Millisecond nonlinear state changes during droplet coalescence identify therapeutic-antibody developability liabilities

Abstract

Apoha's Liquid State Machine records ellipsometric waveforms from injections depositing sub-microgram quantities of antibody drop-by-drop onto a liquid reservoir. We previously showed that a behavioural feature extracted from the waveforms, VIBE1, identified antibodies carrying multiple biophysical liabilities in an industrial dataset of 71 monoclonal antibodies, and enriched for clinical failure across a larger dataset of 235 therapeutic antibodies. Here, we use an auxiliary coalescence-sensor channel to decode VIBE1 by separating the coalescence event from its propagation through the substrate. The per-titration drop-to-drop standard deviation of pinch-off time, {sigma}{tau}, explains most of VIBE1's variance across the dataset (R2 = 0.92, n = 1182). High-speed imaging at 10,000 frames per second reveals that all imaged drops initially thin at the same Newtonian capillary-inertial rate while the neck remains wide. In drops from certain antibodies, the thinning bridge then decelerates as internal strain builds in the narrowing neck. This elastocapillary stiffening response has a timescale {lambda} that decreases as pinch-off time {tau}i increases across the imaged set. {sigma}{tau} is therefore a readout of the antibody's propensity to undergo a transient gel-like stiffening response during coalescence, and that variability is what VIBE1 captures. The signal is concentration dependent, and absent in bovine serum albumin (BSA) tested at up to an order of magnitude higher concentration than the antibodies, despite BSA being a strongly surface-active globular protein. The instrument is configured so that complex behaviours of this kind appear in its recorded waveforms; the gel-like coalescence response we identify here is one such phenomenon.