Part 3 — Reading the evidence

10. How to read a scientific paper

You will never read papers the way a specialist does, and you don't need to. You need to extract the claim, the evidence, and the weakness. Here is a method that works.

Read in this order — not front to back:

flowchart TD
    A["1. <b>Title + Abstract</b><br/>What do they claim?"] --> B["2. <b>Figures + captions</b><br/>What did they actually show?<br/><i>This is where the truth lives</i>"]
    B --> C["3. <b>Last paragraph of Discussion</b><br/>What do they admit are the limits?"]
    C --> D["4. <b>Methods</b> — only if it matters<br/>How many? What controls?"]
    D --> E["5. <b>Intro</b> — only for background"]

The figures are the paper. Text is interpretation; figures are data. If a strong claim in the abstract isn't visible in a figure, be skeptical.

Six questions to ask of any paper:

Practical access tips:

• Many papers are paywalled. Try PubMed Central for free full text, or Europe PMC.
• Preprints (bioRxiv, medRxiv) are papers posted before peer review. Often the newest science — and explicitly not yet vetted. Treat accordingly, but read them: in a fast-moving area they're where the field actually is.
• Authors will frequently send you a PDF if you email and explain you're a parent. This works far more often than people expect.

Key terms:

• Abstract — the summary at the top.
• Peer review — evaluation by independent scientists before publication.
• Preprint — posted publicly before peer review.
• Control — the comparison condition.
• n — the number of independent samples.
• Replication — independent repetition of an experiment.
• Impact factor — a journal-level metric; a weak proxy for the quality of any individual paper.

References:

• PubMed · PubMed Central · Europe PMC · bioRxiv

11. Study types and the evidence hierarchy

Not all evidence is equal. This ladder is how scientists implicitly rank it — and knowing where a claim sits tells you how much weight it can bear.

flowchart BT
    A["<b>In silico</b> — computer prediction<br/><i>a hypothesis generator, not evidence</i>"] --> B["<b>In vitro</b> — cells in a dish<br/><i>where most rare-disease programs begin</i>"]
    B --> C["<b>In vivo, animal</b> — mouse / zebrafish<br/><i>a whole body, wrong species</i>"]
    C --> D["<b>Case report / N-of-1</b><br/><i>one human, no comparison group</i>"]
    D --> E["<b>Open-label trial</b><br/><i>humans, but everyone knows who got the drug</i>"]
    E --> F["<b>Randomized controlled trial</b><br/><i>the gold standard</i>"]
    F --> G["<b>Meta-analysis</b><br/><i>many trials pooled</i>"]

The honest situation in ultra-rare disease: you will often never reach the top of this ladder. There aren't enough patients on Earth for a randomized trial in Xq25 duplication syndrome. This is not a failure — it is a structural feature of ultra-rare disease, and regulators increasingly accept it (Module 24).

What follows is that the lower rungs have to be unusually rigorous, because they carry more weight than they normally would. That is precisely why the quality of the cell work matters so much here.

The attrition problem — internalize these numbers. Of compounds that look promising in cells, only a minority work in animals; of those, a minority work in humans; of those, a minority are approved. Roughly 90%+ of drugs entering human trials fail. For repurposed drugs the odds improve (safety is already known), but the efficacy failure rate is still high.

This is not a reason for despair. It's a reason to:

• expect and plan for multiple hits to fail,
• never reorganize your family's hopes around a single compound,
• and push hard on the quality of the early evidence, since that's the one lever that improves the odds.

Key terms:

• In silico / in vitro / in vivo — in computer / in glass / in a living organism.
• Randomized controlled trial (RCT) — participants randomly assigned; the strongest single design.
• Open-label — everyone knows who receives the treatment.
• Placebo effect — improvement from expectation alone; the reason blinding exists.
• Blinding — concealing treatment assignment from participants and/or assessors.
• N-of-1 — a formal study design in a single patient.
• Effect size — how big the difference is, as opposed to how confident we are it's non-zero.

References:

• ClinicalTrials.gov — Learn About Studies — plain-language explanation of trial types and phases.

12. Reversibility, and the MECP2 duplication precedent

This is the module that should give you grounded hope. Not vague optimism — specific, cited, and appropriately bounded.

The question every parent asks. A child with the duplication has had too much STAG2 in every cell, throughout brain development, from the beginning. By the time a diagnosis arrives — often at school age or later — is it too late?

The honest answer is probably not entirely too late — and the strongest evidence comes from the closest sister disease.

Why MECP2 duplication syndrome is the right comparison. Look at how closely the two conditions parallel:

The two diseases are structurally almost the same problem. What works there is genuinely informative here.

