Introduction

Integrin proteins are ubiquitous, heterodimeric, transmembrane glycoprotein receptors that primarily act as signalling proteins in mammals1. Each consists of an α-subunit and a β-subunit, of which there are 18 and 8 variants, respectively, creating the 24 known heterodimers (Fig. 1). The α- and β-subunits are bound in a noncovalent complex with the ligand-binding site at the interface. Integrins act as adhesion receptors, with the unusual ability to signal in both directions across the plasma membrane2. These events are called ‘inside-out’ signalling3 and ‘outside-in’ signalling4, resulting either from binding to extracellular ligands or from interacting with the cytoskeleton via the integrin intracellular domains. Integrins can therefore enable human cells to respond to changes in the extracellular environment (via outside-in signalling) and can influence the extracellular environment itself (via inside-out signalling). Information from outside the cell is communicated intracellularly when the ligand binds to the receptor, resulting in changes in cell polarity, cytoskeletal structure, gene expression, cell survival and proliferation5. In the opposite direction, intracellular activators such as talin-1 (ref.6) bind to the cytoplasmic tail of the β-subunit, evoking a conformational change that shifts the integrin into a high-affinity state, which more readily binds to extracellular ligands and thus promotes cell migration and extracellular matrix (ECM) assembly and remodelling7.

Fig. 1: The integrin family and targeted therapies.
figure 1

All 24 distinct integrin heterodimers, formed from one α-subunit and one β-subunit, are represented and grouped according to their broad classification by cognate ligand or cellular expression. Therapeutically targeted integrins are highlighted in blue along with the therapeutic areas that are of current interest. Additionally, a select number of therapies in development, and marketed or terminated drugs and their modalities, are shown. Arrows are intended to demonstrate which integrin targets are thought to be key and their purpose is not to capture all known integrin activities. See Tables for additional information. aApproved for ulcerative colitis and Crohn’s disease. bBeing investigated clinically for ulcerative colitis. cAlso approved for Crohn’s disease in the USA.

The integrin proteins are classified into families that consist of receptors with related properties (Fig. 1). For example, all eight members of the RGD-binding family of integrins recognize the amino acid binding motif Arg–Gly–Asp (RGD) in their endogenous ligands. The related integrins α4β7 and α4β1 are therapeutic targets that are expressed on leukocytes and also recognize short peptide sequences, one of which is Leu–Asp–Val (LDV)8,9. In addition, families of integrins that bind to either collagen10 or laminin11 have wide-ranging roles in disease, but to date have not been extensively targeted12,13.

The inhibition of integrins has led to several marketed drugs, and many others are being investigated preclinically in both academic and industry settings. Since 2015, there have been at least 130 clinical trials of integrin-targeted therapies (https://www.clinicaltrials.gov/.clinicaltrials.gov and https://www.clinicaltrialsregister.eu/ctr-search/search using search term “integrin”). (This number is an estimation because there are clinical trials with integrin molecules that are not returned with this search term. Databases interrogated December 2020.) In total, six integrin inhibitor drugs, targeting four integrins (αIIbβ3 (also known as glycoprotein IIb/IIIa), α4β7, α4β1 and αLβ2), have been marketed (Table 1). Three of these drugs are antibodies and three are small molecules, but none is delivered by the oral route — contributing factors for the lack of orally available small molecules include the polar pharmacophore in those molecules coupled with complex pharmacology for the target pathway14. Intravenously administered inhibitors of the RGD-binding integrin αIIbβ3 were some of the first developed, exemplified by two small molecules, tirofiban (Aggrastat) and eptifibatide (Integrilin), alongside the antibody abciximab (ReoPro). All three therapies are prescribed for acute coronary syndrome and for the treatment of thrombotic cardiovascular events15 (Box 1; Table 1). Additionally, drug discovery programmes centred on the integrin αLβ2 (which is expressed on leukocytes) delivered a marketed small molecule, lifitegrast, for the topical treatment of dry eye disease16. Inhibitors of αLβ2 have also been investigated for autoimmune diseases and inflammatory disorders17.

Table 1 Approved integrin-targeting drugs

The remaining marketed integrin drugs, vedolizumab (Entyvio) and natalizumab (Tysabri), are antibodies that act principally on the leukocyte integrins α4β7 and α4β1 and are used for treating ulcerative colitis, Crohn’s disease18 and multiple sclerosis19. Combined sales of these two molecules have reached more than $US4 billion per year20,21, underlining the impact of integrin inhibitors in treating disease. However, this class of molecules suffered a setback in 2009 when efalizumab, which targeted αL integrins (these are also expressed predominantly on leukocytes), was withdrawn from the market because of multiple cases of progressive multifocal leukoencephalopathy (PML), believed to be associated with inhibition of α4-containing integrins and αLβ2 (ref.22) (Box 2). Similarly, natalizumab was withdrawn from the market in early 2005 after its use was associated with PML in patients with multiple sclerosis; natalizumab returned the following year with a black box warning about the increased risk of PML, after a detailed review of all clinical trial data. These high-value biological targets continue to be investigated clinically with the risk of PML in mind. Indeed, two orally delivered antagonists of α4β7 are in clinical trials: a small molecule from Morphic Therapeutics (which is in phase I) and a peptide from Protagonist Therapeutics (which is in phase II) (Table 2).

Table 2 Selected clinical studies with pending data

Excitingly, new molecules that target αv-containing integrins are now entering clinical trials for fibrotic diseases, including idiopathic pulmonary fibrosis (IPF) and nonalcoholic steatohepatitis (NASH), which have high and increasing23,24 unmet medical need. Because integrin proteins are pivotal to numerous biological pathways25 and they bind to a variety of endogenous ligands, inhibition of a single integrin or integrin family could treat a range of diseases26, such as multiple fibrotic diseases or multiple types of cancer.

It is therefore timely to discuss the progress made in drug discovery for the RGD-binding integrins, mainly the αv integrin subfamily, as detailed in this Review. Many of the lessons learned from this class of molecules are also pertinent to drug discovery and integrins in general. We discuss the different modalities adopted in integrin drug discovery, alongside their mechanisms, and diseases that might be treated by inhibition of these integrins. We also assess the factors to consider in integrin inhibitor drug design — the unusual physicochemical properties, how to achieve selectivity, the optimal modes of administration and the pharmacology. The latest developments will be discussed alongside the learnings from past programmes and clinical terminations to provide insight on the progress made to identify new, safe and effective treatments. We conclude with a summary of the key challenges and prospects for therapies that target integrins.

