Artificial Wombs When?

What should we expect in the (near) future of reproduction?

Artificial wombs, superbabies, extended fertility?

Should we expect advances in reproductive technology to fundamentally change how we have babies? To raise birth rates? To change the social role of sex and gender?

People like to debate the bioethical and societal implications of reproductive technology, especially as these and other advances start to look increasingly plausible. Just last September, advisors to the FDA met to discuss artificial wombs, which — according to a flurry of news articles — might be available in just a few years. But first it’s helpful to understand what present-day technology and near-future research directions can and can’t do. In my investigation of the field, I found that some technologies seem overhyped relative to their actual capabilities, while others are quietly making huge progress. What does the landscape look like today?

Artificial wombs present the prospect of allowing mothers to skip pregnancy altogether and gestate the fetus in a machine outside the body.

For now research into artificial wombs is focused on continuing gestation of extremely premature infants born “at the cusp” of viability. Children born at this stage — between 22 and 28 weeks — have high mortality rates and carry high risks of disability if they survive. And 6% of all US preterm births fall into this extremely premature category. Rather than replacing pregnancy, these incubators aim to reduce infant mortality and long-term disability.

The main challenge for an incubator is to replicate the circulation and gas exchange in the uterus. In a normal pregnancy, the fetus is connected via an umbilical cord to the placenta, which exchanges blood between the fetus and mother ¹ to keep the fetus supplied with oxygenated blood. Failed attempts to build artificial wombs, historically, have come from (animal) fetuses failing to get enough circulation and oxygenation outside the womb.

Incremental progress in adjusting the designs of these “artificial placenta” systems has pushed the date of viability earlier and earlier. A successful prototype, created by a team at the Children’s Hospital of Philadelphia in 2017, kept premature lamb fetuses alive for four weeks in a “biobag” circulating synthetic amniotic fluid. ² Other groups have kept premature lambs alive with the help of artificial placenta devices. These include scientists at the University of Western Australia ³ and the University of Michigan, both in 2019, ⁴ among others, though none have surpassed the four weeks of gestation achieved by the Philadelphia team.

The Philadelphia researchers are currently in discussions with the FDA about taking the first steps toward testing the “biobag” on extremely premature human infants.

Before we get our hopes up: These types of “artificial placenta” or “ectogenesis” devices do not even attempt to replicate all the functions of the uterus. Current artificial wombs are primarily mechanical — not chemical — devices. They pump the fetus’s deoxygenated blood into an oxygenator and then freshly oxygenated blood back into the fetus, just as the placenta normally does in utero and the lungs and heart do in healthy infants. In contrast, a biological uterus has complex signaling mechanisms involving hormones, immune system modulation, gene expression, and epigenetic modifications, all of which contribute to fetal development. Many of these processes are poorly understood, and more are certainly unknown. 

Current paradigms of ectogenesis cannot, even in principle, substitute end-to-end for a natural pregnancy, from zygote to birth. They are effective only on a fetus whose circulatory system has already developed, to connect the “plumbing” of the artificial blood circuit to existing veins and arteries. 

When people talk about “artificial wombs” being imminent, they almost always mean ectogenesis for prematurity — essentially an advanced form of life support. There’s not a natural incremental path whereby this will lead to full end-to-end artificial wombs. That’s because the “biobag” doesn’t address the hardest part, which is early development in the first weeks of pregnancy. Guiding a newly fertilized egg to grow into an embryo with all the correct anatomical structure, in a petri dish rather than in a uterus, remains an unsolved problem.

In human embryonic development, conception begins with a single fertilized egg or zygote. In natural conception, this process starts when the zygote is in the fallopian tube. It’s not until about five days later, when the new embryo has developed into a spherical, hollow blastocyst, that it implants in the uterus. With in vitro fertilization, or IVF, an egg is fertilized outside the body, and the zygote is allowed to divide until it reaches the blastocyst stage, at which point it can be implanted in a human uterus. 

But unlike the case of artificial gestation for extremely premature infants, which is a present medical need, there isn’t a practical use for incrementally extending human in vitro embryonic development past the blastocyst stage. “Growing” the embryo further in vitro would mean it could no longer implant in the uterus and result in a healthy pregnancy. The only way to actually provide any usable “artificial womb” functionality would be to “grow” the fetus entirely outside the body, all the way to “birth.”

