This is a high-level roadmap of the technical aspects of how strong germline engineering technology can be developed for humans. I’ll start with a quick introduction, but you can also skip down and play with the roadmap.

The idea of germline engineering is:

Make a healthy baby who has a genome that the parents nudged in some directions they’ve chosen.

How can we do this? There are three basic elements:

• Figure out what genome you want. (Polygenic scores)
  • To do this, we have to understand which genetic variants are associated with various traits, such as lower risk of various diseases, longer lifespan, more capacity to flourish and achieve, and so on. Then, parents have to learn about these traits, consider the tradeoffs and uncertainties, and make a choice.
• Make a cell with the genome you want. (Genomic vectoring)
  • To do this, we have to manipulate DNA somehow, so that we make a cell with the DNA you were targeting.
• Make a healthy baby from that cell. (Epigenomic correctness)

All three steps are important, and there’s lots more work to be done on them. But if there is one bottleneck to strong germline engineering, it’s actually the third step: making a healthy baby from a given cell.

That’s not too hard if you already have a normal healthy embryo. You just let it grow and then implant it. Thus, there are already several companies that offer genomic screening for embryos:

That’s genuine germline engineering, in the broad sense, and it can already have some quite beneficial effects by avoiding high risks of many diseases. However, due to Reasons, the strength of simple embryo selection is limited:

How do we get stronger germline engineering? We have to actively change the DNA, and then we have to make a healthy baby from the cell. And to make a healthy baby, the cell has to have the right epigenomic state for healthy embryonic development.

What does that mean? Different cells have different epigenomic states. The cells in your liver or skin, your brain cells, the stem cells in your immune system—they’re all different from each other in terms of their shape, and what proteins they produce, and how they move around, and what chemical reactions they do. The underlying reason for all those differences is that the DNA inside the nucleus of each cell produces different amounts of different RNAs. These RNAs, and the proteins that many of them are translated into, are what do the work of constructing and operating the cell and its particular functions.

And yet, all these cells have basically identical genomes; their complement of DNA is the same. So it’s not differences in the DNA itself that make different cells produce different RNAs.

Instead, the main reason is epigenetic marks: chemical modifications to DNA and histones. (Histones are proteins that DNA wraps around.) These epigenetic marks persist for days, months, or years on cellular DNA. The marks control how much each gene is expressed by giving directions, so to speak, to transcription machinery in the cell, like “this gene I’m attached to, transcribe it into RNA a lot!” or “this gene I’m attached to, stay away from it for now!”. (Levels of gene expression in a cell are also affected by how packed up the cell’s DNA is, and which transcription factors are floating around in the cell. These gene regulation mechanisms are themselves largely controlled by epigenetic marks. Besides epigenetic marks within the cell, signaling molecules coming in from other cells also sometimes control the cell’s gene expression.)

(Diagram from Gilchrist, Daniel A. ‘Histone’. Accessed 3 April 2025. https://www.genome.gov/genetics-glossary/histone.)

To make a healthy baby from a cell, the cell has to have the right epigenomic state. Natural sperm DNA has a specific epigenomic state; natural oocyte (egg) DNA has a specific epigenomic state; and these combine to produce a specific epigenomic state in the embryo. That specific epigenomic state makes the cells in the early embryo produce a whole bunch of RNAs in a specific pattern, in order to quickly grow the baby, both in size and in terms of all the different kinds of tissue organized in the typical healthy baby configuration. If, hypothetically, you tried to make a baby from a cell that’s not in the right epigenomic state, then probably what would happen is, usually the embryo just dies, occasionally you get a baby, and often you get a deformed baby. So don’t do that.

If we want to make a cell with a genome that’s been nudged, and then make a healthy baby, we therefore have to understand what the normal healthy epigenomic state of a sperm / egg / embryo is supposed to be, and we have to make our cell have the right epigenomic state. This is the biggest bottleneck to strong germline engineering.

The whole problem of germline engineering can be broken down like this: You have to know and make the target genomic and epigenomic states in your cell.

