Co-Inhibitory Checkpoints Beyond PD-1: CTLA-4, LAG-3 and TIM-3

PD-1 gets the attention, but it is one brake among many — and the others do not work the same way. The co-inhibitory receptors are routinely drawn as interchangeable stop signs on a T cell. They are nothing of the kind. They read different ligands, engage different intracellular machinery, and in one case do not signal into the T cell at all so much as strip the antigen-presenting cell of the molecules it needs to stimulate. Understanding which mechanism each receptor uses is the difference between a rational combination and a hopeful one.

CTLA-4 does not compete — it confiscates

The textbook account has CTLA-4 outcompeting CD28 for shared ligands. It does bind CD80 and CD86 with much higher avidity, but competition is the smaller half of the story. CTLA-4 captures those ligands and internalises them, pulling them off the antigen-presenting cell surface and degrading them — trans-endocytosis. The consequence is unusual among checkpoints: the suppression is cell-extrinsic. A regulatory T cell expressing CTLA-4 can strip an APC of co-stimulatory ligand and thereby dampen every other T cell that later engages it, including cells that never made contact with the CTLA-4-bearing cell at all.

CTLA-4 also recruits PP2A, a serine/threonine phosphatase, rather than relying solely on the tyrosine phosphatases the other receptors use. Between the ligand depletion and the distinct phosphatase, there is a mechanistic basis for why anti-CTLA-4 and anti-PD-1 behave so differently in vivo — and why their toxicity profiles differ too. One alters the shared antigen-presenting environment; the other tunes an individual synapse.

LAG-3: two ligands, one of them from the liver

LAG-3 binds MHC class II, and does so with higher affinity than CD4 itself, with a strong preference for stable peptide–MHC complexes. That gives it a natural role in restraining CD4 responses. The second ligand is the more interesting one for tumour immunology: FGL1, fibrinogen-like protein 1, secreted by hepatocytes and often elevated in cancer. It is a soluble, liver-derived checkpoint ligand, which means LAG-3 engagement is not solely a function of what the local APC or tumour cell displays.

LAG-3 also associates with Bat3 (BAG6) at its cytoplasmic tail. Bat3 acts as a restraint: while bound, inhibitory signalling is held back, and its displacement is part of what licenses the receptor to act. That shared dependence on Bat3 links LAG-3 and TIM-3 mechanistically, despite their unrelated ligands.

TIM-3: four ligands and no ITIM

TIM-3 breaks most of the rules the other receptors follow. Its cytoplasmic tail contains no classical ITIM, so it cannot dock SHP-1 or SHP-2 the conventional way. And it has four structurally unrelated ligands: galectin-9, which binds the IgV domain through carbohydrate; CEACAM1, which can act in cis on the same cell or in trans; HMGB1, through which TIM-3 interferes with nucleic-acid sensing by dendritic cells; and phosphatidylserine, exposed on dying cells. Four ligands with four different biological contexts, feeding one receptor.

The signalling mechanism runs through Bat3 release. In the resting state Bat3 occupies the tail and recruits an active form of LCK, keeping the receptor permissive. Ligand engagement and tyrosine phosphorylation displace Bat3, and the receptor switches to inhibition. This is why TIM-3 can look either stimulatory or suppressive depending on context in published work — the direction depends on the state of the tail, not just on whether the receptor is present.

The six mechanisms at a glance

Set side by side, the differences are the point. Two receptors sharing a phosphatase are not interchangeable if they read different ligands in different compartments — and one of them does not signal into the T cell at all.

Read down the last two columns and the combination logic falls out. Pairing two ITIM receptors that both feed SHP-1 is largely redundant. Pairing ligand confiscation with synapse inhibition — CTLA-4 with PD-1 — is not, because neither compensates for the other.

BTLA, VISTA and CD200R — the three usually left out

BTLA (CD272) answers HVEM, a TNF-receptor-family member, and the interaction is bidirectional: HVEM engagement can deliver signals back into the cell presenting it. VISTA is distinctive for working preferentially at acidic pH, which is exactly the condition inside a solid tumour, and answers VSIG-3. CD200R1 reads the broadly expressed ligand CD200 and is a strong myeloid brake as well as a T-cell one. None of the three appears in most checkpoint figures, and all three have blockable or measurable reagents available.

