Editing life

Foundations

Where the tool came from

Bacteria are attacked by viruses, and some of them keep a library.

CRISPR sequences are stretches of bacterial DNA containing fragments of viruses the lineage has survived. Paired with Cas proteins, they let the cell recognise a returning virus and cut its DNA. It is an immune system with a memory, running in organisms with no cells to spare for one.

In 2012 came the result that turned it into a tool. Cas9 could be directed to cut any chosen DNA sequence by supplying a matching guide RNA.

That is the whole invention. The system is billions of years old; the programmability is ours. And it made genome editing cheap enough for an ordinary laboratory, which is why the eleven years from that paper to an approved medicine is one of the fastest translations in the history of biology.

The distinction that carries the entire ethics

Before anything else, one line that everything depends on.

A somatic edit changes cells in one person’s body. It affects that person, it can be consented to, and it dies with them.

A germline edit changes eggs, sperm or early embryos. It is inherited by every descendant, forever, and the people most affected by it do not exist yet and cannot agree to it.

Nearly every argument in this field turns out, on inspection, to be about which of these two is under discussion. Keep them separate and the subject becomes tractable.

The achievement, which is real

In 2021 a trial reported results for patients with sickle cell disease and beta-thalassaemia treated with a CRISPR-Cas9 therapy. Patients who had lived with transfusions and pain crises no longer needed them.

And the way it works is the most elegant thing in this lesson.

Sickle cell disease is caused by a single letter change in the adult haemoglobin gene, which makes the protein polymerise and distort red cells into a rigid sickle shape. The obvious fix is to correct that letter.

That is not what the therapy does.

Before birth, we make a different haemoglobin — fetal haemoglobin — which is switched off shortly after birth by a regulatory system involving a gene called BCL11A. People who happen to keep making fetal haemoglobin into adulthood have much milder disease, and this has been known for decades.

So the therapy takes the patient’s own blood stem cells, uses CRISPR to disrupt the regulatory element that switches fetal haemoglobin off, and puts the cells back. The broken gene is left exactly as it is. The body starts making the other protein again.

It is a repair that routes around the fault instead of fixing it, and it works because evolution left an unused part in place. Lesson 5.3 said natural selection has no foresight and builds by modifying what is already there. This is a therapy that exploits precisely that fact.

The sentence this lesson exists for

Now look at where the disease is.

Sickle cell disease is concentrated in sub-Saharan Africa and India. The allele persists at high frequency because carrying one copy confers protection against malaria — a textbook case of balancing selection, and a direct consequence of where malaria has been endemic.

Within India the distribution is highly uneven, with elevated frequencies in central and eastern India and particularly among Scheduled Tribe communities. Mapping it properly required combining hundreds of local surveys.

Now look at what the cure requires. Harvesting the patient’s blood stem cells. Editing them in a specialised facility. Myeloablative conditioning — chemotherapy that destroys the patient’s existing bone marrow. Weeks of inpatient care in a transplant unit with the infection control that implies. Then long-term follow-up.

And the price. Economic analyses of a hypothetical curative therapy for sickle cell disease found that prices in the low millions of dollars per patient could be justified as cost-effective — in a high-income health system, against that system’s lifetime costs of treating the disease.

Put the two maps together. A cure exists for a disease whose burden falls overwhelmingly on the people and health systems least able to deliver a transplant ward or pay a seven-figure price.

This is not an argument against the technology. The science is superb and the patients in that trial are genuinely better. It is a statement about what “we can now cure X” means, and it is the same structure as lesson 8.4’s Green Revolution: a real achievement whose distribution is a separate problem that the achievement does not solve.

Now dismantle the film version

The public conversation about gene editing is almost entirely about designer babies: choosing a child’s intelligence, height, appearance.

The main obstacle is not law or ethics. It is that the biology does not work that way.

Height and cognitive test performance are polygenic. Not a few genes — thousands of variants, each shifting the trait by a tiny amount, interacting with each other and with environment.

Which produces four separate problems, and each is enough on its own.

There is no small set of edits that produces the trait. You would need thousands, each with its own risk of going wrong.

