Date of Award

January 2026

Document Type

Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Jeffrey Ishizuka

Abstract

Background or scientific premise: Midkine (MDK) is a secreted, heparin binding cytokine implicated intumor immune resistance to immunotherapy and reported to engage multiple receptors including low density lipoprotein receptor related protein 1 (LRP1) on dendritic cells to suppress antigen presentation. Research aims: To enable future MDK antagonists that disrupt receptor engagement, nanobody (VHH) binders were designed against a hotspot patch on the MDK C terminal domain using a structure guided computational pipeline coupled to experimental validation. Two differentiable objectives were evaluated, a BindCraft style loss centered on interface predicted TM score (ipTM) and a modified BindEcraft loss that re- places the ipTM term with an interface objective computed from predicted aligned error logits (iptmenergy); see Chapter 4 for mathematical details. Hypothesis or question: Can predictor guided nanobody design rapidly generate high affinity MDK binders, and does an interface objective that penalizes distribution level uncertainty provide a practical alternative to optimizing scalar confidence summaries alone. Methods or approach: Designs were generated from a fixed VHH framework with masked CDRs using an ESM2 protein language model (ESM2) to propose plausible CDR residues and to provide position specific sequence priors during optimization. Candidates were evaluated with AlphaFold3 in MDK only complexes and in receptor context models containing MDK with LRP1 and integrin, and were triaged using confidence metrics, receptor context displacement proxies, PyRosetta interface scoring, sequence clustering, and developability heuristics. Ten candidates were synthesized and tested by biolayer interferometry (BLI). Results: Multiple nanobodies bound recombinant human MDK with nanomolar apparent affinities, with reported equilibrium dissociation constants (KD ) of approximately 6 to 34 nM and best KD approximately 6 nM. Scientific impact and relevance for communities of interest, including patients: These results demonstrate the feasibility of rapidly generating MDK binding nanobodies using modern predictor guided design workflows and provide MDK binding reagents that motivate follow up studies to test receptor competition and functional antagonism. If receptor blocking activity is confirmed, such reagents could help advance strategies to restore antigen presentation and improve immunotherapy responsiveness in MDK high tumors.

Comments

This thesis is restricted to Yale network users only. It will be made publicly available on 07/14/2028

Download

Share

COinS