header
Science

Why Scientists Used Artificial Neural Networks to Decode the Origins of Menopause

Since there exists no known human population without menopause, there is no direct way of studying the emergence of the phenomenon.

Bengaluru: Except for killer whales and pilot whales, females in no other species – not even among apes, our closest relatives – lose the ability to reproduce as early in their lives as humans do. And no wonder, because if the transmission of genes to offspring (i.e. reproduction) is the primary driving force behind evolution, why would any species adopt a mechanism that stops it? In other words, why did menopause emerge among humans?

Scientists in France, led by evolutionary biologist Carla Aime-Jubin of the University of Montpellier, Montpellier, employed artificial intelligence to simulate a population in which menopause had not yet developed in humans. Since their experiment was virtual, they were able to tweak the conditions that the simulated population lived in and note the settings under which menopause developed. Their results, published last week in PLOS Computational Biology, seem to give strength to one of the three most popular hypotheses that explain menopause.

According to the ‘maternal hypothesis’ proposed by Jocelyn Scott Peccei in 1995, menopause is a response to the increased risk of dying during childbirth in older women. Peccei suggested that natural selection started to favour women who became prematurely fertile because they were better off spending their energy on existing children than on late pregnancies. But in Aime-Jubin’s simulated population, this hypothesis did not fit. Even in models where there was no increased cost of reproduction with age (the main assumption of Peccei’s hypothesis), menopause continued to emerge.

Besides, the maternal hypothesis suffers from one major deficiency. Though it seems to reasonably explain why menopause emerged, it doesn’t answer the other aspect of the puzzle: why live so long after reproduction stops? This is why this theory failed to overturn the older and more dominant ‘grandmother hypothesis’. The grandmother hypothesis states that a long life after menopause allows women to focus their attention on their own children who are of reproductive age and rear their grandchildren.

The value of grandparenting

On the surface of it, this hypothesis may not strike one as particularly convincing, but it turns out that grandparents have been proven to have enormous impacts on the lives of growing children. In the 1980s, anthropologist Kristen Hawkes studied the division of labour in a tribe of hunter-gatherers in Tanzania and what she saw highlighted this. “With no young children of their own, they (grandmothers) help feed their daughters’ and nieces’ offspring. This help is especially important for the nutritional welfare of weaned children when their mothers forage less at the arrival of a newborn,” she noted in one of her papers. By using their strength to keep their grandchildren safe and healthy, grandmothers are indirectly ensuring that their genes get reproduced – despite not being fertile themselves. Hence, Hawkes argues that it may not be accurate to call menopausal women ‘post-reproductive’ after all.

Aimes-Jubin’s computer models partially supported the grandmother hypothesis but they also emphasised that it is not just physically that grandparents are helping children but also cognitively. That is the essence of a more recent update on the grandmother’s hypothesis, called the ‘embodied capital model’ (ECM).

Like in the grandmother’s hypothesis, the main idea of ECM is that grand-mothering plays a pivotal role in the evolution of menopause. But it also acknowledges the role of cognitive resources. Aime-Jubin explained via email that investing in neural development at one stage of life will return increasing benefits along the way by promoting the accumulation of experience and skills. “These delayed benefits allow ageing people to survive and still acquire resources (more than they need for survival) from their environment even if their physical condition decreases. These surplus resources could be used for having new children or for grandchildren care,” she said.

Aime-Jubin said that though the link between ECM and menopause was first suggested in 2012 by Hillard Kaplan, an evolutionary anthropologist, theirs is the first study to formally test for the relation. While the paper stresses on what this could mean for a better understanding of why women undergo menopause, equally important is the scope for the methodology Aime-Jubin and colleagues have used here.

The pros and cons of modelling

Mathematical modelling studies are considered problematic because the results they gives are only as good as the assumptions that are used. Indeed, some concerns have cropped up with the assumptions made in this one, too. In an article in Popular Science, anthropologist Lorena Madrigal of University of South Florida wrote that some of the parameters in Aime-Jubin’s studies may not fit in with all communities and ecologies. She called for more cross-cultural research across different ecologies and time periods. Nevertheless, in the case of menopause studies, they present an opportunity.

Artificial neural networking, the modelling technique used in the new study, is a computer programme designed to mimic the way the brain works. It ‘learns’ by detecting patterns and relationships between individuals in a simulated population and eventually uses its experience to predict outcomes. In 2012, a study from Trinity College used this method to show evidence that cooperation among individuals played a role in driving the evolution of intelligence and larger brains. It showed how artificial intelligence-based models are capable of addressing fundamental questions about human evolution.

Aime-Jubin’s study is a newer example of this. Since there exists no known human population without menopause, there is no direct way of studying the emergence of the phenomenon. “Artificial neural networking allows observing the evolution of menopause from its emergence, by simulating populations where menopause does not necessarily exist and then testing the conditions for its emergence,” said Aime-Jubin. “It is almost like an evolutionary ‘time machine’.”

In this case, they could simulate individuals who can extract resources from their environment as well as decide how they want to allocate resources – whether for increasing survival or for reproduction. Their decisions will determine which individuals will survive and reproduce so that their “genes” (represented in the artificial neural network as numbers) can become more frequent in the population. It is exactly the same process as followed in real populations, i.e. evolution, pointed out Aime-Rubin.

The authors see a future where artificial neural networks could similarly be used to study other life history traits like ageing. But before that, they need to address the primary shortcoming of their study: that it is a one-sex model. This is inadequate because ECM is a two-sex model, meaning it predicts that both ageing women and men may stop reproducing to allocate resources to existing children and grandchildren. Her next objective is to design a two-sex model of this experiment so that they can begin to address the next question: why don’t men lose the ability to reproduce like women do?