Predictive scores built from many variants explain only a limited fraction of the variation, so even selecting among existing embryos gives far less than people imagine.

Those scores transfer poorly between populations, because they are built largely from European-ancestry datasets and the statistical associations do not carry over. For most Indians, the predictive power is worse still.

And many variants do more than one thing. Editing for a trait means accepting unknown effects on everything else that variant touches.

So the honest summary: single-gene disease is tractable. Complex traits are not, and not because we lack ambition.

The near-term ethical questions are more mundane and more real than the film version — who gets access, what counts as a disease worth editing, and who decides.

The line that was crossed

In 2018, a researcher announced the birth of children whose embryos he had edited. The response from the scientific community was immediate and overwhelmingly condemnatory.

The condemnation was not merely about procedure, though the procedural failures were severe. The substantive objections were:

The edit targeted CCR5, aiming at HIV resistance, in a situation where established methods already prevent transmission from an infected parent. There was no unmet need.

CCR5 does other things. Disabling it has consequences that are not fully characterised.

The technical risks were live: off-target edits elsewhere in the genome, and mosaicism, where an edit reaches some embryonic cells and not others, producing a patchwork organism.

And the edits are heritable by people who cannot consent.

What makes this worth teaching is that a scientific statement three years earlier had already recommended against exactly this, pending better understanding and broader public discussion. The community had drawn a line in advance, and the line held afterwards — the response was near-unanimous. That is a rare and instructive example of self-governance in science actually functioning.

The one that cannot be taken back

Save the biggest for last.

A normal gene is inherited by half of an organism’s offspring. A gene drive is engineered to copy itself onto the partner chromosome, so it is inherited by nearly all of them. Release it into a wild population and it spreads on its own, generation after generation, whether or not it helps the organism carrying it.

In caged experiments, a CRISPR drive targeting the doublesex gene in Anopheles gambiae — the principal African malaria vector — caused complete population suppression, with no resistance appearing during the experiment.

Malaria kills hundreds of thousands of people a year, most of them children. The case for using this is not weak.

And this is the first technology deliberately designed to alter a species in the wild, permanently, without a recall mechanism. It does not respect national borders. There is no meaningful way to un-release it. Whatever ecological role that mosquito plays is removed by the same action.

Which makes the decision to release one unlike any biotechnology decision made before, and it is worth noticing that no existing institution is obviously the right one to make it.

And the version already growing

None of this is as new as it feels, which is the last point.

Lesson 7.2 described domestication: ten thousand years of humans reshaping the genomes of wheat, rice, cattle and dogs by selective breeding. We have been editing genomes for a hundred centuries. What changed is precision, speed, and the ability to move sequence between lineages that could not exchange it.

India’s own live case is Bt cotton, carrying a bacterial gene for an insecticidal protein — the one genetically modified crop approved for commercial cultivation there. It was adopted by the large majority of Indian cotton farmers, and panel data studies report associated yield and profit increases.

It is also one of the most bitterly contested technologies in Indian agriculture, over seed pricing and control, farmer debt, pest resistance emerging over time, and trust in the regulatory process.

Both of those are true at once, and a course that reported only one would be doing propaganda. The economic studies say what they say. The controversy is about ownership, dependency and governance, and those are not questions the yield data answers.

Which is the pattern of this entire chapter. The technical questions are hard and are being answered. The questions about who benefits are harder and are not.

Deeper

Why fetal haemoglobin works as a target

Haemoglobin is a tetramer. Adult haemoglobin uses two alpha and two beta chains; fetal haemoglobin substitutes gamma chains for beta. The sickle mutation is in the beta chain, so gamma-containing tetramers are unaffected, and their presence interferes with polymerisation of sickle haemoglobin. Individuals with hereditary persistence of fetal haemoglobin have long been observed to have milder disease, which identified the target long before a tool existed to act on it. Editing the erythroid-specific enhancer of BCL11A rather than the coding sequence limits the effect to the red cell lineage, which matters because BCL11A has other roles including in neural development. The design is a good example of choosing a target for its regulatory specificity rather than its obviousness.

