Parent-Of-Origin Effects And Genomic Imprinting

Why Allele Origin Matters

At a biallelic locus, an offspring with one copy of an allele may have inherited it from the mother or from the father. If the two copies have different effects depending on their parental source, the usual genotype dosage does not retain enough information to represent that difference. This is the statistical idea behind a parent-of-origin effect. Genomic imprinting is a biological mechanism in which expression depends on parental origin; a statistical parent-of-origin contrast can be consistent with imprinting, but does not by itself establish the molecular mechanism.

The key comparison is between offspring with the same heterozygous genotype but reciprocal inheritance: one received the allele of interest from the mother, and the other received it from the father. Ordinary dominance does not distinguish these two cases. Identifying the cases generally requires parental genotypes, and sometimes additional transmission or phase information.

A Model That Keeps the Parental Copies Separate

Consider one biallelic locus and label the allele of interest as 1 and the other allele as 0. For offspring $i$, let $M_i$ be the count of allele 1 inherited from the mother, and $P_i$ the count inherited from the father. Each is either 0 or 1. Define total dosage $G_i=M_i+P_i$, heterozygosity $H_i=I(G_i=1)$, and the origin contrast $O_i=M_i-P_i$. Thus, $O_i=1$ for a heterozygote with the allele inherited maternally, $O_i=-1$ for a heterozygote with it inherited paternally, and $O_i=0$ for either homozygote.

Definition: A simple parent-of-origin mean model for a quantitative offspring phenotype is $$ E(Y_i\mid M_i,P_i,\mathbf{C}_i) = \mu+\beta_G G_i+\beta_H H_i+\beta_O O_i+\boldsymbol{\gamma}^{T}\mathbf{C}_i, $$ where $\mathbf{C}_i$ denotes any included covariates. The coefficient $\beta_O$ is the parent-of-origin contrast parameter. The coefficient $\beta_H$ represents a heterozygote shift shared by the two reciprocal heterozygote classes, as in ordinary dominance. The model separates these two features rather than treating them as interchangeable.

For clarity, first consider the mean model without covariates. The four genotype classes have means

\begin{aligned} \mu_{00}&=\mu,\\ \mu_{10}&=\mu+\beta_G+\beta_H+\beta_O,\\ \mu_{01}&=\mu+\beta_G+\beta_H-\beta_O,\\ \mu_{11}&=\mu+2\beta_G, \end{aligned}

where the first index is the maternal allele count and the second is the paternal allele count. In particular, the subscripts 10 and 01 denote reciprocal heterozygotes. Their mean difference is $2\beta_O$, so $\beta_O$ is half that difference under this coding. The ordinary dominance term $\beta_H$ shifts both heterozygote means equally; it cannot create a difference between them.

The model assumes that the conditional mean is adequately represented by these terms and any specified covariates. For standard linear-model inference, errors should have conditional mean zero; independence and the variance assumptions used by the chosen standard error or test must also be appropriate. In family data, related observations may require an analysis that accounts for their dependence. The coefficient describes a conditional association unless the study design and assumptions justify a causal interpretation.

Worked Example: Separating Dosage, Dominance, and Origin

Suppose a hypothetical study estimates these adjusted phenotype means, in millimetres: $\mu_{00}=10.0$, $\mu_{10}=14.0$, $\mu_{01}=11.0$, and $\mu_{11}=16.0$. Assume the four classes are defined correctly by parental transmission and the stated mean model is suitable. Find the dosage, dominance, and parent-of-origin coefficients.

The homozygote means give the dosage coefficient because $G$ changes from 0 to 2 while $H=O=0$:

\hat{\beta}_G=\frac{\mu_{11}-\mu_{00}}{2} =\frac{16.0-10.0}{2}=3.0\ \text{mm per dosage unit}.

The reciprocal heterozygote difference identifies the origin coefficient:

\hat{\beta}_O=\frac{\mu_{10}-\mu_{01}}{2} =\frac{14.0-11.0}{2}=1.5\ \text{mm}.

To find the shared heterozygote shift, compare the average heterozygote mean with the midpoint of the homozygote means. That midpoint is $(10.0+16.0)/2=13.0$ mm, while the heterozygote average is $(14.0+11.0)/2=12.5$ mm. Thus $\hat{\beta}_H=12.5-13.0=-0.5$ mm. Substitution checks the result: the model predicts 14.0 mm for class 10, 11.0 mm for class 01, and 16.0 mm for class 11. The difference between reciprocal heterozygotes is 3.0 mm, whereas the shared dominance shift is $-0.5$ mm.

The Origin Test and Its Justification

The direct test for a parent-of-origin contrast is $H_0:\beta_O=0$, against an alternative such as $H_a:\beta_O\ne 0$. Under the model, this null states that the two reciprocal heterozygote classes have equal conditional means. It does not state that there is no genetic effect: dosage and dominance coefficients may still be nonzero. A test of $\beta_H$, by contrast, concerns the shared heterozygote deviation and is not a test of imprinting.

The contrast follows directly from the model. For a maternal-origin heterozygote, $(M,P)=(1,0)$, so its mean is $\mu+\beta_G+\beta_H+\beta_O$. For a paternal-origin heterozygote, $(M,P)=(0,1)$, so its mean is $\mu+\beta_G+\beta_H-\beta_O$. Subtracting cancels the intercept, dosage, and shared dominance terms:

\mu_{10}-\mu_{01} = (\mu+\beta_G+\beta_H+\beta_O) -(\mu+\beta_G+\beta_H-\beta_O) =2\beta_O.

