Protein Medicines That Biology Can’t Get Its Hands On

Experimental data from D-amino-acid protein therapeutics reveal protease stability, immune stealth, and enhanced tissue access.

Protein therapeutics inherit both the power and limitations of natural biology. They can bind their targets with extraordinary precision, but once inside the body, they are subject to the same degradation, immune surveillance, and tissue barriers as any other protein.

For decades, scientists have overcome some of these constraints through engineering and optimization. But it’s worth asking whether incremental change is the only option. What if some of the most persistent limitations of biologics could be addressed by rethinking their fundamental building blocks?

Beyond Natural Biology

Many successful biologics incorporate non-canonical amino acids or chemical conjugates to improve stability and extend half-life. These approaches extend what is possible within natural protein biology by making modifications along an otherwise L-amino-acid scaffold.

Proteins composed entirely of D-amino acids are a fundamentally new modality. Rather than modifying the scaffold, it’s inverted completely.

At the chemical level, D- and L-amino acids are mirror images of one another, and over millennia, biology has evolved mechanisms that are only compatible with one. Enzymes, proteases, and immune receptors have evolved to interact with L-amino-acid proteins — much like a right-handed glove is shaped for a right hand.

D-proteins, alternatively, are the wrong hand for biology’s glove. They occupy an orthogonal biological space and therefore they are unaffected by the biological machinery governing degradation and recognition. It’s for this same reason that they are also impossible to discover using biological screening techniques that rely on organisms such as yeast to make millions of proteins and test them for binding.

Flagship-founded Abiologics, an “abiological” AI company, designs and builds protein therapeutics within this space, turning what was once science fiction into science. Because existing protein design algorithms are trained on natural proteins and recombinant expression systems cannot produce D-proteins, accessing this biology required rebuilding AI protein design from first principles, and developing robotic systems to chemically synthesize the in silico-generated prototype designs.

Abiologics has developed an integrated platform that combines:

• the first generative AI model capable of designing synthetic D-protein therapeutics (Synteins™) de novo

• the only automated chemical synthesis and wet lab screening process capable of testing hundreds of candidate Synteins™ every week

• a proprietary experimental dataset of D-proteins interacting with the L-world (D-to-L), continually learning and optimizing the rules of D-to-L protein design.

Together, these capabilities make it possible to specify, synthesize, and test functional protein therapeutics composed entirely of D-amino acids that interact with L-biology. To be clear: these molecules are not simply mirror images of existing proteins. Under most circumstances, a simple mirror of an L-protein binder, for example, would not engage its natural target.

Instead, the resulting Synteins™, which include binders, inhibitors, and agonists, are designed from scratch rather than adapted from natural templates. This is where a generative design platform that operates beyond the L-world is essential to creating new molecules capable of productive interaction.

Synteins™ maintain the same potential for therapeutic efficacy as standard protein therapeutics but, because of their chirality, possess exceptional pharmacological properties that are difficult, or impossible, for most therapeutics to achieve even through complex optimization. This includes protease stability (Figure 1), immune stealth (Figure 2), and enhanced tissue access (Figure 3).

Intrinsic Stability

Protein therapeutics face intense stability challenges inside the human body. Their ability to evade proteases long enough to exert their therapeutic effect is determined by a combination of their sequence, their structural features, and the pathways that neutralize them within the body.

To evaluate if Synteins™ operate differently than natural proteins, sequence-matched L-proteins and Synteins™ were tested in models of harsh environments where proteases are abundant, and L-proteins are quickly neutralized.

In rodent serum, L-proteins were degraded within hours, while Synteins™ remained intact for at least seven days (Fig. 1A). In simulated intestinal fluid, L-proteins were rapidly degraded, so much so, that they were undetectable after only two minutes, whereas Synteins™ remained stable for at least five days (Fig. 1B).

Figure 1. Protease stability. Synteins™ and sequence-matched L-proteins were spiked into rodent serum (A) or simulated intestinal fluid (B). Intact peptide was quantified using mass spectrometry at various time points.

For D-protein therapeutics, resistance to degradation is directly derived from the incompatibility of their backbone with the active sites of protease enzymes. Stability under harsh conditions, such as in the blood, gut, liver, and other protease-rich tissues, becomes a fundamental property.

This intrinsic durability expands the range of delivery routes that may be feasible for Synteins™, including oral administration.

Reduced Immune Recognition

Immune recognition remains a challenge for protein therapeutics, particularly for drugs that are repeatedly dosed, where antibody responses can undermine both safety and long-term efficacy.

To examine how Synteins™ are perceived by the immune system, they were evaluated side by side with sequence-matched L-proteins in a repeat-dosing immunization paradigm.

