In recent years, peptide–drug conjugates (PDCs) have advanced from theoretical interest to legitimate therapeutic agents. This was achieved by conjugating the pharmacological efficacy of cytotoxics to the concise addresses offered by cell-penetrating peptides (CPPs). The most prominent feature of these hybrids is a molecular weight that allows them to pass through size-restricted anatomical barriers that are largely impenetrable by monoclonal antibodies (mAbs) but is still sufficiently high to prevent their rapid renal clearance of high potency payloads. As a result, modern medicinal chemistry no longer considers the peptide moiety as merely a delivery vehicle but rather as a versatile carrier whose binding avidity, hydrophobic character and even immunogenicity can be finely tuned, in parallel to the linker dynamics. Early phase clinical reports have confirmed objective responses across a range of histologically different malignancies suggesting that the platform is nearing a point of ubiquitousness where tumor classification is superseded by surface-ligand expression. However, those studies also highlight that tumor microenvironments adapt survival capabilities more rapidly than static conjugates can be reprogrammed, mandating the next generation of designs to include real-time sensing, combinatorial logic and self-immolative safety locks. If these challenges can be overcome without exacerbating cost or regulatory complexity, peptide-driven chemotherapy may soon graduate from last-resort salvage therapy to upfront precision medicine.
Fig. 1. Schematic structure of a peptide–drug conjugate (PDC).1,5
The challenge in cancer treatment remains the most robust in medical therapeutics. Cancer utilizes cellular plasticity at the genetic, epigenetic, metabolic and anatomical level. While cytotoxic agents remain the basis of current systemic therapies, their potency is already at a ceiling; dose escalation faces constraints of bone marrow, neural and epithelial toxicity and dose reduction often results in selection. First generation small molecules were initially heralded to be mutation-specific, however the duration of this response is usually short-lived; reconstitution of the oncogenic circuit via feedback, bypass signaling and trans-differentiation reconstitute the effector pathway, often in a form that is unresponsive to the initial inhibitor. Antibody–drug conjugates (ADCs) had tried to address the former by imposing spatial specificity, but their size impedes diffusion to fibrotic or anoxic cores, and Fc-mediated recycling may also inadvertently sensitize normal tissues. Immune checkpoint inhibitors have shifted the paradigm of some cancers that are negative for oncogenes, but cases of hyper-progression or autoimmune side effects are still stochastic. As such, there is an unmet need for a modular system with the properties of high-affinity and deep penetration, yet rapid systemic clearance of free drug and flexibility to interchange payloads to target the evolving resistance mechanisms. PDCs have emerged at this junction of needs, with synthetic feasibility and an immunological profile that is sufficiently tolerable to allow repeated administration. The goal of developing PDCs should not be to out-compete existing modalities, but rather to act as an adaptable platform that chemotherapy, immunotherapy and radiotherapy could be re-designed around to act synergistically with the tumor's vulnerabilities.
The major risk inherent in any ligand-based approach is the ability of the tumor to edit the very antigen that was selected for targeting. The malignant clone doesn't need to incur the genetic insult of a catastrophic deletion; more subtle mechanisms, such as alternative splicing, glycosylation masking or promoter methylation, can lower the surface density below the limit needed for the formation of a productive conjugate. Monitoring over time demonstrates that this drift can be recognized in a single window between cycles, suggesting that a library of peptides may be obsolete before the lots expire. One potential solution is a polyvalent scaffold that can simultaneously bind to two epitopes that are positioned on non-overlapping extracellular loops, such that the simultaneous loss of both targets might incur a fitness cost too great for clonal persistence. A second strategy uses allosteric peptides whose affinity is actually increased when the antigen is in a conformation associated with ligand deprivation, and thereby turns tumor adaptation into a pharmacodynamic benefit. Regardless of the specific approach, the broad point is that antigenic escape should be considered an engineering variable rather than a clinical surprise, and needs longitudinal liquid biopsies and machine-learning algorithms that can predict drift trajectories before they become apparent as radiological progression.
