Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Ig-like V-type domains of cancer cell surface proteins: PD-L1 crystal structure (PDB:4zqk; hotspots: I54, Y56, V68, M115, Y123), VTCN1 (B7-H4) crystal structure (PDB:4gos; hotspots: L72, L79, Y131), CD276 (B7-H3) AF2 model (hotspots: I66, L75, F123, F129), and Nectin-4 crystal structure (PDB: 4frw; hotspots: A66, L81, F132). b) Histograms summarizing the result of multiple RFdiffusion design runs for different targets. c) Block diagram presenting the evolutionary refinement algorithms described in this work. d) Schematic explaining the introduction of variations into initial AI-minibinder designs by either partial diffusion (option I) or sequence manipulation (option II). e) Boxplot showing pAE interaction scores of initial RFdiffusion designs and GA-refined (partial diffusion) designs. Median pAE scores are indicated and percentages of pAE<= 5 or pAE>5. Wilcoxon test. f) Histogram showing pAE interaction scores of the initial RFdiffusion design runs for Nectin-4. The designs that were used as inputs for the refinement algorithms are highlighted by a red box. g) Histogram showing the refinement results using option I of the evolutionary algorithm. Insert: Cartoon representation of Nectin-4 (cyan) and a set of diverse AI-minibinder designs (green).

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Blocking Assay, Diffusion-based Assay, Sequencing

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Boxplot showing differences in ipTM scores between Nectin-4 minibinder initial designs and option II GA-redefined designs. Median ipTM scores are indicated. Two-sided Wilcoxon test. b-d) Three examples of predicted structures of option II GA-refined Nectin-4 minibinders in complex with Nectin-4. Respective ipTM and pTM scores are indicated. e) Boxplot showing differences in ipTM scores between Nectin-4 minibinder initial designs and option II GA-redefined designs with implemented pTM binder penalty and with or without target isoelectric point (pI) and hydrophobicity (GRAVY score). Pairwise two-sided Wilcoxon test with Benjamini & Hochberg correction for multiple testing. f-h) Three examples (from group 3) of predicted structures of option II GA-refined Nectin-4 minibinders with pTM penalties. Respective ipTM and pTM scores are indicated.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques:

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Scheme illustrating the experimental screening pipeline using the mammalian cell-surface display for AI-minibinder presentation, fluorescence-activated cell sorting and next-generation sequencing to identify candidates. b) Scatter plots showing enriched minibinders for human, murine, rat and monkey Nectin-4 (colored dots). c) Cartoon showing the experimental set-up of surface presentation of the AI-minibinder on HEK293T cells and detecting the binding to the target protein. Staining of the HA-tag was used to normalize the protein binding to the surface expression of the AI-minibinder. d) Heatmap summarizing the binding to human, murine, rat and monkey Nectin-4 Fc-fusion proteins of 40 individually validated Nectin-4 AI-minibinders using HEK293T cell-surface display. Data are shown as geometric mean fluorescence intensity (gMFI) normalized to the surface expression detected with the HA-tag signal. e) Representative histograms showing flow cytometry analyses of the binding to Nectin-4 Fc-fusion proteins from the different species for 8 AI-minibinder candidates. f) Quantification of the experiment described in e. g) Representative histograms of minibinder 2-17-59 showing selective binding to Nectin-4 Fc-fusion protein from murine, rat and monkey origin (left) and minibinder 4-21-32 demonstrating preferential binding to rat Nectin-4 Fc-fusion protein (right). h) Quantification of the experiment described in g for binder 2-17-59 (left) and 4-21-32 (right). Data are shown as mean ± SD performed in biological triplicates (n = 3). **p<0.01; ***p<0.001; ****p<0.0001; two-way ANOVA with Tukey’s multiple comparisons test.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Fluorescence, FACS, Next-Generation Sequencing, Binding Assay, Staining, Protein Binding, Expressing, Flow Cytometry

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Scatter plot showing the enriched minibinder population for human Nectin-4 (green dots). b) Scatter plots of the enriched minibinder population detected from the screen with murine, rat and monkey Nectin-4 Fc-fusion proteins overlaid into the human Nectin-4 enrichment plot (colored dots).