Targeting αv-containing integrins

Historically, much of RGD-binding integrin drug discovery has focused on αvβ3 for cancer27, ophthalmology28 and osteoporosis29,30; αvβ3 remains the most studied integrin in the past two decades (see Supplementary information for data comparing publication trends in integrin research). The rationale for targeting αvβ3 in cancer comes from its known role in tumour angiogenesis and its upregulation on endothelial cells; inhibitors of αvβ3 have had positive effects in preclinical models31. However, these studies have not translated effectively to the clinical setting22. For ophthalmic diseases, integrin inhibitors could reduce various pathologies associated with eye disease, such as inflammation, vascular leakage, angiogenesis and fibrosis. Integrin inhibitors have been effective in several preclinical models, and promising results have been reported thus far from clinical trials32. Indeed, most of the current αvβ3 clinical investigations centre on treating eye diseases (age-related macular degeneration (AMD) and diabetic macular oedema (DME))33 using topically dosed or intravitreally injected small molecules and peptides, although these molecules also inhibit other αv integrins and/or α5β1 to varying degrees (Table 3). The molecules that have progressed the furthest in the clinic are risuteganib (Luminate, Allegro Ophthalmics, structure 4 in Fig. 2), the fluorinated MK-0429 analogue SF0166 (Scifluor Life Sciences; now OT-166, OcuTerra Therapeutics, 11) for treating AMD and DME34, and the pan-αv/α5β1 inhibitor, THR-687 (Oxurion/Galapagos), which is currently in phase I trials for DME. JSM-6427 (Takeda, proposed structure shown in 3) is a peptidic small molecule that has been evaluated preclinically for treating ocular neovascular diseases35 and progressed to phase I clinical trials for AMD. However, no data have been reported since 2010. Most recently, a 20-mer synthetic peptide that targets both αvβ3 and α5β1 (AXT-107, developed by AsclepiX Therapeutics), derived from the non-collagenous domain of collagen IV, has entered clinical trials for treating retinal vascular diseases36,37.

Table 3 Selected integrin inhibitors in clinical trials with reported clinical safety and efficacy data
Fig. 2: Selected small molecules with parenteral properties that bind to RGD integrins.
figure 2

The small molecules shown here have either been evaluated in the clinic or have been examined in preclinical models and are administered by a parenteral route with advantageous potency or selectivity profiles. The molecules are: 1 cilengitide, 2 tirofiban, 3 JSM-6427 (proposed structure); 4 risuteganib (Luminate), 5 GSK3008348, 6 αvβ6 integrin small-molecule inhibitor, 7 GLPG0187, 8 acyclic peptide selective for αvβ8, 9 αvβ1 integrin small-molecule inhibitor, 10 CWHM-12, 11 SF0166 (now OT-166), 12 RUC-4. RGD, Arg–Gly–Asp.

Although molecules that target αvβ3 integrins generally have an acceptable safety profile (Table 3), interest in using them to tackle cancer has waned, mainly owing to lack of efficacy. Broad reasons for failure may encompass several factors, such as redundancy, promiscuity and compensation mechanisms38. The most studied αvβ3 inhibitor molecule and furthest progressed is the cyclic peptide, cilengitide (Merck KGaA, 1), which also inhibits αvβ5 and α5β1 and has been assessed in approximately 30 different clinical trials for cancer. Ultimately, however, cilengitide fell short in phase III trials owing to a lack of efficacy against glioblastomas, with no improvement in overall survival39,40. Antibodies that target αvβ3, such as etaracizumab (MEDI-522; Abergrin) also advanced to clinical trials for several diseases, including cancer41, but progression has halted. GLPG0187 (7), a broad-spectrum αv inhibitor, also failed to show signs of efficacy in a phase Ib trial in patients with solid tumours, again, despite being well tolerated42.

Meanwhile, alternative approaches to αvβ3 inhibition are offering greater promise for treating the same or similar diseases: there are numerous promising therapies for treating glioblastomas43, cathepsin K inhibitors could be useful in osteoporosis44 and agents targeting vascular endothelial growth factor A (VEGFA), such as ranibizumab, are already available for treating AMD45. Any new drug discovery programme focused on the αvβ3 integrin will therefore require convincing target validation and demonstrable evidence that preclinical models are predictive of effects in the clinic. Academic research on the role of αvβ3 in cancer and other areas continues, aided by several high-quality tools that are available to test integrin-mediated mechanisms in new disease models to potentially deliver more effective therapies.

In recent years, αvβ6 and the well-established target α4β7 have attracted substantial interest as therapeutic targets. Several new inhibitors of the αvβ6 and/or αvβ1 integrins (most of which are small molecules) have progressed to clinical trials in the past 5 years, and α4β7-directed therapies (mostly antibodies) have advanced to late-stage clinical trials. The change in focus from αvβ3 to αvβ6 also switches the focus of downstream effector pathways from angiogenesis to modulation or inhibition of the transforming growth factor β (TGFβ) pathway (Box 3). The increased interest in α4β7 has presumably come from the clear benefit to patients from prescribed drugs, such as natalizumab, and the need for alternative therapies after several terminated clinical trials and withdrawals, specifically in ulcerative colitis and multiple sclerosis.

Modalities and mechanisms

Integrins can be targeted with a range of mechanisms designed to activate (as agonists) or inactivate (as silent antagonists or inhibitors) the integrin complex, inhibit a secondary biological process initiated by the integrin (as functional antagonists), deliver a cytotoxic drug (as a drug conjugate) in a cell-specific manner or direct chimeric antigen receptor T (CAR T) cells for immunotherapy46,47,48. The majority of integrin drug discovery initiatives have set out to target the orthosteric binding sites (endogenous ligand-binding sites) on integrins, which are formed when the α- and β-subunits bind to each other in a noncovalent complex. However, several drug discovery programmes that targeted the orthosteric site have not been successful, which highlights the potential risks of this approach. Some of the approved therapies exert their effects via allosteric interactions with the α-subunits of the integrins they target; for example, natalizumab exclusively targets the α4-subunit of α4β1 at an allosteric site (ref.49) (Table 1).

Integrins can exist in an activated or inactivated state, in which they demonstrate high or low affinity for ligands, respectively50. Extracellular ligand binding to the orthosteric site induces intracellular signalling, but also shifts the integrin from a low-affinity to a high-affinity state7. Under normal physiological conditions, this enables cells to respond to changes in the extracellular environment, but this response can become exaggerated and unwanted under pathophysiological conditions51. Therefore, an antagonist molecule designed to bind to an integrin and prevent an endogenous ligand from binding may also have a direct agonist effect on the integrin if it causes this shift in affinity. These effects have been observed for marketed αIIbβ3 RGD mimetics such as eptifibatide52, which induced severe thrombocytopenia in a small group of patients53, a phenomenon that was particularly evident during clinical trials with oral inhibitors54,55. In a related observation, low concentrations of αvβ3 integrin RGD mimetics stimulate, rather than inhibit, tumour growth and angiogenesis in preclinical models56,57. Agonist effects can therefore also occur at low receptor occupancy for multiple members of the integrin family, perhaps in a manner analogous to the two-state model of activation of another family of membrane-bound receptors, the G protein-coupled receptors58. Importantly, these agonist effects may have caused or contibuted to the clinical failure of cilengitide (1)56. Paradoxically, however, low-dose activation of an integrin receptor may indeed have therapeutic potential: in preliminary studies with cilengitide at low doses, its pro-angiogenic effects enhanced the delivery and potency of the chemotherapy agent gemcitabine to tumours59.

