1620) and the rest defined as non-smokers (= 5702). These data were all collected from a self-reported questionnaire. N N N N N
Genetic instruments
We constructed both weighted and unweighted GRSs using the three sets of SNPs mentioned above in the two-sample MR section, resulting in six different GRSs for coffee consumption. We constructed an unweighted GRS as the regression coefficients obtained in the coffee consumption GWAS might not accurately reflect effect sizes during pregnancy. These GRSs explained approximately 1.94% (eight SNPs), 1.81% (six SNPs) and 0.82% (three SNPs) of the variance in coffee drinking according to the effect estimates reported in the exposure GWAS.The weighted risk scores were created by multiplying the coffee-increasing alleles by their effect size from the exposure GWAS whereas the unweighted risk scores were obtained by adding the number of coffee consumption increasing alleles together. 121 121
Statistical analysis
UKBB
Linear regression analyses were used to assess the relationship between the maternal GRSs for coffee consumption (both weighted and unweighted) and self-reported and maternal coffee intake (cups/day) in addition to the number of spontaneous miscarriages, number of stillbirths and birthweight of the first offspring, adjusting for maternal age at birth (to reflect age at the pregnancy in question). Maternal age was only available for birthweight as women only reported their total number of miscarriages or stillbirths. To match the inclusion criterion of the GWAS of birthweight, if the birthweight of the offspring was <5 or >10 pounds (equivalent to approximately <2.3 or >4.5 kg), the value was excluded from the analysis. Gestational age and pre-term birth were only available in a small subsample in UKBB, and therefore were not included in the analyses here.
Additionally, we utilized the mother–offspring pairs in UKBB and performed a linear regression of maternal unweighted GRS for coffee consumption and offspring own-reported birthweight, adjusting for offspring unweighted GRS. We also conducted linear regression of maternal weighted GRS for coffee consumption on offspring own-reported birthweight, adjusting for individual offspring SNPs.A description of the mother–offspring pairs is provided in(available asatonline). 121 Supplementary Note S4 Supplementary data IJE
One of the strongest instruments for coffee intake, rs2472297, is also associated with smoking in the female population in UKBB (=0.002;, available asatonline). Therefore, when we identified an association between the GRSs of coffee intake that incorporated rs2472297 (i.e. the two sets of GRSs constructed using eight and six SNPs) and pregnancy outcomes, we conducted sensitivity analysis by including smoking as a covariate in the analysis. P IJE Supplementary Table S3 Supplementary data
A flowchart showing the number of individuals included in each analysis is presented in(available asatonline). Supplementary Figure S2 Supplementary data IJE
ALSPAC
We constructed three weighted and three unweighted GRSs of coffee consumption as described above. We examined the association between the GRSs and maternal coffee intake (cups/day) at Week 32 during pregnancy in 6845 mothers. We also examined the association between the GRSs of coffee consumption and three pregnancy outcomes, with (= 4510 for birthweight and 2989 for gestational age and pre-term birth) and without (= 3714 for birthweight and 2456 for gestational age and pre-term birth) conditioning on child’s unweighted GRS or individual SNPs as described above in UKBB analysis. Covariates included maternal age at the last menstrual period and child’s sex. Analyses were conducted using a linear regression model for birthweight and gestational age, and a logistic regression model for pre-term birth. N N
Lastly, we examined the association between maternal GRS quartiles and two potential confounders (alcohol drinking and smoking during pregnancy) using chi-square tests.
A flowchart showing the number of individuals included in each analysis is presented in(available asatonline). Supplementary Figure S3 Supplementary data IJE
Associations between self-reported coffee consumption and pregnancy outcomes in the ALSPAC study
To explore whether previous observational associations regarding coffee consumption and birthweight were applicable to ALSPAC, we examined the phenotypic association between maternal coffee intake at Week 32 during pregnancy and the three pregnancy outcomes, including (i) child’s birthweight in 4302 mother–child pairs, (ii) weeks of gestation at birth, pre-determined by ALSPAC using estimated due date and (iii) pre-term birth defined as <37 weeks of gestation in 4231 mother–child pairs. We did not include analysis of stillbirth and miscarriage as we did not have information regarding coffee consumption in the pregnancy that ended in these outcomes.
Analyses were conducted using a linear regression model for birthweight and gestational age, and a logistic regression model for pre-term birth. Covariates included maternal age at the last menstrual period, maternal smoking and alcohol intake during pregnancy, and child’s sex.