How do we actually do this? The visual roadmap below gives a broad outline of how science and biotech research could go, building up to the full technology of strong germline engineering in humans. It’s organized like a plan: time flows downward, with higher nodes unlocking the lower nodes that they point to. The red regions in the left half of the roadmap address the problem of epigenomic correctness (knowing and making the right epigenomic state), corresponding to the lower row in the above 2-by-2 scheme. The blue regions in the right half of the roadmap address the problem of genomic vectoring (knowing and making your target genomic state), corresponding to the upper row in the 2-by-2 scheme.

Most nodes in the roadmap represent open questions, achievements that haven’t been reached, and projects that haven’t been done. (Some exceptions: There’s substantial progress on polygenic scores to predict traits from genomes; there’s lots of research using CRISPR to edit DNA; and there’s some progress in epigenomic correction in mice.)

Much more detail, as well as sources for many of the images used below, can be found in my book “Methods for strong human germline engineering”.

Enjoy!

graph TB subgraph sub_GV[ ] style sub_GV fill:#a4c2f4; header_GV["genomic vectoring problem (GV)

Step: make a cell with a genome you want. Genomic vectoring (GV) is the task of altering a cell's genome toward some target. GV requires that we know what the target looks like (polygenic scores), and that we have a way of making a cell move toward a target (a GV method). Since there are several genomic vectoring methods that should work well, GV is less likely to be a bottleneck for strong germline engineering compared to the epigenomic correctness problem.

" ] style header_GV fill:transparent; header_GV ~~~~ sub_GV_methods subgraph sub_GV_methods[ ] header_GV_methods; GV_IMS; GV_editing; GV_CS; GV_synthesis; style sub_GV_methods fill:#9ad2ff; style header_GV_methods fill:transparent; header_GV_methods[GV methods

Step: make a cell with a genome that scores well on some criterion. The other nodes in this blue box are different possible strong genomic vectoring methods. Each one would be strong on its own; they could also combine to be even stronger. A genomic vectoring method is a biotechnology method that makes a cell with a genome that's been moved toward some target. There are two broad classes of GV methods: editing and selection. There's also DNA synthesis. Embryo screening is already doable, but it's a fairly weak GV method. Gamete selection—selecting the highest-scoring sperm and eggs—would be a bit more powerful; perhaps 1.4 times as powerful. But it's probably not feasible before stronger methods, and it's also not a strong GV method. Strong GV matters. The strong GV methods are: chromosome selection, iterated CRISPR editing, iterated recombinant selection, and whole genome synthesis.

] GV_CS[chromosome selection

Method: take the highest-scoring chromosomes from several cells and put them into one cell. Chromosome selection involves moving target chromosomes between cells. Chromosome selection applied to recombined DNA, from a sperm or egg, is fairly powerful. For parents who want to exchange chromosomes with other parents, chromosome selection is very powerful. This genomic vectoring method might not be feasible to implement because manipulating tiny individual chromosomes is hard; on the other hand, chromosome selection may be able to bypass most of the epigenomic correction problem by using already-correct DNA. This method might be able to bypass the EC problem by using natural reproductive DNA. See the sections “Method: Chromosome selection” and “Appendix: Best crossover”.

] GV_editing[iterated CRISPR

Method: edit stem cells at dozens or hundreds of loci. In iterated multiplex CRISPR editing, you repeatedly edit the DNA in stem cells with CRISPR, and periodically sequence clones of cells to screen for successful edits and against DNA damage. While this method will be laborious, and will have challenges such as avoiding mutation during many culture passages, it would be a medium or strong genomic vectoring method and it is already minimally working technology. This method might be able to use EC methods that pause natural reprogramming. See “How to Make Superbabies” by GeneSmith and kman, and the section here: “Method: Iterated multiplex CRISPR editing”.

] GV_IMS[iterated recombination

Method: run a sort of evolution in vitro, really fast, by dividing and fusing cells and selecting the high-scoring ones. Iterated recombinant selection involves repeatedly dividing cells into haploid cells so that the chromosomes crossover, exchanging DNA, and then fusing haploid cells to form diploid cells with new genomes. By selecting cells that got desired crossovers, target DNA segments are assembled into one cell. This genomic vectoring method will be complex to implement, and the fundamental component of inducing meiosis is not yet feasible at scale; but this method would ultimately be very powerful. This would require full epigenomic correction. See “Meiosis is all you need” by Metacelsus, and the section here: “Method: Iterated recombinant selection”.