In light of this potential issue, academic groups have used structure-guided design to identify silent small-molecule integrin antagonists. Initial research hypothesized that the limitations manifested by thrombocytopenia and/or increased bleeding times with αIIbβ3 inhibitors were partially because the inhibitors induced high-affinity conformations of the integrin60. Inhibitors that minimize the agonist effects by stabilizing the low-affinity conformation have been designed and identified. One of the first breakthroughs in this area came with the amine ligands RUC-1 (ref.61) and RUC-2 (ref.62), which are non-RGD mimetics that bind to the orthosteric site in αIIbβ3 and induce smaller conformational changes in the β3-subunit than marketed agents do. This observation has been backed up preclinically with the next molecule in the series, RUC-4 (12), which has reduced bleeding and the appropriate pharmacokinetic properties required for an antithrombotic drug, and may therefore be effective in the clinical setting63. RUC-4 is currently in a phase II clinical trial (NCT04284995) by CeleCor Therapeutics64 to assess the pharmacokinetic and pharmacodynamic properties of a single subcutaneous injection in patients with a myocardial infarction65. This is a promising approach, and applying this binding principle to other αv-containing integrins could provide opportunities to develop new non-zwitterionic chemotypes that do not induce potentially undesired conformational states. To our knowledge, such molecules are still in preclinical development, but the results are eagerly anticipated66.

Another group, meanwhile, reported the first crystal structure of a mutant of fibronectin bound to αvβ3 that acts as a ‘pure’ antagonist67. Other pure αvβ3 small-molecule antagonists have been designed, using cryogenic electron microscopy imaging of integrin conformations, that had limited access to the high-affinity conformation of αvβ3 and did not enhance angiogenesis at low concentrations60. Using similar methodology, small-molecule pure antagonists for αIIbβ3 were designed that have reduced bleeding in preclinical models68. These molecules are traditional RGD ligand mimetics, unlike the RUC series of compounds. Additionally, a high molecular weight polypeptide disintegrin, TMV-7, which recognizes the αIIb β-propeller domain, does not induce a conformational change in the β3-subunit, and maintains the antithrombotic effects with little tendency for bleeding69. Several exciting avenues are currently being investigated in the pursuit of safer and more effective integrin inhibitors.

Inhibitors such as TMV-7 that bind allosterically may have fewer unwanted side effects. With this approach in general, the affinity state of the receptor is also probably less relevant, because the binding site is distinct from the orthosteric site. Allosteric molecules could block integrin activation either by occluding the orthosteric site or by inducing a conformational change that shifts the integrin to a low-affinity state. However, and especially for antibodies, reduced selectivity may result if the integrin being targeted contains either an α- or β-subunit that pairs with multiple other β- or α-subunits. For example, abituzumab, an antibody that binds to an allosteric site on the αv-subunit and blocks the RGD site, likely inhibits all αv integrins, and this may broaden the risk of on-target toxic effects and thereby potentially reduce the therapeutic window.

The levels of several integrins are increased on tumours, so targeting these integrins with a conjugated cytotoxic molecule could be an effective strategy to specifically target tumour cells70. Instead of directly inhibiting an integrin, this novel approach, using drug conjugates that target αvβ6, takes advantage of high local expression in diseased tissue compared with relatively low levels elsewhere. Interestingly, the β7 subunit, in combination with a number of tumour cell antigens, has also been targeted with a CAR-T cell approach in multiple myeloma46 in an effort to improve the immunosuppressive microenvironment of tumours (Table 2).

Modifying molecules to reduce their half-lives in the systemic circulation can also ameliorate some potential toxicity. For example, the high-affinity αvβ6-binding small-molecule RGD mimetic, GSK3008348 (5), is internalized and degraded in cells, thereby reducing both lung and systemic drug load following inhaled administration71. This molecule is both an agonist of αvβ6 and a functional antagonist of TGFβ, as it indirectly inhibits TGFβ signalling by reducing TGFβ activation. Ligand-induced internalization could also be exploited to improve selectivity. For example, if αvβ6 was the only αv-containing integrin to be internalized quickly and return to the cell surface slowly following RGD binding, inhibitors targeting this integrin via an RGD-mimetic interaction would have an additional selectivity bias. The functional consequences of this approach would depend on the type and duration of signalling initiated as a result of internalization, which are not known, but to date there has been no evidence of negative effects.

The general advantages and disadvantages of small versus large integrin inhibitor molecules have been described elsewhere72. However, given the seemingly delicate balance between beneficial and detrimental effects derived from full and partial engagement of integrins, from a safety perspective, small molecules may well be better than antibodies because they are cleared from the body in hours, whereas antibody clearance takes days or weeks. It is also easier to have periods of time when no drug is present (for example, if adverse effects arise) and manage receptor occupancy levels with small molecules than with antibodies. Small molecules can also be dosed at home via the oral or inhaled route of administration, whereas antibodies are often dosed at a clinical site by subcutaneous or intraperitoneal injection.

With the breadth and complexity of biology and mechanisms at play within the integrin family, it is likely that only a small component of integrin biology is understood, meaning further research is required to fill the gaps. Do the phenomena observed with certain integrins after drug binding — agonism, internalization and downregulation — also occur with other integrins? More research and an improved set of tool molecules73 are required to further dissect these complexities. Indeed, current RGD-binding integrin drug discovery efforts, irrespective of their clinical successes, are likely to catalyse future research because they will deliver a new set of well-characterized tools. These investigations could lead to the successful generation of integrin-targeting drugs.

Diseases involving integrins

For the subfamily of integrins that contain the αv-subunit, a plethora of target validation studies have been completed in fibrotic diseases and oncology. In this section, we summarize the target validation studies that support the hypotheses that integrin inhibition will have a therapeutic benefit in fibrotic diseases, cancer and viral infections.

Pulmonary fibrosis

Currently, clinical investigations for the antifibrotic potential of integrin inhibitors centre on IPF, a debilitating chronic condition of unknown aetiology. Several αv-containing integrins (αvβ1 (refs74,75), αvβ5 (ref.76) and αvβ6 (ref.77)) are upregulated in IPF. αvβ1 and αvβ6 are the most thoroughly validated targets; levels of αvβ6 also have potential prognostic value78. Multiple integrins may be involved in IPF: αvβ6 drives TGFβ activation in alveolar epithelial cells while αvβ1 mirrors this in myofibroblasts, which characterize and contribute to the development of fibrotic diseases. The contribution of the integrins and cell types to IPF depends on the disease stage; αvβ6 is implicated in the early phase, during epithelial damage, after which αvβ1 in myofibroblasts drives the fibrotic foci75.