What conditioning means in practice

Autologous stem cell therapies require making space in the marrow for the edited cells to engraft, which is done with chemotherapy that destroys existing haematopoietic cells. The patient is then profoundly immunosuppressed until engraftment, with the infection risk and supportive care requirements that implies, and the conditioning agents carry their own long-term risks including infertility and secondary malignancy. This is why the access problem is not solved by lowering the price of the edit itself: the surrounding procedure requires infrastructure that most of the affected population does not live near. Non-genotoxic conditioning is an active research area precisely because of this.

Why polygenic scores transfer badly

A polygenic score weights variants by associations estimated in a particular sample. Those associations depend on linkage disequilibrium patterns, allele frequencies and environmental context, all of which differ between ancestries. A score trained in one population therefore loses predictive power in another, often substantially. Since the large biobanks are heavily European in composition, this is a systematic problem for most of the world, and it is not fixed by larger samples of the same population. It bears directly on embryo selection claims: a service offering polygenic selection to Indian parents is applying weights estimated somewhere else.

How a gene drive spreads and what stops it

The drive element encodes a nuclease and a guide targeting the equivalent site on the homologous chromosome. In the germline it cuts the wild-type allele, and homology-directed repair copies the drive across, converting a heterozygote to a homozygote. Inheritance therefore approaches one hundred per cent. The standard failure mode is resistance: end-joining repair produces sequences the guide no longer recognises, and those spread instead. The doublesex target was chosen because the sequence is functionally constrained, so most resistant variants are themselves non-functional. That the caged experiments showed no resistance is a strong result at that scale, and cage experiments are not field releases.

Frontier

Reaching the patients

The technical direction that would change the access picture is in vivo editing, delivering the editor to the patient’s own marrow rather than removing, editing and returning cells, and non-genotoxic conditioning to remove the chemotherapy requirement. Both are active. Whether either arrives at a price and a complexity appropriate to a district hospital in Chhattisgarh is the question that determines whether this technology reaches the people who need it most.

Governing releases

There is no settled international mechanism for deciding whether a gene drive is released, and the affected populations span borders. Proposals range from treaty instruments to community-level consent processes in the areas concerned. This is an unresolved governance problem attached to a technology that already works in the laboratory.

Transcript and shot list

Target runtime about 11 minutes. Built to lead with the real achievement, dismantle the film version, then spend the second half on access and on the one technology that cannot be taken back. Shot types: ANIM is rendered from code with Manim, AI is a generated shot from Higgsfield with the shared style reference, REAL is agency imagery with credit, CAM is the narrator.

[00:00] S01 · ANIM · A bacterium storing fragments of viral DNA from surviving infections; a returning virus recognised and its DNA cut. Then 2012: a guide RNA supplied by a human, and Cas9 directed to any chosen sequence.

Bacteria get attacked by viruses, and some of them keep a library. CRISPR sequences are stretches of bacterial DNA holding fragments of viruses the lineage survived, and paired with Cas proteins they let the cell recognise a returning virus and cut it. An immune system with a memory, in an organism with no cells to spare for one. Then in 2012 came the result that turned it into a tool: Cas9 could be directed to cut any chosen sequence by supplying a matching guide RNA. That is the whole invention. The system is billions of years old. The programmability is ours.

[00:47] S02 · CAM · Narrator direct to camera.

Before anything else, one distinction that everything in this subject depends on. A somatic edit changes cells in one person’s body. It affects that person, they can consent to it, and it dies with them. A germline edit changes eggs, sperm or embryos. It is inherited by every descendant, and the people most affected by it do not exist yet and cannot agree to anything. Almost every argument in this field turns out to be about which of those two is being discussed.

[01:25] S03 · ANIM · Sickle cell disease: a single letter change in the adult haemoglobin gene; the protein polymerising; red cells distorting into rigid sickles; blocked capillaries and pain crises.

In 2021 a trial reported results for patients with sickle cell disease and beta-thalassaemia treated with a CRISPR therapy. People who had lived with transfusions and pain crises no longer needed them. Sickle cell disease comes from a single letter change in the adult haemoglobin gene, which makes the protein polymerise and bend red cells into a rigid sickle shape.

