Th17 and IL-23: Why Blocking IL-17 Sometimes Backfires

Two drugs aimed at the same axis can give opposite results in the same disease. Blocking IL-23p19 is highly effective in psoriasis and works in Crohn’s disease. Blocking IL-17A directly is highly effective in psoriasis and made Crohn’s disease worse in trial. Both agents hit the same pathway. The reason they diverge is that Th17 cells are not one population doing one job — they are a spectrum, and IL-23 is the cytokine that decides where on that spectrum a cell sits. This is the most consequential thing to understand about the axis, and it is invisible if you treat “Th17” as a single lineage.

The Th17 / IL-23 axis, induction through to tissue response. Open the interactive version to click any protein for its role and the matching validated reagent.

Differentiation and pathogenicity are two separate decisions

The first decision happens in the lymph node. A dendritic cell presenting antigen alongside IL-6 and TGF-β1 pushes a naive CD4 cell towards RORγt. IL-1β, through IL-1R1 and the MyD88–NF-κB route, lowers the threshold and makes commitment faster and more robust. At the end of this step you have a Th17 cell.

The second decision happens later, in the tissue, and it belongs to IL-23. Here is the structural detail that makes the whole axis a feed-forward loop: IL-23R is not present on a naive cell. It is induced by the very STAT3 signalling that IL-6 started. IL-23 therefore cannot initiate Th17 differentiation — it can only act on cells that have already committed. IL-6 and TGF-β open the door; IL-23 decides what walks through it.

What IL-23 does to a committed cell is convert it from a transient, IL-10-producing, tissue-tolerant phenotype into a stable, GM-CSF-producing one. In mouse models of CNS and joint autoimmunity, that second phenotype is the pathogenic one. Cells that never see IL-23 make IL-17 and go away. Cells that do see it make IL-17, GM-CSF, and trouble.

Why anti-IL-23 and anti-IL-17 behave differently

Once you separate the two decisions, the clinical divergence stops being a puzzle. Blocking IL-23 removes the pathogenicity signal but leaves the baseline Th17 population and its barrier functions largely intact. Blocking IL-17A removes the effector cytokine from every source, including the homeostatic ones — and at the gut barrier, IL-17 signalling into epithelium is part of how the barrier is maintained.

In skin, where the pathology is the IL-17 effector programme and the barrier does not depend on it in the same way, both approaches work and IL-17 blockade works faster. In gut, removing IL-17 removes a repair signal. Same pathway, opposite tissue economics. Any model built to test one of these agents needs to state which of the two decisions it is interrogating.

The shared p40 chain, and why it matters commercially

IL-23 is a heterodimer: IL-23p19 paired with p40, the chain it shares with IL-12. That sharing is the single most instructive piece of structural biology in the axis, because it produced a natural experiment. Anti-p40 agents remove IL-12 and IL-23 together, taking out the Th1 arm alongside the Th17 arm. Anti-p19 agents remove IL-23 only.

In psoriasis the selective p19 agents proved more effective, not less — which is the clearest available evidence that the IL-23 arm, not the IL-12/Th1 arm, is the one driving that disease. The lesson generalises to model design: if a p40 intervention works and a p19 intervention does not, the phenotype was at least partly IL-12-dependent and calling it “Th17-driven” is wrong. Running both is how you find out, and it costs one extra arm.

One STAT, two outcomes

IL-6Rα, IL-23R and IL-12Rβ1 all converge on JAK2 and TYK2, and therefore on STAT3. There is no separate “pathogenic” transducer. The same molecule carries both messages, and what distinguishes them is signal strength and duration rather than identity.

That is also why SOCS3 matters more than its supporting-cast status suggests. STAT3 induces SOCS3, and SOCS3 shuts JAK2 down again. A normal response is self-limiting because of this loop. A chronic one is, in large part, a loop that has stopped closing — which is a different problem from too much cytokine at the top.