This cancellation is why the comparison isolates origin in this parameterization. It also shows why a model using only total dosage $G$ and an ordinary heterozygote indicator $H$ cannot estimate $\beta_O$: both predictors have the same values for classes 10 and 01. The origin variable $O$, or an equivalent reciprocal-heterozygote contrast, is needed.

Worked Example: Testing the Reciprocal-Heterozygote Difference

In a hypothetical study, correctly classified maternal-origin and paternal-origin heterozygotes have sample sizes 25 and 25, respective phenotype means 14.0 mm and 11.0 mm, and a common within-group standard deviation of 4.0 mm. Assume independent observations, approximately normal errors within these groups, and a common variance. These assumptions support the pooled two-sample $t$ procedure for this unadjusted contrast.

The estimated difference is $14.0-11.0=3.0$ mm. Its standard error is

SE=\sqrt{\frac{4.0^2}{25}+\frac{4.0^2}{25}} =\sqrt{1.28}=1.131\ \text{mm}.

The test statistic for equal means is

t=\frac{3.0-0}{1.131}=2.652,\qquad df=25+25-2=48.

The two-sided $p$-value is approximately 0.0108. At $\alpha=0.05$, this provides evidence against equal mean phenotype values in the two reciprocal heterozygote groups under the stated assumptions. The estimated origin coefficient is half the difference, $1.50$ mm, with standard error $1.131/2=0.566$ mm. This is an association in the study population; the $p$-value is not the probability that the null hypothesis is true, and the result alone does not demonstrate an imprinting mechanism.

Why Parental Genotypes Are Needed

An offspring genotype of 0/1 identifies a heterozygote but not which parent transmitted allele 1. Parental genotypes can identify the source when the mating configuration makes transmission unambiguous. If the mother is 0/0 and the father is 0/1, a 0/1 child must have received allele 1 from the father. If the mother is 0/1 and the father is 0/0, a 0/1 child must have received allele 1 from the mother. Genotyping only the offspring would not distinguish these cases.

Some parental-genotype configurations do not resolve the source on their own. If both parents are 0/1 and the child is 0/1, either parent could have transmitted allele 1. Direct transmission information, informative linked markers or a suitable statistical method may be needed to assign origin. Merely recording that both parents are heterozygous does not reveal which allele each transmitted. Homozygous offspring likewise do not provide a reciprocal-heterozygote comparison at this locus.

Worked Example: Classifying Informative and Ambiguous Families

Consider three hypothetical parent-child trios. The allele of interest is coded 1, and each child is heterozygous.

Trio A: Mother 0/0, father 0/1, child 0/1. The mother cannot transmit allele 1, so the child's allele 1 is paternal: this is class 01 and $O=-1$.

Trio B: Mother 0/1, father 0/0, child 0/1. The father cannot transmit allele 1, so it is maternal: this is class 10 and $O=1$.

Trio C: Mother 0/1, father 0/1, child 0/1. Either parent could have transmitted allele 1. Without transmission or phase information, the origin and thus $O$ cannot be assigned from these genotypes alone.

Trio A and Trio B can contribute to the reciprocal-origin comparison; Trio C cannot be assigned to either group from the stated data. Treating Trio C as if its origin were known would misclassify the predictor, potentially obscuring or distorting the estimated contrast.

Common Mistakes and Interpretation

Calling a heterozygote difference “dominance.” Ordinary dominance is a shared deviation for heterozygotes relative to the homozygote-based dosage pattern. A difference between reciprocal heterozygotes is an origin contrast.
Using offspring dosage alone. Both reciprocal heterozygotes have dosage 1. Without origin information, the two classes are merged and the parent-of-origin contrast is not identifiable.
Assuming parental genotypes always resolve transmission. When both parents carry the allele, a heterozygous child may be compatible with either transmission. Additional information or an explicit inference procedure is needed.
Equating statistical evidence with a molecular explanation. A nonzero origin contrast is consistent with parent-dependent effects, but other design or modeling issues can also produce differences. It does not by itself prove allele-specific expression or establish causation.

A parent-of-origin effect should also be kept conceptually distinct from an effect of the mother's genotype or environment. The contrast here concerns the parental source of an allele inherited by the offspring; it is not, by itself, an estimate of a maternal environmental or genetic effect.

Key takeaway: Model maternal and paternal transmitted alleles separately when testing parent-of-origin effects. In the stated coding, the contrast between reciprocal heterozygote means is $2\beta_O$; ordinary dominance shifts both heterozygote means together. Estimating that contrast requires reliable origin assignment, usually using parental genotypes and, for ambiguous transmissions, additional information.

Check Your Understanding

Answer each question using the definitions and model above.

For a 0/1 offspring, what values do $G$, $H$, and $O$ take when allele 1 came from the mother? What changes if it came from the father?
Why can a model with dosage and a heterozygote indicator estimate ordinary dominance but not a parent-of-origin contrast?
If the reciprocal heterozygote means are 8.2 and 7.0 units, what is the estimated $\beta_O$ under this coding?
A mother and father are both 0/1, and their child is 0/1. Why may the child's allele origin remain unknown from these genotypes alone?
What conclusion is supported by a statistically significant test of $\beta_O=0$, and what biological conclusion does that test not establish on its own?