A flowchart showing the number of individuals included in each analysis is presented in(available asatonline). All statistical analyses were performed in R (version 3.4.3). Supplementary Figure S3 Supplementary data IJE
Power calculations
We were interested in the statistical power of our study to detect the causal effect of maternal coffee consumption on different pregnancy outcomes. For the unconditional GRS analyses, we used the ‘Genetic Power Calculator’ ().For risk of miscarriage, we used the MR power calculator ().For all calculations, we assumed the absence of offspring genetic effects and a Type 1 error rate of α = 0.05 (the presence/absence of offspring genetic effects has little influence on the power to detect maternal genetic effects so long as the proportion of variance explained by the offspring genetic effects is small). We estimated the size of the causal effect on each pregnancy-related outcome that we had 80% power to detect in our analyses. We note that whilst these asymptotic calculations assume slightly different underlying models/tests to those employed in our manuscript (e.g. the genetic power calculators test for association between the GRS and outcome assuming a certain size of causal effect of the exposure on the outcome; the MR power calculator uses a one-sample MR model to estimate asymptotic power, whereas we use two-sample MR in our study, etc.), the calculations should be good enough to provide useful approximations of the power of our study. https://zzz.bwh.harvard.edu/gpc/qtlassoc.html↗ https://shiny.cnsgenomics.com/mRnd/↗ 121 121 121
Results
Two-sample MR analyses
We observed an effect of increased maternal coffee consumption on increased offspring birthweight in the analysis including six SNPs (IVW Effect size: 0.050 SD birthweight per cup of coffee/day, 95% CI: 0.001, 0.098) (;, available asatonline). The effects estimated using eight or three SNPs were in the same direction, but the effect sizes were attenuated. The 95% CIs around the estimated causal effect on the number of reported stillbirths, risk of sporadic miscarriages, pre-term births and gestational age overlapped zero. SNPs included in the GRSs are listed in(available asatonline). Figure 2 Supplementary Table S5 Supplementary data Supplementary Table S2 Supplementary data IJE IJE
Sensitivity analyses using MR–Egger, Weighted Median, Simple and Weighted Mode showed similar trends to the IVW estimates (, available asatonline); however, due to the small number of variants, these sensitivity analyses are included only for completeness. The IVW heterogeneity analysis showed no strong evidence of heterogeneity in the causal effect estimate for coffee consumption on number of stillbirths, spontaneous miscarriage, gestational age, pre-term birth or birthweight, except for the birthweight analysis with all eight SNPs included in the model (heterogeneity analysis for IVW had a=0.004). Likewise, we found no evidence of directional pleiotropy for maternal coffee consumption on any of the outcomes (, available asatonline). However, due to the small number of variants in the analysis, the MR–Egger results should be carefully interpreted. Theof >0.99 was observed for all analyses, well above the cut-off of 0.9, which suggests little influence of weak instrument bias in the MR–Egger analysis. Supplementary Table S5 Supplementary data Supplementary Table S6 Supplementary data IJE P IJE I 2 GX
![Overview of the effect estimates and 95% confidence intervals for the two-sample MR analyses per cup of coffee consumed per day. We did not observe a causal effect of coffee consumption (defined as cups per day) on any of the traits analyzed, although there was evidence of a small effect of increased coffee consumption on increased birthweight for the six SNP analysis. Birthweight was analyzed as a Z-score.Miscarriages were defined as the log(OR) of having had one to two spontaneous miscarriages [case control Genome Wide Association Study (GWAS)],stillbirth was defined as the number of self-reported incidents of stillbirth [(10 December 2020, date last accessed)], gestational age was defined as self-reported length of gestation and preterm birth was defined as the log(OR) (log of the odds ratio) of having any birth at <37 weeks of gestation (case control GWAS).The units for the causal effects are per unit increase in the outcome per extra cup of coffee consumed per day. MR, Mendelian randomization; SNP, single nucleotide polymorphism. [121] [121] [121] http://www.nealelab.is/uk-biobank](https://europepmc.org/articles/PMC9908064/bin/dyac121f2.jpg.jpg "Click to view full size")
