Immunotherapies that activate costimulatory receptors on T cells have failed so far in the clinic, but the first phase III trial of an agonist antibody in cancer has just begun and second-generation candidates are advancing.

Nature Reviews Drug Discovery, January 2020.

Blocking antibodies have transformed oncology over the past quarter century. But agonist antibodies that boost downstream signalling have proved much harder to develop, despite great promise. In November, GlaxoSmithKline opened a phase III trial for its ICOS-binding GSK3359609, however, making it the first ever anti-cancer agonist antibody targeting a costimulatory receptor on T cells to make it to pivotal trials.

The outlook, from a purely historical perspective, isn’t good. Over the past 15 years, agonist antibodies against costimulatory receptors including ICOS, GITR, OX40, CD27 and 4-1BB have uniformly failed to live up to high expectations in earlier stage trials. Taha Merghoub, a cancer immunologist at the Memorial Sloan-Kettering Cancer Center, recently published preclinical data on GITR agonists, and experienced the resistance the field faces. “We got reviewer comments on this paper and another paper on OX40 [commenting] ‘Why do we even work on these drugs? We know that they will never work.’”

GSK and some other pharmas and biotechs reject this nihilism, persevering with OX40, GITR, 4-1BB, and ICOS antibody agonists (Table 1). Second-generation antibodies will have better efficacy and lower toxicity, they expect. “As a field of immuno-oncology, we have not drugged costimulatory receptors very well,” says Tom Schuetz, CEO of Compass Therapeutics, one of the companies in this space. “We can do better.”

That will require a more careful approach, says Axel Hoos, vice president for oncology R&D at GlaxoSmithKline. “For antagonists it’s easy, all you have to do is bind the thing and block it,” he explains. But for agonists, discovery considerations include affinity, epitope selection, valency, cluster formation, Fc gamma receptor interaction, receptor occupancy and the isotype of the antibody. In clinical trials, dosing, scheduling and combination strategies also need to be carefully considered.

“We think the biology around agonists is compelling and makes sense,” Hoos adds. “But it needs to be handled differently than antagonists.”

Agonizing history

The anti-cancer rationale for agonizing costimulatory receptors is clear. As the inevitable auto analogy goes, while inhibitors of the CTLA4 and PD1/PDL1 checkpoint receptors release the brakes on T cell activation, agonist antibodies to costimulatory receptors press on the gas. Either approach should boost antitumour immunity.

But whereas checkpoint inhibitors have transformed cancer medicine, costimulatory agonists have a perilous history. CD28 was the first target. Constitutively expressed on the T cell surface, it provides the second signal following T cell receptor (TCR) engagement by the antigen-presenting cell (APC). Once both TCR and CD28 are triggered, a T cell is primed to secrete immunostimulatory interferons and to release perforin and granzymes to lyse tumour cells.

However, boosting CD28 signalling has led to disaster. In 2006, TeGenero’s CD28 superagonist antibody caused massive cytokine storms and multi-organ failure in six healthy phase I volunteers. Because the antibody activated T cells without TCR engagement, it unleashed indiscriminate immune attacks.

Other costimulatory receptors are safer targets, probably because they are only expressed once T cells are activated against tumour antigens. These receptors include 4-1BB, OX40, GITR, CD27 and ICOS. When these costimulatory receptors bind their ligands, the receptors crosslink and cluster, initiating downstream signalling to deepen the T cell response and boost proliferation, survival, memory cell formation and cytokine secretion. In the highly immunosuppressive tumour microenvironment, agonist antibodies to these receptors should rescue and amplify antitumour immunity.

Extensive mouse data confirm this therapeutic potential. Genentech, for example, combined an OX40 agonist antibody and an anti-PDL1 antibody to achieve complete responses in nine of ten colorectal tumours in one mouse model.

But clinical trials have fallen far short of such deep effects. In Genentech’s phase Ib trial of its OX40 agonist MOXR0916, combined with the PDL1 inhibitor atezolizumab, only 2 of 51 (4%) treated patients had a partial response. Genentech discontinued its OX40 programme in May 2019. Agonist antibodies against other costimulatory receptors have fared similarly, many of their programmes being discontinued or suspended.