αvβ6 was the first αv integrin to be identified as crucial in IPF. Gene knockout and pharmacological intervention studies in the bleomycin-induced mouse model of lung fibrosis with a selective αvβ6 monoclonal antibody, BG00011 (ref.79) (known as STX-100 and 3G9 preclinically), demonstrated that αvβ6 deletion or inhibition could prevent the development of fibrosis or reverse established fibrosis, respectively. Later work showed that blocking αvβ1 with a small-molecule inhibitor (9) in the bleomycin model could reverse established fibrosis75, but the lack of selectivity of this molecule, which also binds to non-αv-containing integrins, suggests that further target validation work is required74. It is worth noting that genetic deletion of αvβ1 is not possible because homozygous β1-null mice do not survive through to birth80 and therefore a key preclinical tool in the target validation armoury for αvβ1 is missing. The pan-αv inhibitor MK-0429 (13 in Fig. 3) also reduces fibrosis progression in the bleomycin-induced mouse model81, supporting the notion that αvβ1 and αvβ6 have important roles in fibrosis, but the data from this inhibitor does not differentiate their contributions. However, as the drug was dosed 5 days after bleomycin treatment in this study, during the inflammatory phase of the model rather than in the subsequent fibrotic stage, the results should be treated with caution.

Fig. 3: Selected small molecules with oral properties that bind to RGD integrins.
figure 3

The small molecules shown here have either been evaluated in the clinic or have been examined in preclinical models and are administered by a parenteral route with advantageous potency or selectivity profiles or are predicted to have these attributes based on their physicochemical properties. The molecules are: 13 MK-0429, 14 SB-273005, 15 PLN-74809 (proposed structure), 16 an example of the series from Bristol-Myers Squibb patent WO2019/094319, 17 an example of the series from St Louis University and Indalo Therapeutics patent WO2018/132268, 18 an example from GlaxoSmithKline’s patent WO2016/046226, 19 an example of the series from St Louis University patent WO2017/117538. RGD, Arg–Gly–Asp.

There is some preclinical evidence that α3β1 (ref.82), α4β1 (ref.83) and α8β1 (ref.84) integrins play a role in pulmonary fibrosis, and that αvβ8 is important in small airway fibrosis associated with chronic obstructive pulmonary fibrosis and asthma85. However, these are challenging integrins to target because few selective small molecules and antibody tools exist, and perhaps, as a consequence, to the best of our knowledge, there are no advanced IPF drug discovery campaigns focused on these integrins.

Hepatic fibrosis

Hepatic fibrosis is associated with the end stage of chronic liver diseases such as chronic virus-induced hepatitis B and hepatitis C, and nonalcoholic fatty liver disease (NAFLD). NAFLD encompasses a range of chronic liver diseases related to obesity, steatosis (accumulation of fat) and NASH that can lead to fibrosis, cirrhosis and hepatocellular carcinoma. In end-stage liver disease, αvβ6 protein levels are increased in bile duct epithelia and transitional hepatocytes, and the αvβ6 mRNA levels increase with disease progression in patients with hepatitis C86. In vivo target validation studies for individual integrins have been confined to less-relevant disease models such as the chemically induced carbon tetrachloride (CCl4) and the surgery-based bile-duct ligation (BDL) models, which are not particularly representative of end-stage human liver fibrosis. As with IPF, both αvβ1 and αvβ6 have been implicated in the development of liver fibrosis, with evidence from genetic deletion or pharmacological inhibition with small molecules or monoclonal antibodies in animal models75,87,88. In addition, pan-αv blockade attenuated fibrosis in a more relevant model of NASH89 — the choline-deficient, amino‐acid-defined, high‐fat diet model — and further interrogation of this model with selective tools may aid selection of the optimum integrin or integrins to target. The most advanced integrin-directed therapy for NASH is the selective αvβ1 inhibitor PLN-1474, which was recently acquired by Novartis AG from Pliant Therapeutics (Table 2).

Chronic kidney disease

In comparison with the target validation data that implicate αv-containing integrins in lung and liver fibrosis, the kidney has attracted less attention, possibly because there are no fibrosis-specific surrogate end points for clinical trials, so demonstrating clinical efficacy in short-term studies is difficult90. Several chronic renal diseases such as glomerulonephritis, diabetes, IgA nephropathy and Alport syndrome are associated with fibrotic changes and increased αvβ6 expression in epithelial cells91. Although some evidence supports a role for αvβ1 in kidney fibrosis92, most target validation studies suggest that αvβ6 is more relevant91,93. A breakthrough in understanding how to test this hypothesis clinically is needed. Interestingly, Merck recently patented an historic pan-αv molecule, MK-0429 (13), for chronic kidney disease94. This patent was supported by preclinical evidence; efficacy was demonstrated in a rat in vivo model of diabetic nephropathy95.

Skin fibrosis

Skin fibrosis may be present in systemic sclerosis (SSc, a connective tissue disease), hypertrophic scarring and keloid lesions. There is some evidence that integrins have a role in skin fibrosis, specifically SSc, as fibroblasts isolated from disease samples showed elevated levels of activated β3-subunits96. In addition, the dual αvβ3/αvβ5 inhibitor cilengitide blocked cutaneous fibrosis when therapeutically administered in a murine model of SSc97. Interestingly, these effects were thought to be due to inhibition of integrin signalling pathways rather than blockade of TGFβ activation. It is possible that a pan-αv integrin inhibitor could be of value in SSc-associated interstitial lung disease (SSc-ILD); in this disease, αvβ6 is hypothesized to contribute to disease in the lung98, and αvβ3 and/or αvβ5 may drive the skin manifestations. However, an investigational study using the pan-αv monoclonal antibody abituzumab was terminated for lack of eligible patients (NCT02745145).

Cancer

Numerous integrins, namely αvβ3, αvβ5 and α5β1, have been investigated as potential therapeutic targets for various cancers for more than 25 years. The role of integrins in cancer22,99,100 has been reviewed extensively elsewhere but we highlight some of the key findings for inhibitors of αv-containing integrins.

Although the rationale for targeting αvβ3 and αvβ5 in cancer was plausible based on the preclinical data and the role of these molecules in tumour angiogenesis, this approach has not led to clinical success. Additional cell types may need to be targeted in the tumour microenvironment for these inhibitors to have a benefit38. αvβ6 is also upregulated in several tumours101 and could also have prognostic potential102,103,104,105. Clinical trials have investigated the pan-αv antibodies abituzumab and intetumumab in colorectal carcinoma106 and melanoma107, respectively, as monotherapies, in combination with chemotherapy or in combination with chemotherapy and an epidermal growth factor receptor inhibitor. Although neither set of studies achieved statistically significant efficacy in its primary end points, there was some evidence of improvement in patients with colorectal carcinoma that expressed high levels of αvβ6 (ref.106). This suggests that selecting patients who have high αvβ6 expression in the target organ could maximize the chances of clinical efficacy. The selective imaging tools now available108 (Table 4) could form part of a triage strategy in future trials and for other indications in which αvβ6 is a potential key driver of disease. With this rationale, selectively targeting αvβ6 with antibodies or peptides may have utility in treating pancreatic cancer109 or breast cancer110 or as conjugates to antitumour agents111.