] GV_synthesis["DNA synthesis

Method: artificially generate chromosomes with any sequence you want, and then put them together in a cell, and use that cell. Artificial DNA synthesis involves synthesizing entire chromosomes. I don't know much about it. In 2008, the J. Craig Venter Institute synthesized a .5 megabase DNA strand by stitching together smaller strands. I don't know if there's been much advance since. There are dual-use risks. If one could synthesize a whole genome with very low error rate, and assemble it into a cell, that would be an arbitrarily powerful genomic vectoring method; i.e. it would be limited only by our understanding of the genetics of traits. This would require full epigenomic correction. (Diagram from https://commons.wikimedia.org/wiki/File:Long_Overlap_Assembly.png.)

" ] header_GV_methods ~~~ GV_IMS GV_editing ~~~ GV_CS GV_synthesis ~~~ GV_CS end GV_cows; GV_mice; GV_monkeys; GV_humans GV_cows["proof-of-concept GV agriculture

Step: make very productive agricultural products using germline engineering. Developing genomic vectoring technology will probably require significant resources. So, it would be helpful to make money along the way. Agriculture is one likely commercial application. Since it's considered ethically acceptable to make e.g. cattle with some risk of birth defects, the epigenomic correctness problem doesn't have to be addressed before applying genomic vectoring to crops or livestock. We can apply a GV method to stem cells; use somatic cell nuclear transfer to make offspring that may have some birth defects; and then produce a second generation from those offspring, which will likely be developmentally normal and also genomically vectored. There are polygenic scores for key traits in high-volume livestock and crops, such as milk production, egg production, and overall monetary value. Thus, this method can produce very valuable products much faster than the current agricultural practice of selective breeding and embryo selection. Besides providing income, applying GV methods in agriculture would also provide a much needed sort of information: What happens when you genomically vector strongly based on a polygenic score? Animal and plant breeding has produced large changes in phenotype, but that's always been over the course of several generations. If we keep selecting for increased size, productivity, etc., across several generations, the way that we're steering the genome might be changing at each generation. In contrast, a stronger genomic vectoring method would make a larger change all in one go, moving a larger step in one direction. It'd be useful to see if the effects are what we predict based on our polygenic scores, and if there are any health problems caused by taking a larger step.

" ] GV_mice["test GV method in yeast, mice

Step: develop the genomic vectoring method by experimenting with easy-to-culture cells like yeast or mouse stem cells. Since primate stem cells are somewhat finicky (they die out or lose their ability to divide), some genomic vectoring methods might be developed by experiments in yeast cells or especially convenient mouse stem cell lines. The mouse lines could also be used to produce live mice via SCNT, mainly as a concrete proof-of-concept for making large changes to the genome of a future organism. These tests aren't necessary, though; just potentially convenient.

" ] GV_monkeys[test GV method in monkeys

Step: refine the genomic vectoring method so that it works efficiently in non-human primate cells, doesn't mess with epigenomic correction, and produces healthy offspring. Some ways of handling the epigenomic correction problem can be solved and validated separately from genomic vectoring. In vitro oogenesis, for example, should be able to take any stem cell, and turn it into an epigenomically competent egg. For these methods, it's not strictly necessary to test the genomic vectoring method in non-human primates: we can genomically vector some human stem cells, and then produce competent eggs from those stem cells using the in vitro oogenesis method that we separately validated. However, some other ways of handling the epigenomic correction problem rely on the genomic vectoring method to not interfere too much with the epigenomic state of the cells being operated on. For example, chromosome selection applied to sperm or egg DNA is supposed to leave the epigenomic state intact; and iterated CRISPR editing embryonic stem cells or spermatogonial stem cells is supposed to combine with some way of maintaining the epigenomic state of those cells. These methods have to be validated as safe in non-human primate models.

