ART and mQTLs

Can assisted reproductive technologies (ART) impact the association between DNA methylation and genotypes?

Julia Romanowska

2026-05-22

Photo by Zoltan Tasi on Unsplash

OUTLINE

Background

DNAm and mQTLs

DNAm reprogramming and ART

methods

some results

discussion

DNA methylation (DNAm)

• CpG - most popular methylation site

• non-uniformly distributed

• ‘marker’ for gene activation

• DNA methylation is a methyl-group attached to DNA;
• in humans, methylation occurs primarily on a cytosine in a CpG dinucleotide;
• these CpGs are not uniformly distributed on the genome and the methylation level marks whether the genes are activated or disactivated

DNAm reprogramming

wave of demethylation followed by rapid re-methylation

• paternal DNA is demethylated faster than maternal

• at blastocyst stage (days 5–9), DNAm level is lowest (ca.20% of all CpGs)

• freezing of eggs could influence de-methylation and differentiation

[src: Albert and Greenberg, Development (2023); Greenberg, Front.Cell Develop.Biol. (2021)]

Blastocyst = naive pluripotency stage; only imprinted regions and transpose elements (TEs) are methylated! days 5–9

However, a lot of signalling is also done by methylation of histones!

DNAm reprogramming & ART

ART = assisted reproductive technology

More and more people postpone the creation of family and thus often when they decide on having a child it might be too late.

ART were in use since 1980’s and are well tested and safe procedure. However, we still don’t know details on how it’s affecting the biological processes that take place in the first days of embryonic development.

The ART procedures require, namely, extracting the sperm and eggs, and growing the fertilized embryos on a petri dish in 3-5 days. After that, some embryos are frozen and the best is transferred to the woman.

DNAm reprogramming & ART

ART = assisted reproductive technology

src: Guo, H., et al. (2014). Nature, 511(7511), 606–610.

The problem is that it’s actually during this time, around days 3-5, the most vurnerable period occurs. The DNA methylation on the entire genome is wiped out

mQTL

methylation quantitative locus

SNP–CpG pair:

association between
methylation status of that CpG and
allele at that SNP

• cis-mQTLs and trans-mQTLs

• mQTL is a methylation quantitative locus;

• an association between SNP and CpG where the change in the alleles on the SNP are correlated with the change in methylation status of CpG;

• this association indicates the functional connection between alleles and methylation, which in reality is probably mediated through various proteins;

• depending on the distance between the SNP and CpG in an mQTL, we have cis and trans;

mQTL

methylation quantitative locus

SNP–CpG pair:

association between
methylation status of that CpG and
allele at that SNP

• cis-mQTLs and trans-mQTLs

• today, I’ll focus only on cis, defined such that the distance is less than 100 kb

Aim of the study

do ART procedures influence re-methylation processes?

(using mQTLs as tool)

mQTL: No

ART-effect: No

mQTL: No

ART-effect: No

mQTL: No

ART-effect: Yes

mQTL: Yes

ART-effect: No

mQTL: Yes

ART-effect: Yes (after fertilization?)

mQTL: Yes

ART-effect: Yes

PoO ART-effect: Yes (maternal line?)

mQTL: Yes

ART-effect: Yes

PoO ART-effect: Yes (paternal line?)

methods

data

• genetic data - MoBa

• DNAm data - START

methods

tools

• PLINK

• R

• JuliaLang

• snakemake

methods

1. prepare SNP-CpG list:

   1. SNP–CpG distance of max.100 kbp
   2. SNP–CpG distance of min. 50 bp
   3. min.10 persons with DD of minor allele
   4. DNAm range > 0.1

2. agnostic search for mQTLs in parents

3. select the significant mQTLs (FDR < 0.1)

4. remove outliers

5. check for association in children

methods – models

5. check for association in children

• quantile regression[*]

• using three models:

  • DNAm \(\sim\) SNP \(+\) case_control_status
  • DNAm \(\sim\) SNP \(\times\) case_control_status
  • DNAm \(\sim\) SNP, stratified by case_control_status

[*] comparing medians instead of means!

Results

mQTLs in parents

How many tests and cis-mQTL pairs?

Timing of runs

How many cis-mQTL pairs?

• significance: FDR \(< 0.1\) per chromosome

Results

mQTLs in children

How many cis-mQTL pairs?

model: additive

Volcano plot for additive model, using DNAm data from children.

Out of 969 associations checked with the additive model, 399 (41.0%) had p-values that were lower than the stringent Bonferroni correction threshold (5.159958720330238e-5).

model: interaction

Volcano plot for interaction model, using DNAm data from children.

Out of 969 associations checked with the additive model, 520 (54.0%) had p-values that were lower than the stringent Bonferroni correction threshold (5.159958720330238e-5).

model: stratified

‘ART’ group

‘non-ART’ group

ART stratum:

Out of 969 associations checked with the stratified model, 300 (31.0%) had p-values lower than the stringent Bonferroni correction threshold (5.159958720330238e-5).

nonART stratum:

Out of 969 associations checked with the stratified model, 493 (51.0%) had p-values lower than the stringent Bonferroni correction threshold (5.159958720330238e-5).

Stratification model

Stratification model

“Interesting” findings

CpG	nrow
InlineStrings.String15	Int64
cg17612569	41
cg20741235	32
cg21139150	17
cg22706458	1

each SNP appear only in one mQTL

Example – genomic location

Example

Example