Table 4 Recent clinical imaging studies targeting integrins

Specific subsets of cancers are targeted by immunotherapies that disrupt the programmed cell death protein 1 (PD1)–PD1 ligand 1 (PDL1) axis. These surface proteins constitute a receptor–ligand pair and are members of a family of checkpoint inhibitors that moderate immune function and halt the development of a T cell response112. This checkpoint protects the host from autoimmunity and inappropriate immune responses but is inappropriately activated in some tumour microenvironments; overexpression of PDL1 in tumour cells induces tolerance by binding to PD1 on T cells, thereby inhibiting cytotoxic T cell activation and proliferation, and cytokine secretion. Tumour cells thus evade immune detection113. A new class of anticancer immunotherapies that target susceptible PDL1-expressing cancers has emerged over the past decade, and several biologic agents that target this pathway have been approved in cancer114.

Recent work highlights three promising independent mechanisms by which integrins may be targeted to treat cancer: reducing the expression of PDL1 in cancer cells, reducing the levels of TGFβ in the tumour microenvironment and targeting leukocyte integrins to prevent T cells from homing to tumours. Integrins participate in regulation of PDL1 expression and are thus an important constituent of the immune evasion apparatus. αvβ3 positively regulates PDL1 expression in the tumour microenvironment115, and in murine studies αvβ3 depletion restricts the growth of primary tumours116. This is particularly relevant because most patients with cancer who receive anti-PD1 or anti-PDL1 therapies do not respond to treatment117; anti-αvβ3 therapy may sensitize tumours to disruption of this axis and therefore be useful in combination with PD1- or PDL1-targeted agents. Because overexpression of αvβ3 is a common feature in some cancer types and is often associated with poor prognosis, targeting αvβ3 in these cancers could unmask tumours protected by PDL1 overexpression, rendering them susceptible to treatment with an anti-PD1 therapy. Selecting patients on the basis of αvβ3 expression could increase the likelihood of a positive outcome. Inhibiting integrins other than αvβ3 may also sensitize tumours to checkpoint inhibitors. For example, inhibition of αvβ6 induced T cell-mediated immunity in immunotherapy-resistant tumour models118.

Similar to αvβ3, blockade of αvβ8 has been shown to potentiate a cytotoxic T cell response in tumours, although these effects seem to be independent of the PD1–PDL1 axis. In contrast to αvβ3, αvβ8 expression in tumours does not usually correlate with PDL1 expression, although αvβ8 expression on cancer cells drives tumour growth in vivo. αvβ8 promotes tumorigenesis through a mechanism different from that of αvβ3 that may involve TGFβ. In this alternative mechanism of immune evasion, active TGFβ is released from its latent form, which is present on immune cells, by binding to αvβ8 on tumour cells119 or potentially on immune cells120,121. Active TGFβ in the tumour stroma can prevent the penetration of T cells into the tumour and thus protect tumours from T cell attack122,123.

The success of strategies that directly target αvβ3 and αvβ8 will likely be tied to the expression profiles of these integrins on individual tumours, whereas improving T cell adhesion and activation by targeting leukocyte adhesion integrins should be less dependent on the precise mechanisms by which individual tumours evade host immunity. Allosteric activation of the leukocyte-specific integrins αLβ2 and α4β1 in T cells with the small molecule 7HP349 enhanced T cell activation and adhesion, and thereby improved the penetration of T cells into tumours in mouse models of melanoma and colon carcinoma124. This compound is now in phase I trials (Table 2).

Efforts are also being made to conjugate integrin-binding small molecules, peptides and antibodies to bioactive moieties to target specific tissues. Integrins are cell surface receptors and are overexpressed in specific diseased tissues, and are therefore perfect candidates for the application of this technology. So far, efforts have centred on RGD-binding approaches to target delivery of drug conjugates to tumours, and αvβ3 has been the primary focus owing to its role in the development of tumour vasculature. Proof-of-concept experiments with nanoparticles as the drug moieties and various cyclic RGD peptides as the αvβ3-targeting component demonstrated that these molecules rapidly localized to tissues and induced targeted cytotoxicity in vivo125,126,127. The potential to deliver other tools, such as diagnostic imaging ligands, to tissues is also of interest99,128. Studies have successfully delivered therapeutic and imaging compounds to specific tumours and improved the uptake and activity of the bioactive cargo in tumours. Altering the targeting moiety and conjugate design can give this system remarkable flexibility and precision, and this conjugate approach may well lead to successful treatments. Indeed, a recent clinical trial (NCT04389632) has been initiated to investigate an antibody–drug conjugate that recognizes β6 to selectively target solid tumours129.

Viral infections

In light of the SARS-CoV-2 pandemic of 2020 and beyond it is appropriate to highlight the less explored role that integrins can have in virus transmission. In this regard viruses use various mechanisms, including binding to integrins, to gain cell entry or attachment. By expressing relevant peptide sequences on their surfaces, viruses can bind to integrins to invade host cells, activate intracellular signalling events and mediate disease pathogenesis130. Viruses commonly co-opt the RGD recognition sequence, which, in principle, can therefore be targeted to intervene in a range of viral infections131. Notably, RGD-binding integrins are used, amongst others, by Zika virus (αvβ5)132, rotavirus (αvβ3)133, Ebola virus (α5β1)134 and foot-and-mouth-disease virus (αvβ6)135 to gain entry to host cells. However, a potential drawback for the design of integrin inhibitors for this use is the risk of redundancy, whereby alternative cell entry mechanisms are available to viruses to enable rapid evasion from drugs.

Recently, it was postulated that RGD-binding integrins are also co-receptors for angiotensin-converting enzyme 2 (ACE2)136, which is used by SARS-CoV-2 for host cell entry; thus, integrin inhibitors could have utility in multiple types of viral infection137. In in vitro experiments, the noncompetitive α5β1 inhibitor, ATN-161, reduced infection138. Interestingly, Pliant Therapeutics is investigating its dual αvβ6/αvβ1 small-molecule inhibitor in a phase II clinical trial (NCT04565249) in patients with COVID-19 and acute respiratory distress syndrome (ARDS).

Drug design for RGD-binding integrins

Small-molecule inhibitor properties

Small molecules are important for two key reasons: first, as tools and probes139 (including imaging molecules; Table 4) and secondly as safe and efficacious marketed drugs. Tool molecules, ideally with suitable drug-like characteristics, can answer key mechanistic and biological questions in preclinical settings. For example, integrin inhibitors from UCSF, c8 (9), and from St Louis University, CWHM-12 (10), have been used extensively in target validation studies for fibrotic diseases. However, converting a tool or an early lead into a drug molecule is a challenging and time-consuming business, and all the knowledge and design is ultimately combined into a single entity, from perhaps many hundreds or even thousands of molecules profiled. Many of the recent integrin clinical candidates are small molecules that have undergone extensive optimization processes and can offer significant benefits, such as in activity and pharmacokinetic profiles, compared with other modalities, including antibodies or larger conjugate molecules. In this section, we therefore discuss the properties of these molecules in detail, the selectivity challenges and how emerging new research could alter future drug design.