] GV_humans[test GV method in human cells

Step: refine the genomic vectoring method so that it works efficiently in human cells and doesn't mess with epigenomic correction. That means the GV method reliably, and at reasonable cost, produces cells with genomes that have a lot of the target DNA, and that have very little or no damage. To ensure genomic integrity, the stem cells will be deeply sequenced and checked for mutations. This step might require long-read sequencing in order to resolve any possible ambiguities such as copy number variation or chromosome breakage. GV methods can in theory combine in several ways with a method for ensuring epigenomic correctness: However, there's no clean separation between GV and EC; see the section “How GV and EC interact”. Therefore, the GV method has to be tested in combination with an EC method. Even if the EC method ought to be independent of the GV method, they still have to be tested together. For example, many GV methods would stress the stem cells being manipulated. Stressed cells could respond differently to the EC method.

] sub_GV_methods -..-> GV_mice sub_GV_methods -..-> GV_monkeys sub_GV_methods ---> GV_humans GV_mice -..-> GV_cows GV_mice -..-> GV_monkeys GV_monkeys -..-> GV_humans GV_verify[check for any harmful genomic aberrations caused by GV

Step: check children to see if they have any DNA damage. Before a live human birth is attempted, the cell line and embryo must be very thoroughly verified as having undamaged DNA. Nevertheless, it will probably be possible for damage to slip through the cracks. One issue is that DNA sequencing is usually not perfect, even with a high read depth, e.g. because of ambiguities in copy number. Another issue is that, even if the epigenomic correctness method has been rigorously validated, it could still produce somewhat stressed or otherwise very slightly epigenomically abnormal cells. In theory, those cells could have higher rates of DNA damage. A third reason is that most assisted reproductive technologies have to accept some small rate of DNA damage; natural reproduction produces children with genomes that have on the order of 100 de novo mutations. If the distribution of de novo mutations in genomically vectored cells is different from the natural distribution, it could in theory have a higher chance of producing harmful mutations. Therefore, babies should probably be sequenced again to triple-check that their DNA is intact, and should definitely have their health thoroughly checked to detect any harmful effects of DNA damage.

] GV_trait_verify[check actual effect on traits

Step: check children to see what effects the genomic vectoring had on their traits. Even if we have very good reason to expect the results to be safe and beneficial, we don't know what exactly the results of genomic vectoring will be. Polygenic predictors are quite far from perfectly predicting traits, and the meaning of any given trait isn't fully understood, and there could be hidden pleiotropies between measured and unmeasured traits. Therefore children with vectored genomes should be checked to see if the genomic vectoring had something like the expected effects on target traits, and to see if the GV may have had surprising effects on any traits, especially harmful ones.

] subgraph sub_PGS[ ] style sub_PGS fill:#c9daf8ff; header_PGS[understanding genetics of traits

Step: find some genetic variants that affect some traits that we care about. To know what genes to vector toward, we have to understand the relationship between genetic variants and traits of interest such as disease risk and cognitive capabilities. There's been a lot of research into these questions over the past two decades, with partial but substantive success, so this step is not a bottleneck for strong germline engineering.

] style header_PGS fill:transparent; header_PGS ~~~ psychometrics geno_pheno[gather genotype and phenotype data

Step: collect a bunch of datapoints that give a person's DNA and also some of their traits. Millions of whole genome sequences and tens of millions of genome-wide SNP-array genotypes have been collected from humans in the past decade. To be directly useful for germline engineering, genome data has to be paired with phenotype data, i.e. measurements of disease traits, other health traits, and cognitive or behavioral traits. There are several large national biobanks doing the work of data collection, so the data collection itself is not a strict bottleneck for strong germline engineering. However, more data is needed in several respects: higher quality phenotype measurements, e.g. good IQ tests and personality tests; whole genome sequences to detect rare variants; and more data from non-European populations, to make the benefits of germline engineering accessible to more populations and to better identify causal variants. Also, the data in major biobanks should be made more accessible to researchers studying the genetics of traits that people care about.

] make_PGS["construct polygenic scores

Step: build statistical models that identify which genetic variants have which effects on traits of interest. There's a large amount of existing work on constructing polygenic scores for disease traits and for some cognitive traits such as intelligence. See the PGS catalog. Thousands of polygenic scores have been constructed.

