In several brain regions, a subpopulation of neurons exists being characterized by the expression of a peculiar form of extracellular matrix, a so-called perineuronal net (PNN). We have previously shown that the PNN can bind large amounts of iron due to its polyanionic charge. Because free iron can generate reactive oxygen species thus being potentially toxic, the PNN may have a protective function by “scavenging” this free iron. Because of this ability, we have hypothesized that PNN-related neurons have an altered iron-specific metabolism.
Thus, to compare the intracellular concentrations of iron containing proteins, specifically, the iron storage protein ferritin H between neurons with and without a PNN, we have used slide-based cytometry with image-based threshold-boundary cell detection on brain sections. In tissue sections, the integrity of the extracellular matrix, especially the characteristic PNNs, is preserved, which is necessary for the identification of the two neuronal subpopulations. A multilabeling approach was chosen to select neurons (neuronal marker NeuN), to classify the neurons according to their subtype (matrix marker Wisteria floribunda agglutinin), and to quantify the protein concentration (protein marker).
Slide-based cytometry (SBC) combines the advantages of flow cytometry and (immuno)histochemistry and therefore enables multiparametric analyses under preserved morphological characteristics. One of the main advantages of SBC is its applicability to analyze fresh or fixed tissue sections on microscopic slides. Thus, it is well suited for in situ identification, classification, and analysis of cells and subcellular structures.
Laser scanning cytometry (LSC), a major representative of SBC, has been used for a plethora of samples and applications ranging from single cells to tissue biopsies and sections (1–4). It enables to focus on different issues: quantitative analyses of cells and cell constituents with single to multicolor staining, determination of polyploidy, cell cycle, and apoptosis (2, 5–9), protein kinetics (10) or immunophenotyping (11, 12). Mosch et al. (2) established a sample preparation procedure applicable for brain tissue, which allows for studying cell distribution within the tissue and the discrimination of neuronal subtypes. However, there are no applications reported so far using SBC for the analysis of intracellular concentrations of a specific protein on the single cell level.
Because we were interested in the concentration of iron transport and iron storage proteins in neurons, we applied LSC for the in situ quantification of these intracellular proteins in rat brain. Thereby, our aim was to compare the iron protein concentrations of two subpopulations of neurons. Both neuronal subtypes are morphologically different. They can be distinguished by the presence or absence of an aggrecan-based extracellular matrix reticular covering the neuronal membrane, a so-called perineuronal net (PNN). The PNN has been shown to be able to bind large amounts of iron in affinity studies using increasing concentrations of colloidal Fe(III)-hydroxide (13, 14). Besides Fe3+, a high affinity is also highly expected for Fe2+, because its binding is related to the polyanionic charge of the PNNs.
As free iron can generate oxidative stress, the PNN may have a protective function by scavenging potentially toxic extracellular iron (15, 16). Because of the interaction of iron ions and the PNN, an altered iron metabolism of the PNN-clad neurons can be assumed. This hypothesis has to be tested by quantifying and comparing the intracellular concentrations of iron-specific proteins in these two subtypes of neurons. Because no intracellular or membrane-bound marker for the PNN-clad neurons is known, the discrimination of both subtypes relies on the identification of the PNNs. Thus, the preservation of the extracellular matrix is mandatory. Therefore, SBC is the method of choice.
We used multicolor tissue analysis (i) to identify neurons, (ii) to detect the extracellular matrix (targeting PNNs), and (iii) to analyze the intracellular iron protein concentration. The application of LSC for the relative protein quantification will exemplarily be demonstrated for the ferritin H concentration in the parietal cortex in rat brain.
Materials and Methods
Animals and Specimen Preparation
Male adult Wistar rats (Rattus norvegicus domestica) were anaesthetized with CO2, transcardially perfused with saline (0.9% NaCl), followed by a fixative solution of 4% formaldehyde and 0.1% glutaraldehyde in phosphate-buffered saline (PBS, pH 7.4) for 30 min. The brains were removed immediately, postfixed for 24 h in 4% formaldehyde in PBS, and stored in PBS with 30% sucrose and 0.1% sodium azide. Frontal sections of the parietal cortex (Bregma -4.1) were cut in frozen state on a microtome. A thickness of 30 μm was chosen in order to assure minimal truncated and less overlaid cells.
