Doppelgängers don’t just look alike; people who appear nearly identical without being related are also astonishingly similar in their genetic makeup and often share lifestyle traits, a study published August 23 in Cell Reports finds.
“This shows a creative approach to finding similar sets of genes in people around the world,” Christopher Mason, a physiologist and biophysicist at Weill Cornell Medicine who was not involved in the study, writes in an email to The Scientist. “The methods were pretty standard in the field, but the application of them here was novel.”
Manel Esteller, a geneticist and cancer researcher at the Josep Carreras Leukaemia Research Institute in Spain who led the new study, previously found that epigenetic changes are responsible for notable (though sometimes slight) differences in appearance in twins with the same genetic makeup. That got him wondering: “What about people who have the same face but are not related?”
To answer that question, Esteller and his colleagues first needed to find doppelgängers. The team collaborated with Canadian photographer François Brunelle, who has been collecting and sharing portraits of look-alikes around the world since 1999 as a part of his “I’m not a look-alike” project. “We ran the pictures of faces through three different facial recognition algorithms, to show they are, in an objective manner, look-alikes,” Esteller explains. The team then asked 16 couples who were highly similar enough to be identified as look-alikes by all three algorithms to answer a questionnaire about their lifestyles and provide biological samples. From these, the team obtained data on sites in their genomes known as single-nucleotide polymorphisms (SNPs), which commonly vary from person to person, as well as data on gene expression (the epigenome) and oral microbiome composition. The researchers also used the sequence data to analyze ancestry, confirming that the look-alikes were, indeed, not related to one another.
The team found that people who look very similar are also very similar genetically, says Esteller. Comparing the 16 pairs of true look-alikes with 16 other pairs photographed by Brunelle that hadn’t been scored as look-alikes by all three algorithms, the researchers found that true look-alikes shared more genetic variants with each other than did the 16 less similar pairs. However, the true look-alike pairs differed when it came to patterns of gene expression and bacterial communities. “The differences we see between look-alikes are more due to the epigenetics and the microbiome,” he says.
The results weren’t wholly surprising to Esteller, who points to genome-wide association studies showing that some variations in genes are associated with facial features. Mario Falchi, who conducts twin research at King’s College London and was not involved in the study, agrees. “The elegant experiment described in the paper . . . shows that look-alike individuals share more genetic variants than randomly-selected individuals.”
The team also found that the resemblance between doppelgängers extended beyond their looks and genetics, Esteller remarks. Although the study was designed to examine faces, he says, the team used “a very extensive questionnaire,” which revealed that physical features such as weight and height also tended to be similar in doppelgängers, as did lifestyle traits, such as smoking habits and educational attainment. That may indicate that similarities “extend to more personality-related traits,” he posits, adding the caveat that such a claim would need to be “carefully evaluated.” Falchi says it’s likely that the reported association between facial features and physical and behavioral phenotypes is likely due to pleiotropic effects, in which a gene affects multiple seemingly unrelated traits, and epistatic effects, in which genes interact with one another.
“It was interesting to see a genetic connection to the facial recognition algorithm; it shows that using both sets of data could be even more powerful in the future,” writes Mason. “This could open up a Pandora’s Box for forensics, but it is exciting.”
The results, Esteller posits, could lead to the use of AI to infer genetic variants based on images of faces. He speculates that scientists could eventually “differentiate from the nose or the mouth that the person is a carrier of a [medically relevant] mutation,” which could result in early intervention to prevent disease development and more tailored care.
The opposite could also become reality, with the results potentially leading to better facial reconstructions from DNA evidence, or as Esteller puts it, “how to draw a face from the genome.”
In a possible step in that direction, Esteller says he and his team are now investigating the complete genome to uncover additional variants involved in facial similarities. He also plans to study the roles of proteomes and transcriptomes in facial variation. Using this multiomics approach, Esteller says, “we will be able to reconstruct the perfect face from biological material.”