Where they converge, and why CD28 is the real substrate

For all the mechanistic variety upstream, the convergence is tight. ITIM and ITSM motifs recruit SHP-1 and SHP-2, and those phosphatases dephosphorylate the proximal machinery — LCK and ZAP-70 — collapsing calcium flux and, with it, NFAT activation.

The detail that reframes the whole picture is which substrate matters most. Work on PD-1 signalling found that its dominant target is not the TCR complex but CD28. The brake is applied to co-stimulation rather than to antigen recognition. That single fact explains a great deal: why co-stimulation agonists combine so well with checkpoint blockade, why CD28-dependent populations are the ones that recover, and why measuring TCR-proximal phosphorylation can understate how hard the brake is being applied.

Exhaustion is a transcriptional programme, not a receptor count

Chronic antigen keeps NFAT signalling running without the co-stimulatory context that normally accompanies it. That sustained, unbalanced signalling induces NR4A1 and drives a transcriptional state built on BLIMP1 and EOMES, which displaces TCF1. The distinction matters clinically as much as mechanistically: the TCF1-high population is the stem-like progenitor that proliferates in response to checkpoint blockade, while the BLIMP1/EOMES-high terminal population largely does not. Two cells with identical surface PD-1 and TIM-3 can sit on opposite sides of that divide.

Effector functions are lost in a consistent order. IL-2 goes first, then IFN-γ and TNF, with granzyme B and perforin persisting longest. The practical implication is uncomfortable for anyone using a single cytokine as a functional readout: an IFN-γ ELISPOT can look reassuring in a population that has already lost its IL-2 capacity entirely.

Studying co-inhibitory receptors in vivo

Two practical points. First, endotoxin. Any lipopolysaccharide contamination in a blocking antibody is a TLR4 agonist, and in a checkpoint experiment the readout is an immune activation state — so contamination does not merely add noise, it produces the direction of effect you were looking for. Low- and ultra-low-endotoxin functional-grade material is not a luxury here.

Second, isotype and combination design. Anti-CTLA-4 efficacy in mouse models depends substantially on Fc-mediated depletion of regulatory T cells, so the isotype is part of the mechanism rather than an inert carrier. And because the receptors converge but do not duplicate, the informative combinations are the ones that hit different mechanisms — ligand confiscation plus synapse inhibition, rather than two ITIM receptors feeding the same phosphatase. Anti-CD3 provides the stimulation arm, and PD-1 and TIGIT each have their own dedicated pathway maps in this library.

Featured products for this pathway

Frequently asked questions

Is blocking one co-inhibitory receptor enough?

Usually not. Because the receptors use different ligands and, in CTLA-4's case, a different mechanism entirely, removing one leaves the others intact. That is the mechanistic argument for combination blockade, and it is why pairs that hit distinct convergence points — CTLA-4 with PD-1, or LAG-3 with TIM-3 — tend to be more than additive.

Why is CTLA-4 blockade so different from PD-1 blockade?

CTLA-4 acts cell-extrinsically. By trans-endocytosing CD80 and CD86, it depletes co-stimulatory ligand from the antigen-presenting cell, affecting bystander T cells that never contacted it. PD-1 acts on the cell that expresses it. One is a change to the shared environment; the other is a change to a single synapse.

What does TCF1 tell me that a receptor stain does not?

TCF1 separates the stem-like progenitor pool from the terminally exhausted pool. Two cells can carry identical PD-1 and TIM-3 levels and behave completely differently depending on TCF1. If you are asking whether a population can still respond to checkpoint blockade, TCF1 and BLIMP1 are more informative than receptor density.

Which readout shows exhaustion earliest?

IL-2. Effector functions are lost in order — IL-2 first, then IFN-γ and TNF, with granzyme B and perforin persisting longest. Measuring IFN-γ alone will make a substantially exhausted population look functional.

Do I need low-endotoxin antibodies for in vivo blockade?

Yes. Endotoxin is a TLR4 agonist, so a contaminated preparation delivers an inflammatory stimulus alongside your blockade and confounds the result — particularly damaging in checkpoint experiments, where the readout is itself an immune activation state. Functional-grade low- and ultra-low-endotoxin material exists for this reason.

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