(Right diagram from Lello, Louis, Timothy G. Raben, Soke Yuen Yong, Laurent C. A. M. Tellier, and Stephen D. H. Hsu. ‘Genomic Prediction of 16 Complex Disease Risks Including Heart Attack, Diabetes, Breast and Prostate Cancer’. Scientific Reports 9, no. 1 (25 October 2019): 15286. https://doi.org/10.1038/s41598-019-51258-x.) Even as of 2018 there was a polygenic predictor for IQ that explained about 10% of the population variance in IQ. The causal nature of the genetic variants that show up in a polygenic score can be validated by looking at non-identical siblings raised in the same family, which naturally controls for environment. There is a problem of missing heritability, meaning that polygenic scores haven't been able to explain nearly as much of the variance in a trait as was estimated to be genetically heritable by other means. To some extent this will be solved with more data collection, and to some extent it means the heritability estimates were mistaken. However, it's plausible that some of the missing heritability could be found by using statistical models that are a little bit more complex than the linear models that have been used so far, e.g. small circuits with, say, up to 1000 latent variables. Such models might be simple enough to make use of the somewhat limited genomic data, without being so complex that they can't be trained.

" ] psychometrics["psychometrics

Step: figure out what cognitive traits there even are, and how to measure them. Disease traits are generally fairly clear: If someone gets sick enough to affect their life, you can at tell how it's affecting them, and if a lot of people have the same disease, we investigate and figure out the underlying causes. At least one cognitive trait is also fairly clear: The g factor, reflecting some aspects of intelligence, is robust (e.g. has test-retest reliability and correlates with many other life outcomes) and measurable. Many other behavioral traits, however, are harder to define, and harder to measure accurately. E.g. personality traits such as the Big Five, or even less well-defined traits that might be interesting for germline engineering, such as bravery, wisdom, kindness, perseverance, curiosity, imaginativeness, justness, insightfulness. Understanding these traits isn't strictly necessary to get large health and capability benefits using traits we already somewhat understand, such as disease risk and IQ. However, in the longer-term, the possibility for parents to choose moral enhancement of various kinds for their future children would help to make a world with germline engineering be a good world, by coupling increased capability with increased saneness and humaneness.

" ] psychometrics -.-> geno_pheno geno_pheno ---> make_PGS end make_PGS ----> GV_humans end sub_GV ~~~ GE_humans GV_humans ---> GE_humans subgraph sub_EC[ ] style sub_EC fill:#ea9999ff; header_EC["epigenomic correctness problem (EC)

Step: figure out how to make a cell that can grow into a healthy baby, other than just the usual natural way. To make a healthy baby from a cell, the cell has to have the right epigenomic state for embryonic develeopment. Epigenetic marks on and around DNA help control which genes are expressed. Natural reproductive cells, namely eggs, sperm, and zygotes, have specific epigenomic states that we'd have to duplicate. If you tried to make a baby from a cell with the wrong epigenomic state, you'd run a large risk of making a baby with major developmental problems. To address the epigenomic problem, we have to both understand what the correct states look like, and also be able to maintain or create the correct states in vitro. This is the main bottleneck to strong germline engineering. See the section “Reproductive GV and epigenomic correctness”.

"] style header_EC fill:transparent; classDef tightSep margin:-80px; header_EC ~~~ header_EC_making subgraph sub_EC_making [ ] style sub_EC_making fill:#f4aaaa; header_EC_making ~~~ group1 subgraph group1[ ] direction TB style group1 fill:transparent,stroke:none; EC_IVO ~~~ EC_crispr end subgraph group2[ ] direction TB style group2 fill:transparent,stroke:none; EC_natural_dna ~~~ EC_IVS ~~~ EC_maintenance end header_EC_making[making a cell epigenomically correct

Step: make a cell with the right epigenomic state so that it could grow into a healthy baby. The other nodes in this red box are different possible ways to get epigenomically correct cells. To make a cell that's epigenomically correct, the main ways are to either recapitulate natural gametogenesis in vitro, or to somehow piggyback off of natural reproduction. Natural gametogenesis is complex: Making an EC cell is the hard part of germline engineering. A haploid epigenomic correction method can make a haploid cell that's like a sperm, or that's like an egg. A diploid correction method can do both, or can directly make a diploic cell that's like a zygote. A diploid EC method would give twice as much genomic vectoring power as a haploid method. To develop EC methods that involve artificially steering the cell through epigenomic states, it would be helpful to understand how natural germline cells move through epigenomic states. To get that understanding, we need good atlases of epigenomic states in cells from different stages and tissues in natural primate reproduction.