Overview of the effect estimates and 95% confidence intervals for the two-sample MR analyses per cup of coffee consumed per day. We did not observe a causal effect of coffee consumption (defined as cups per day) on any of the traits analyzed, although there was evidence of a small effect of increased coffee consumption on increased birthweight for the six SNP analysis. Birthweight was analyzed as a Z-score.Miscarriages were defined as the log(OR) of having had one to two spontaneous miscarriages [case control Genome Wide Association Study (GWAS)],stillbirth was defined as the number of self-reported incidents of stillbirth [(10 December 2020, date last accessed)], gestational age was defined as self-reported length of gestation and preterm birth was defined as the log(OR) (log of the odds ratio) of having any birth at <37 weeks of gestation (case control GWAS).The units for the causal effects are per unit increase in the outcome per extra cup of coffee consumed per day. MR, Mendelian randomization; SNP, single nucleotide polymorphism. [121] [121] [121] http://www.nealelab.is/uk-biobank
Individual-level data MR analysis using GRS (ALSPAC and UKBB)
UKBB
An overview of the demographics of the women included in the different analyses can be found in(available asatonline). The GRSs for coffee consumption were tested for association with the number of cups of coffee consumed per day in all female individuals from UKBB. An association was found between coffee consumption (= 194 196) and the different unweighted and weighted GRSs (, available asatonline), suggesting that having more coffee consumption increasing alleles identified in the sample of both men and women would increase a woman’s coffee consumption. Supplementary Table S7 Supplementary data Supplementary Table S8 Supplementary data IJE N IJE
GRS analyses for the number of miscarriages (= 78 681) and stillbirths (= 78 867) showed no association (). An association between increased genetically predicted maternal coffee consumption and increased offspring birthweight was observed in the analyses using GRSs of eight and six variants (= 155 242) () and the association remained after adjusting for smoking in the sensitivity analysis. N N N Table 1 Table 1
| Unweighted GRS | Weighted GRS | ||||||
|---|---|---|---|---|---|---|---|
| Effect estimate | SE | P | Effect estimate | SE | P | ||
| Miscarriage | 8 SNPs | −0.003 | 0.002 | 0.225 | −0.030 | 0.025 | 0.224 |
| 6 SNPs | −0.004 | 0.003 | 0.079 | −0.037 | 0.025 | 0.145 | |
| 3 SNPs | −0.001 | 0.004 | 0.699 | −0.017 | 0.037 | 0.651 | |
| Stillbirth | 8 SNPs | 0.001 | 0.001 | 0.439 | −0.001 | 0.008 | 0.894 |
| 6 SNPs | 0 | 0.001 | 0.807 | −0.003 | 0.008 | 0.684 | |
| 3 SNPs | 0.002 | 0.002 | 0.106 | 0.007 | 0.012 | 0.59 | |
| Birthweight | 8 SNPs | 2.64 | 0.698 | 1 × 10−4 | 30.5 | 7.98 | 1 × 10−4 |
| 6 SNPs | 2.21 | 0.805 | 0.006 | 27.4 | 8.25 | 9 × 10−4 | |
| 3 SNPs | 1.66 | 1.15 | 0.148 | 20.09 | 12 | 0.094 | |
| Birthweight adjusted for smoking | 8 SNPs | 2.64 | 0.699 | 1 × 10−4 | 30.4 | 7.99 | 1 × 10−4 |
| 6 SNPs | 2.24 | 0.806 | 0.006 | 27.3 | 8.26 | 9 × 10−4 | |
Mother–offspring pairs
To investigate whether adjusting for offspring GRS would impact our results, we performed the GRS analysis in a subset of individuals in UKBB where both maternal and offspring GRSs were available in addition to offspring self-reported birthweight (= 2965). We were not able to detect any associations between maternal GRS and offspring birthweight in these analyses (). N Table 2
| Unweighted GRS | Weighted GRS | ||||||
|---|---|---|---|---|---|---|---|
| Effect estimate | SE | P | Effect estimate | SE | P | ||
| Maternal GRS | 8 SNPs | 2.44 | 4.24 | 0.565 | 50.4 | 49 | 0.303 |