There are regulatory and ethical challenges here in addition to scientific ones. Currently, the International Society for Stem Cell Research ⁵ guidelines for human research do not allow long-term in vitro human embryo culture from a zygote because it could enable human cloning. That means, to comply with the guidelines, research on advanced in vitro embryo development will primarily have to focus on animals.

In 2019 a Chinese group was able to develop macaque monkey fetuses in vitro as far as 20 days, just beyond gastrulation (the formation of the primitive gut), at which point their structure collapsed. ⁶ A 2022 study from Spain, similarly, developed sheep embryos for 14 days, observing the early stages of gastrulation. ⁷

A 2022 study went even further with mouse embryos, developing a “beating heart-like structure,” a neural tube, a gut tube, a tail bud, and “headfolds” containing forebrain and midbrain regions. (These embryos developed for eight and a half days. Mouse fetuses, of course, develop much faster than human fetuses; mouse gestation is only 20 days in total.) ⁸

Amazingly, all of this mouse embryonic organ development happened spontaneously, without any external developmental signals in the culture medium. We’ll never know if these fetuses would have been viable, but sequencing and microscopic imaging suggest they were developing normally.

This kind of progress points toward a path to true artificial wombs: developing longer and longer in vitro development of embryos and fetuses until they can reach the 22–28 week stage that we know can be brought safely to the point of viability with “biobag”-like advanced incubators.

Still, one should caution that human pregnancy and fetal development are very different from that of almost all other mammals: The applicability of even a successful “artificial womb” in mice doesn’t tell us that much about its likelihood of success in humans. 

As one example, humans and great apes are unique in their unusually invasive placentas, which become deeply integrated with the uterus after the fertilized zygote implants. Other mammals don’t even have a blastocyst implant directly in the uterus. Humans are also remarkable for their long development periods, especially in the brain, so it may be especially difficult to get ex utero development “right” in a human to the point that it doesn’t result in intellectual disability. In general, experiments on early embryonic development in any common lab animal, including primates (there are lab macaques but not lab gorillas or chimpanzees), are importantly unlike their equivalents in humans.

The bottom line is that artificial wombs to replace pregnancy are not going to be developed any time soon and face challenging obstacles. 

The “artificial wombs” that you’ve heard about in the news are an important breakthrough that could save the lives of many premature infants, but they aren’t even potentially an alternative to natural pregnancy. 

Screening fetuses for genetic disorders is nothing new. Amniocentesis, or taking a sample of fetal cells from amniotic fluid, has been widely used to determine fetal sex and detect chromosomal disorders since the 1960s.

A newer method dates to 1997, when it was first discovered that cell-free DNA from the fetus circulates in the mother’s bloodstream. Screening fetuses for genetic defects using this method allows for detection of chromosomal disorders in the first trimester (before the 15th week of pregnancy there usually isn’t sufficient amniotic fluid to perform amniocentesis safely). The first commercially available cell-free DNA fetal screening tests came out in 2011. ⁹ Fetal genetic screening is often used to make decisions resulting in selective abortion. (For instance, 67% of fetuses with a prenatal diagnosis of Down syndrome were aborted between 1995 and 2011.) ¹⁰

Genetic screening has been used for IVF since the 1990s. In order to maximize the chance of a healthy live birth, three to five-day-old IVF embryos are sampled and screened for genetic disorders, so only the (relatively) disorder-free embryos get implanted in the uterus.

Thanks to improved screening, ovarian stimulation, and embryo culture, IVF has gotten vastly more successful since its origin in 1978, when each cycle of IVF had only a 6% chance of resulting in pregnancy. Today, it’s about 20%–30% per cycle, and across all cycles it’s successful 50% of the time for women under 35. ¹¹ , ¹²  

What’s new in the world of IVF genetic screening is whole genome sequencing, which has gotten vastly cheaper in recent years.

You don’t need to sequence every A, T, C, and G of the genetic code to detect when a whole chromosome is missing or duplicated. So when DNA sequencing is expensive and scarce, genetic screening mostly looks for chromosomal anomalies or very large deletions. These tend to be unambiguously and reliably disabling — every individual with three copies of chromosome 21 will have Down syndrome.

