imaging is emerging as a powerful tool to monitor biological processes in living organisms 1
. In vivo
probes can reveal cellular behavior and protein activity in their native context, amidst intact physiologic regulation at the level of interacting molecules, cells and organs. The discovery of novel in vivo
probes, however, can be challenging. Typically, a pre-specified target protein (e.g.
, one that is differentially expressed across cell types) is screened against a library of ligands (antibody, peptide or small molecule). In subsequent steps, ligands that emerge from a screen are often conjugated to a nanoparticle, such as a quantum dot or magneto-fluorescent nanoparticle, to confer imaging capabilities and favorable pharmacokinetics. 2
While this target-based approach has generated many in vivo
probes, it can only be applied when there is a priori
knowledge of a suitable candidate targeting ligand.
An alternative approach to discovering new imaging probes is systematically screening a library of imaging probes for useful phenotypes in cultured cells in vitro 3–5
, a phenotype-based, as opposed to a target-based approach). This approach does not rely upon pre-specified imaging targets, and by analogy to forward genetic screens, can lead to the discovery of imaging probes that act through novel mechanisms and reveal new biology. However, this approach presumes that in vitro
cell lines (typically single samples) can serve as faithful surrogate models for complex in vivo
processes; in light of data suggesting that in vitro
phenotypes can change rapidly upon cell culture 6
, the idiosyncrasies of a particular cultured cell line may limit the ultimate in vivo
applicability of probes arising from these screens.
We sought to develop a more generalizable, robust approach to the discovery of in vivo imaging probes in order to enable integrative studies of biology in intact organisms. We frame the problem of probe discovery as an exercise in class distinction: that is, a promising imaging probe is one that maximizes signal in a desired class of cells (target cells), while minimizing signal in an alternative class of cells (background cells). We screen a library of imaging probes for their phenotypes in multiple distinct samples of each class of cells (target vs. background), and utilize statistical metrics to select those probes that best discriminate between target vs. background cells. The rationale for profiling imaging probes across multiple isolates of target and background cells is twofold. First, we reason that individual samples in a class would differ broadly in many phenotypes, but would still share phenotypes common to their identity as either target or background cells. Second, we hypothesize that comparing imaging probe phenotypes across all members of the target class vs. across all members of the background class would identify probes that most robustly distinguish between these classes without being overly dependent on any single isolate. We also utilize low-passage, primary isolates of human cells (as opposed to immortalized cell lines), to attempt to minimize (but not eliminate) altered cellular phenotypes caused by cell culture.
Here, we illustrate this approach by screening for imaging probes with enhanced binding to vascular endothelial cells. Vascular endothelial cells line the luminal surface of blood vessels, and are critical mediators of vascular homeostasis, inflammation, and repair. Despite their importance in health and disease, there are currently limited options for imaging the endothelium 3, 7
. We screened a library of small-molecule-modified magneto-fluorescent nanoparticles against a large number of primary cell isolates belonging to two general classes: endothelial cells (target class) vs.
macrophages (background class; the unmodified nanoparticle localizes almost exclusively to macrophages 5
). We then calculated a metric that reflects the extent to which nanoparticles bind differentially to endothelial cells vs.
macrophages, across all cell lines tested; permutation testing assigns statistical significance to these differences. We demonstrate that this approach yields small-molecule-modified nanoparticles with significantly enhanced binding to endothelial cells in vitro
(while keeping macrophage uptake constant). Furthermore, we validate the in vitro
findings using ex vivo
incubation with human carotid artery specimens, and in vivo
intravital microscopy in mice. These results may speed the development of a novel imaging probes for endothelial cells in vivo
, and suggest that profiling libraries of imaging probes across multiple cell lines can be a generalizable method to discover new imaging probes.