It is often the case that affinity ligands and bioconjugation strategies must be optimized for each new preparation to maximize the binding properties of the targeted conjugates14,15
. To streamline these efforts, there remains a critical need to develop advanced conjugation techniques that simplify operation, as well as extend detection limits by improving the efficiency of targeting and amplifying marker signals. Moreover, successful translation into clinical settings will require simple scale-up to successfully process tens or hundreds of samples. Here, we explore a modular and broadly applicable targeting platform based on bioorthogonal chemistry. We were particularly interested in (i) using a biocompatible chemistry with a fast reaction rate, (ii) using reaction partners with very small ‘footprints’ to maximize the number of covalent binding sites, and (iii) developing universal labelling approaches that build upon the vast array of available monoclonal antibodies. We reasoned that such a strategy would be valuable in further improving nanoparticle targeting.
We and others have recently described a covalent, bioorthogonal reaction between a 1,2,4,5-tetrazine (Tz) and a trans
-cyclooctene (TCO) and used it for small molecule labelling16–18
. The [4 + 2] cycloaddition is fast, chemoselective, does not require a catalyst, and proceeds in serum. We hypothesized that this chemistry could be adapted to targeting nanoparticle sensors in different configurations to improve binding efficiency and detection sensitivity. We have named this technique ‘bioorthogonal nanoparticle detection’ (BOND).
summarizes the chemistry, comparative molecular species dimensions, and experimental approaches of the different BOND strategies. We used magneto-fluorescent nanoparticles (MFNPs) to assess the performance of BOND using established fluorescence techniques and a novel miniaturized diagnostic magnetic resonance detector system developed for clinical point-of-care use19,20
. To explore BOND in a biologically relevant system, we chose monoclonal antibodies as the scaffold for nanoparticle attachment due to their large size and the availability of numerous primary amine functional groups. For example, the monoclonal antibody trastuzumab, which is used clinically to treat breast cancers expressing HER221
, has approximately 90 lysine residues that could be converted into nanoparticle reaction sites (). We then comparatively tested BOND for targeting extracellular receptors on cancer cells using two assay types. In one setting, we directly coupled MFNPs to antibodies before cell exposure (BOND-1). In another setting, we used a two-step strategy (BOND-2) in which TCO-modified antibodies were used for primary target binding followed by covalent reaction with Tz–MFNP ().
We first determined the extent to which TCO modification of antibodies promotes nanoparticle binding under the BOND-2 format. summarizes the results for three antibodies that were separately used to target HER2, EpCAM (CD326) and EGFR on cancer cell lines. For each antibody, TCO loading was modulated using various concentrations of an amine-reactive TCO. This resulted in a range of TCO valencies between approximately one and 30 per antibody (see Supplementary Methods
and Figs S1–S3
). Following sequential incubations with TCO–antibody and Tz–MFNP, we found that nanoparticle binding increased with successive TCO loading until saturating around at 20 TCO per antibody for HER2. TCO modifications had little effect on anti-HER2 affinity until loading levels reached 30 TCO per antibody (Supplementary Fig. S4
). Furthermore, TCO modification did not affect the level of non-specific binding of MFNPs to control NIH/3T3 fibroblasts (Supplementary Fig. S4
Effect of TCO loading on nanoparticle binding using BOND-2
To further confirm the above results and to determine the spatial distribution of targeted nanoparticles, we performed confocal microscopy on live cancer cells (). In these experiments, the cells were similarly incubated with TCO–antibody followed by Tz–MFNP. In all cases, a strong fluorescence signal was detected at the cell membranes; this was not observed for a TCO-modified control antibody. These data establish that the Tz/TCO cyclo-addition used for BOND-2 is sufficiently rapid and chemoselective to effectively target nanoparticles to live cells.
We next determined the comparative performance of BOND-2 relative to direct labelling with MFNP immuno-conjugates produced via maleimide/thiol chemistry and BOND-1 (). For BOND-2, the optimized preparations determined in the above experiments were used. We found that BOND-2 consistently yielded higher nanoparticle binding to cells compared to either of the immuno-conjugates, higher by more than a factor of 15 for HER2 and approaching 10 for the other cases. The BOND-1 and maleimide/thiol immuno-conjugates bound to a similar level, but tended to vary across the different markers depending on antibody affinity (5 nM for cetuximab, 9 nM for trastuzumab22,23
). These observations suggest that TCO-decorated antibodies can serve as scaffolds for subsequent nanoparticle attachment. This strategy effectively amplifies the achievable signal, the extent of which increases with the number of available TCO reaction sites.