As genetic testing gets cheaper, it is becoming possible to look at the much wider (and more uncertain) world of genetic risk factors for disease. Most diseases that are partially heritable, from schizophrenia to heart disease, are statistically linked to many, many genes, though each with an individually small effect. With large datasets of the genomes of people with and without each disease, it’s possible to generate a polygenic risk score, a function of your genes that gives an estimated probability of carrying a given disease. 

Consumer genetic testing companies like 23andMe or Nebula offer polygenic scores for disease risk, but this doesn’t have much practical impact beyond satisfying curiosity and maybe helping individuals decide what lifestyle choices or health screening exams are worth prioritizing. 

Polygenic prenatal screening, by contrast, does have a practical upshot — people might choose which children to have based on their risk scores.

Startups like Orchid and LifeView offer genetic tests for IVF embryos that include risk scores for diseases like Alzheimer’s, cancer, and diabetes. These tests allow parents to select the healthiest embryos for implantation. Already, this has sparked controversy — a recent Nature editorial worried about the “alarming rise of complex genetic testing in human embryo selection” and called for stricter regulation.

Though nearly all of the focus to date has been on using polygenic prenatal screening to prevent susceptibility to disease, the same techniques could be used to select for other traits. Perhaps the most controversial of these is IQ. 

But to what extent can you even predict IQ from genes?

Genome-wide association studies examine the associations between single base pair variations in DNA called single nucleotide polymorphisms, or SNPs (pronounced snips), and particular traits across the human genome. They typically require very large sample sizes, and on IQ, GWAS have gotten really big (one 2018 meta-analysis covered 269,867 individuals). ¹³ At such massive scales, there’s enough statistical power to confidently tell genetic association from noise. Educational attainment (years spent in school, something of a proxy for cognitive ability) is more studied than IQ and can be found in even bigger GWAS studies — a 2022 study looked at about three million people. ¹⁴

So you can compute a polygenic prediction score for IQ (or educational attainment) from a big genetic study. How much does that matter? A common metric is variance explained — how much of the variation in the population is accounted for by the variation in genetic risk scores? Recent IQ GWAS studies report a very modest predictive ability — the polygenic score can explain 2%–5% or so of the IQ variance in their samples. ¹⁵ The biggest educational attainment study, more optimistically, reports a 12–15% variance explained by the polygenic score. ¹⁶

What does all this cash out to in terms of the expected impact of embryo selection on IQ? 

Independent researcher Gwern Branwen’s ¹⁷ statistical analysis suggests that using polygenic scores to select the best-scoring out of 10 embryos (one IVF cycle usually yields closer to five embryos) might be expected to yield about a three IQ point improvement. An upper bound based on the overall variance in IQ explained by genes would be more like a nine to 11 IQ point gain out of 10 embryos (or an eight-point gain out of a more realistic five embryos).

Assuming this analysis is right, it’s not clear that you would typically notice a difference in a child conceived via embryo selection for IQ compared to one who wasn’t.

In a family with three children, one of which was conceived with an embryo selection procedure that had an expected IQ gain of five points, the embryo-selected child would only have the highest IQ of the three about half the time. And while IQ differences of five points have statistical associations with life outcomes at the population level, it’s not at all unusual for two individuals whose scores differ by five points to have no obvious differences in income, education, or other “real-life” properties associated with intelligence. 

In other words, you wouldn’t get genius babies out of this procedure even in the best-case scenario — and there would be quite a lot of families who get no noticeable benefit from IQ-selecting their children.

Many people also have ethical concerns about attempting to create “designer babies.” Would the rich have “better genes” than the poor? Would embryos selected for a “good” polygenic score on one dimension actually be more at risk for some other disease or disability that correlates with that “high-scoring” gene signature? Or would “choosier” approaches to IVF simply delay pregnancy longer, requiring more cycles of IVF and more cumulative health risks? 

It remains to be seen what regulatory regime will end up being tolerable to experts and the general public.

A less-developed but potentially more impactful approach to selecting embryos for polygenic traits is “massive multiple embryo selection.” This process would involve generating very large numbers of eggs (or egg precursor cells) in vitro, which could be genetically screened and fertilized, allowing for selection based on a polygenic score from a very large batch of genetically diverse gametes and/or embryos.

The larger the batch of individuals to screen from, the more selective the polygenic predictive score, and thus the higher the expected effect on the trait you’re selecting for.

In principle, this could allow for double-digit expected IQ point gains from embryo selection, which would be an unmistakable effect. (It could similarly allow for dramatic changes in any other polygenic heritable trait, such as height.)

