The Adenosine Axis: How CD39 and CD73 Switch Immunity Off

The most abundant immunosuppressive molecule in a solid tumour is not a protein. It is adenosine — a metabolite, produced from ATP in two enzymatic steps, at concentrations orders of magnitude above those found in healthy tissue. It is not secreted by a specialised cell type, it cannot be knocked out, and it acts on every immune cell that carries a receptor for it. Understanding how a danger signal becomes its own antidote is the most useful thing you can know about the tumour microenvironment that checkpoint biology does not tell you.

The adenosine axis, end to end. Open the interactive version to click any protein for its role and the matching validated reagent.

The same molecule means two opposite things

Extracellular ATP is a danger signal. Healthy cells keep it inside; dying, hypoxic and mechanically stressed cells let it out, through pannexin-1 channels, through connexin-43, and simply through ruptured membranes. Once outside, ATP is read as damage. It gates P2X7, drives inflammasome priming, recruits myeloid cells and promotes dendritic cell maturation. Everything about extracellular ATP is pro-inflammatory.

Adenosine is ATP minus three phosphates, and it does the reverse of all of it. It suppresses T-cell proliferation, blunts NK cytotoxicity, pushes myeloid cells towards a suppressive phenotype and favours regulatory T cells. The immunological meaning of the entire extracellular purine pool therefore comes down to one variable: how fast the phosphates come off. That rate is set by two surface enzymes, and both of them are drug targets.

Two enzymes, one committed step

CD39 (ENTPD1) takes ATP and ADP down to AMP. CD73 (NT5E) takes AMP to adenosine. Because CD73 catalyses the last step and has no redundant partner of comparable throughput, it is the committed step of the pathway — which is why it, rather than CD39, has attracted most of the therapeutic attention.

The cascade is not a single linear route, though, and that matters when an inhibitor underperforms. ENPP1 hydrolyses ATP to AMP in parallel with CD39, so CD39 blockade alone does not stop AMP being made. The CD38–CD157 pair generates AMP from NAD⁺ rather than from ATP — a completely separate substrate pool, and one of the reasons anti-CD38 therapy has an adenosinergic rationale on top of its depleting one. On the clearance side, adenosine deaminase converts adenosine to inosine and is the only brake on the metabolite itself; it is held at the cell surface by CD26 (DPP4), which localises breakdown to the places where DPP4 is expressed.

The uncomfortable part: CD39 and CD73 are co-expressed on regulatory T cells, on tumour cells and on tumour endothelium. The tissue that most needs an inflammatory response is the tissue best equipped to abolish one.

Four receptors that do not agree

Adenosine signals through four G-protein-coupled receptors, and they pull in opposite directions. A2AR (ADORA2A) and A2BR (ADORA2B) couple to G_s, activate adenylate cyclase and raise cAMP. A1R and A3R couple to G_i and lower it.

They also differ in affinity, and this is the part most often glossed over. A2AR is high-affinity and dominant on T cells and NK cells, so it responds to modest rises in adenosine. A2BR is low-affinity, meaning it only engages at the concentrations found in tumours and ischaemic tissue — it is effectively a sensor for pathological adenosine, and it is the receptor most associated with myeloid suppression and VEGF release. Which receptor a cell reads therefore depends on how much adenosine is present. This is a genuine concentration-dependent switch, not redundancy.

The practical implication: any experiment reporting "adenosine signalling" without naming the receptor is reporting the average of opposing effects, and a pan-adenosine readout can move in either direction for reasons that have nothing to do with your intervention.

cAMP is the suppression; PKA is how it lands

Raised cAMP activates PKA, and PKA does two things that matter. It phosphorylates CREB, driving a transcriptional programme; and it activates Csk, which phosphorylates the inhibitory tyrosine of Lck. That second step is the mechanism people usually skip. Lck is the proximal kinase of TCR signalling, so switching it off blunts T-cell activation at the very first step — upstream of everything a checkpoint inhibitor acts on.

What follows is exactly what you would predict from a cell that cannot signal through its TCR: less IL-2, less IFN-γ, less perforin and granzyme B, and downregulated NKG2D on the NK compartment. Epac1 carries the cAMP-dependent effects that do not run through PKA, which is why PKA inhibition alone does not fully reverse adenosine suppression.