Cytotoxic warhead choice is no longer simply a matter of tubulin versus DNA damage; it is now understood that the mechanism of action of the payload can be married with intrinsic tumor susceptibilities and other treatments given concurrently. As an illustration, transient chromatin looseners prime the target cell for immune-mediated clearance, while drugs that inhibit oxidative phosphorylation will enhance the redox burden that has already been laid down by radiotherapy. Furthermore, the linker can play a role in temporal synchronization of treatments; the kinetics of release of the payload should be matched with the pharmacodynamic window of activity of the other treatment modality in question, providing a period of time in which the tumor's defences are down. Resistance can be further delayed by metabolic primers (small molecules that reset the tumor pH or oxygenation status) given prior to the cytotoxic drug, shutting off the gene expression programmes associated with hypoxia which would otherwise attenuate drug killing. Crucially, these payloads are also to be selected for their clearance kinetics; rapid renal clearance of metabolites precludes off-target accumulation, and binding motifs that mediate reversible binding to albumin can be engineered into the peptide backbone to increase retention in the tumor bed. This would lead to a positive feedback loop in which every round of treatment not only kills tumor cells, but modifies the microenvironment in such a way as to support the next round, and resistance would thus be indefinitely forestalled.
The selectivity of a peptide–drug conjugate is conceptually predicated upon a unique and identifiable surface landmark that is either quantitatively or qualitatively elevated only on malignant cells and does not exist (or only in a level that is functionally silent) on their normal counterparts. Target surfaces within the thorax are understood to be typically graded over-expressions/mutations/neoglyco-processes rather than categorical on/off flags; the overall ratio, however, has to be far enough skewed towards binding to favor the malignant state. Target identification is no longer a linear pursuit, fixated on fold-change in transcriptomic assays, and has embraced other rich single-cell proteomic imaging, secretome deconvolution, and extracellular vesicle surveying technologies to generate antigen maps that are annotated with intensity of expression as well as internalization/recycling kinetic rates. The "sweet spot" in this spectrum is an epitope whose occupancy as a function of cell-cycle rate, both spatially and temporally, meets a three-pronged criterion: the first prong is a Kd value that matches the sub-μM concentration exposure in plasma, the second is a lysosomal targeting index that beats the pace of competing efflux pumps, and the third is a biochemical threshold that is large enough to buffer against mutations without collapsing quickly. This threshold is never reached without the co-evolution of both the peptide ligand and linker chemistry, i.e. payload release should not (theoretically) be the rate-limiting factor.
Fig. 2 Schematic outline of targeting a tumor-expressed G protein-coupled receptor for anti-cancer drug delivery with a peptide-drug conjugate. 2,5
Arguably the most druggable class of receptors is the family of integrins, specifically those isoforms that bind the arginine-glycine-aspartic acid (RGD) motif. Integrins are upregulated by malignant pneumocytes to aid in dissociation from the basement membrane during metastasis and can be spatially co-distributed, which creates a higher avidity in the use of multivalent cyclic peptides. Somatostatin receptors have been interrogated in the past using radio-peptides, but are re-emerging as a means to introduce chemotherapeutics as agonist-induced internalization of the receptor creates a delivery conduit into a cathepsin-rich lysosomal pathway that requires no additional cell-penetrating sequence. The zinc-dependent aminopeptidase N ectoenzyme that can also cleave components of the extracellular matrix, counterintuitively creates a cryptic pocket when bound to ligands, with the ability to bind longer peptide backbones; binding to this pocket has shown to increase residence time, resulting in higher payload delivery. A more recent target is the presentation of mutant p53 on the cell surface that occurs via complexation with chaperones in the endoplasmic reticulum, and has been shown to be a neo-epitope not present on the wild-type protein, thus allowing allele-specific targeting without the associated gene-editing off-target effects. As biomarker validation now can move beyond biopsy-based endpoints, identification of these same receptors on circulating extracellular vesicles provides an opportunity to be serially sampled and could be used to dynamically recalibrate dosing of these conjugates to meet real-time antigen expression. Additionally, receptor co-expression is now being modeled as a Boolean network to rationally add bivalent ligand chimeras that can bind to complementary epitopes on the cell surface, diluting the selective pressure that drives antigenic escape.
