Eggs are expensive, sperm are cheap. It’s a fundamental fact of biology . . . for now.

Currently, embryo selection can improve any heritable trait, but the degree of improvement is limited by the number of embryos from which to select. This, in turn, is because eggs are rare.

But what if we could select on sperm instead? We could choose the best sperm from tens or even hundreds of millions, and use that to make an embryo. However, any method that relies on DNA sequencing must destroy the sperm. Sure, you can identify the best one, but that’s of limited value if you can’t use it for fertilizing an egg.

There have been a few ways proposed to get around this:

1. Nondestructive sperm testing. Technically challenging: sperm DNA is packaged tightly and you would have to partially denature it without killing the cell. Selection based on total DNA content (separating X and Y bearing sperm) is possible but only useful for choosing the sex of the baby. Phenotypic selection (swim rate, etc) is not very useful because sperm phenotypes don’t correlate well with sperm genotypes.
2. Doing in vitro spermatogenesis, and keeping track of which sperm came from where.^[1] There are four sperm produced from each spermatocyte, and three of them could be destructively sequenced to deduce the genotype of the remaining one. Challenging (nobody has done human in vitro spermatogenesis yet) and low throughput.

Here, I propose a different approach, which I call androgenetic haploid selection.

Androgenetic haploid selection

1. Make a bunch of eggs. The chromosomes and imprinting don’t have to be correct (we’ll get rid of them in the next step), so even a low quality in vitro oogenesis method would work. Something like Hamazaki’s approach would work well here.
2. Remove the chromosomes from the eggs. This can be done at large scale through centrifugation: spin the eggs hard enough, and the DNA will fall out.
3. Add an individual sperm to each egg and establish haploid stem cell lines. This recent paper is an example of doing this for cows and sheep. These cell lines are called “androgenetic” and retain the DNA imprinting patterns of sperm.
   1. Notably, Y-bearing sperm cannot make viable haploid stem cell lines because many essential genes are on the X chromosome.
4. Sequence many cell lines and choose the best one. Because the cells divide, it’s possible to destructively sequence some of the cells from each line without destroying all the cells.
5. Collect eggs the normal way, and “fertilize” them with nuclei from your chosen androgenetic cell line.
   1. Optionally: perform additional selection based on the embryo genome.

Comments on this approach

1. This method could give high genetic optimization for the paternal half of the genome. At scale, I estimate an overall $200/sample cost for cell line establishment and sequencing, so taking the best of 100 cell lines could be performed for around the cost of a normal IVF cycle (~$20,000). For a perfectly heritable trait with a perfect polygenic score, this would give (+2.5 SD * 0.5) = +1.25 SD from sperm selection alone. (Gains will be lower for less heritable traits and less accurate predictors.)
2. This would only work for daughters (sorry Elon!) Although genetic engineering could make XX males by adding SRY, this would probably not be a good idea.
3. This would make even a low-quality in vitro oogenesis method valuable. More broadly, it’s not necessarily required that the recipient cells be eggs per se, as long as they express the correct factors for zygotic genome activation.

1. ^{^}

   This would have to be done at the spermatid stage, before the sperm swim away.