Rethinking 4-1BB antibodies

Evidence points to the drugs, not the targets, as the problem. For example, drug developers have built the 4-1BB cytoplasmic tail into several highly effective chimeric antigen receptor (CAR) T cell therapies, including Novartis’s FDA-approved tisaglenleucleucel. 4-1BB “is a very powerful costimulatory motif, particularly in terms of permitting the persistence of the infused CAR T cells,” says cancer immunologist Ignacio Melero of the Clinica Universidad de Navarra. In fact, CAR T cells didn’t work well until investigators incorporated 4-1BB into their products, a decade ago. Given this surrogate clinical validation, agonistic anti-4-1BB antibodies delivered systemically, investigators reason, should also enhance T cell survival and antitumour immunity.

Antibodies against 4-1BB, also known as CD137, have yet to deliver. The first one, Bristol-Myers Squibb’s urelumab, entered human trials in 2005. In 2008 the company halted trials owing to liver toxicity, after two patients died. This was an animal-to-human translation error; researchers later discovered that the antibody’s affinity for monkey 4-1BB is at least an order of magnitude lower than for human 4-1BB, leading investigators to use too high a dose in their phase I trial. The company restarted trials at lower doses, but antitumour activity has been modest. (Urelumab continues to be used in a few investigator-initiated trials, mostly in lymphoma or via intratumoural injection.) Pfizer’s utomilumab, which targets a different epitope on 4-1BB, proved safe but similarly ineffective. “It’s very clear that utomilumab is a weak agonist,” says Melero.

Other companies are pushing a new wave of 4-1BB antibodies forward, looking to avoid liver toxicity while boosting potency.

Bispecific antibodies that target both PDL1 and 4-1BB are already in early-stage trials. Because tumours often express high levels of PDL1, these should concentrate 4-1BB activity at the tumour. But bispecifics — which are also in development for several other costimulatory receptors — have pitfalls. 4-1BB is most active when it forms a trimer on the surface of the T cell, and bispecifics may not be as efficient as standard bivalent monospecific antibodies at inducing this clustering, because they carry only one antigen binding region for the receptor[Sorry, I am not sure about the biophysics here]. Bispecifics may also have trouble binding both targets in the space between the APC and the T cell, or the T cell and the tumour. And bispecifics do not permit sequential dosing of the component parts, a potentially more rational approach given the sequential nature of T cell priming and activation.

Compass originally set its sights on an anti-4-1BB bispecific, but “the data didn’t take us in that direction,” says Schuetz. “The combination of monoclonals was always superior to bispecifics in preclinical modelling.” So Compass learned from the field’s possible mistakes. To begin with, it wanted its antibody to have an IgG4 isotype, instead of the IgG1s more typical of first-generation antibodies.       The structure of the antibody’s heavy chain defines the isotype and its properties. IgG1s efficiently recruit effector cells, including NK cells, that drive antibody-dependent cellular cytotoxicity (ADCC). And this, says Schuetz, is a problem for agonist antibodies: “You’re going to be killing the cells that you actually want to stimulate”.

The low-depleting istotypes, IgG2 and IgG4, also have possible drawbacks. Their Fc domains bind weakly to Fc receptors on nearby immune cells. And such Fc receptor binding is crucial for the crosslinking and clustering of the T cell target receptor, in Compass’s case, 4-1BB. “You need crosslinking, otherwise 4-1BB stimulation doesn’t work,” says Melero. But while IgG2s and IgG4s do this poorly compared with IgG1s, they do it well enough, says Schuetz. “Even the minimal binding [of IgG4] compared with IgG1 is actually, believe it or not, enough to mediate clustering,” he says.\

Compass also wanted its antibody to bind a different epitope on 4-1BB than its predecessors, in the hopes of avoiding urelumab’s liver toxicity and utomilumab’s weak agonism. Compass’s chosen epitope does this, says Schuetz, while still allowing the endogenous ligand to bind 4-1BB. This boosts forward signalling into the T cell and also enables reverse signalling into the APC, since 4-1BB’s ligand, expressed on the APC, has its own intracellular signalling domain. That could boost APCactivity, not just T cells.