| 6 SNPs | 4.81 | 4.94 | 0.33 | 61.9 | 50.9 | 0.224 | |
| 3 SNPs | −0.996 | 6.96 | 0.886 | 19.9 | 73.5 | 0.787 | |
| Maternal GRS, adjusting for offspring genetics | 8 SNPs | 5.32 | 4.96 | 0.284 | 94.1 | 57.4 | 0.101 |
| 6 SNPs | 8.34 | 5.74 | 0.146 | 106 | 59.3 | 0.072 | |
| 3 SNPs | 1.8 | 7.99 | 0.822 | 53.4 | 84.8 | 0.529 | |
ALSPAC
An overview of the demographics of the women included in the different analyses can be found in(available asatonline). Both weighted and unweighted GRSs of maternal coffee consumption constructed using either eight or six SNPs were associated with increased coffee consumption at Week 32 of gestation (GRSs of eight and six SNPs account for 0.37% and 0.35% of the variance, respectively;0.003) in ALSPAC. Although the GRSs constructed using three SNPs were not associated with coffee consumption during pregnancy (0.457), the direction of the effect was the same but the effect size was attenuated. There was no association between any of the GRSs and maternal alcohol drinking or smoking during pregnancy (>0.05;, available asatonline). Supplementary Table S7 Supplementary data Supplementary Table S9 Supplementary data IJE P ≤ P P IJE ≥
Maternal GRSs were not associated with offspring birthweight, gestational age or pre-term birth either with or without adjusting for offspring genetics (). We also conducted the analysis without adjusting for offspring sex to mimic the conditions in UKBB and obtained similar results (, available asatonline). Table 3 Supplementary Table S10 Supplementary data IJE
| Unweighted GRS | Weighted GRS | ||||||
|---|---|---|---|---|---|---|---|
| Effect estimate | SE | P | Effect estimate | SE | P | ||
| Birthweight, maternal GRS (= 4510)N | 8 SNPs | 0.134 | 3.505 | 0.97 | −14.200 | 39.852 | 0.72 |
| 6 SNPs | 0.567 | 4.064 | 0.889 | −13.238 | 41.128 | 0.748 | |
| 3 SNPs | 0.121 | 5.844 | 0.984 | −34.437 | 60.867 | 0.572 | |
| Birthweight, maternal GRS, adjusting for offspring genetics (= 3714)N | 8 SNPs | −4.337 | 4.537 | 0.339 | −55.786 | 51.429 | 0.278 |
| 6 SNPs | −3.597 | 5.269 | 0.495 | −51.005 | 53.559 | 0.341 | |
| 3 SNPs | −1.549 | 7.563 | 0.833 | −59.261 | 78.71 | 0.452 | |
| Gestational age, maternal GRS (= 2989)N | 8 SNPs | 0.014 | 0.02 | 0.464 | 0.216 | 0.222 | 0.331 |
| 6 SNPs | 0.011 | 0.023 | 0.625 | 0.184 | 0.231 | 0.426 | |
| 3 SNPs | −0.005 | 0.033 | 0.885 | −0.099 | 0.343 | 0.774 | |
| Gestational age, maternal GRS, adjusting for offspring genetics (= 2456)N | 8 SNPs | 0.007 | 0.025 | 0.769 | 0.012 | 0.282 | 0.966 |
| 6 SNPs | 0.005 | 0.029 | 0.873 | 0.007 | 0.295 | 0.981 | |
| 3 SNPs | −0.025 | 0.042 | 0.556 | −0.427 | 0.434 | 0.326 | |
| Pre-term birth, maternal GRS (= 2989) N 4 | 8 SNPs | −0.021 | 0.039 | 0.588 | −0.537 | 0.445 | 0.228 |
| 6 SNPs | −0.008 | 0.046 | 0.867 | −0.4804 | 0.4645 | 0.301 | |
| 3 SNPs | 0.031 | 0.066 | 0.641 | 0.021 | 0.686 | 0.976 | |
| Pre-term birth, maternal GRS, adjusting for offspring genetics (= 2456) N 4 | 8 SNPs | 0.004 | 0.052 | 0.945 | 0.124 | 0.598 | 0.836 |
| 6 SNPs | −0.009 | 0.061 | 0.88 | 0.031 | 0.625 | 0.961 | |
| 3 SNPs | 0.045 | 0.088 | 0.609 | 1.107 | 0.921 | 0.229 | |
Associations between self-reported coffee consumption and pregnancy outcomes in the ALSPAC study
A higher maternal self-reported coffee consumption at Week 32 of gestation was associated with a lower offspring birthweight (effect size: −9.63 grams/cups, 95% CI: −17.57, −1.69); however, when adjusting for smoking and alcohol consumption during pregnancy, this association attenuated and the 95% CI overlapped zero (effect size: −4.52 grams/cups, 95% CI: −14.92, 5.88). Maternal self-reported coffee consumption was associated with a higher risk of pre-term birth both when adjusting for smoking and alcohol (odds ratio: 1.11, 95% CI: 1.03, 1.19) and without adjustment (odds ratio: 1.08, 95% CI: 1.02, 1.14). We found a similar trend for the association with gestational age with (effect size: −0.03 weeks/cups, 95% CI: −0.07, 0.01) or without adjusting for alcohol and smoking during pregnancy (effect size: −0.02 weeks/cups, 95% CI: −0.05, 0.01), even though their 95% CI overlapped zero.