] style header_EC_making fill:transparent; EC_IVO[in vitro oogenesis

Method: artificially make eggs from stem cells. In natural oogenesis, germline stem cells in the fetus are epigenomically reprogrammed over the course of a female's life, eventually becoming mature eggs. If we can mimic this process of oogenesis in the lab, we can turn stem cells into eggs that can make babies. This would be a full haploid epigenomic correction method that could combine with any genomic vectoring method. In vitro oogenesis has sort of been done in mice, but this method wouldn't transfer well to humans. There's been a lot of research on human in vitro oogenesis, but there's a gap in the crucial part of oogenesis where the epigenetic imprints are established.

Several academic labs have been working on this, and there are some companies working on commercializable aspects of IVO in order to treat infertility:

See the section “In vitro oogenesis”.

] EC_IVS[in vitro spermatogenesis

Method: artificially make sperm from stem cells. Like with oogenesis, if we could mimic natural spermatogenesis in the lab, we could turn stem cells into sperm cells with DNA that's competent to make a healthy baby. In vitro spermatogenesis would be a full haploid epigenomic correction method that could combine with any genomic vectoring method. In vitro spermatogenesis has sort of been done in mice. But for human in vitro spermatogenesis there's still a gap in the crucial part where the epigenetic imprints are established.

Several academic labs have been working on this, and one company is working on IVS in order to treat infertility: See the section “In vitro spermatogenesis”.

] EC_crispr[epigenetic CRISPR

Method: use epigenetic CRISPR to add in the necessary epigenetic marks. In theory, it should be possible to use CRISPR to edit the necessary epigenetic imprints. Doing full epigenetic correction with CRISPR would be difficult because it would require dozens of epigenetic edits of different kinds, and it would require a detailed knowledge of the necessary imprints. Epigenetic CRISPR might be helpful in combination with other epigenomic correction methods, to correct small epigenetic aberrations. See the section “Epigenetic CRISPR editing”.

] EC_natural_dna[natural reproductive DNA

Method: use DNA taken directly from eggs, sperm, or embryos.

The DNA in natural eggs, sperm, and embryos is usually in the right epigenomic state to make a developmentally healthy baby. If a genomic vectoring method can operate quickly on DNA in eggs, sperm, or embryos, then the epigenomic state would probably be preserved. So the result cell could still be used as a gamete or embryo. E.g. chromosome selection applied to gamete chromosomes, or a single round of CRISPR editing applied to a very early embryo. These EC bypass methods might be feasible sooner than other GV methods because they don't require epigenomic correction. See the section “Using natural reproductive DNA”.

] EC_maintenance["pausing natural EC

Method: maintain natural reproductive cells in their epigenomic state while you work on their genomes. This way of handling epigenomic correction involves taking natural germline cells, keeping them in stasis while you apply a genomic vectoring method, and then resuming their natural flow through the reproductive cycle. For example, it may be possible to grow spermatogonial stem cells in vitro, or to grow early embryonic stem cells. These cell cultures could for example be editing with CRISPR. However, when you culture stem cells, they tend to lose their stem-ness, and tend to lose some of their imprinting. So an EC pause method would have to solve the imprint maintenance problem. There's research looking into maintaining embryonic stem cells in their naive state, such as the super-SOX modified Yamanaka factor pictured here, but imprinting is usually lost to some extent. (Diagram from MacCarthy, Caitlin M., Guangming Wu, Vikas Malik, Yotam Menuchin-Lasowski, Taras Velychko, Gal Keshet, Rui Fan, et al. ‘Highly Cooperative Chimeric Super-SOX Induces Naive Pluripotency across Species’. Cell Stem Cell 31, no. 1 (4 January 2024): 127-147.e9. https://doi.org/10.1016/j.stem.2023.11.010.)