But Compass’s antibody’s most important feature, in Schuetz’s opinion, is its mid-range affinity for 4-1BB. The company was wary of going too high. “If you’re … providing a constant stimulation signal to the T cell, that is a great way to induce T cell exhaustion,” says Schuetz. Overstimulation can also cause activation-induced cell death (AICD), he adds, again depleting the same cells you want to harness.

The kinetics of receptor occupancy are also important. “When we compared different antibodies of very different affinities, the ones that were the best had very fast on-rates, and they had off-rates that were also intermediate,” says Schuetz. That, he says, gives the receptor time to recover.

Melero agrees that the affinity of an agonist antibody for its target needs to be thoughtfully optimized, not maximized. “But in my view what will probably become more important is not affinity — it’s going to be intermittent exposure,” he adds. That’s because activation of costimulatory receptors results in their internalization into endosomes, and continuous drug exposure prevents receptor re-expression on the cell surface. “If you let the cell rest a little bit from 4-1BB exposure, and re-expose again, you get again a very high, a very strong, signal activity,” Melero says. “Perhaps we have to let 4-1BB expression reload, and then pull the trigger again.”

Learning from OX40 failures?

OX40 agonists have fared even worse than 4-1BB agonists. “So far clinical activity is very modest, if any,” says Melero.

Here, the dosing strategy may be a key factor. “One of the reasons why OX40 didn’t work is that people do dose escalation, and [then use] the highest tolerable dose,” says Merghoub. “What they should have done is give the lowest bioactive drug amount.” That, he says, would reduce the risk of AICD.

Hoos agrees. “All the OX40s have failed,” he says. “We clearly treated them like antagonists, where we dosed as high as we could, as an industry. And when you dose as high as you can, you might end up exhausting the cell, and losing the effect.” Four years into clinical trials, GSK’s OX40 agonist remains in phase I. “We have not fully sorted out the biology of OX40 in patients,” says Hoos.

Scheduling may matter with OX40 as well, says Providence Cancer Institute tumour immunologist and AgonOx president Andrew Weinberg. In 2016 AstraZeneca launched a phase I trial of its OX40 agonist MEDI0562, which AgonOx licensed out in 2011. Dosing was every 2 weeks, similar to PD1 blockade. Other companies followed suit. “In hindsight this was not the optimal dosing scheme,” says Weinberg. Whereas every 2 weeks makes sense when blocking a negative signal like PD1, he explains, when an antibody enhances a positive signal this runs the risk of overstimulating the T cell and causing exhaustion or cell death. Instead, he suggests, “give a big bolus early [and] let the immune system run its course for a month or two.” Patients that have stable or regressing disease can then be re-treated later on.

A decade ago Weinberg’s group conducted a phase I trial of its own mouse OX40 agonist antibody, and 12 of 30 (40%) patients experienced at least some tumour shrinkage after 3 doses over 5 days.

Combination sequencing is another crucial consideration. Many trials of agonist antibodies are already combining these agents with PD1 or PDL1 inhibitors, since stepping on the gas while releasing the brake should have additive if not synergistic effects. Most of these trials have dosed the agonist antibody simultaneously with the PD1 or PDL1 blocker. Weinberg instead favours sequential dosing, boosting costimulatory signalling before blocking PD1.

As evidence, in 2017 two groups published data from mouse models showing that OX40 agonists and a PD1 inhibitor cancelled out their positive effects when dosed simultaneously, because of either T cell exhaustion or programmed cell death. Sequential dosing, on the other hand, boosted the antitumour activity of the agents. One hypothesis is that costimulation must first boost tumour-specific T cells to the point where checkpoints are inhibiting the immune response, and only then can checkpoint inhibitors be effective. Now that AstraZeneca has jettisoned its OX40 agonists, Weinberg hopes to test new dosing schemes in future AgonOx-sponsored trials.

In general, “these agonist antibodies need to be fine tuned, and dosing schemes need to be carefully constructed,” he says.

GITR agonists have also disappointed, especially as single agents. AstraZeneca dropped its GITR programme earlier this year, for example. And in 2016 Leap Therapeutics reported minimal efficacy for its TRX518 antibody, given alone, in phase I. That doesn’t mean GITR agonists don’t work though, says Merghoub. “The issue is that we are expecting single-agent activity,” he says. “If you have a tumour that is established, then you are not able to treat with a single agent.” Merghoub’s group recently published mouse data showing that TRX518 by itself could not reverse T cell exhaustion in established tumours, but that adding PD1 blockade overcomes that hurdle. Leap Therapeutics is now testing TRX518 in a phase Ib/II triple combination trial with the PDL1 inhibitor avelumab and the alkylating agent cyclophosphamide.

