Is DRD2 Really a Depression Drug Target? How Colocalization Demoted the Top Signal and Put SLC12A5 (KCC2) Forward

This is an accessible companion to our preprint. Preprint (Research Square, v1, not yet peer-reviewed): https://www.researchsquare.com/article/rs-9880173/v1 · Code & extraction data: available from the author on request (every input is public — see Resources). Everything here is computational, hypothesis-generating work — not clinical advice or an approved treatment claim.

TL;DR (Quick Answer)

We re-analyzed the latest major depressive disorder (MDD) GWAS to ask a practical question: which genes are real, causal drug targets — not just statistical shadows of nearby variants?

• The top hit was a trap. DRD2 (the dopamine D2 receptor) was the single strongest brain-TWAS signal (P = 4.43×10⁻²⁸), and its direction implied agonism would help, not the blockade used by antipsychotics. But under colocalization it did not share the same causal variant as the GWAS signal (COLOC PP4 ≈ 0) — so we demote DRD2-based dopaminergic drugs from a high-confidence shortlist to an exploratory tier.
• One gene passed every test. Of 208 TWAS genes tested, only 30 (14%) colocalized, and SLC12A5 (KCC2) alone cleared all three (SMR, HEIDI, and coloc PP4 = 0.996) — making it our confirmed causal lead.
• Honest caveats. "Did not colocalize" means not confirmed via cis-regulation, not proven non-causal. SLC12A5's causality is confirmed but the direction of a useful drug effect is not. FURIN and DCC are weaker, provisional candidates. This is a preprint; replication and fine-mapping are next.

The methodological takeaway for anyone doing post-GWAS drug-target work: a TWAS top hit is a hypothesis, not a target — confirm it with colocalization before you build a drug story on it.

Background: the MDD drug-target gap

Depression is a leading cause of global disability, yet drug development has stalled: only ~50–60% of patients respond to first-line antidepressants, ~30% become treatment-resistant, and no novel mechanistic class was approved between SSRIs (1987) and esketamine (2019). Human genetics is one of the few levers that reliably de-risks this — genetically supported drug targets are about twice as likely to reach approval (Nelson et al. 2015).

So the prize is clear: turn GWAS hits into causal genes and then into directionally correct drug hypotheses. The problem is that the usual shortcut — TWAS (transcriptome-wide association) — is easily fooled.

What TWAS gets right, and where it breaks

TWAS (S-PrediXcan) predicts gene expression from genotype and tests it against disease, turning thousands of variants into gene-level signals. We ran it across 13 GTEx v8 brain tissues on the European-ancestry PGC MDD2025 GWAS (412,305 cases; effective N ≈ 1.15M). It is powerful — but a strong TWAS hit can simply reflect a nearby variant in linkage disequilibrium (LD), not the gene itself.

Colocalization is the fix. It asks whether the GWAS signal and the eQTL (expression) signal are driven by the same underlying variant. We used a confirmation gate of three independent tests against BrainMeta brain cis-eQTL (n = 2,865):

• SMR — is there a shared causal effect?
• HEIDI — or is it actually two different linked variants (heterogeneity)?
• coloc.abf — Bayesian posterior (PP4) that one shared variant explains both.

Only genes passing this gate are treated as confirmed causal. That single step changed the conclusions.

The DRD2 twist

DRD2 was the headline: strongest gene, and a negative TWAS Z (Z = −10.99 in nucleus accumbens) implies that lower expression tracks risk — so increasing D2 signaling (agonists/partial agonists like bromocriptine, pramipexole, aripiprazole) would be the therapeutic direction, the opposite of antipsychotic D2 blockade. Our direction-aware filter against DGIdb turned 678 raw drug matches into 102 (28 approved), explicitly excluding D2 antagonists.