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques:

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Coomassie-stained SDS-PAGE of purified Nectin-4 AI-minibinder proteins that were overexpressed in E. coli . Left: original designs, right: biotinylated versions of minibinders including an engineered mutant cysteine residue on the ‘backside’ of the AI-minibinder, opposite of the interaction interface. b) Analytical SEC of purified AI-minibinder proteins using a Superdex Increase 75 3.2/300 column. SEC traces of a commercial standard are included as molecular weight reference. c) Coomassie-stained SDS-PAGE of E. coli supernatants after the overexpression of AI-minibinders with (+) and without (−) incubation at 94 °C for 20 min. The boiled samples were centrifuged and the cleared supernatant was loaded onto the gel. d) Nano differential scanning fluorimetry (nanoDSF) analysis showing the thermal stability of the AI-minibinder proteins. The y-axis shows the first derivative of the corresponding melting curve. The two panels show the unfolding (left) and refolding (right) phases of the experiment. e-k) SPR (surface plasmon resonance) sensorgrams showing raw binding data (colored dots) and fitted curves (black) used to determine binding affinities (K D ) of Nectin-4 AI-minibinders. The concentration series of analyte injections are indicated in the legend. l) Structural model of 3-8-16 and 3-24-26 Nectin-4 minibinders (shades of green) in complex with Nectin-4 (cyan) as determined in Chai-1 shown as overlay.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Staining, SDS Page, Purification, Mutagenesis, Residue, Molecular Weight, Over Expression, Incubation, Nano Differential Scanning Fluorimetry, SPR Assay, Binding Assay, Concentration Assay

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Schematic illustration of the assembly of biotinylated Nectin-4 AI-minibinders with fluorophore-conjugated streptavidin as tetravalent quattrobinder. b-c) Representative histograms showing flow cytometry analyses of AF647-conjugated Nectin-4 quattrobinders in comparison to a conventional biotinylated Nectin-4 antibody with a secondary AF647-streptavidin staining on target expressing urothelial cancer cells (RT4) and non-expressing cancer cells (T24) as well as on HT-1376 WT and the corresponding knockout cell line (HT-1376 Nectin-4 KO). d-e) Quantification of the experiment described in b and c. Data are shown as mean ± SD performed in biological triplicates (n = 3). ns – non-significant; *p<0.05; ***p<0.001; ****p<0.0001; two-way ANOVA with Šídák’s multiple comparisons test.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Flow Cytometry, Comparison, Staining, Expressing, Knock-Out

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Representative histograms showing flow cytometry analyses of AF647-conjugated Nectin-4 quattrobinders in comparison to a conventional biotinylated Nectin-4 antibody with a secondary AF647-streptavidin staining on target overexpressing CHO-Nectin-4 cells and parental CHO-K1 cells. b) Quantification of the experiment described in a. Data are shown as mean ± SD performed in biological triplicates (n = 3). c) Western blot analysis displaying Nectin-4 expression levels of various urothelial carcinoma cell lines used in this study. d) Next-generation sequencing analysis of genomic DNA from polyclonal HT-1376 WT and Nectin-4 KO cells prior and after cell sorting as well as from monoclonal HT-1376 cell lines. e) Histograms showing a concentration titration of the conventional biotinylated Nectin-4 antibody with secondary AF647-streptavidin staining and four AF647-conjugated Nectin-4 quattrobinders (3-8-16, 2-40-23, 2-36-30, 2-28-14) on HT-1376 WT and the corresponding knockout cell line (HT-1376 Nectin-4 KO). ns – non-significant; *p<0.05; ***p<0.001; ****p<0.0001; two-way ANOVA with Šídák’s multiple comparisons test.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Flow Cytometry, Comparison, Staining, Western Blot, Expressing, Next-Generation Sequencing, FACS, Concentration Assay, Titration, Knock-Out