" ] end EC_mice[test EC method in mice; health studies of offspring

Step: apply your epigenomic correctness method in mice, and see if the mouse babies are healthy. Mouse models are a convenient way to test ways of addressing the epigenomic correctness problem. Since they are widely studied, there are good methods for handling them in the lab; they are relatively inexpensive; they have a relatively fast lifecycle; and their reproduction is relatively less sensitive to epigenetic aberration, so we can get live births more easily. There ought to be deep health studies of multiple generations of mice derived through in vitro gametogenesis, or other EC methods. Fast feedback about the health of offspring producing by some method should help us understand what methods work, how to make them work, what can go wrong, and how to tell at earlier stages of development when something is going wrong. However, the epigenetics of human reproduction are substantively different from that of mice, so results in mice won't fully generalize to humans; just because a method works well enough in mice, doesn't mean it'll work well enough in humans.

] EC_monkeys[test EC method in monkeys; health studies of offspring

Step: apply your epigenomic correctness method in monkeys, and see if the eggs / sperm / embryos look epigenomically identical to natural monkey eggs / sperm / embryos, and see if the monkey babies are healthy. Primate models are less convenient than mice. They have slower lifecycles, they're more expensive, they carry greater ethical weight, etc. But, their reproduction is much closer to human reproduction. Probably epigenomic correction methods such as in vitro oogenesis and in vitro spermatogenesis will be tested first in primate models, both to refine the method and then later to demonstrate that the method is safe enough to try in humans. One key way to guide experiments and validate results is to compare against the gold standards derived from primate epigenomic atlases. We can also compare morphology against normal developmental morphology. Since it's generally considered ethically acceptable to run small-scale experiments on some non-human primates, we can also get valuable end-to-end feedback about EC methods in primate models: We can make live births and then check to see if they're normal and healthy.

] EC_humans[test EC method in human cells

Step: apply your epigenomic correctness method in human cells, and see if the eggs / sperm / early embryos look epigenomically identical to natural human eggs / sperm / early embryos. As with monkeys, you guide experiments and validate results by comparing against the gold standards derived from human epigenomic atlases. It's not ethical, and is actually quite dangerous, to try to make a live human birth from a cell that we don't have very good reason to think is epigenomically competent. However, once the epigenomic state is normal, and the morphology of the early embryo is normal, and the EC method produces healthy live births in non-human primates, then it will be acceptable to try making a human live birth. In order to get better information about the health of a fetus produced by some EC method, it would be helpful to be able to study natural human fetuses for a little bit longer than the currently allowed 14 days. Therefore the 14-day limit should be extended to a 28-day limit.

] sub_EC_making -..-> EC_mice sub_EC_making ---> EC_monkeys EC_mice -..-> EC_monkeys ---> EC_humans EC_humans ~~~ dummy_EC[ ] style dummy_EC fill:transparent,stroke:none; subgraph sub_EC_standard[ ] style sub_EC_standard fill:#f4cccc; header_EC_understanding[understanding primate reproductive epigenomics

Step: figure out what the germline cells in natural reproduction look like, epigenomically. A crucial element of addressing the epigenomic correctness problem is understanding what epigenomic changes happen during natural reproduction, in the first place. Germline cells, destined to become eggs or sperm, are widely reprogrammed during fetal development, wiping the slate clean. Then they're reprogrammed more, both genome-wide and also in some sex-specific regions. The sex-specific epigenetic imprints are especially important for healthy development. But it's not fully understood how germline cells differentiate, which exact imprints are necessary for development, and what epigenomic states are normal and healthy in a developing embryo.

] style header_EC_understanding fill:transparent; class primateEpiAtlas donehalf; primateEpiAtlas[epigenomic sequencing; atlas of primate reproduction

Step: read the epigenomic states of as many cells as possible that are involved in natural reproduction. In the past decade, tools have emerged to read various aspects of cell state, including epigenetic marks. These methods can even apply to single cells, and potentially to multiple aspects of single cells. Now that we have these tools, researchers can build an atlas of epigenomic states of cells involved in reproduction, from fetal development, to the growth of ovaries, to various stages in gametogenesis. Some work is being done in this direction, but more is needed. We should map out both non-human primate reproduction and also human reproduction. Fetal tissue is not very accessible for humans, though.