Two receptor details are worth carrying. IL-6 signals in cis through membrane IL-6Rα but also in trans, using soluble IL-6Rα to act on cells that do not express the receptor at all — so IL-6Rα staining understates who can respond. And TYK2 carries a common human loss-of-function variant that protects against psoriasis and inflammatory bowel disease, which is the human genetics validating this branch of the map.

Four transcription factors, four different jobs

RORγt is the lineage-defining factor and opens the Il17a/Il17f locus. RUNX1 co-operates with it at that locus — the two act together rather than in series. AHR is ligand-activated by dietary and microbial metabolites and biases output towards IL-22, which is how diet and microbiome exert direct transcriptional control over this axis. IRF4 reads TCR signal strength and effectively sets how much of the programme gets written at all.

The practical consequence of four inputs is that “RORγt-positive” is not a sufficient description of a cell. Two RORγt⁺ cells with different AHR and IRF4 status make different cytokines in different proportions, and that difference is what a cytokine panel is actually measuring.

The Treg branch is the same signal minus one cytokine

TGF-β1 on its own drives Foxp3 and a regulatory phenotype. Add IL-6 and you get RORγt instead. The two factors antagonise each other directly, and cells expressing both exist as a genuinely unstable intermediate rather than an artefact.

This has a direct experimental consequence that catches people out: depleting CD25⁺ regulatory cells reliably amplifies a Th17 phenotype, so a Treg-depletion model is not a clean background for asking whether an intervention is Th17-dependent. It is also why IL-10 deserves more attention than it usually gets here — non-pathogenic Th17 cells make it themselves, and its loss is often what converts a tolerogenic response into a destructive one, with no change in IL-17 at all.

The output is a tissue programme, not a killing programme

IL-17A and IL-17F signal through IL-17RA and the Act1–TRAF6 relay into NF-κB in epithelial and stromal cells — not in immune cells. What comes out is CXCL1 and G-CSF, which recruit and expand CD11b⁺ neutrophils, and CCL20, which pulls in more CCR6⁺ Th17 cells. That last edge is a genuine self-reinforcing loop: the tissue recruits more of the cells that inflamed it.

IL-22 is the exception and is frequently mis-filed. Its receptor is essentially restricted to epithelium, so it cannot act on immune cells at all. What it drives is repair and MUC2 production. Blocking it can therefore worsen barrier disease while looking, on a cytokine panel, like successful anti-inflammatory treatment. IL-21 closes another loop, feeding straight back onto STAT3 as an autocrine amplifier.

Three ways this axis gets measured wrong

Reading IL-17A as a measure of pathogenicity. It is a measure of Th17 activity. Non-pathogenic and pathogenic cells both make it. GM-CSF, not IL-17, is the effector most tightly tied to disease severity in CNS and joint models — and it is the one most often left off the panel.

Assuming IL-23 blockade should reduce Th17 numbers. It need not. IL-23 acts on already-committed cells, so its removal changes what those cells produce more than how many of them there are. A frequency readout can be flat while the functional effect is complete.

Treating a cytokine panel as a lineage readout. Th17 cells are plastic. Under inflammatory conditions they convert towards an IFN-γ-producing phenotype, and the cell that caused the damage may no longer look like a Th17 cell by the time you sample it. Fate-mapping answers this; a terminal cytokine panel does not.

Which reagent answers which question

The In Vivo layer on this map sits at the edges of the axis rather than its centre, and that is worth saying plainly rather than padding the map. The core cytokines here — IL-6, IL-23, IL-17A, IL-17F — are read with ELISA. What functional-grade blockade gives you is the ability to change severity and to remove brakes, which is a different and often more informative set of questions.

Two notes on reagent specificity

The RUNX1 antibody linked from this map is pan-RUNX1/2/3 and does not discriminate between family members. The IL-23R node links to a recombinant Fc-tagged receptor protein rather than to an antibody. Both are flagged in the tooltips on the interactive version, and both are worth knowing before you design around them.

For research use only. Not for use in diagnostic or therapeutic procedures.