[01:53] S04 · CAM · Narrator direct to camera.

So the obvious fix is to correct that letter. And that is not what the therapy does.

[02:00] S05 · ANIM · Fetal haemoglobin produced before birth, switched off after by a regulatory system involving BCL11A. People who keep making it having milder disease. Then the therapy: patient’s own stem cells taken, the regulatory element disrupted, cells returned, fetal haemoglobin restarting — the broken gene untouched.

Before birth we make a different haemoglobin, fetal haemoglobin, and it gets switched off shortly after birth by a regulatory system involving a gene called BCL11A. People who happen to carry on making fetal haemoglobin into adulthood have much milder disease, and that has been known for decades. So the therapy takes the patient’s own blood stem cells, uses CRISPR to disrupt the regulatory element that does the switching off, and puts the cells back. The broken gene is left exactly where it is. The body simply starts making the other protein again. It is a repair that routes around the fault instead of fixing it, and it works because evolution left an unused part lying about. Lesson 5.3 said natural selection has no foresight and builds by modifying what is already there. This is a therapy that runs on that fact.

[03:06] S06 · ANIM · The global sickle cell burden map: sub-Saharan Africa and India. The malaria overlay explaining the allele’s persistence through carrier protection. Then India’s internal distribution, elevated in central and eastern India and among Scheduled Tribe communities.

Now look at where the disease actually is. Sickle cell is concentrated in sub-Saharan Africa and India. The allele persists at high frequency because carrying one copy protects against malaria — a textbook case of balancing selection, and a direct consequence of where malaria has been endemic. Inside India the distribution is very uneven, elevated in central and eastern India and particularly among Scheduled Tribe communities. Mapping it properly meant combining hundreds of local surveys.

[03:41] S07 · ANIM · What the cure requires, itemised: stem cell harvest, editing in a specialised facility, myeloablative conditioning, weeks of transplant-unit inpatient care, long-term follow-up. Then the modelled price in the low millions per patient, justified against a high-income system’s lifetime costs. The two maps overlaid.

Now look at what the cure requires. Harvest the patient’s blood stem cells. Edit them in a specialised facility. Then myeloablative conditioning — chemotherapy that destroys the existing bone marrow. Then weeks of inpatient care in a transplant unit with the infection control that implies. Then long-term follow-up. And the price: analyses of a hypothetical curative therapy found that prices in the low millions of dollars per patient could be justified as cost-effective — in a high-income health system, against that system’s lifetime cost of treating the disease. Put the two maps on top of each other. A cure exists for a disease whose burden falls overwhelmingly on the people and the health systems least able to provide a transplant ward or a seven-figure sum. That is not an argument against the technology. The science is superb and those patients are genuinely better. It is a statement about what we can now cure X actually means. And it is the same structure as the Green Revolution in lesson 8.4: a real achievement whose distribution is a separate problem the achievement does not solve.

[05:05] S08 · ANIM · The designer-baby premise taken apart: thousands of variants of tiny effect for height and cognitive scores; four failures listed — no small edit set, limited variance explained, poor transfer across populations, variants doing several things at once.

Now the film version. Public conversation about gene editing is almost entirely about designer babies — choosing a child’s intelligence, height, appearance. And the main obstacle is not law and it is not ethics. It is that the biology does not work like that. Height and cognitive test performance are polygenic. Not a few genes. Thousands of variants, each shifting the trait a tiny amount, interacting with each other and with environment. Which gives you four separate problems, any one of which is fatal. There is no small set of edits that produces the trait; you would need thousands, each with its own chance of going wrong. Predictive scores built from many variants explain only a limited fraction of the variation, so even selecting among existing embryos gives you far less than people imagine. Those scores transfer badly between populations, because they are built largely from European-ancestry data and the statistical associations do not carry across — for most Indians the predictive power is worse again. And many variants do more than one thing, so editing for a trait means accepting unknown effects on everything else that variant touches.

[06:32] S09 · CAM · Narrator direct to camera.