Contemporary sequences also break from the linear lock and key metaphor, instead using conformationally restricted macrocycles which are better modelled as crescent wrenches. Non-natural amino acids (with orthogonal side-chain reactivity) can be used to staple α-helical turns to create peptide barrels which have hydrophobic sides which fit into receptor grooves and exposed charged residues to keep the molecule water soluble. This two-surface strategy may help to avoid off-target adhesion to serum albumin, which has in the past led to sequestration in pharmacologically inert depots. In a similar vein mirror image phage display has been used to screen for D ligands with the hope that they may be resistant to proteolytic shaving by lung-elastases, to increase systemic half-life without using PEGylation, which has gained notoriety due to potential immunogenicity (acceleration of clearance via anti PEG antibodies). As a complementary approach to variability in receptor conformation, "evolvable" peptide libraries are being prepared with photosensitive protecting groups (photocages) on certain residues. A short exposure to near infrared light (delivered locally using a bronchoscope) releases a thiol that is able to undergo disulfide exchange allowing local reconfiguration of a ligand's binding affinity without requiring a new systemic dose. Deep learning methods have been trained on co-crystal structure data to predict optimal residue spacing between pharmacophore elements, minimizing the number of iterations during directed evolution and design by ensuring selectivity is built into the primary sequence rather than added later in post-synthetic modifications.
Peptide–drug conjugates targeted to a single antigen have started to yield early clinical experiences across histologically distinct cancers that may inform future approaches beyond proof of concept. Common to these case studies is an unmet clinical need, a specific surface antigen, a highly stable lysosome-targeted linker and a payload that can exploit the tumor's inherent synthetic vulnerability. The clinical relevance of these case studies is less likely to be blockbuster agents but to instead inform broader efforts by identifying challenges of antigen loss, stromal barriers, and immunogenic excipients in a compassionate use setting. Information gathered from these small patient populations is beginning to be used to inform second-generation designs, including those that leverage multi-valent binding, imaging reporters and payloads to delay resistance.
In one example, a cyclic peptide targeting a neural-crest restricted receptor was coupled to a transcriptional CDK inhibitor and given to a patient with an ultra-rare fusion-driven sarcoma. The target, which is epigenetically silenced in adult mesenchyme, was re-expressed as part of the oncogenic transcriptional programme, creating a lineage-specific bull's-eye. Pharmacodynamic biopsies showed near-complete receptor occupancy within hours, with concomitant down-regulation of the fusion oncoprotein itself (a downstream consequence of CDK inhibition). Resistance was selected through amplification of a phosphatase that reactivated downstream signalling, and the mechanistic understanding of this allowed immediate re-engineering of the payload to a parallel epigenetic pathway. In short, even vanishingly small patient populations can produce generalizable design principles when the antigen–payload axis is defined by the molecular logic of the disease.
In a separate exploratory cohort, high-grade pulmonary neuro-endocrine tumors were identified that had preserved high surface expression of a bombesin-family receptor despite having downregulated other more classical endocrine markers. In this setting, the conjugate was armed with a topoisomerase poison that was only activated by the mildly acidic pH found in the interior of neuro-endocrine secretory vesicles, creating an additional, intracellular-based gatekeeper mechanism that protected adjacent receptor-negative cells. durable metabolic responses were observed, though eventual escape was associated with compensatory overexpression of a different topoisomerase isoform. This case illustrated the potential benefit of combining receptor specificity with organelle-conditional payload activation, an approach now being applied to other secretory malignancies.