Comparison of different nanoparticle targeting strategies
To determine whether the amplification was unique to the bioorthogonal chemistry applied or simply a consequence of using a two-step labelling strategy, we tested avidin/biotin as the secondary coupling mechanism. We used avidin/biotin for this purpose because it is the gold standard for biological, non-covalent binding interactions24
. However, it is known that the large molecular size of avidin (~6 nm, ~67 kDa) and its potential for eliciting immune responses in vivo
limit its use for many clinical applications25
. We biotinylated antibodies using similar procedures to achieve a range of loadings (see Supplementary Methods
and Figs S1–S3
). Although nanoparticle binding using avidin/biotin exceeded the direct conjugates and could be improved by increasing biotin valency, the overall signal remained considerably lower compared to BOND-2, despite higher biotin loadings (). We believe that this finding can be attributed to the large size of avidin, which could potentially mask adjacent biotin sites. Second, biotin must associate within a deep cleft inside the avidin protein, which could physically or spatially constrain certain binding configurations. Conversely, Tz is a small molecule and can interact with TCO on the surface of the antibody without physical limitation or encroachment on neighbouring TCO sites. Finally, the Tz valency (84) was considerably higher than the avidin valency (8) on the nanoparticle due to its smaller size, increasing binding probability. The above arguments are supported by the fact that nanoparticle binding saturated at a lower TCO valency (~20) in comparison to biotin (>30). Thus, relatively higher quantities and more diversely spaced reaction sites are required to further increase nanoparticle binding for avidin/biotin. The net result is that the small-molecule bioorthogonal chemistry allows nanoparticles to pack more densely onto the antibody scaffolds, yielding greater signal amplification. It should be noted that the MFNP, although larger than avidin (~28 nm versus ~6 nm hydrodynamic diameter), is not limiting because the bulk of the size can be attributed to a dextran matrix, which can promote binding at a longer range through the presentation of extended reactive linkages. summarizes the results from the various labelling techniques used in this study.
Comparison of nanoparticle targeting strategies
The ability to rapidly profile cancer cells in peripheral blood26,27
or fine needle aspirates20
has important clinical applications for early cancer detection and in devising treatment decisions28
. We therefore adapted BOND-2 to molecular profiling of small populations of cancer cells by diagnostic magnetic resonance (). MFNPs were targeted to tumour cells using BOND-2, and the transverse relaxation rate (R2
) was measured for ~1,000 cells using a miniaturized diagnostic magnetic resonance device20
. At these scant sample sizes, which are in line with clinical specimens, fluorescent signal detection was difficult. However, parallel magnetic resonance measurements could be performed rapidly and at good signal-to-noise levels (). As expected, markers signals were near normal levels for benign fibroblasts and leukocytes (with the exception of CD45, naturally expressed in the latter). Tumour cells showed considerable heterogeneity in the expression of the different markers, a finding that correlated well with the actual expression levels that were independently determined by flow cytometry using larger sample sizes (). (Marker expression levels are listed in Supplementary Table S1
.) The sensitivity of magnetic detection, including both diagnostic magnetic resonance and magnetic resonance imaging, using BOND-2 and the other targeting techniques are presented in Supplementary Fig. S5
. Collectively, these data demonstrate the feasibility of profiling scant cell populations using the efficient and modular nanoparticle targeting strategy of BOND-2.
Profiling cancer cells using diagnostic magnetic resonance
The bioorthogonal [4 + 2] cycloaddition chemistry between TCO and Tz results in higher nanoparticle binding to mammalian cells compared to other standard techniques. This was achieved because the high valencies and small sizes of the reactants promoted attachment of multiple nanoparticles to each antibody scaffold, amplifying the signal per marker. In contrast, direct nanoparticle immuno-conjugates are limited to at most one nanoparticle per marker, and potentially less than one due to multivalent binding. Because the TCO–antibody is applied in excess before nanoparticle exposure and contains numerous TCO moieties, most marker sites should be occupied by separate antibody scaffolds (Supplementary Fig. S4
), and crosslinking of neighbouring antibodies by a nanoparticle consumes an additional TCO rather than an entire marker. We therefore speculate that a portion of the amplification observed for BOND-2 (and avidin/biotin) can be attributed to more efficient antigen recognition. BOND-2 also outperformed a similar two-step targeting strategy using avidin/biotin, most likely as a result of biotin masking or steric constraints that are imposed by the large footprint of avidin. Consequently, the covalent Tz/TCO reaction permits more nanoparticles to bind per antibody scaffold. The increased detection sensitivity resulting from amplification and the modular nature of the BOND-2 technique make it ideally suited for clinically oriented molecular profiling applications. We have demonstrated such an application here with magnetic detection of tumour cells using a miniaturized magnetic resonance detector that was designed for point-of-care clinical use.
We expect that the described BOND platform will have widespread use in diverse nanoparticle targeting applications, including alternative bioorthogonal small-molecule chemistries, affinity molecule scaffolds (proteins, peptides, aptamers, natural products, engineered hybrids) and nanoparticle sensors such as quantum dots, carbon nanotubes, gold nanoshells and polymer matrices/vesicles.