So: single-gene disease is tractable. Complex traits are not, and not because we lack ambition. Which means the near-term ethical questions are more mundane and much more real than the film version. Who gets access. What counts as a disease worth editing. Who decides.

[06:53] S10 · ANIM · 2018: germline-edited births announced; near-unanimous condemnation. The substantive objections listed — no unmet need given existing prevention, CCR5’s other functions, off-target edits, mosaicism, heritability without consent. Then the 2015 statement that had already drawn the line, three years earlier.

In 2018 a researcher announced the birth of children whose embryos he had edited, and the response from the scientific community was immediate and overwhelmingly condemnatory. Not just about procedure, though the procedural failures were severe. The substantive objections were that the edit targeted CCR5 for HIV resistance in a situation where established methods already prevent transmission, so there was no unmet need; that CCR5 does other things and disabling it has consequences that are not fully characterised; that the technical risks were live, including off-target edits elsewhere in the genome and mosaicism, where an edit reaches some embryonic cells and not others and you get a patchwork; and that the edits are heritable by people who cannot consent. What makes this worth teaching is that a scientific statement three years earlier had already recommended against exactly this, pending better understanding and public discussion. The line was drawn in advance and it held afterwards. That is a rare and instructive case of scientific self-governance actually working.

[08:09] S11 · ANIM · Normal inheritance at fifty per cent; then a drive element copying itself onto the partner chromosome and approaching one hundred per cent. It spreading generation on generation through a wild population, crossing borders, with no recall.

Now the biggest one. A normal gene is inherited by half an organism’s offspring. A gene drive is engineered to copy itself onto the partner chromosome, so it is inherited by nearly all of them. Release it into a wild population and it spreads on its own, generation after generation, whether or not it helps the organism carrying it.

[08:36] S12 · ANIM · Caged Anopheles gambiae; a drive targeting doublesex; complete population suppression with no resistance appearing. Then malaria mortality, mostly children. Then the four irreversibilities: wild, permanent, borderless, no recall.

In caged experiments, a CRISPR drive targeting the doublesex gene in Anopheles gambiae, the main African malaria vector, caused complete population suppression, with no resistance appearing during the experiment. Malaria kills hundreds of thousands of people a year, most of them children. The case for using this is not weak. And this is the first technology deliberately designed to alter a species in the wild, permanently, with no recall mechanism. It does not respect borders. There is no meaningful way to un-release it. And whatever ecological role that mosquito plays goes with it. Which makes that decision unlike any biotechnology decision anyone has made before, and it is worth noticing that no existing institution is obviously the right one to make it.

[09:33] S13 · ANIM · Ten thousand years of selective breeding reshaping wheat, rice, cattle and dogs, recalled from lesson 7.2. Then what is actually new: precision, speed, and moving sequence between lineages that could not exchange it. Then Bt cotton in India — majority adoption, reported yield and profit gains — set beside the contested ground of seed pricing, debt, resistance and regulatory trust.

And none of this is as new as it feels. Lesson 7.2 was ten thousand years of humans reshaping the genomes of wheat, rice, cattle and dogs by selective breeding. We have been editing genomes for a hundred centuries. What changed is precision, speed, and the ability to move sequence between lineages that could not exchange it. India’s own live case is Bt cotton, carrying a bacterial gene for an insecticidal protein, the one genetically modified crop approved for commercial cultivation there. It was adopted by the large majority of Indian cotton farmers, and panel studies report associated yield and profit increases. It is also one of the most bitterly contested technologies in Indian agriculture, over seed pricing and control, farmer debt, pest resistance emerging over time, and trust in the regulatory process. Both of those are true at once, and a course that told you only one would be doing propaganda. The economic studies say what they say. The controversy is about ownership, dependency and governance, and the yield data does not answer those.

[10:54] S14 · CAM · Narrator direct to camera, then calendar returns.

Which is the pattern of this whole chapter. The technical questions are hard and they are being answered. The questions about who benefits are harder and they are not. Next: whether any of this travels off the planet.

Shot list

Zero AI shots, ten code-rendered shots, four narrator shots. Generated imagery of patients or of edited children would be indefensible in this lesson; the diagrams do the work.