Power calculations
(available asatonline) shows approximate causal effect sizes that we have 80% power to detect in our main analyses. As expected, the power to rule out negative effects depends critically on the amount of variance explained in coffee consumption by our GRSs. In the case of conservative estimates for the amount of variance explained in coffee consumption (i.e. a GRS explaining 0.37% of the variance), we are well powered to rule out moderate to large adverse effects of coffee consumption on many pregnancy-related outcomes (i.e. coffee consumption explaining a 26.9-gram difference in offspring birthweight per cup of coffee per day, 0.03 more still births per cup of coffee per day, 0.09 more miscarriages per cup of coffee per day or 0.11 increased log odds of miscarriage per cup of coffee per day). Indeed, if the GRS explains a more substantial proportion of the variance in coffee consumption (i.e. ∼2%), then we are well powered to rule out the action of much smaller negative causal effects (, available asatonline). Supplementary Table S11 Supplementary data Supplementary Table S11 Supplementary data IJE IJE
Discussion
We did not find strong evidence for a causal effect of maternal coffee consumption on the number of miscarriages, stillbirths, gestational age or pre-term birth in any of our analyses conducted using either individual-level data from UKBB and ALSPAC or summary level data from 23andMe or publicly available sources. We constructed three sets of genetic instruments associated with coffee consumption in men and non-pregnant women, and found that the instruments containing either eight or six SNPs were associated with self-reported coffee consumption at Week 32 of pregnancy and therefore potentially good instruments for coffee consumption during pregnancy. Our results suggest previously reported observational associationsmay be driven by confounding factors such as smoking and alcohol consumption during pregnancy.Our results are consistent with a recent MR analysis by Yuanusing data from UKBB and the FinnGen Consortium, who also found no strong causal relationship between caffeine consumption and pregnancy loss.Yuandefine ‘pregnancy loss’ as a history of having stillbirth, spontaneous miscarriage or termination (UKBB), or as spontaneous abortion as classified by the International Classification of Diseases 8th to 10th codes (FinnGen). Our results confirm and extend the findings of Yuanby investigating the effect of coffee consumption on risk of miscarriage and number of stillbirths, and by examining the relationship between coffee consumption and other perinatal outcomes (i.e. birthweight, gestational age, pre-term birth). 121 121 121 121 et al. et al. et al.
We were able to replicate the previously reported observational association between increased maternal coffee consumption during pregnancy and decreased offspring birthweight/higher risk of pre-term birth in ALSPAC, when we did not include smoking and alcohol as covariates. However, the observational association with birthweight attenuated when these variables were adjusted for, suggesting these factors may have confounded previous observational relationships. Additionally, our MR analyses did not support a causal relationship for either outcome.
We showed a causal effect of increased maternal coffee consumption on increased offspring birthweight in the analyses using either eight (both two-sample and one-sample analysis) or six SNPs (only one-sample analysis). The effect sizes were ∼0.05 for the two-sample MR, corresponding to an increase in birthweight of ∼24 grams per cup of coffee drunk each day, whereas the one-sample analysis showed effect sizes corresponding to an increase of ∼16 grams per coffee-drinking-increasing allele. In the analysis with eight SNPs, we included variants in both thegene and thegene known to be associated with type 2 diabetes and body mass index (BMI) and smoking behaviour, respectively. As expected, the SNP in thegene was associated with birthweight, whereas the SNP in thegene was associated with both smoking, BMI and birthweight. Both these variants were removed from the analysis when using either six or three SNPs. The inconsistency could also be partly due to the variance in coffee consumption explained by the three SNP GRSs being less than half that explained by the six or eight SNP GRSs. Additionally, these results could not be replicated in the mother–offspring analysis either in UKBB or in ALSPAC, although this could be due to the effect being too small to be detected with the substantially smaller sample size. Overall, the birthweight results should be interpreted with caution, as the results from the eight and six SNP GRSs could reflect pleiotropy from smoking or alcohol-related SNPs. GCKR BDNF GCKR BDNF
One of the strengths of our two-sample MR analysis is the use of partitioned maternal genetic effects from a birthweight GWAS, allowing estimation of the maternal genetic effects on birthweight conditional on the fetal genome and using these effects in the two-sample MR, as well as using conditional mother–offspring analysis for the GRS analysis of birthweight (UKBB and ALSPAC) and analysis of gestational age and pre-term birth (ALSPAC). However, these partitioned genetic effects are naturally not available for stillbirths and miscarriages, and we therefore could not perform MR analysis of the maternal contribution alone.
