Antibodies have been an invaluable tool for protein localization in situ
ever since their introduction for biological imaging [8
]. In recent years, the rapid accumulation of proteomics data has prompted the emergence of novel antibody-based imaging methods that aim to address the new challenges (). A variety of approaches have been explored to achieve multiplexed proteomic imaging of tissues. As a general rule, however, the higher the resolution of the imaging method is, the lower its capability for multiplexing. This is a new variation of the old conundrum faced by immunoelectron microscopy: methods that increase the ultrastructural preservation of tissues, decrease their antigenicity, and thus a delicate balance between the quality of ultrastructure and the level of antigenicity has to be found. Similar to immunoelectron microscopy, multiplexed proteomic imaging requires good antigenicity, and the ability to resolve the protein location with great precision requires good ultrastructural preservation.
Comparison of antibody-based proteomic imaging methods.
On one end of the spectrum of proteomic imaging methods are those that can resolve groups of cells, individual cells or even subcellular structures, without reaching the level of individual synapses. Multiplex tissue immunoblotting is based on protein transfer from a tissue onto a stack of membranes, which can then be labeled with multiple antibodies and protein distribution can be determined at multicellular level [9
]. While this method has only been used with up to 20 antibodies, it can potentially allow labeling with a much larger number of antibodies, by using multiple immunofluorescence on each membrane and then eluting and restaining the membranes. Another approach to proteomic imaging takes advantage of the stereotypy of organisms like Drosophila and uses sophisticated 3D registration atlases that can align an unlimited number of similar tissues immunostained or labeled by a different method, thus basically achieving unlimited multiplexing capabilities with a cellular resolution [12
]. On the other hand, multi-epitope-ligand cartography, or toponomics, exploits the bleaching property of fluorophores which can allow multiple fluorescently labeled antibodies to be applied sequentially to the same tissue, thus visualizing the distribution of up to a hundred of proteins within the same tissue with subcellular resolution [13
]. The high multiplexing potential of all of these methods has fostered a new appreciation for the complexity of the nervous tissue [12
] and has already hinted at the enormous diversity at synaptic level, even without the capabilities of individual synapse resolution [14
On the other end of the spectrum of proteomic imaging methods are those that can resolve individual synapses and subsynaptic structures. Array tomography [15
] has allowed the imaging of up to 24 different antibodies at individual synapses within brain tissue, and the potential for multiplexing is greater. The high spatial resolution of this wide-field fluorescence microscopy based method is enabled by the use of ultrathin serial sections (70 nm) of resin-embedded tissue, and its multiplexing capabilities are due to the possibility for multiple rounds of immunofluorescent labeling, imaging and antibody elution on such sections. This approach is currently limited to a lateral resolution of ~200 nm, the diffraction limit of conventional light microscopy.
Recent fluorescent microscopy methods have broken the diffraction limit, allowing imaging at resolutions intermediate between conventional light microscopy and electron microscopy. Two approaches in particular, STED and STORM, have shown significant potential in the analysis of neurons and synaptic structures. STED (stimulated emission depletion microscopy) is a confocal scanning method that achieves diffraction-unlimited resolution by spatially confining the emitting fluorophores with a second overlapped “depletion” beam that forces excited molecules back to the ground state [17
]. As a scanning method, this approach has the speed of conventional confocal microscopy, but a resolution typically <50–80 nm in the lateral dimension, though preserving conventional axial resolution. To improve the axial resolution, STED was integrated with ultrathin sectioning and image alignment to achieve a reconstructed image of fixed hippocampal neurons stained with two different antibodies with <80 nm resolution in all dimensions [18
]. While these experiments were carried out with sections stained prior to embedding, this approach seems well suited to integration with array tomography sections to increase its multiplexing capabilities. Implementing multicolor imaging with STED though is not as straightforward as with wide-field fluorescence, because it involves multiple wavelengths (excitation, emission and depletion of each fluorophore). A simultaneous three-color STED was demonstrated only recently in cultured cells immunostained with antibodies [19
] (stochastic optical reconstruction microscopy) is a superresolution fluorescence microscopy method that achieves resolution beyond the diffraction limit (typically ~30 nm lateral resolution) by single-molecule imaging of photoswitchable fluorescent probes. STORM has enabled the mapping of the immunofluorescence of 13 antibodies onto a common synaptic coordinate system [21
], albeit across multiple distinct sections. Multiplexable STORM probes have been demonstrated using distinct pairs of activator and reporter dyes, which would allow for at least 3 color, and potentially up to 9 color simultaneous superresolution labeling [22
]. Even though the full potential of STORM for antibody multiplexing has not yet been explored, it is easy to see how STORM could be a great asset in proteomic imaging, for example by using it with array tomography sections.
With STORM, localization works best for objects near the focal plane, because under single photon switching molecules outside the focal plane are activated, contributing out-of-focus background and introducing photobleaching artifacts. Two-photon photoactivation, which confines the activated fluorophores within the focused optical section [23
] substantially enhances the performance of single-molecule based methods like STORM in thicker tissue specimens, though at the cost of speed, since each cycle of photoactivation involves scanning the two-photon excitation, and is limited by the speed of the scan-head. An improvement that overcomes the logistical scanning limit of this approach was implemented using temporal focusing of the photoactivation beam [24
]. In this approach, a high speed galvo-mirror scans the beam along one axis, while structuring of the excitation pulse results in a spatial displacement of the focused beam across the other axis. Hence, the speed of photoactivation is limited by only one physically scanning mirror, decreasing the scan time and maintaining a resolution of <50 nm lateral and <100 nm axial. Generally, these approaches have been used in cells, not yet in tissues. Recently, however, a four-color super-resolution imaging based on single molecule switching and using only a single laser for excitation was applied to imaging in tissues as well [25
Finally, the highest resolution (0.5 nm lateral resolution) has been achieved by ATEM (automated transmission electron microscopy) [26
]. Although originally not a strictly speaking proteomic, but metabolomic imaging method, ATEM is based on small molecule immunostaining of ultrathin sections imaged by light microscopy intercalated within a large series of ultrathin sections imaged by automated electron microscopy. Alignment of the immunostained sections with the EM sections allows mapping of the immunolabels onto the tissue ultrastructure. Up to 11 different antibody labels have been used in this manner and in recent studies those have included antibodies not only for small molecules but also for proteins [28
]. Remarkably, ATEM, despite being an electron microscopy method, also has the capability of imaging large tissue volumes. For example, it has allowed the imaging and full reconstruction at a 2 nm resolution of a retinal circular segment with a diameter of 0.22 mm and an approximate thickness of 0.03 mm [27
]. To better illustrate the scale of this effort, the 16.5 terabyte imaged volume comprised more than 350, 000 image tiles that were captured over five months at a rate of 3,000 images/day.
The high-resolution methods, such as array tomography, STED, STORM and ATEM, are finally beginning to enable researchers to directly address long-standing questions about the scope of synaptic diversity, the plasticity of the synaptic molecular architecture and the complexity of synaptic connections in mammalian brain and retina.