Finding 1 — Reversal has been shown in animals. In mouse models of MECP2 duplication, lowering MeCP2 back toward normal using antisense oligonucleotides improved or reversed neurological abnormalities — including in animals treated after symptoms had developed. This is the core evidence that a "too much of a gene" brain disorder is at least partially correctable after development, rather than being a fixed structural outcome.

Finding 2 — It works in human patient-derived neurons. ASO treatment of iPSC-derived neurons from MECP2 duplication patients knocked down MECP2 and revealed gene-expression programs responsive to MeCP2 levels. That is exactly the experiment that patient-derived neurons make possible for STAG2 — and it is a strong methodological template to point a collaborating lab toward.

Finding 3 — and this is the big one — it is now in humans. A first-in-human trial is dosing children:

Why this matters so much here, stated precisely:

1. The concept is no longer theoretical. "Partially lower an overexpressed, dosage-sensitive gene in a child's brain" is a thing that is being done, in children, right now, with regulators on board.
2. The platform is retargetable in principle. Cas13-based RNA editing works by recognizing a target mRNA sequence. Pointing it at a different mRNA is, at the platform level, a change of targeting sequence rather than a new invention. This does not mean it is easy — it still requires target validation, dosing, delivery, safety, manufacturing and regulatory work. But it means there is an existing platform and an existing organization for whom STAG2 would be an adjacent problem, not a foreign one.
3. The age range is broad. Enrolment is 2–18. That is a meaningful signal about what is currently considered a reasonable treatment window — it is not restricted to infants.
4. RNA editing is reversible-ish and dose-tunable — attractive given the tumor-suppressor constraint from Module 4.

The honest bounds. Preclinical reversal in mice is not proof of benefit in a human child. The HERO trial has not reported efficacy results. MECP2 is not STAG2, and biology has repeatedly punished assumptions of transferability. The correct posture is: this substantially raises the plausibility of the whole strategy, and it identifies a concrete organization and modality worth approaching — not this is a treatment for your child.

What "reversible" realistically means. The most defensible expectation is meaningful improvement and halted decline, not a return to an unaffected baseline. This condition is developmental (stable) rather than degenerative (worsening), which has a useful corollary: against a steady baseline, a real treatment effect should be visible rather than lost in noise. That is also exactly why documenting your child's baseline now matters so much (Module 25).

Key terms:

• MECP2 duplication syndrome — the closest sister disease; the key precedent.
• Rett syndrome — caused by too little MeCP2; the opposite-direction disease.
• ASO (antisense oligonucleotide) — short synthetic DNA/RNA that reduces a specific mRNA.
• Cas13 — a CRISPR enzyme that targets RNA rather than DNA.
• AAV (adeno-associated virus) — a common, generally well-tolerated gene-therapy delivery vehicle.
• ICV (intracerebroventricular) — injection into the fluid-filled ventricles of the brain.
• Orphan Drug Designation — FDA status for rare-disease drugs, giving incentives.
• Rare Pediatric Disease Designation — FDA status that can yield a priority review voucher.

References:

• NCT06615206 — HERO trial record — the primary source. Check it periodically for status changes.
• HuidaGene — Nature Neuroscience publication announcement for HG204
• HuidaGene — FDA Rare Pediatric Disease + Orphan Drug designations for HG204
• Antisense oligonucleotide therapy in a humanized mouse model of MECP2 duplication syndrome — Science Translational Medicine
• Modeling ASO therapy in MECP2 duplication iPSC-derived neurons — Human Molecular Genetics 2024 — the closest methodological template for patient-derived neurons carrying an Xq25 duplication.
• Baylor College of Medicine — ASOs to treat MECP2 disorder

13. The databases you should know how to use

Five resources cover almost everything you'll need. Learning to use them yourself removes your dependence on others for basic facts.

Also worth knowing:

• gnomAD — how often variants occur in the general population. Useful for judging whether something is actually rare.
• GeneCards — fast orientation on any gene.
• Google Scholar — set up an alert for STAG2 duplication and for cohesinopathy therapy. New papers arrive in your inbox automatically. Do this today — it takes two minutes and pays off for years.

A power technique: citation chaining. Take the Kumar 2015 paper and look at who has cited it since. On PubMed and Google Scholar there's a "Cited by" link. Anyone citing that paper is, by definition, working on or near this condition. This is the single most efficient way to find both new science and potential collaborators, and it costs nothing.

Key terms:

• PMID — PubMed's unique article identifier.
• DOI — a permanent identifier for a publication.
• Citation chaining — following citations forward or backward to map a field.
• Allele frequency — how common a variant is in the population.

References:

• Kumar 2015 "Cited by" on PubMed — scroll to the citation section and work through it periodically.