Two adjacent brain slices (for control and analysis) were put on each microscopic slide in order to provide identical poststaining and analysis conditions. All sections were washed thoroughly in PBS for subsequent histochemistry.
Experiments were carried out in accordance with the European Council Directive (1986; 86/609/EEC) and had been approved by the local authorities of the University of Leipzig.
Free-floating sections were treated with 2% bovine serum albumin in PBS-Tween (0.05%; pH 7.4) at room temperature (RT) for 1 h to permeabilize the sections and to block nonspecific protein interactions. The sections were incubated at 4°C overnight with a cocktail of the following primary antibodies and lectins in PBS-Tween: (1a) monoclonal mouse anti-NeuN (neuronal nuclei) IgG (Chemicon, 1:150), (1b) biotin-conjugated Wisteria floribunda agglutinin (WFA; Sigma, 1:300), and (1c) polyclonal goat anti-ferritin heavy chain (Y-16) IgG (SantaCruz, 1:200). After repeated washing in PBS-Tween, a cocktail of the following secondary antibodies/reactives was added to the sections at RT for 1 h in the dark: (2a) Cy2-conjugated donkey anti-mouse IgG (JacksonImmunoResearch, 1:250), (2b) Cy3-conjugated streptavidin (JacksonImmunoResearch, 1:250), and (2c) Cy5-conjugated donkey anti-goat IgG (JacksonImmunoResearch, 1:200). After repeated washing in PBS-Tween and PBS, the sections were mounted on glass slides (SuperFrost Plus, Menzel), dehydrated, and covered with DePeX (Fluka) and a cover slip. The set of associated control slices was stained and treated equally; however, the incubation with primary antiferritin H antibody was omitted.
In previous Western blot analyses of protein extractions from the rat brain, the specificities of antibodies were proven. However, this does not exclude unspecific binding of the secondary antibody to nonprotein structures in histochemical slices. Even if no apparent unspecific staining of the secondary antibody in the immunohistochemically stained control tissue was visible by eye, the sensitive photomultiplier tubes (PMTs) might provide a measureable signal. Thus, the purpose of the control slice on each slide was to determine the unspecific Cy5 fluorescence signal and the autofluorescence for compensation in the iron protein related channel. Moreover, it served as internal staining control.
Excitation and Fluorescence Detection
The intracellular protein concentrations were analyzed using the iCys® Research Imaging Cytometer (CompuCyte Corp., Westwood, MA). Core of the iCys is an inverted fluorescence microscope (Olympus). It is equipped with three different lasers for excitation (17) and an automated high-precision X–Y stage.
Multicolor analyses always bear the risk of spectral crosstalk. Hence, compensation of spillover signals is common practice and should be performed for proper identification of specific fluorescence signals, especially if the signal intensity is to be quantified.
Thus, fluorescence dyes (Cy2 and Cy3), known to generate spillover, were used to stain markers needed to identify both neuronal subtypes (NeuN and WFA, respectively). Both dyes were excited with the 488 nm laser line, whereas Cy5 was measured in a separate pass under excitation by a helium–neon laser (633 nm). The two-laser pass scan and the distance between the Cy5 excitation and the 488 nm excitable dyes exclude spectral overlap into the Cy5 channel and enables therefore a proper quantification of the iron protein marker (see dotted spectra in Fig. 1). The fluorescence light of the three markers was split by three optical filters, a green filter (530/30 nm) for the Cy2-channel, an orange filter (580/30 nm) for the Cy3-channel, and a long red filter (650 nm long pass) for the Cy5-channel, respectively (Fig. 1). The fluorescence intensities of the dyes were detected in the respective channels by PMTs. The resulting grayscale images served as basis for analysis (Figs. 2A–2C). Bleaching of the dyes is not an issue of concern when comparing the two-cell populations' protein concentrations, because the cells are intermingled on the same slice and thus treated and scanned simultaneously.
