Rapid and accurate in silico solubility screening of a monoclonal antibody library

Abstract

Antibodies represent essential tools in research and diagnostics and are rapidly growing in importance as therapeutics. Commonly used methods to obtain novel antibodies typically yield several candidates capable of engaging a given target. The development steps that follow, however, are usually performed with only one or few candidates since they can be resource demanding, thereby increasing the risk of failure of the overall antibody discovery program. In particular, insufficient solubility, which may lead to aggregation under typical storage conditions, often hinders the ability of a candidate antibody to be developed and manufactured. Here we show that the selection of soluble lead antibodies from an initial library screening can be greatly facilitated by a fast computational prediction of solubility that requires only the amino acid sequence as input. We quantitatively validate this approach on a panel of nine distinct monoclonal antibodies targeting nerve growth factor (NGF), for which we compare the predicted and measured solubilities finding a very close match, and we further benchmark our predictions with published experimental data on aggregation hotspots and solubility of mutational variants of one of these antibodies.

Figure 1

Simultaneous screening of affinity and solubility of antibody libraries. The screening of the antibodies derived from an in vitro discovery experiment (e.g. phage display) can be performed using two parameters: (1) a measured binding strength (e.g. binding affinity or off-rate), and (2) a predicted solubility score (e.g. the CamSol intrinsic solubility; a.u. stands for arbitrary units as the scores are dimensionless). The latter is readily computed from the amino acid sequence, thus enabling the selection from the initial screening of lead antibodies with high affinity and solubility.

Figure 2

Comparison of the monoclonal antibody variants used in this study. (a) Structural model of a monoclonal antibody with the constant domains coloured in grey and the variable domains in orange (VL) and light blue (VH). (b) Table summarising the number of mutations distinguishing any two mAb variants employed in this study; the VH domain is shown in the lower-left half and the VL domain in the upper-right half.

Figure 3

PEG-precipitation assay on the 9 mAbs targeting NGF analysed in this work. Plot of soluble mAb concentrations (y-axis) versus percent PEG amounts (weight/volume, x-axis). Data points are fitted with a sigmoidal curve (broken lines) and the value of the fitting parameter PEG _1/2 is used as a proxy for the solubility (see Methods and Figure S2).

Figure 4

Correlation between the measured solubility and corresponding sequence-based computationally predicted value. Scatter plots of the observed PEG _1/2 as a function of the intrinsic CamSol solubility score. Panel (a) reports the CamSol scores of the VH domain only, as most mutations are found here (see Figure 2). Panel (b) reports a combined score reflecting both VH and VL-domain contributions (see Methods). Regression lines, reported Pearson’s coefficients of correlation (R) and corresponding p-values (p) have been calculated by excluding the outlier point circled in red (mAb3).

Figure 5

Aggregation hot-spot analysis and validation on mutational variants or mAb2. (a) Structurally-corrected (solid lines) and intrinsic (broken lines) CamSol solubility profiles for the VH domain of mAb1 (blue) and mAb2 (red). Residue positions at which the two sequences are different are labelled with a black dot above the profiles, CDR positions (IMGT annotation) with grey boxes. Square markers on the structurally corrected profile denote the positions on mAb2 (W30, F31, L57) that have been experimentally identified as aggregation hot-spots. (b) The structurally corrected solubility profile is color-coded on the surface of homology models (built with SabPred) of the VH/VL domains of mAb2 (left) and mAb1 (right). Aggregation-promoting regions are in orange/red while aggregation-protecting ones in light blue/blue. The experimentally-identified self-association hotspots on mAb2 and the corresponding residues on mAb1 are labelled and represented as balls-and-sticks. (c) Measured HP-SEC monomer retention time for various mAb variants as a function of their combined-chain solubility score calculated from the sequence alone. mAb2 has the residue types W, F, and L at the hot-spot positions 30, 31 and 57 respectively, while mAb1 has S, T, and T. The six variants between mAb2 and mAb1 are named according to which mAb2 position(s) have been mutated to the corresponding mAb1 amino acid (e.g. WFT is mAb2 L57T). The line is a guide for the eyes.