The first step of testing our project was asking whether our split-GAL4 logic controller can cleanly separate healthy-like (OFF) from disease-like (ON) conditions when driving a pseudotherapeutic (NanoLuc) payload. In HEK293 cells, the three sensors (stress-responsive promoters driving Gal4DBD, VP16AD, Cre respectively) were stimulated to model disease context, and the gate output was read on a luminometer. A first experiment set out to see whether all the plasmids functioned appropriately. This was done by (1) using a cell culture media supplemented with fetal calf serum (FCS) which is likley to stimulate the targeted cell signaling pathways and (2) providing the three sensors independent of extracellular environment (F1).

Fig. F1
HEK293T cells were transfected with P1, P2, P3, P4 (S), VP16Gal4, Cre, P4 (C-poz), or pcDNA3.1 (C-neg) using FCS(10%) complemented cell culture media.
The second experiment laid out to test the specificity of the gate. The “inputs” were varied in two ways: (1) genetically (only transfection of some of the sensors in extracellular environment dependent (II) or independent manner (I)); (2) by stress (where all wells were transfected with all sensors and output plasmid, and the treatment was applied or not (III)). In the first approach, the gate produced a strong ON state while either incomplete input remained close to baseline, yielding an ON/OFF separation consistent with low leak (≤5–10% of ON). Assay quality, computed plate-wise on dedicated control wells, gave Z′ factors up to 0.67 with a typical range ≥0.50, indicating excellent dynamic separation and reproducibility. This established that our safety layer can reliably restrict therapeutic expression to the intended context while preserving a broad signal window for downstream readouts (F2a ). In the second approach, cells transfected with all detection components were either treated (with EGF for JAK-STAT pathway; Menadione for oxidative stress (ROS); tunicamycin for ER stress (CHOP)) or left untreated in low FCS (2%) complemented cell culture media. The modest signal-to-noise ratio reflects residual leakage within certain sensors, highlighting opportunities for optimization of this proof-of-concept design. (F2b ).

Fig. F2a:
Z′ factor per plate with 95% CI and acceptance line at 0.5.

