Engineered proteins are useful in a multitude of applications including molecular imaging and therapy. Although these proteins can be engineered for high molecular specificity with high affinity, physiological mechanisms can result in off-target delivery. Non-specific localization can hinder the diagnostic value of molecular imaging, create dose limitations for nuclear imaging, and cause substantial side effects in therapeutic applications. Two common organs for such unintended accumulation are the kidneys and liver. As many of the engineered proteins scaffolds fall below the ~50 kDa glomerular filtration limit, renal uptake is especially problematic. Yet the factors impacting such off-target delivery are poorly understood and strategies to modulate biodistribution would be valuable for molecular imaging and drug delivery.
As we tested the generalizability of the fibronectin domain scaffold for PET tracers, we discovered that clones A′ and D′ have vastly different renal and hepatic signals despite 79% sequence identity (Fig. ). One hypothesis for this discrepancy is their differing hydrophobicity. Thus, we systematically quantified kidney and liver signal in mice by PET imaging with 64Cu-labeled fibronectin domains with a range of hydrophilicities.
The selection criteria (>20% SASA, non-structural, Kyte-Doolittle hydrophobicity >−0.7, mutation to phylogenetically prevalent residue) to identify the primary mutants was relatively effective as 7 of 10 identified sites yielded improved hydrophilicity upon mutation with at least 50% yield. Two exceptions, F48Y and I90Q, have low yield, which could have been predicted by infrequent phylogenetic occurrence of hydrophilic residues at these positions. Mutant analysis reveals a strong correlation (Pearson's correlation coefficient = 0.91) between the logarithm of relative yield and phylogenetic frequency of amide or hydroxyl residues with the exception of V11T (Fig. ). It is noteworthy, though, that the outlying reduced yield of V11T in A′ (65%) is not observed when V11T is applied to the A12N mutant (168% yield) or the L19T mutant (100% yield). The other ineffective mutant, V66T, has high yield but lacks an increase in hydrophilicity. Mutation to an inherently more hydrophilic residue, V66Q, provides a 0.3-min hydrophilic shift with good yield (199%). Only one of five mid-hydrophilic sites provided a significantly useful mutation, as the other four did not yield a substantial increase in hydrophilicity. Yet T16N did provide a 1.2-min shift thereby justifying screening mid-hydrophilics experimentally. It is worth noting that four of the five mid-hydrophilic sites fit the yield:phylogenetic frequency correlation, which could help to guide mutant selection. Overall, the current approach enables rational identification of hydrophilic mutants with strong accuracy. These correlations can also be used to guide combinatorial library design and selection of lead clones. Taken in tandem with the strong correlations of renal and hepatic uptake with hydrophobicity, we demonstrate that designed mutation can be used to tailor biodistribution. The opposite impacts of hydrophilicity on renal and hepatic uptake do not facilitate dual reduction of these off-target sites, but rather enable modulation of the more clinically or scientifically impactful site. This decision is dependent on multiple factors including physiological localization of the molecular target and application-dependent impact of off-target delivery (e.g. radiation dosimetry or therapeutic index). For example, for molecular imaging applications in the abdomen it would likely be best to minimize hepatic uptake while maximizing renal clearance.
Fig. 6. Relation of yield to phylogenetic frequency of hydrophilic residues. The logarithm of the yield of a mutant relative to the yield of A′ is plotted on the ordinate. The phylogenetic frequency of hydrophilic amide or hydroxyl residues (N, Q, S, (more ...)
Applicability to alternative protein scaffolds must be evaluated and is the subject of future studies. The strong negative correlation between kidney uptake and hydrophobicity is in agreement with results for 99m
Tc-labeled cyclic RGD peptide variants (Kunstler et al., 2010
) and two metronidazole derivatives (Giglio et al., 2011
) but directly contrary to the inverse trend observed for albumin derivatives (Ono et al., 2002
) and the lack of a trend observed for variants of octreotide peptide (Schottelius et al., 2002
), Fab fragments (Ono et al., 2002
), and cyclin-dependent kinase 4 inhibitors (Koehler et al., 2010
) (Supplementary Fig. S1
). The strong positive correlation between liver uptake and hydrophobicity is in agreement with octreotide (Schottelius et al., 2002
F-labeled cyclic RGD peptides (Glaser et al., 2008
), and two small molecule organics (Giglio et al., 2011
) but directly contrary to the negative correlation for albumin derivatives (Ono et al., 2002
), the variance of Fab fragments (Ono et al., 2002
) and the lack of a trend for 99m
Tc-labeled cyclic RGD peptides (Kunstler et al., 2010
). In both cases, the contrary results could result from other unquantified molecular variables. Future studies on the physiological mechanisms responsible for organ retention may enlighten these discrepancies.
Previous reports have implicated charge in renal uptake. Thus, we also explored the ability to modulate renal uptake through mutagenic removal of charged amino acids. Eight mutations were tolerated as inferred by their phylogenetic frequency. Note that mutants of R33 were not expressed well, which is perhaps predictable by the phylogenetic lack of hydrophilic side chains (only 6% Q and no occurrence of N, S or T). Likewise, E38 only exhibits 12% S, and E38S is expressed at <10% of the parental clone. Conversely, the phylogenetically frequent (70%) E38P is well expressed (90% relative to parental). Despite the removal of eight and nine charged residues from A′ and D′, respectively, the combination mutants A-22 and D-14 did not exhibit modified renal uptake. Reduction in the net positive charge of both mutants, to A-42 and D-12, decreased kidney retention by 36% in each case. Thus, charge modification, via removal of charged residues and consideration of net charge, can be implemented to alleviate renal retention of engineered protein scaffolds. The impact of both net and total charge will be further elucidated through study of more mutants. Future studies will also investigate the impact of charge residue location and test the generalizability to other protein scaffolds.
In addition to engineering binding paratopes to provide molecular specificity, protein hydrophilicity and charge can be engineered to modulate in vivo biodistribution. Effective mutation can be guided by structural and phylogenetic data.