3D multicolor super-resolution imaging offers improved accuracy in neuron tracing

Abstract

The connectivity among neurons holds the key to understanding brain function. Mapping neural connectivity in brain circuits requires imaging techniques with high spatial resolution to facilitate neuron tracing and high molecular specificity to mark different cellular and molecular populations. Here, we tested a three-dimensional (3D), multicolor super-resolution imaging method, stochastic optical reconstruction microscopy (STORM), for tracing neural connectivity using cultured hippocampal neurons obtained from wild-type neonatal rat embryos as a model system. Using a membrane specific labeling approach that improves labeling density compared to cytoplasmic labeling, we imaged neural processes at 44 nm 2D and 116 nm 3D resolution as determined by considering both the localization precision of the fluorescent probes and the Nyquist criterion based on label density. Comparison with confocal images showed that, with the currently achieved resolution, we could distinguish and trace substantially more neuronal processes in the super-resolution images. The accuracy of tracing was further improved by using multicolor super-resolution imaging. The resolution obtained here was largely limited by the label density and not by the localization precision of the fluorescent probes. Therefore, higher image resolution, and thus higher tracing accuracy, can in principle be achieved by further improving the label density.

Figure 1. Comparison between cytoplasmic and membrane labeling for neuron imaging.

(A) STORM images of microtubules demonstrating the effect of label density. In the first panel the localizations from only the first few hundred frames of a STORM movie are included in the reconstructed image to simulate the effect that would be observed in the case of low label density. In the last panel localizations coming from the entire STORM acquisition are included to simulate the effect that would be observed in the case of high label density. The panels in between include progressively increasing number of localizations in the final reconstructed image. It is not possible to reconstruct the actual microtubule structure from the first image due to the low number of localizations, whereas the ability to reconstruct the microtubule structure increases with increasing number of localizations. (B) 2D STORM image of a neural process expressing YFP in the cytoplasm. The YFP was immuno-labeled with antibodies conjugated to photoswitchable A405-A647 pair for STORM imaging. The zoomed-in view shows a region with small neural processes. The small volume of these processes results in a low localization density in STORM images. (C) 2D STORM image of a neural process expressing mCherry attached to the membrane through a palmitoylation sequence. The mCherry was similarly immuno-labeled with antibodies conjugated to photoswitchable A405-A647 pair. The zoomed-in view shows a region of small neural processes. The membrane targeting resulted in a 3.6-fold improvement in label density.

Figure 2. Single color 3D imaging of hippocampal neurons by STORM and confocal.

(A) Mosaic 3D STORM image of hippocampal neurons. The color indicates z-position information according to the colored scale on the right. This image spans a volume of 147×80×1.4 µm (B) A zoomed-in view showing 2D maximum intensity projection of a neural process in confocal (left), confocal after deconvolution (middle). and STORM (right). The left graph shows the intensity profile in the deconvoluted confocal image (grey plot) and the STORM image (black plot) across the red line indicated on both images. Similarly, the right graph shows the intensity profile in the deconvoluted confocal image (grey plot) and the STORM image (black plot) across the green line indicated on both images. The diameter of the neural process at the measured locations is on average 63 nm (FWHM) in STORM and 250 nm (FWHM) in confocal. (C) A zoomed-in view showing 2D maximum intensity projection of neural processes imaged by confocal (left), confocal after deconvolution (middle) and STORM (right). Two neural processes in close proximity are resolved in the STORM image but are not as clearly resolved in the confocal image. (D) The graphs show the intensity profile plotted across the red line shown in (C) for the confocal image after deconvolution (grey plot) and the STORM image (black dotted plot). Two peaks are visible in the STORM plot indicating the two distinct neural processes in the STORM image. (E) xy cross-section of a 100 nm thick slice of a small neural process taken from the midpoint image of a confocal (left) and STORM (right) stack. The middle panel shows the confocal slice after deconvolution. The membrane boundaries contain more labels and are clearly evident in the STORM slice. (F) Intensity profile across the cyan line shown in (E) for the confocal image after deconvolution (grey plot) and the STORM (black plot) image. The two membrane boundaries appear as two well-separated peaks in the STORM plot. (G) Vertical cross-section images across the three yellow lines shown in (E) for the confocal image after deconvolution and the STORM image. The STORM cross-sections look hollow in the middle, as expected for membrane labeling.

Figure 3. Tracing of hippocampal neurons.

(A) Z-stack showing two neural processes that cross each other at different heights in confocal (upper panels), confocal after deconvolution (middle panels), and STORM (lower panels). (B) The xz cross-sections are plotted across the three green lines in (A) for the confocal image after deconvolution (left) and the STORM (right) image. The xz cross-section of the “top” and “bottom” neural processes cannot be easily discerned in the confocal images at the crossing point (left) as they merge together. Thus the two neural processes (red and green circles) appear to merge into one process (yellow circles). On the other hand, the membrane that separates the two neural processes is clear in the STORM cross-sections and a “top” (red oval) and “bottom” (green oval) neural process can be identified at all locations. (C) xy cross-section taken from the midpoint image of a 3D confocal (left) and STORM stack (right). The middle panel shows the confocal slice after deconvolution. (D) The graphs show the intensity profile across the red rectangle shown in (C) for the confocal image after deconvolution (grey plot) and the STORM (black plot) image. Three clearly separable peaks are seen in the STORM plot. The first peak is the membrane edge of the first neuron and the last peak is the membrane edge of the second neuron. The peak in the middle is the membrane boundary that separates the two neurons. (E) The difference in tracing results for this region in confocal (left) and STORM (right). The confocal tracing leads to one parent process splitting into two branches (red) whereas the STORM tracing leads to two neural processes (red and green) in close proximity. (F) Tracing results for an identical region of neurons in confocal (left) and STORM (right). Distinct processes are assigned different colors.

Figure 4. Two-color and multicolor (Brainbow-like) imaging of hippocampal neurons by STORM.

(A) A zoomed-in field of view of neural processes with (left) and without (right) the color information. The neurons were separately transfected with YFP and mCherry, mixed and co-cultured. For the STORM imaging, each fluorescent protein was immuno-stained with antibodies conjugated to different dye pairs. Neural processes that are clearly distinct in the two color images (left, arrows) are difficult to distinguish in the absence of color (right, arrows). (B) STORM image of neural processes labeled with a combination of three fluorescent proteins. The neurons were co-transfected with a mixture of the three fluorescent proteins. The co-transfection resulted in co-expression of different amounts of each fluorescent protein inside individual neurons and hence to different color combinations. Each fluorescent protein was immuno-stained with antibodies conjugated to different dye pairs. The arrow and arrowhead point to two neural processes that show different color combinations. (C) The STORM image of the same region of neural processes (upper panels) is shown in the presence (left) and absence (right) of color. The tracing results for these two cases are shown in the bottom panels. (D) Tracing results with (left) and without (right) color information for the image shown in (B).