Fig. F2b:
(I) Specificity of stress-independent gate, obtained by VP16Gal4 and/or Cre expressing plasmids via transfection; (II) Specificity of stress-dependent gate, obtained by selective transfection of P1-P3 together with P4;
(III) Logic gate response under stress conditions (EGF 2.56nM (JAK-STAT pathway, 5h), 10µM Menadione (Osidative Stress, 3h), 5µg/mL Tunicamycin (ER stress, 5h)), all plasmids transfected. Data
was normalized to average of the respective off states to clearly show fold.
Design and biophysical plausibility (dry lab). We designed TRI-LYTAC as a three-arm chimera: AT8 scFv (binds phospho-tau epitope S202/T205), HAI/TfR-binding peptide(blood–brain barrier
shuttle), and IGF-2 dimer (CI-M6PR engagement for lysosomal routing) linked by (G4S)₃ spacers. AlphaFold produced models in which each domain retained its canonical fold and the linkers remained solvent-exposed
without creating persistent steric clashes. Docking analyses supported both recognition events: IGF-2 binding poses on CI-M6PR consistent with sub-100 nM interaction affinity and an AT8-epitope engagement that positioned CDR loops as expected
for selective binding. (Figures F3–F4). TRI-LYTAC Domain and AlphaFold Ribbon (AT8 scFv, HAI/TfR, IGF-2 dimer; (G4S)₃ linkers). Fig. F3:
(A) Schematic comparison of the three TRI-LYTAC constructs (G1–G3, controls G-drop-TfR, G-drop-Lys, G-drop-Bind) illustrating modular differences in the tau-targeting domain while maintaining the same receptor-targeting and secretion framework.
(1) G1 (AT8 scFv TRI-LYTAC): Incorporates the AT8 single-chain variable fragment recognizing the phosphorylated tau AT8 epitope (pSer²⁰²/pThr²⁰⁵). (2) G2 (Tau-VHH TRI-LYTAC): Employs a nanobody (VHH) specific for misfolded or aggregated tau conformers,
offering broader recognition of extracellular tau seeds. (3) G3 (W-Tau control): Uses a non-pathological wild-type tau fragment as a negative-control binder to assess non-specific receptor activation. Each variant includes an N-terminal Ig κ (IGK)
signal peptide for secretion, flexible (G₄S)₃ linkers between domains, a HAIYPRH transferrin receptor (TfR)-binding motif for blood–brain barrier transcytosis, an IGF-2(F19L) dimertargeting the Insulin-like Growth Factor 2 Receptor (IGF-2R / CI-M6PR)
for lysosomal routing, and a C-terminal HiBiT tag for luminescent quantification. (B) AlphaFold2 (ColabFold) predicted tertiary structures of the TRI-LYTAC variants, colored by domain (blue = tau-binding module (AT8), grey = linkers, green = HAI/TfR motif,
pink = IGF-2 dimer, orange=HiBiT tag). All three models adopt semi-extended configurations with minimal steric interference, preserving accessibility of both the tau-binding and receptor-binding regions. The consistent fold across variants supports the
feasibility of modular replacement of the tau-recognition arm without compromising overall structural stability.
Docking using HADDOCK, Showing TRI-LAC with TfR and IGF-2R Fig. F4:
(A–B) Docking of TRI-LYTAC to the human transferrin receptor (TfR) using HADDOCK 2.4. The model reveals the HAIYPRH motif positioned at the receptor binding pocket, forming complementary electrostatic and van der Waals contacts while leaving the antibody and IGF-2 domains solvent-exposed. Cluster 1 (size = 43, Z = −2.0) exhibited a HADDOCK score of −79.3 ± 2.2, E_vdW = −46.5 ± 5.9 kcal mol⁻¹, E_elec = −140.7 ± 20.4 kcal mol⁻¹, E_desolv = −13.2 ± 1.5 kcal mol⁻¹, and a buried surface area ≈ 1,397 ± 76 Ų, supporting a stable, well-packed interface. Panel B shows the close-up of the TfR binding pocket, with hydrogen bonds and hydrophobic contacts highlighted. (C–D) Docking of the TRI-LYTAC IGF-2(F19L) dimer to the Insulin-like Growth Factor 2 Receptor (IGF-2R / CI-M6PR) domain 11. The complex reached the semi-flexible (it1) refinement stage; although water refinement failed, the pre-water model aligns accurately with the canonical IGF-2 binding pocket from the reference crystal structure (PDB 2V5P). Panel D provides a close-up view of the predicted interface. Together, these docking models validate the biophysical plausibility of TRI-LYTAC receptor engagement through both its TfR- and IGF-2R-binding modules.
TRI-LYTAC internalization (wet lab).The first step was to produce TRI-LYTAC conditioned media (CM) by transfecting HEK293 cells with the corresponding mammalian expression designed plasmids. As shown in figure F5a, all constructs led to extracellular secretion of HiBit tagged TRI-LYTACs. To verify the first step of the intended mechanism—target-bound uptake into the endo-lysosomal pathway - we measured over time the HiBiT luminescence in the extracellular media of BV2 microglia exposed to similar concentrations of TRI-LYTACs (approximated by RLUs). Initially, we performed a long time course experiment up to 120 minutes (F5c). We noticed a significant decrease of the signal in the first 20 minutes followed by a plateau or a partial restoration of the signal for the different LYTACs (F5b). This data might be in line with our strategy of TRI-LYTAC recycling but more data are needed to prove this hypothesis. Fig. F5a
Quantification of LYTAC secretion (relative luminescence units, RLU) measured at 24 and 72 hours for each construct. Values represent the mean ± standard deviation of three replicates. Secretion increased over time in all variants, with G2, G3, and G-drop-bind exhibiting exceptionally high luminescence signals at 72 h, suggesting enhanced expression or stability of the secreted construct.
Fig. F5b
Short-term internalization dynamics of TRI-LYTAC constructs (0–21 min). Each construct (G1–G3, G-drop-TfR, G-drop-lys, G-drop-bind) is shown individually and as a normalized overlay. Most variants internalized rapidly within 6–12 min, with G-drop-bind
showing the highest uptake (45%) and G2 remaining the most stable (94.9%). This timeframe captures the early endocytic phase dominated by internalization, with minimal recycling observed.
Fig. F5c
Long-term trafficking and recycling kinetics of TRI-LYTAC constructs over a 120-minute period. Each construct’s individual curve and normalized comparison plot reveal a biphasic trend: initial internalization followed by progressive
recovery, indicating recycling or re-expression at the cell surface. G-drop-lys exhibited the deepest internalization (29.6% at 15 min) and strongest recycling (46.8% recovery by 120 min), while G1 showed a comparable 45.4% recovery.
Overall, F5b illustrates the complete trafficking cycle of TRI-LYTACs, highlighting receptor-specific differences between lysosomal, binding, and TfR-only constructs.
TRI-LYTAC