In mice, repeat administration of the L-protein with a strong adjuvant elicited a robust IgG antibody response, while no measurable IgG response was detected following administration of the Syntein™ (Fig. 2).

Figure 2. Immunogenicity. BALB/c mice received either a Syntein™ or a sequence-matched L-protein in a prime-boost-boost regimen using Complete Freund’s adjuvant. Total IgG levels were quantified in mouse serum by ELISA.

Immunogenicity is rarely binary, so the near absence of a detectable antibody response despite the identical protein sequences is notable and explained by their chirality. Proteins composed entirely of D-amino acids fall outside many immune recognition pathways that have evolved to engage with fragments of natural L-amino acid proteins.

Here again, reduced immune recognition appears to be an advantage afforded by designing therapeutics with D-amino acid building blocks, something that could not be achieved by modification of L-proteins alone. If borne out across additional targets and dosing regimens, this property could support improved safety profiles and sustained efficacy and tolerability for repeated or chronic administration of protein medicines.

Deep Tumor Penetration

Even when protein therapeutics bind their targets with high affinity, reaching those targets in meaningful concentrations within the right areas of complex tissues remains a major challenge. Size, fragility, and vascular and fibrotic barriers often limit tissue access, particularly in solid tumors.

To examine whether D-protein therapeutics have advantaged tissue distribution in vivo, fluorescently labeled Synteins™ designed to bind to a tumor antigen, and a monoclonal antibody targeting the same antigen were compared in a tumor xenograft mouse model.

Synteins™ localized to tumors within 1.5 hours of injection. Near-infrared imaging revealed Synteins™ distributed throughout the tumor mass (Fig. 3A, top row). In stark contrast, the antibody remained largely confined to perivascular regions (Fig. 3B, bottom row).

Figure 3. Tumor penetration. Either a fluorescently labeled Syntein™ (A) or monoclonal antibody (B) binding the same cancer target was injected into tumor-xenograft–bearing mice. Whole-body fluorescence imaging was performed from 20 minutes to 8 days post-injection. Tumor tissue was harvested, fixed, sectioned, and stained with DAPI. Gross tumor section scale bar = 800 µm; magnified tumor section scale bar = 50 µm.

De novo design of Synteins™ enables the creation of simple, small proteins that remain durable in vivo. Penetration into dense tissues, such as solid tumors, becomes a design choice rather than a roadblock.

While these data are specific to a tumor model, they illustrate a broader point: When stability and size are tunable design parameters rather than modifications for constraints, tissue access becomes a controllable variable rather than a fixed limitation.

Structural Confirmation

The functional data show what a molecule does. Structural validation tells you whether it’s doing it the way it was designed. For any de novo designed protein therapeutic, computational predictions are only meaningful if they correspond to experimental reality. Structural validation therefore represents a critical benchmark for determining whether designed molecules look and bind as predicted.

To confirm the accuracy of our approach, we compared the predicted Syntein:target complex to the experimental ground truth via X-ray crystallography. The resulting structure demonstrated sub-angstrom agreement between the computationally designed interaction and the experimentally determined crystal structure (Fig. 4), providing direct validation of the platform’s ability to accurately design D-protein therapeutics de novo and synthesize them chemically.

Figure 4. Structural confirmation. Video of a de novo designed all-D Syntein™ (pink) engaging with its L-protein target (white). Confirmed Syntein™ crystal structure (purple mesh) has 0.89 Å RMSD interface accuracy and 1.00 Å RMSD whole-protein accuracy.

This level of structural fidelity is particularly notable because the interaction occurs across biological chirality, in which a synthetic D-protein engages a natural L-protein target. Existing protein design systems and algorithms were not built for this.

To our knowledge, this represents the first experimentally confirmed structure of a de novo AI-designed protein composed entirely of D-amino acids binding to a natural biological target.

Where to Next?

Biologics have already transformed medicine, enabling therapies defined by specificity, potency, and entirely new mechanisms of action. The data presented here build on that legacy, while also challenging long-standing assumptions about the trade-offs inherent to protein therapeutics.

These results represent early experimental data generated in specific models and contexts. But they suggest that some of the most persistent limitations of biologics, including instability, immune recognition, and restricted tissue access, may not be biological necessities.

If that’s true, the next chapter of medicine may come from pushing the boundaries of AI to unlock a completely untouched design space, preserving what has made biologics powerful, while unleashing supranatural properties. In doing so, we can open a portal into an equally infinite but completely different abiological protein space that will complement and expand our toolkit against the world’s most challenging diseases.

For more information, visit www.abiologics.com.