There is a minor spillover of fluorescence light from the Cy2 green channel (identifying neurons) into the Cy3 orange channel (disclosing PNNs). This overlapping Cy2 fluorescence was compensated by generating two virtual channels. The first represents the percentage of the spillover of the Cy2 signal. The second virtual channel (Cy3mod) is generated by subtracting the first virtual channel from the Cy3 channel. The correct level of compensation was achieved when the neurons (Cy2 signal) just disappeared in the Cy3mod image (Fig. 2C). Cy3mod was used for depicting PNNs only, but not for quantification purposes. For the Cy5 channel, a spillover from the Cy2 and Cy3 channel was securely excluded. This proves the Cy5 channel an appropriate choice for quantification purposes in tissue stained with these dyes. The PMT-generated images were cytometrically analyzed using the iCys Cytometric Analysis Software 3.2.5.
The automatic cell recognition requires an accurate trigger with an adequate threshold setting. We used the neuronal marker NeuN instead of the commonly used DNA binding stains DAPI, PI, or Hoechst (1, 2, 8, 17, 18). This is a key point when assessing the intracellular concentrations of proteins in neurons. NeuN labels the nucleus and the soma of neurons but excludes neurites and glia cells (19–21). Thus, the troublesome exclusion of glia cells is obsolete, and the overall cell triggering allows selecting the whole soma of neurons as trigger contour (Fig. 2A, red contour). The minimum area for the trigger signal was set to 20 μm2. Because there were always neurons in the brain slices that overlap or appear in close vicinity, the Watershed segmentation algorithm to separate cell doublets (implemented feature of iCys software) was generally applied. The iCys preset values for the Watershed parameters were adapted to the given tissue characteristics; thus, with the following settings, the cells could be separated without generating artifacts: sensitivity: 0.97, ridge length: minimum three pixels, maximum eight pixels.
The analysis contour was set two pixels outside the trigger contour (Fig. 2, green contour). This included, besides the whole soma of the neuron, a perineuronal region where the Cy3-labeled PNN is detectable for the classification of neurons into PNN-clad (Cy3-positive) and PNN-less (Cy3-negative) neurons. For protein quantification, the ubiquitous background fluorescence has to be taken into account. Therefore, the average background fluorescence per pixel was determined within a two-pixel wide ring around each cell (in eight pixel distance; Fig. 2, double blue line contour). It was used to correct the fluorescence intensities of the Cy2 and Cy5 channels. The background correction was not applied to the Cy3mod channel since also neurites are surrounded by the PNN. The correction would otherwise lead to a reduction of the PNN signal, which could spoil the separation of the two cell populations.
For each individual analysis contour (i.e., neuron), the following information was extracted from the three fluorescence channels: (1) integral (sum of the fluorescence intensity of all pixels), (2) area (area in μm2 of each event), (3) max pixel (fluorescence intensity of the brightest pixel within the cell), and (4) the x- and y-coordinates (position of the event/cell on the slide), and the total number of analyzed events. The associated control tissue was analyzed the same way.
For LSC measurements, the interactive tissue scan was used. First, a high-speed overview scan of the slide was done using the 20× objective with the 488 nm laser line, which allowed to locate the brain sections on the slide. Within this low resolution overview (pixel size 20 μm × 20 μm), the brain region of interest was manually encircled to define the target area, the parietal cortex, for LSC analysis (Fig. 3A). The overview scan was also used to optimize the focus, to set the threshold for triggering, and to adjust the amplification of the photomultipliers. All these parameters, which have to be adopted for each individual encircled region of interest, were saved and later used to restore the settings for the measurement of the corresponding brain region in the associated control tissue.
The region of interest, defined in the overview image, was scanned at high resolution (40× objective, pixel size 0.5 μm × 0.25 μm). If assembling the PMT generated scan field images, the mosaic image depicts the scan area, that is, the cells that are included into analysis (Fig. 3B, neurons in green and the PNNs in red). All scan field images were processed. At the beginning of each analysis, a threshold was optimized.
The position of each single trigger contour is displayed in an X–Y-coordinate dot plot (Fig. 3C; Cy3-negative, green; Cy3-positive, red). Thus, the regional distribution of the cells within the analyzed brain region becomes apparent and allows for reliable cell recognition or artifacts.