] header_EC_understanding ~~~ primateEpiAtlas primate_gold_standard["gold standard healthy reproductive epigenomic state

Step: figure out what exactly cells should look like, epigenomically, in order to act exactly like normal reproductive cells. By examining epigenomic sequencing data from healthy non-human primates and humans, we can derive gold standard epigenomic states to aim for. If we take some cell, such as a stem cell from a very young embryo, we can sequence some epigenomic information about it (such as patterns of RNA expression, methylation, histone modifications, and/or chromatin accessibility). Then we can compare that to a gold standard for how an early embryonic stem cell ought to look, epigenomically speaking, or at least how they normally look in normal reproduction. The comparison tells us whether our cell is normal and therefore most likely healthy enough, or abnormal and therefore potentially unhealthy. We could use gold standards for what a normal sperm and egg should look like; we could also use gold standards for epigenomic (as well as morphological) changes throughout fetal development. Such gold standards are crucial to have: they tell us how specifically our EC methods are failing and what needs fixing, they tell us when our EC methods are working properly, and eventually they will provide evidence to show regulators and parents that the EC methods are safe.

" ] primateEpiAtlas ---> primate_gold_standard primate_epi_landscape["steer or maintain cells in differentiation landscape

Step: figure out how to make cells have the epigenomic state you want. Most ways of handling the epigenomic correction problem require nudging the epigenomic state of cells, either through a developmental progression of states (e.g. for in vitro oogenesis) or to keep them steady in one state (e.g. maintaining spermatogonial stem cells in culture). To nudge the epigenomic state of a cell, it helps to understand what exactly are the states you want to reach. If we had higher resolution epigenomic atlases, we'd be able to track progress with better accuracy, and potentially we could infer which transcription factors or other molecules would nudge cells the way we want to.

" ] primateEpiAtlas ---> primate_epi_landscape end primate_epi_landscape ---> EC_monkeys primate_epi_landscape ---> EC_humans primate_gold_standard ---> EC_monkeys primate_gold_standard ---> EC_humans EC_verify[check for any harmful effects of epigenomic abnormality

Step: check children to see if they're developing normally. Children born from epigenomically corrected cells should have their health monitored, and in particular checked for any health problems or developmental abnormalities that are associated with epigenomic aberrations. E.g., if they have any of the health problems associated with any uniparental disomies.

] end GE_humans[strong human germline genomic engineering

Step: make a live human baby with a genome that's been nudged in some directions, and with an epigenome that's competent to grow healthily. Once the methods for epigenomic correctness and genomic vectoring have been worked out and validated as safe in animal models, they can be tried in humans. The parents and scientists will construct a genomic vectoring target that the parents want, and that will be safe as a GV target. Existing embryo selection companies, such as Orchid, Lifeview, Heliospect, and GenEmbryonics, could be asked to provide lessons from their experience producing practical combined polygenic scores and working with parents. The embryo and fetus should be tested as much as is feasible to check that it is developing normally and healthily. The embryo can be sequenced for genomic integrity and for epigenomic normalness, comparing against gold standards. Likewise, the embryo's morphology can be checked for health, as with any assisted reproductive technology. Since it's hard to access a fetus growing in a womb, monitoring will be somewhat limited, but in the future artificial wombs could help gain more understanding of healthy development and ability to monitor growing fetuses.

] style GE_humans fill:#ffccff EC_humans ---> GE_humans followup_humans["followup studies on health and traits

Step: followup with people born from genomically vectored cells to check how they're doing. People who are born from cells that were genomically vectored and epigenomically corrected should be studied so that any deleterious effects of the assisted reproductive technology can be noticed, and intended effects can be validated. Data privacy should be especially protected.

" ] GE_humans -------> followup_humans followup_humans ---> GV_verify followup_humans ---> GV_trait_verify GV_trait_verify ~~~ GV_verify GV_humans -.- GV_verify sub_PGS -.- GV_trait_verify followup_humans ---> EC_verify EC_verify -.- EC_humans classDef donehalf stroke:orange;