This dream of tumor targeting on a grand scale is often dashed in the face of the complex fractal nature of human cancer. Pan-cancer antigens are almost invariably also required for fundamental housekeeping functions, so even a moderate systemic disruption can lead to on-target, off-tumor toxicities that swamp any therapeutic benefit. Furthermore, genetic divergence of primary and metastatic lesions means that a ligand optimized to a given tissue context might be faced with a significantly altered antigen conformation or glycosylation pattern that decreases its affinity by many fold. Attempts to address this by affinity maturation can often exacerbate cross-reactivity with normal tissues, as can a broad-spectrum payload. For these reasons, pan-cancer PDCs are at risk of becoming a pharmacological sledgehammer that misses the basic therapeutic requirement of selective vulnerability.
Spatially, antigen density can vary at different coordinates within the same patient, forming so-called "islands" of high-expression tumor cells in a "sea" of low antigen-expressing cells. A pan-cancer conjugate would therefore not only kill the susceptible subpopulation but also provide the resistant subpopulation a free pass to grow unchecked. The ensuing clonal bottleneck would not only obviate the initial benefit but also select for the expansion of lineages that are capable of seeding antigen-loss metastases, in effect turning a cytoreductive strategy into a growth-promoting one. Attempts to avert this selection by using a polyvalent ligand, on the other hand, run up against manufacturing inconsistency, where small differences in stoichiometry translate into large pharmacokinetic differences in the heterogeneous tumor microenvironment.
Typically, such promiscuous antigens do not completely cease their physiological functions, but are instead sequestered away to hidden anatomical sites such as stem-cell niches or regenerating epithelial tissues. A conjugate intended to target all such antigen-expressing cells is therefore likely to inhibit tissue regeneration, leading to clinical sequelae such as cytopenias, mucosal shedding or neuro-inflammatory demyelination. Safety margins are therefore minimal since the therapeutic index is set not by maximal tolerated doses, but by the lowest doses that still have the capacity to inhibit such vital physiological processes. Attempts to increase this safety margin by using pro-drugs or transient antigen-blocking have often created pharmacokinetic lags that allow tumor re-growth to outpace the initial cytotoxic burst.
The shift in personalized oncology from fixed snapshots of the cancer genome to living therapeutics that evolve in parallel with clonal selection is well underway. Peptide–drug conjugates are a natural fit for such individualization, as swapping or tuning either the targeting ligand or the payload is a relatively straightforward process that does not require a revalidation of the entire PDC manufacturing process. Antigen drift can be monitored using real-time liquid biopsies, while machine-learning tools can be used to predict linker stability under an individualized proteomic landscape. Such tools, when combined, could allow PDCs to be the first truly "living" therapeutic, whose molecular identity is updated at a faster pace than the tumor can evade it, and precision medicine will no longer be a once-and-done compass but a continuously recalibrated GPS.
Patient-specific mutations produce peptide fragments that are absent from the germline repertoire, creating an immutable antigenic bull's-eye. Immunizing phage libraries against such neo-epitopes presented on autologous dendritic cells can identify high-affinity peptides whose cognate sequences are known to be tumour-restricted. Coupled to a payload that requires intracellular processing, only cells that present the exact neo-peptide will be poisoned, synthetically mimicking adoptive T-cell specificity without the logistical challenges of cellular manufacture. Antigen-loss resistance would require driver mutation loss, a manoeuvre that usually compromises tumor fitness.
The addition of mid-course biopsies with rapid synthesis has streamlined the adaptive trial design so that a peptide that is no longer effective can be replaced in a matter of weeks, not months. Compact solid-phase synthesizers now in academic hospital pharmacies can produce GMP-compliant batches of the drug tailored to the specific profile of receptors that need to be targeted next, and the regulatory bodies are acknowledging that a reduced package of toxicology data may be sufficient if the structural modifications are below a certain similarity index. These developments give PDC therapy the properties of a dialogue between physician and tumor, where every cycle of treatment refines the question asked of the cancer and delays the day when all therapeutic choices have been exhausted.
Our platform enables the design of tumor-selective PDCs by leveraging biomarker insights, receptor targeting, and innovative payload strategies. We help partners translate tumor biology into effective therapies, supporting development pipelines from discovery through trials. Request a consultation on our tumor-targeting PDC expertise.
References
