First small-molecule modulators of SOX2 — found inside living cells

SOX2 (SRY-box transcription factor 2) is a pioneer transcription factor and a master regulator of stem cell identity and differentiation. It is overexpressed in multiple cancers, including glioblastoma and squamous cell lung carcinoma, where it sustains the self-renewal programs that drive tumor growth and treatment resistance. That biology makes it a high-interest oncology target.

It is also a textbook hard-to-drug target. SOX2 has no defined binding pockets. Large parts of it are intrinsically disordered, and its function is executed through a DNA-binding interface and protein–protein interactions rather than an active site. A peptide modulator, P42, has been reported, but at the outset of this work no small-molecule modulator of SOX2 had been described.

There is a further, less obvious obstacle. Any screening campaign against a DNA-binding protein using a DNA-encoded library must contend with the target’s affinity for DNA itself. The positively charged interface that recognizes the SOX2 consensus motif will also engage the library’s coding DNA, generating enrichment that looks exactly like a hit and is nothing of the kind. Distinguishing compounds that bind SOX2 from compounds whose barcodes bind SOX2 is a prerequisite, not a downstream concern.

Vipergen screened SOX2 using DELs in Cells, built on our cellular Binder Trap Enrichment (cBTE) technology. cBTE is the only DEL screening format that operates inside a living cell. The protein of interest is expressed as a Prey fusion protein in Xenopus laevis oocytes; the DNA-encoded library and a Bait-DNA are then microinjected into that same cell. The affinity of the Prey protein for the small Bait molecule is used to label the target with DNA in vivo, and binding events are captured by emulsification followed by DNA ligation — so compound–target complexes are encoded as a ligated DNA molecule that can be identified by sequencing.

For a target like SOX2 this matters in three ways:

• No purified protein is required. The target is expressed in the cells as an intrinsic part of the project, which removes the purification and stability problems that intrinsically disordered proteins routinely present. Around 95% of proteins tested to date are amenable to expression in this system.
• Physiological screening conditions. Binding is measured in a crowded intracellular environment at native ionic strength, with the target’s folding, post-translational modifications and cofactors intact. 
• Competition can be run in-cell. Because the target is screened inside an intact cell, competitors such as a consensus DNA duplex, or a resynthesized compound can simply be co-injected and the whole hit landscape re-read in their presence. This is the feature that made the SOX2 campaign tractable.

The SOX2 screens produced a weak but reproducible signal, strongest with the C-terminal Prey construct. Across the six experiments, 25 hits were identified and Tanimoto clustering resolved them into three chemical series: C1, C2 and C3.

Adding consensus DNA changed the picture decisively. In its presence, C1 was maintained while C2 and C3 were repressed. Adding a control DNA sequence produced no major change in any cluster. The interpretation follows directly: C2 and C3 were driven by SOX2’s engagement with library DNA and were competed away by an authentic, sequence-specific DNA duplex; C1 survived that competition and therefore represents compounds binding SOX2 on their own merits. Four compounds (22A and 24A from C1, 27A from C2 and 28A from C3) were resynthesized off-DNA.

SOX2 has everything that makes a target intractable: no pocket, extensive disorder, a function that depends on a protein–DNA interface, and a DNA-binding activity that generates false positives in a DEL screen. DELs in Cells returned genuine chemistry against it, and the platform’s own competition formats — consensus DNA in Screening I, resynthesized compounds in Screening II — did the triage in-cell, before recombinant protein entered the workflow. The compounds that survived that filter then reproduced their behavior in an orthogonal biochemical assay.

The mechanism is worth stating plainly, because it is an unusual one. Where transcription factors have been drugged successfully, it has mostly been by disrupting coactivator or corepressor recruitment, by targeted degradation, or – for nuclear receptors – by exploiting a genuine ligand-binding domain. SOX2 offers none of those handles readily. It has no ligand-binding domain, and both PPI disruption and degradation depend on first having a molecule that binds the protein at all. Altering the sequence selectivity of a transcription factor is a rare mechanism rather than the field’s default, so 22A and 24A are intriguing molecules with an interesting perspective. More investigations are currently ongoing at Vipergen to resolve the mechanism of binding.

What the data supports at this stage: The compounds engage SOX2 inside living cells and alter its DNA sequence specificity. What remains open is whether re-tuned selectivity can be converted into a therapeutically useful effect on the SOX2 transcriptional program. These are the first small molecules that make that question answerable.

That said, redirecting a transcription factor’s specificity may prove to be a difficult pharmacological principle to build on: reprogramming which sequences a TF prefers is a subtle, context-dependent effect, and the therapeutic window is far from obvious. Other uses of the same binder — a PROTAC starting point, for example — may ultimately prove more tractable.