It is evident from the low number of molecules that have been developed from research activities and are clinically successful that small-molecule integrin inhibitor drug design is not straightforward. However, for integrins, emerging insights from structural biology and pharmacology research may be game changing. Clearly, the validation of the target in clinically predictive disease models is crucial, but additionally, a key factor to now consider is how molecules affect the integrin conformational states and the relationship — if any — between these states and safety and efficacy. This relationship has not been routinely investigated and research is ongoing, but ideally such efforts would be part of any integrin lead optimization programme, although such endeavours are likely to be resource intensive. The conformational states induced by αvβ3 and αIIbβ3 inhibitors and subsequent signalling is potentially paralleled by other integrins, including other RGD-binding integrins and α4β7.

There are a substantial number of inhibitors described for integrins, a small subset of which are suitable for clinical evaluation. The vast majority of RGD-binding integrin inhibitors are RGD mimetics with physicochemical properties that are not compatible with oral bioavailability. Much of modern drug design is facilitated and informed by heuristics: Lipinski’s ‘Rule of Five’140,141 is perhaps the most widely used set of guidelines for drug design. Although these tools are blunt and simplified142, they are useful because they are easily applied and allow the rapid design of drug-like molecules. There is also a growing recognition that molecules outside Lipinski drug space, so called ‘beyond Rule of Five’, are of value for unusual or less tractable targets143,144. Compared with marketed oral drugs, most αv integrin inhibitors, including those that can be orally dosed, display molecular descriptors that conform less to the Rule of Five and Veber’s rules145 owing to the intrinsic polar nature of the RGD pharmacophore, with increased molecular weight, an increased number of hydrogen bond donors or acceptors, a higher number of rotatable bonds and larger polar surface area (PSA) (see Supplementary information for a plot of the physicochemical properties of integrin molecules compared with FDA marketed oral drugs). There are also large differences in the properties of integrin inhibitors according to the route of administration — those delivered parenterally (Fig. 2) and those delivered orally (Fig. 3). Only molecules from the parenteral class of RGD-binding integrin inhibitors have become marketed drugs (namely tirofiban and eptifibatide), which may reflect the challenges of obtaining high-quality orally bioavailable inhibitors in this physicochemical space (as discussed earlier). Compromises often have to be made to the preferred pharmacological profile to obtain molecules with sufficient permeability and/or bioavailability for clinical use. Approaches to mask the polarity as ester prodrugs have also enabled moderate oral bioavailability for several αIIbβ3-inhibitor small molecules146,147, which demonstrates that ionization, polarity and lipophilicity are important for oral absorption of integrin inhibitors. Although prodrugs are more complex to progress to the market than single drug entities, they can maintain the key potency and selectivity requirements that may otherwise be compromised in favour of oral absorption.

RGD versus non-RGD

Almost all RGD-binding integrin inhibitors are RGD mimetics featuring a mimetic for the guanidine of arginine and a β-arylpropionic acid or α-amino-carboxylic acid to replace the aspartic acid72 (Figs 2,3). Such compounds are zwitterionic or amphoteric and thus usually charged at physiological pH; this constrains molecular design, and obtaining good oral pharmacokinetics can be challenging. We are unaware of any small-molecule modulators of αv integrins that bind outside the orthosteric ligand-binding site, although several large molecules — including an αvβ6 antibody (BG00011, 3G9; Biogen)81 and ProAgio (an αvβ3-binding protein)148 — have been described. Also, as previously mentioned, ATN-161 is a small-molecule peptide that binds outside the RGD-binding site of α5β1. Consequently, new chemotypes that are non-zwitterionic — either basic or acidic but not both — would be useful. Such chemotypes would have different physicochemical characteristics from traditional RGD mimetics, and would almost certainly change the nature of the pharmacokinetic design challenge. However, few non-RGD-mimetic compounds have progressed. Many feature functionality that is incompatible with oral bioavailability or are unlikely to demonstrate high passive permeability, and some include structural motifs that are found in pan-assay interference compounds (PAINS)149.

RUC-4 (12) is a notable exception. This αIIbβ3 inhibitor does not have a carboxylic acid (which otherwise forms a strong interaction with the magnesium ion in the binding site), and thus is basic and has non-zwitterionic properties. In the ligand binding site, the primary amine displaces the magnesium ion, locking the receptor in an inactive conformation62,63. In addition, GlaxoSmithKline (GSK) designed potent and selective αvβ1 inhibitors150 that, whilst conforming to the general RGD-mimetic motif, were non-zwitterionic, featuring a phenyl urea as a non-basic arginine mimetic. Non-zwitterionic molecules may offer good permeability and absorption while still acting as ligands in the RGD-binding site.

Numerous approaches to identify new small-molecule chemotypes can be considered. Structure-based design could be useful because αvβ3 and αvβ6 crystal structures, as well as homology models for other RGD-binding integrins, are available151. Crystal structures of integrins in the inactive unbound forms through to the activated ligand-bound forms have been published and these structures have elucidated fundamental principles of integrin activation4,49,50,61,62,67,68,152,153. The conformation of the tertiary structure of integrins can change considerably, although the actual binding site regions change less dramatically. Recent publications suggest that ligands for different conformations of αvβ3 and αIIbβ3 can be designed, including peptidic pure antagonists67 or small molecules60,61,62. It remains to be seen what effect these compounds have clinically, but pure antagonism of these integrins could overcome previously encountered problems and thus trigger significant renewed interest. In our own experience, however, even if the potency and selectivity of RGD-integrin inhibitors can be rationalized from modelling studies in hindsight, the design of inhibitors de novo from modelling and docking studies is challenging154. Similarly, despite NMR studies that identify ligand-binding interactions with αvβ6 (ref.155), αIIbβ3 (ref.156) and αvβ3 (ref.157) in a cellular environment — which more closely reproduces physiological binding — these studies have predominantly provided insights retrospectively.

Combined computational and NMR studies enabled the design of a potent and selective αvβ6 small peptide from a nonapeptide158. In this novel approach, a computationally driven algorithm, using a docking score as proxy for binding activity, successfully predicted active analogues. But by far the most common and successful approach (based on new patent filings) has used existing, published knowledge159. As a result, many of these chemotypes look similar to each other, and the structural space is becoming crowded.