On the transcriptional side, CREB activity favours FoxP3 and CD25. The suppression is therefore not only a brake on effectors — it is an active push towards regulatory phenotypes, and towards higher expression of PD-1 and CTLA-4. That is the argument for combining adenosine-axis blockade with checkpoint blockade rather than choosing between them: this pathway raises the very receptors the antibodies target.

The loop is the point

HIF-1α and HIF-2α transcriptionally upregulate CD39 and CD73 under hypoxia. The adenosine that results promotes Tregs, myeloid arginase-1 and IL-10, and A2BR-driven VEGF — and VEGF builds the disorganised, leaky vasculature that sustains the hypoxia in the first place. TGF-β1 closes a second loop by inducing CD39 and CD73 directly.

Three feedback arms means single-point intervention tends to disappoint, and it explains a pattern in the literature: monotherapy data for this axis are modest while combination data are not. If you are designing an experiment here, assume you are pushing against a loop, not a line.

Why this behaves differently from a checkpoint

It is tempting to file the adenosine axis alongside PD-1 and CTLA-4 as "another inhibitory pathway". Three differences make that framing misleading.

It is soluble, not contact-dependent. PD-1 needs its ligand presented on an apposed membrane; the interaction is spatially confined to a synapse. Adenosine diffuses. A cell does not need to touch anything to be suppressed, which means suppression extends into the space between cells and affects populations that never engage the tumour directly — including cells that have only just arrived and have not yet found a target.

It acts upstream of the TCR, not on it. Checkpoint receptors recruit phosphatases to a signalling complex that has already formed. The cAMP–PKA–Csk route disables Lck before that complex assembles. Functionally, the difference is between turning the volume down and unplugging the input, and it is why adenosine suppression is not rescued by blocking a checkpoint receptor.

It has a metabolic source that regenerates. A checkpoint ligand is a gene product with a measurable expression level. Adenosine is a product of ongoing turnover: as long as cells are dying and the enzymes are present, it is replenished. There is no expression level to knock down, only a flux to slow.

Those three properties together explain the clinical pattern. Adenosine-axis agents look weak as monotherapy and useful in combination, because they do not remove a brake so much as raise the ceiling on what removing other brakes can achieve.

The same axis, read the other way round

Everything above treats adenosine as a problem, which is the oncology reading. In every other setting it is a repair signal, and the sign flips. In ischaemia–reperfusion injury, A2AR signalling limits neutrophil-driven tissue damage and is protective. In colitis and graft-versus-host disease, CD39 on regulatory T cells is part of how those cells work. In asthma and fibrosis, A2BR on myeloid and mesenchymal cells contributes to remodelling that is pathological for a different reason.

This is worth stating plainly because it constrains the therapeutic window. Blocking CD73 or A2AR does not remove a bug; it removes a brake the tissue installed deliberately. The relevant question in a model is never "is adenosine signalling on" but "in this tissue, at this concentration, is the brake helping or hurting" — and the answer changes with the hypoxia status of the tissue you are looking at.

Three ways this gets measured wrong

Measuring adenosine itself. Adenosine has a half-life measured in seconds in whole blood, and adenosine deaminase in your sample continues working after collection. A low reading may mean low production or fast clearance during handling. Measuring the enzymes and the downstream second messenger is more reproducible than chasing the metabolite.

Reading CD73 expression as CD73 activity. Surface abundance and catalytic output are not the same. Shed, soluble CD73 is enzymatically active and is not counted by flow cytometry of the cell surface; conversely a cell can be CD73-bright and substrate-limited if CD39 or ENPP1 upstream is not supplying AMP.

Attributing a cAMP change to one receptor. With A1R and A3R lowering cAMP while A2AR and A2BR raise it, a small net change can hide two large opposing ones. Receptor-selective tools, or at minimum a stated assumption about which receptor dominates on your cell type, are not optional here.

Which reagent answers which question

The hardest problem in this axis is attribution. "Less adenosine", "fewer Tregs" and "better effectors" produce the same tumour-growth curve, and separating them needs the right control set rather than a bigger n.

A note on reading the map

Five nodes on the interactive version carry the blue In Vivo marker — CD73, CD8, CD25, PD-1 and CTLA-4. That is a smaller functional-grade layer than some of the other pathways in the library, and it reflects a real feature of the biology rather than a gap in the catalogue: most of this axis is enzymes and second messengers, which are measured rather than blocked with an antibody. The ELISA layer is where the mechanistic readouts live.

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