It is a tidy story — and colocalization did not support it. At the DRD2 cis-eQTL locus, COLOC PP4 ≈ 3×10⁻¹² (≈ 0): the GWAS and eQTL signals point to distinct variants. We therefore move DRD2-based dopaminergic repurposing to an exploratory tier.

The careful wording matters: this is absence of cis-colocalized support, not proof that DRD2 is non-causal. Trans-regulatory or post-transcriptional effects aren't excluded. But you should not anchor a drug program on it on this evidence.

The confirmed lead: SLC12A5 (KCC2)

SLC12A5 encodes KCC2, the neuron-specific potassium–chloride cotransporter that sets the chloride gradient making GABA inhibitory in mature neurons — a master regulator of excitation/inhibition balance, already implicated in epilepsy and neurodevelopment. It was the one gene that passed everything:

Gene	SMR	HEIDI	COLOC PP4	Verdict
SLC12A5 (KCC2)	p = 3.6×10⁻¹⁴	p = 0.20 (pass)	0.996	Confirmed causal lead
FURIN	significant	fail	1.00	Locus too complex — needs fine-mapping
DCC	—	—	0.72	Suggestive only
DRD2	significant (TWAS)	—	≈ 0	Not cis-colocalized → exploratory

SLC12A5's causality is confirmed; the direction of a beneficial drug effect is not — the TWAS Z-sign and the preclinical literature disagree, so whether you'd want to enhance or inhibit KCC2 is genuinely open. FURIN colocalizes strongly but fails HEIDI (its locus is shared with other psychiatric traits, so it needs SuSiE-coloc fine-mapping and conditioning on schizophrenia). DCC is suggestive.

Why this matters

• For target selection: the single most cited TWAS hit (DRD2) is the one we'd deprioritize, and a transporter most people wouldn't have guessed (KCC2) becomes the lead. Colocalization, not TWAS rank, picked the winner.
• As a method: "direction-aware + confirmation-gated" is a reusable recipe — keep only drugs that move expression the therapeutic way, and only after the gene survives SMR/HEIDI/coloc.

Honest Limitations

1. Preprint, not peer-reviewed. Treat conclusions as provisional.
2. European-ancestry only (no-23andMe PGC subset). Non-European validation is a priority next step.
3. Direction unresolved for SLC12A5 — causality ≠ knowing which way to drug it.
4. FURIN is COLOC/HEIDI-discordant — provisional pending fine-mapping (SuSiE-coloc) and mtCOJO conditioning on schizophrenia.
5. Hypothesis-generating, not clinical. Nothing here recommends any drug for any patient.

FAQ

Q: Does this say antipsychotics treat depression?

No — the opposite. The DRD2 signal implied dopamine agonism, not the blockade antipsychotics use. And because DRD2 didn't colocalize, we don't recommend acting on it at all yet.

Q: What does "didn't colocalize" actually mean?

That the GWAS risk variant and the expression-controlling variant at that locus appear to be different variants — so the TWAS association is likely LD-driven, not the gene acting through cis-regulation. It is not proof the gene is irrelevant.

Q: Why is SLC12A5 (KCC2) plausible for depression?

KCC2 sets the chloride gradient that makes GABA inhibitory. Disrupt it and excitation/inhibition balance shifts — a mechanism already tied to seizures and neurodevelopment, and a reasonable candidate for mood circuitry. Genetics here says causal; biology still has to say which direction.

Q: Can I reproduce this?

Yes — every input is public (PGC MDD2025, GTEx v8 PredictDB, BrainMeta brain eQTL, DGIdb v5), so the pipeline (MAGMA → S-PrediXcan → SMR/HEIDI → coloc) can be rebuilt from them; the analysis code is available from the author on request.

Resources

• Preprint (Research Square, v1): https://www.researchsquare.com/article/rs-9880173/v1
• Code & extraction data: available from the author on request
• Key methods: MAGMA, S-PrediXcan (TWAS), SMR/HEIDI, coloc; target rationale: Nelson et al. 2015.