Our approach has a number of limitations, which we discuss in the remaining paragraphs. First, the SNPs used in the analysis were identified in the general population rather than in pregnant women. Nevertheless, we have shown that GRSs consisting of these SNPs are associated with coffee consumption in women during pregnancy, although the effect sizes on coffee consumption appear to be smaller in pregnant women in ALSPAC compared with non-pregnant women in UKBB. This could be a consequence of women drinking fewer cups of coffee during their pregnancy and genetic effect sizes being proportionally reduced.
Ideally, our genetic instruments would explain as much variance as possible in coffee consumption in order to have maximum power to estimate any causal effects of maternal coffee drinking on pregnancy-related outcomes. Variance in coffee consumption explained by the three SNPs was 0.82%, which is much lower than that explained by six SNPs [1.81% (or 0.35% in pregnant women in ALSPAC)] or eight SNPs [1.94% (or 0.37% in pregnant women in ALSPAC)]. The GRSs with three SNPs were associated with coffee consumption in non-pregnant women in the large UKBB but not with coffee consumption at gestational Week 32 in the smaller ALSPAC cohort, suggesting they are not good instruments for the ALSPAC analysis. However, analysis in ALSPAC showed that the GRS constructed of six SNPs had a strong association with coffee consumption in pregnancy (explaining 0.35% of the variance) and no association with drinking alcohol or smoking in pregnancy, suggesting that this set of SNPs might be the most appropriate to use in the context of coffee consumption in pregnancy.
Power analyses suggest that UKBB has 80% power (alpha = 0.05) to detect an effect of maternal coffee consumption equivalent to a 26.9-gram difference in birthweight per cup of coffee per day, or 0.03 more stillbirths per cup of coffee per day, if we conservatively assume that the coffee GRSs explain ∼0.37% of the variance in maternal coffee consumption during pregnancy. These asymptotic results and others suggest we have adequate power in our analyses to exclude large adverse effects of maternal coffee consumption on many pregnancy-related outcomes, but not the smaller magnitude causal effects that were estimated in this study. In this context it is also important to note we do not find strong evidence of an inverse effect of coffee consumption on birthweight, as previous observational analyses have suggested. Given our power calculations, we can be confident that a large inverse effect of coffee consumption on offspring birthweight is unlikely.
Lastly, although coffee is the primary source of caffeine in most populationsother beverages such as tea and soft drinks or foods such as chocolate could contribute to an individual’s total caffeine intake. Importantly, GRS for coffee consumption have previously been shown to explain some of the variation in tea consumption in UKBB.This suggests that the effect observed could be partially mediated through tea consumption as well as coffee consumption, potentially through caffeine, as the GRSs were only associated with standard tea drinking, but not other forms of tea such as herbal or green tea.More detailed phenotyping in regard to actual caffeine consumption may be needed to properly proxy total caffeine intake. 121 121 121 121 ,
Taken together we have performed a large MR study using three different sets of genetic variants in three cohorts, with additional publicly available data. All of our analyses show similar results and suggest that if any adverse effects of coffee consumption on birthweight, gestational age, pre-term birth, stillbirth or miscarriage exists, it is likely to be small.
Conclusion
In conclusion, we did not find strong evidence of an adverse causal relationship between maternal coffee consumption and birthweight, gestational age, the number of stillbirths, risk of pre-term birth or spontaneous miscarriages. Notably, a null association may be due to a lack of power and needs to be validated in even larger studies.
Ethics approval
This project received ethical approval from the Institutional Human Research Ethics committee, University of Queensland (Approval Number 2019002705).