ICOS: differing on depletion

ICOS stands apart from the other costimulatory receptors. It belongs to the same family as CD28 (the others are in the TNF receptor family), but has a milder costimulatory effect than CD28 and is only inducible upon T cell activation.

The first agonist anti-ICOS antibodies entered the clinic in 2016 with high expectations. Jounce Therapeutics, co-founded by M.D. Anderson Cancer Center immunologist Jim Allison, recipient of the 2018 Nobel Prize for Physiology or Medicine, developed an IgG1 ICOS antibody called vopratelimab.

The company’s preclinical data showed that the drug both activates T effector cells and depletes intratumoural regulatory T cells (Tregs) that dampen immune responses. But in a phase I/II trial, vopratelimab monotherapy resulted in one partial response out of 67 patients treated (1.5%), and 8 partial responses out of 106 patients (8%) when used in combination with the PD1 inhibitor nivolumab. It was not clear that vopratelimab added anything to PD1 blockade. Jounce’s stock price plunged 63% on the data, and has not recovered.

Jounce is refining its approach, now testing vopratelimab in phase II in combination with the CTLA4 inhibitor ipilimumab in lung and urothelial cancer. This new strategy is based on preclinical and clinical data showing that anti-CTLA4 treatment induces high ICOS expression on CD4 cells, which is strongly associated with response. Vopratelimab treatment should boost proliferation and activity of these CD4 cells. “We’ve developed ways of targeting ICOS to increase efficacy of this CTLA4 blockade,” said Allison at the 2019 ASCO meeting.

Based on tumour biopsies and other data, Jounce no longer considers Treg depletion to be a part of vopratelimab’s mechanism of action, the company wrote in an email.

Kymab, by contrast, is fully embracing Treg depletion with its ICOS agonist. Kymab selected ICOS as a target because of its high expression on intratumoural Tregs relative to T effector cells. “ICOS is probably the best target in terms of differential expression,” says Kymab’s senior director for translational medicine Richard Sainson. The company’s IgG1 antibody KY1044 is designed to deplete Tregs in the tumour microenvironment, potentially making the tumour more responsive to checkpoint blockade. “We want to have [Treg] depletion,” says Sainson. “That’s a key part of our mechanism of action.” He points out that non-depleting antibodies could in some tumours stimulate these Tregs and dampen antitumour immunity.

But GlaxoSmithKline’s ICOS agonist GSK3359609, the first anti-cancer agonist antibody to make it to phase III trials, takes a non-depleting approach. GSK3359609 is an IgG4 antibody, engineered to eliminate any residual ADCC activity. As with 4-1BB agonists, IgG1-based ICOS candidates might be working against themselves, suspects Hoos. “The cell you’re binding is a T effector cell, and we need that cell. Having ADCC is counterproductive,” he says. (Jounce says that its IgG1 ICOS agonist does not deplete T cells in humans.)

GSK reported phase I head and neck cancer data for GSK3359609 in September at the 2019 Congress of the European Society for Medical Oncology. While only 1 patient of 16 (6%) achieved an objective clinical response to monotherapy, 8 out of 34 (24%) responded to the antibody combined with pembrolizumab in PD1/PDL1 inhibitor-naive patients.

This open-label trial did not have a placebo arm, so it’s impossible to know how much effect GSK3359609 added to pembrolizumab, already a first-line treatment for advanced head and neck squamous cell carcinoma. But the evidence was enough for the company to open a randomized 600-patient phase III combination trial in head and neck cancer. “The data we have are still early, we recognize that,” says Hoos. “We think the value will mostly come from improving patient survival, rather than shrinking tumours.”

The non-depleting approach makes sense, he adds. “In most cancers, based on what we know, it is the effector cells that carry most of the ICOS,” he says.

With patient screening already underway, the stakes are high. “Our focus is to make one agonist work,” says Hoos. “If you have one success, many others will follow. If you don’t figure it out, you may have a dead field.”