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Scheme illustrating the T cell engager (TCE) construct. AI-TCE constructs were transiently transfected into HEK293T cells and secreted into the supernatant. b) Western blot analysis showing the expression of TCE constructs in HEK293T cell lysates after transient transfection. c) Cartoon (left) showing the experimental set-up. Quantification of CD69 + CD8 + cells (middle) and CD25 + CD8 + cells (right) after 48 h co-culture of HT-1376 WT and Nectin-4 KO cells with human PBMCs and AI-TCEs or mock-TCE. Stimulation with an OKT3 antibody was used as positive control. d) Cartoon (left) showing the experimental set-up. Quantification of CD69 + CD8 + cells (middle) and CD25 + CD8 + cells (right) after 48 h co-culture of HT-1376 WT and Nectin-4 KO cells with human PBMCs and AI-TCEs or mock-TCE. Target cells were pre-incubated with TCEs, followed by the removal of the supernatant and an additional washing step. Stimulation with an OKT3 antibody was used as positive control. e-f) Dot plot showing the correlation between frequency of CD69 + CD8 + cells and CD25 + CD8 + cells from experiment described in d and their binding affinities (K D ) determined by SPR. g) Representative curves showing co-cultures of human PBMCs with HT-1376 WT or Nectin-4 KO cells and AI-TCEs or mock-TCE measured as cell index value. Dashed line indicates the timepoint of effector addition. h) Quantification of the cell index value of experiment described in g at timepoint 72 h after effector addition. Cell index value was normalized to the control condition (co-culture without TCE). i) Representative curves showing co-cultures of human PBMCs with HT-1376 WT or Nectin-4 KO cells and AI-TCEs or mock-TCE measured as cell index value. TCE supernatant was pre-incubated on target cells and removed before effector cells were added. Dashed line indicates the timepoint of effector addition. j) Quantification of the cell index value of experiment described in i at timepoint 72 h after effector addition. Cell index value was normalized to the control condition (co-culture without TCE). Data are shown as mean ± SD performed in biological triplicates (n = 3). ns – non-significant; *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001; two-way ANOVA with Šídák’s multiple comparisons test.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Construct, Transfection, Western Blot, Expressing, Co-Culture Assay, Positive Control, Incubation, Binding Assay, Control

Journal: bioRxiv

Article Title: Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics

doi: 10.64898/2026.03.04.709551

Figure Lengend Snippet: a) Quantification of CD69 + CD4 + cells (left) and CD25 + CD4 + cells (right) after 48 h co-culture of HT-1376 WT and Nectin-4 KO cells with human PBMCs and AI-TCEs or mock-TCE. Stimulation with an OKT3 antibody was used as positive control. b) Quantification of CD69 + CD4 + cells (left) and CD25 + CD4 + cells (right) after 48 h co-culture of HT-1376 WT and Nectin-4 KO cells with human PBMCs and AI-TCEs or mock-TCE. Target cells were pre-incubated with TCEs, followed by the removal of the supernatant and an additional washing step. Stimulation with an OKT3 antibody was used as positive control. c) Curves of replicate 2 and 3 showing co-cultures of human PBMCs with HT-1376 WT or Nectin-4 KO cells and AI-TCEs or mock-TCE measured as cell index value. Dashed line indicates the timepoint of effector addition. d) Curves of replicate 2 and 3 showing co-cultures of human PBMCs with HT-1376 WT or Nectin-4 KO cells and AI-TCEs or mock-TCE measured as cell index value. TCE supernatant was pre-incubated on target cells and removed before effector cells were added. Dashed line indicates the timepoint of effector addition. Data are shown as mean ± SD performed in biological triplicates (n = 3). ns – non-significant; *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001; two-way ANOVA with Šídák’s multiple comparisons test.

Article Snippet: Per sample, 4 × 10 6 cells were blocked with human Fc receptor blocking solution (Biolegend, #422302) diluted 1:200 in PBS and afterwards incubated with 10 μg/mL of recombinant, Fc-tagged Nectin-4 protein from human, murine, rat, and cynomolgus origin (ACROBiosystems, Supplementary Table 1) in FACS buffer (PBS, 2% (v/v) FCS, 2 mM EDTA).

Techniques: Co-Culture Assay, Positive Control, Incubation