What this requires, however, is the ability to turn an egg or somatic cell ¹⁸ into an egg-generating cell. This is known as oogenesis (making eggs). Making sperm from other kinds of cells, analogously, is called spermatogenesis; making either eggs or sperm is called gametogenesis.

Beyond embryo selection, there are other practical applications to artificial gametogenesis.

Oogenesis could allow two genetic males to have genetically related offspring together (by creating eggs from one of their cell samples), while spermatogenesis could allow two genetic females to have genetically related offspring together (by creating sperm from one of their cell samples). Also, women who no longer have viable eggs in their ovaries could use artificial oogenesis to produce eggs from their other cells, potentially allowing them to have genetically related children after menopause.

How far away are we from artificial gametogenesis becoming a reality?

There’s a fair number of experiments so far, including some that produced sperm-like or egg-like cells from adult stem cells. ¹⁹ There have even been animal experiments from which in vitro differentiated stem cells resulted in both eggs and sperm that produced live offspring. There have also been animal studies producing viable artificial sperm from a female and fertilizable eggs from a male. ²⁰

One promising example comes from a 2012 study from Kyoto University where female mouse stem cells, differentiated in the presence of samples derived from ovarian tissue, were able to mature into eggs that could be fertilized in vitro, transplanted into female mice, and developed into healthy offspring. 

Much more recently, a preprint from the Church Lab at Harvard demonstrated successful in vitro initiation of meiosis from human adult stem cells, creating cells that appear equivalent to oogonia, the precursor cells to eggs.

If this research is borne out, it should be possible to create large numbers of artificial eggs, fertilize them in vitro, and then select only the highest-scoring embryos on a polygenic predictive score for implantation.

Artificial gametogenesis is a live area of scientific research and is being pursued in new startups, including Conception Bio ($20 million in funding, founded 2018), Gameto ($73 million in funding, founded 2020), Ivy Natal ($1.8 million in funding, founded 2020), and Dioseve ($3 million in funding, founded 2021).

Bioethical and regulatory controversy aside, this is real science, where the incremental trajectory seems to be rapid recent progress toward better and better control of stem-cell-to-gamete in vitro development. 

In the body, stem cells have the potential to differentiate into all of the cell types we have, including eggs and sperm. What tells a stem cell what kind of cell to become? A complex vocabulary of signaling molecules in its environment and/or local mechanical or electrical forces that the cell’s biochemical sensors can pick up. The better we understand how stem cells develop into eggs and sperm in the human body, the better we can develop cell culture protocols that replicate those signals.

Of course, to be an actual fertility method, any reproductive technology has to do more than produce examples of “viable” offspring. The practical standard for safety would be quite high. How many parents would accept, say, a fertility method that had 10 times the risk of infant mortality relative to traditional childbirth? 

At present, experimental methods of producing artificial gametes are much more prone to resulting in defective cells or embryos than natural gametogenesis and fertilization are, so the safety has to improve a lot before it’s ready for use in humans. But it’s not at all impossible to get there over a multiyear time horizon.

As women age, their chances of successful pregnancy diminish. A 37-year-old is about half as fertile as a 20-year-old. Is it possible for medical interventions to extend fertility?

At puberty, a girl typically has hundreds of thousands of follicles that can develop into eggs — but a woman will go through only fewer than a thousand monthly menstrual cycles in her lifetime. 

Menopause and age-related fertility decline is, therefore, not due to a woman “using up” her initial follicles as time passes. Rather, it’s a result of gradual loss and degradation of follicles, known as “diminished ovarian reserve.” Over time, more and more cycles will fail to produce a fertilizable egg or will result in defective embryos that end in early miscarriage rather than pregnancy and birth. 

In principle, we can study how to slow that gradual decline in follicle quantity and quality.

One approach, currently being developed by the biotech firm Oviva, involves a version of anti-Müllerian hormone. AMH is naturally produced by ovarian follicles and declines steadily throughout adulthood before disappearing about five years before menopause. It is the best-known proxy measure for predicting individual fertility. ²¹ A 2022 mouse study shows that administering AMH to female mice increases the number of oocytes and their chance of being fertilized and developing into embryos, so it’s possible AMH has a causal impact on fertility as well. ²² AMH seems to block follicles from developing into oocytes, which slows the rate at which the ovarian reserve is “used up” with age. ²³

Another company, Celmatix, is looking at other targets that regulate ovarian function. These include FSHR, AMHR2, and MTNR1A/B — all receptors for hormones ²⁴ whose levels are associated with ovarian reserve in women and which play a role in ovarian follicle differentiation. Their hope is that women can produce (or retain) more oocytes by replicating the effect of youthful levels of these hormones in the ovary.