Selectivity

The desired integrin selectivity can greatly affect the design of integrin inhibitors. No αv-targeting molecules are clinically proven, so pharmacological effects cannot be assigned to specific selectivity profiles. In treating fibrotic diseases, for example, it is unclear which αv-containing integrins are pivotal in any particular disease in humans and this is complicated by differing integrin expression levels across the tissues and organs, and in animal models. Validation studies conducted with non-selective tools may therefore shed little light on the precise selectivity requirements to treat a specific disease. However, because non-selective molecules appear to be safe in the clinic (Table 3), investing the time and resources to identify the specific selectivity profile in validation studies may not always be necessary. For some applications, though, selectivity is probably important.

Many inhibitors of αv-containing integrins share key binding characteristics72. The homologous RGD pharmacophore recognition sequence in these integrin binding sites, especially in integrins in which the α-subunit binding domains or the β-subunits are similar, makes the design of highly selective small molecules difficult, even with the help of ligand–protein X-ray crystal structures and the construction of homology models. As a result, a range of selectivity profiles is attainable, but designing molecules that target one specific integrin and not others is challenging.

Inhibitors of αv integrins can often be assigned to one of four broad ‘selectivity buckets’, which indicate likely selectivity profiles: selective dual αvβ3 and αvβ5 inhibitors; pan-αv inhibitors; selective dual αvβ6 and αvβ8 inhibitors; and selective αvβ1 inhibitors (selective for αvβ1 over the other αv-containing integrins). Molecules in this final group may inhibit other β1-containing integrins, and it is too early to assign additional αv selectivity patterns to these inhibitors, but will also constitute a bucket. Although most compounds fall into one of these buckets, the selectivity profile of a compound may have been either intentional or merely tolerated. Obtaining selectivity for αvβ3 and αvβ5 over αvβ6 and αvβ8 or vice versa is relatively straightforward, but obtaining selectivity within these buckets or pairs is challenging, owing to the subtle differences in binding site architecture, and is less well understood. However, for examples where selectivity has indeed been achieved within a bucket, changes are often made to the parts of the molecule that bind to the ‘specificity-determining loop’72.

Few small molecules with selectivity for αvβ5 (particularly over αvβ3) or αvβ8 (particularly over αvβ6) exist, although an acyclic peptide (8)160 that is selective for αvβ8, and αvβ5-selective molecules161, have been identified. Interestingly, the current clinical molecule, SF0166 (11), is reported to be selective for αvβ3 over αvβ5 by virtue of only small structural changes34. Additionally, a few αvβ6 inhibitors have good selectivity over αvβ8, such as compounds from GSK (5, 6)154,162. Antibodies that bind to the various RGD integrins have already been described and have a range of selectivity profiles72.

Interestingly, some of the small-molecule αvβ3 inhibitor lead molecules came from αIIbβ3 compounds and some of the αvβ6 leads originated from in-house αvβ3 leads163, so molecules with different selectivity profiles from the desired profile can be useful starting points.

Given the current clinical potential of αvβ6 inhibitors, can selective αvβ6 small-molecule inhibitors be developed using the wealth of information already available for published αvβ3 molecules? Some specific design tricks could be used to switch selectivity. Many αvβ6 inhibitors (for example, molecules 5, 15 and 18) have basic linker regions, whereas αvβ3/αvβ5 inhibitors (for example, 13 and 14) often have neutral, non-ionizable linkers. The basic group interacts with a threonine residue that is present in αvβ6 but absent in αvβ3 and αvβ5. The molecule (6), which is analogous to the inhaled candidate (5), contains a methyl quaternized nitrogen162 in the linker region, and the presence of this ionized nitrogen increases the potency and selectivity for αvβ6. Unfortunately, these advantages come at the expense of low permeability and reduced oral bioavailability, so this molecule (6) and similar molecules are more suitable for parenteral administration. As with all small-molecule drug design, the properties of molecules are intertwined, and structural changes to optimize selectivity may make optimization of other parameters, such as oral bioavailability, more challenging.

Similarly, certain structural motifs have affinity for specific subunits. The bulky sulfonamide contained in (9), or similar analogues, binds strongly to the β1-subunit and consequently this molecule interacts with several other β1-containing integrins, such as α4β1 (ref.164), α2β1 (ref.165), and αvβ1 (ref.166).

In the absence of an αvβ1-selective antibody, small-molecule tools (such as 9) have proved invaluable for initial target validation studies in fibrotic diseases. However, because it also binds to other β1-containing integrins, this molecule may be less useful to pinpoint the importance of specific integrins as relevant targets74. This integrin cross-reactivity highlights the need for comprehensive cross-screening. However, such screening is less available: unlike kinase screening panels, which are commercially available (example firms are MRC PPU and Eurofins; see Related links); integrin screening panels are not. Furthermore, the observed potency and selectivity of ligands also depend on the type of assay, conditions and set-up, which can make inter-compound comparisons less robust. We recently tested a comprehensive set of tools that bind αv-containing integrins in the same assays72, and the profiles of peptidic tools have also been compared167.

Several molecules have activity at multiple αv integrins, including αvβ6, and are orally bioavailable. Such molecules have been developed by Pliant (15)168, GSK (18)169 and St Louis University (19)170, and show good permeability. In each of these examples, the linker is either neutral, or the pKa of the basic nitrogen in the linker is modulated by virtue of a proximal fluorine or ether functional group. Molecular conformational flexibility can also allow intramolecular folding to ‘hide’ ionized functionality171,172, but this can complicate the molecular design even further.

Clinical data

The key integrin inhibitors in ongoing clinical studies are summarized in Table 2 and those that have completed clinical trials are listed in Table 3 with highest clinical dose, safety and efficacy information. The majority of studies target or have targeted the αv-containing integrins, particularly αvβ6 and αvβ3. The most advanced clinical molecule currently in clinical trials is the oral dual αvβ1/αvβ6 inhibitor, PLN-74809 (15; Pliant Therapeutics), which is currently in phase II for IPF (NCT04072315). Targeting the lung specifically has also been investigated with the inhaled and selective αvβ6 small molecule GSK3008348 (5; GSK)71 but this programme has been strategically placed on hold, despite observed target engagement in a phase Ib study of IPF using an imaging readout173 (Table 4). The development of the αvβ6-selective antibody from Biogen (BG00011) was recently terminated because of undisclosed safety concerns174, and the development of the pan-αv integrin inhibitor IDL-2965 (NCT03949530) for NASH was also terminated. It remains to be seen whether these setbacks will deter companies such as GSK, Pliant Therapeutics, Bristol-Myers Squibb and Morphic Therapeutics, which hold patents describing molecules with varying αv activities (molecules 1518)159.

