Quantitative high-throughput methods for antibody fragment optimization and discovery

摘要 单克隆抗体和抗体片段是一类重要的治疗药物， 行业然而，用于发现和优化抗体以具有所需亲和力的方法通常是已知的。 费力的实验室程序，需要几个月的动手研究， 人员（例如噬菌体展示、杂交瘤、单细胞）。此外，选择要向前推进的线索 治疗开发管道通常必须用有限的信息来制作 对应于定量结合亲和力。为了应对这些挑战，Protillion已经将Prot商业化， MaP是一种用于测量105至109个变体的大型文库中的定量蛋白结合的平台， 自动化仪器，获得结果的时间约为2天。我们通过生成 通过原位转录过程将蛋白质直接固定在Illumina DNA测序流动池上 和翻译。该平台允许直接、定量测量荧光抗原结合至 整个蛋白质库在前所未有的规模-一个规模，终于是一个匹配的稀疏蛋白质 在氨基酸突变空间中的作用。在第一阶段，我们采用Prot-MaP来显示VHH （纳米抗体）能够结合SARS-CoV-2刺突（S1）受体结合结构域（RBD）蛋白。我们 多步优化首先全面识别“有益”突变，然后将其组合成 第二个组合文库。该策略鉴定了数以万计的亲和力上级的蛋白质变体 与野生型相比，最好的表现出VHH与该靶标的最高报道的结合亲和力，是野生型的100倍。 从起点改进。我们还开发了一种策略来使这种纳米抗体人源化， 保持高亲和力的接近人类的序列。在第二阶段，我们将首先改善自动化， 我们的仪器的商业可扩展性，并为图书馆设计开发深度学习模型， 选择治疗电极导线。我们接下来还将优化其他SARS-CoV-2 S1 RBD结合纳米抗体， 作为能够结合PD-L1的纳米抗体，PD-L1是与癌症免疫治疗相关的靶标。我们将开发一个 用于鉴定高亲和力、人源化、临床相关VHH试剂的通用管道。我们将 还将我们的展示能力扩展到更大的scFv结构域，并针对两个 分离的靶配体，包括SARS-CoV-2 S1 RBD。最后，我们将调整我们的方法，以显示多达109 在NovaSeq测序芯片上检测不同的蛋白变体，这一规模足以从 天然人源化VHH文库。本提案中概述的活动将使显示多种类型的 抗体片段，优化亲和力和人源化它们的序列，并清楚地定义 功能蛋白质序列。从非靶向文库重新发现新结合剂的能力将 使Protillion平台成为一个垂直整合的“一站式商店”， 非靶向文库，以及这些变体的详细突变分析和优化。

Abstract Monoclonal antibodies and antibody fragments are an important class of therapeutics comprising a $150B industry. However, methods for discovering and optimizing antibodies to have desired affinity are generally laborious laboratory procedures that require months of hands-on research performed by highly skilled personnel (e.g. phage display, hybridoma, single cell). Additionally, the selection of leads to move forward in the therapeutic development pipeline often must be made with limited information that does not necessarily correspond to quantitative binding affinity. To address these challenges, Protillion has commercialized Prot- MaP, a platform for measuring quantitative protein binding across large libraries of 105 to 109 variants on automated instrumentation, with a time-to-result of approximately 2 days. We achieve this by generating immobilized proteins directly on Illumina DNA sequencing flow cells through a process of in-situ transcription and translation. This platform allows for direct, quantitative measurements of fluorescent antigen binding to entire protein libraries at unprecedented scale—a scale that is finally a match for the sparseness of protein function in amino acid mutation space. In our Phase I period, we adapted Prot-MaP to display VHHs (nanobodies) capable of binding the SARS-CoV-2 spike (S1) receptor binding domain (RBD) protein. Our multi-step optimization first comprehensively identified “beneficial” mutations, which were then combined into a second combinatorial library. This strategy identified tens of thousands of protein variants with affinity superior to wild type, with the best exhibiting the highest reported binding affinity for a VHH to this target, a 100-fold improvement from the starting point. We also developed a strategy to humanize this nanobody, producing a near-fully-human sequence that maintained high affinity. In Phase II, we will first improve automation and commercial scalability of our instrumentation, and develop deep learning models for library design and selection of therapeutic leads. We will next optimize other SARS-CoV-2 S1 RBD-binding nanobodies, as well as nanobodies capable of binding PD-L1, a target relevant to cancer immunotherapy. We will develop a universally applicable pipeline for identifying high-affinity, humanized, clinically-relevant VHH reagents. We will also extend our display capabilities to larger, scFv domains, and carry out scFv affinity optimization against two separate target ligands, including SARS-CoV-2 S1 RBD. Finally, we will adapt our methods to display up to 109 distinct protein variants on a NovaSeq sequencing chip, a scale sufficient to identify binders de novo from naïve humanized VHH libraries. The activities outlined in this proposal will enable display multiple types of antibody fragments, optimize affinity and humanize their sequences, and clearly define the landscape of functional protein sequences. The capability of de novo discovery of new binders from untargeted libraries will make the Protillion platform a vertically integrated “one stop shop” allowing both identification of “hits” from untargeted libraries, as well as detailed mutational analysis and optimization of these variants.