In academic and clinical research, there are still other approaches that seem potentially promising for preserving or restoring female fertility. 

The common supplement DHEA has been claimed to increase ovarian reserve and cumulative pregnancy rates in infertile women. ²⁵ In animal studies, it seems to be possible to extend ovarian lifespan with some of the same pathways and methods being used to extend overall lifespan and prevent age-related disease. Rapamycin, the immune-suppressant drug that has been studied extensively for its anti-aging effects, also seems to extend ovarian lifespan (in mice and rats). ²⁶ , ²⁷ Other compounds with a similar mechanism of action (mTOR inhibition) likewise seem to extend ovarian function in aged mice. ²⁸ And, like most organs, ovaries become fibrotic (scarred due to chronic inflammation) with age. Removing fibrotic tissue from mouse ovaries, as well as administering anti-fibrosis drugs, can recover ovulation in old female mice. ²⁹

On the other hand, there are challenges that mean it may be a long time before we have working therapies to delay ovarian aging.

One obstacle is, once again, the uniqueness of human reproductive biology — while most mammals show gradual decline in fertility with age, female humans are nearly unique in having such a long post-reproductive lifespan. We are totally unable to reproduce for almost half our lives. The few other animals with similar “menopause-like” phenomena include elephants and whales — not easy to run lab experiments on. So research on extending fertile lifespan is mostly limited to rodents and human tissue samples, which are decidedly imperfect models.

Rodents, the most common type of mammal used in lab experiments, have evolved for fast life histories — early sexual maturity, lots of babies, and early infertility and death. This is even more true for lab mice and rats, which are bred to be fecund for the convenience of researchers. As a side effect, they are shorter-lived and quicker to go infertile than their wild counterparts. Now, a major determination of how early a female mammal becomes infertile is how soon she “uses up” her ovarian reserve by developing follicles into oocytes. That’s a life-history parameter — “faster-living” animals use up their ovarian reserve sooner, while “slower-living” ones preserve their fertility longer. 

The good news is that life-history parameters are usually hormonally regulated and thus have a good chance of being modifiable through drugs — in lab mice. The bad news is that humans have about the slowest life histories of any mammal on Earth already, so something that seems to help mice stay fertile longer is unlikely to have similarly strong effects on us. Life history–prolonging interventions are effectively “flipping a switch” in mice that’s already “flipped” in humans. A lot of the classic anti-aging interventions (like caloric restriction or rapamycin) more or less seem to function by telling the body, “Food is scarce; don’t be in such a hurry to grow and reproduce,” ³⁰ and they reliably have more modest effects in larger mammals and humans than they do in lab rodents.

Another challenge is that preventing or delaying female age-related fertility decline necessarily takes a long time to test — a human study would have to last many years. So the incentive for a drug company is to produce short-term increases in ovarian function or fertility in infertile women, which may not be the same research program as trying to slow ovarian aging to prevent infertility in the first place.

But overall it’s not implausible that women alive today might be able to have children at older ages (with or without IVF) than was previously thought possible.

As a 35-year-old woman, which of these reproductive technologies would I actually expect to be able to use during my own remaining reproductive lifespan?

Artificial wombs for rescuing extremely premature infants — yes. These are entering the clinic imminently.

Artificial wombs as an alternative to natural pregnancy — no. As yet, nobody has gotten a mammal embryo to develop more than 20 days post-conception outside the uterus.

Prenatal polygenic genetic screening — yes. This is already on the market.

Embryo selection for IQ — technically yes. It’s on the market, but the effect size is small.

In vitro gametogenesis (for more effective IVF, high-impact embryo selection, and/or LGBT conception) — maybe. It’s making rapid progress and has already seen success in animals.

Extending female fertility or delaying menopause — probably not. There’s research and investment going on, but it takes a very long time to run trials on preventing age-related diseases and conditions.

For a person born today, however, and expecting to have children in the 2040s–2060s, all of these technologies might become realities. The future of reproduction has the potential to become very interesting.