Both pan-αv antibodies and small molecules appear to be well tolerated in humans (Table 3), although the lack of efficacy for several molecules suggests that target engagement may have been insufficient to test for potential mechanism-based toxicity. Although the clinical data from BG00011, which was terminated for undisclosed safety reasons, raises potential safety concerns about selective αvβ6 inhibition, small doses of the inhaled selective inhibitor 5 were well tolerated. It is possible that sustained αvβ6 inhibition detrimentally affects the anti-inflammatory and protective roles of this integrin, and in an already compromised fibrotic lung, this could become a risk. The corollary, therefore, is that there is only a small therapeutic window for targeting αvβ6 in the lung, so the dose and therapeutic modality are particularly important. A study in chronic allograft dysfunction was also withdrawn for BG00011 (then named STX-100, NCT00878761) before recruitment of patients, potentially as a result of preclinical work indicating a protective role for αvβ6 (ref.175).

Although the marketed drug abciximab therapeutically targets αIIbβ3, it also appreciably inhibits αvβ3 (ref.176), adding further evidence that targeting αvβ3 is safe in humans. Cilengitide, too, has been widely explored clinically and seems to have failed for efficacy rather than safety reasons. Although the dosing route is not correlated with toxicity, inhaled dosing may be advantageous for treating diseases of the lung, as it could result in lower systemic drug levels.

As for other therapeutic areas, the generation of translational biomarkers, the demonstration of target engagement and the development of pharmacokinetic and pharmacodynamic relationships are key to understanding the chances of success. In the integrin field, there are few historical studies in which these characteristics have been evaluated, but studies have recently begun to take this translational approach. For example, the target engagement of GSK3008348 (5) was demonstrated using an αvβ6-specific radiolabelled peptide ligand in bleomycin-treated mice via single-photon emission computerized tomography (SPECT)71 and confirmed in patients with IPF via positron emission tomography (PET)173. In addition, the functional consequences of αvβ6 integrin inhibition (using levels of phosphorylated SMAD, a marker for TGFβ activation) have been measured in bronchoalveolar lavage fluid from healthy subjects or patients with IPF, and used as a pharmacodynamic measurement to optimize the dose177.

Reagents to image αv-containing integrins (Table 4) have been used to define integrin expression levels and to determine whether a drug engages its target. As integrins themselves could have prognostic value — for example, levels of αvβ6 in IPF and cancer — imaging an integrin in patients could be used to track disease progression. This approach could also be used to select for patients with either high levels of target expression or rapidly progressing forms of disease. For example, a retrospective analysis of αvβ6 expression in patients with colorectal carcinoma who received the pan-αv antibody abituzumab suggested that individuals with high expression levels were more likely to benefit from the therapy106.

Challenges and prospects

Considering how many drug discovery projects and clinical studies have focused on integrins over the past 30 years, the number of approved therapies has been disappointing5,26. Translating the potential, and indeed the validity, of the preclinical integrin data into clinically efficacious drugs is clearly not without complexity. These clinical failures likely reflect a combination of challenges that are generic to drug development and those that are target-class specific. Many of the integrin inhibitors that were tested historically had suboptimal pharmacokinetics, with routes of administration and dosing regimens that led to poor exposure and target coverage, which is a generic challenge for small-molecule drug development14,72. This, coupled with a lack of clinical pharmacodynamic biomarkers to measure target engagement, left the relevance of integrins in the disease of interest unknown, as molecules often failed to test the hypothesized mechanism. Conversely, target coverage as a result of poor pharmacokinetics was less of an issue for the large molecules (such as antibodies) that were tested in the clinic, and indeed these have arguably shown more potential. A retrospective analysis of the abituzumab clinical trial data, for example, suggested that the drug had some clinical benefit in patients with higher expression of αvβ6 (ref.106). Therefore, identifying patients with higher target expression levels might improve success, and although this seems be a logical approach for any target class, it has rarely been employed for integrins. This is probably because, in practice, it is difficult to measure expression levels in the clinic without negatively affecting patient recruitment and the time needed to complete studies, and because the ultimate market for the drug will be smaller if the label granted is narrowed to only those patients who express high levels of integrins.

As with any drug discovery area, some potentially interesting compounds have not progressed because of strategic decisions. Even if a compound has demonstrated some potential and there is a clear next step, the programme may no longer be attractive. Projects can also be placed on hold or terminated if a company refocuses their goals, as when AstraZeneca shelved their dual αvβ6/αvβ8 monoclonal antibody (RAD264)178, or loses confidence in a target because of negative clinical results or safety concerns from competitor trials.

A further generic challenge is the translation of preclinical in vivo pharmacodynamic studies to the clinic. These studies are still of value for investigating pharmacokinetic and pharmacodynamic relationships in a whole-body system, but their capacity to accurately represent human biology, especially human diseases, is often questionable and the data generated can therefore become supportive instead of essential179. Even safety studies in animal models of the disease are not ideal because the models themselves are not necessarily translatable to the human disease and the animals can be highly compromised. These limitations to preclinical pharmacodynamic and safety studies can result in on-target safety issues in the clinic, such as those observed with the α4 inhibitors and the αvβ6 monoclonal antibody BG00011, which can be difficult to predict or discover early in the process. Studies in diseased human tissue, especially in fibrosis, may therefore further aid validation and translation as more relevant disease systems and surrogates are developed71,180,181.

There are also class-specific challenges to therapeutically targeting integrins. The lack of selective drug-like tools for this target class, especially because the same subunit is part of multiple different integrins and there are many subunits within integrin subfamilies, has made interpretation and therefore translation of preclinical data difficult. Similarly, measuring expression levels of integrins within these subfamilies is difficult if heterodimer-specific antibodies do not exist. This makes correlating expression levels of the key integrins in the organ of interest in in vivo models with levels in diseased human tissue a challenge, which likely contributes to the failures in this field. Developing small molecules with high affinity and selectivity for an integrin, while maintaining the required physicochemical properties and optimal pharmacokinetic profile for oral dosing, has also been difficult. Even circumventing this with large molecules, such as biologics, is not always successful.

New target validation tools for drug discovery using technologies based on genomics, transcriptomics, proteomics and metabolomics will allow more detailed and global analyses of disease cohorts. For example, genome-wide association studies (GWAS) can identify genetic variations that occur more frequently in people with a disease than in those without. Targets with genetic associations are estimated to be twice as likely to succeed in clinical development182. Although there are few associations between diseases and changes in integrin-related genes, there is an indirect association between αvβ6 and IPF. A variation of AKAP13 was identified as a potential genetic risk factor in IPF183, and AKAP13 mediates activation of TGFβ downstream of αvβ6 (ref.184). Although these tools have yet to be fully applied to integrin target validation, they will undoubtedly be an important part of this process in the future.

Although integrins fell out of favour as targets for drug development, there has been renewed interest and investment in the integrins as drug targets, particularly for fibrotic diseases. To be successful, current integrin programmes should note the key lessons from the many integrin-targeted trials that have already been completed. Future integrin-targeted programmes should focus on the development of molecules that enable hypothesized mechanisms to be fully tested; robust target validation; clinical studies that measure target engagement; and translational biomarkers to measure clinical efficacy. With these changes, indications including fibrosis and cancer, which were previously targeted with little success, may be re-evaluated with improved therapeutic agents.