Dot plots represent a group of data with common criteria from which a subgroup of data can be selected by choosing specific criteria presented in new dot plots. Thus, by plotting the area against the Cy2 integral fluorescence intensity of the NeuN marker, cell doubles and debris can be excluded by their size (Fig. 3D). Region 4 (R4) defines the group of confirmed events to be cells from which the histogram of the Cy3 fluorescence intensity is plotted to distinguish Cy3-negative cells (PNN-less neurons) from Cy3-positive cells (PNN-clad neurons) (Fig. 3E). The threshold between these two populations is not obvious. It is adjusted under visual control of the selected cells, which are presented in the center of each gallery image (Fig. 3H). The gallery images provide visualization only. There is no tool to exclude or include individual cells. Hence, the Cy3-based separation is crucial.
Sometimes, it occurs that a PNN-less neuron is detected as a PNN-clad neuron if its soma is close to a PNN-ensheathed neurite or cell body (Fig. 3H, right) and therefore receives a false-positive Cy3-signal. However, PNN-clad neurons seem to be smaller than the average PNN-less neuron and can be excluded by selecting an upper limit in the area histogram of PNN-clad neurons (Fig. 3F). A final evaluation of the correct identification of PNN-clad neurons is done by browsing the gallery images for false classifications. Once the data set is classified into PNN-less and PNN-clad neurons, both groups are analyzed according to their iron protein-related Cy5-fluorescence signals (Fig. 3G).
Analysis and Statistics
The automated fluorescence analysis of the iCys software provides a variety of measures available for further data processing from which the integral, that is, the total intensity of the fluorescence within the analysis contour, and the area of the analysis contour were the most important ones. The fluorescence integral value itself reflects the absolute protein content within the analysis contour (ideally the whole cell) but is not an appropriate measure to assess the intracellular protein concentration, because—and this could be a major pitfall—the integral value also depends on the area of the analysis contour. Therefore, the integral value of each analyzed cell was normalized by the area of the corresponding analysis contour. The derived value, fluorescence integral per area, represents the fluorescence (i.e., protein) density.
For the two subpopulations of neurons, we calculated the arithmetic means of the Cy5-fluorescence densities FPNN and F#PNN, respectively. Whereby “PNN” denotes the group with a PNN, the other group having no PNN is labeled “#PNN.” The two mean values reflect the ferritin H concentration. However, they still contain fluorescence contribution from autofluorescence and nonspecifically bound secondary antibody. The analysis of the associated control tissues revealed a substantial contribution (mean values F and F) to the intracellular Cy5 fluorescence. Hence, the measure of the intracellular ferritin H concentration in neurons was calculated as the difference (F-FC) between the Cy5 fluorescence density of the tissue section for analysis and the control tissue.
Differences between the iron protein concentration of the two neuronal subtypes were assessed by calculating the ratio of the mean fluorescence density of neurons with a PNN to that of neurons without a PNN, (FPNN − F)/(F#PNN − F). From this ratio, a relative difference in the protein content was achieved.
Student's t-test was applied for statistical analysis. Because of the additional arithmetic means of the associated control tissue, the t-test is a four sample statistical test with unequal sample sizes. Such a type of multiparameter test is not very common. The equations are therefore shown [Eqs. (1) and (2); derived from Ref.22], with n the number of analyzed cells, σ the standard deviation, F fluorescence density, and the indices “C” for the control tissue, “PNN” for cells with a PNN, and “#PNN” indicating cells without a PNN.
Results and Discussion
The LSC study of the neuronal ferritin H content in the parietal cortex included 295 neurons clad with a perineuronal net and 6307 net-free neurons (n = 3 rats). The concentration of ferritin H in PNN-clad neurons was 12% higher than in neurons without a PNN (confidence level of 99%). This supports the hypothesis of an altered iron metabolism in PNN-clad neurons. It is assumed that potentially toxic extracellular free iron ions are trapped by the PNN, then picked up by transport proteins, and internalized into the cell where it is stored in a nontoxic form inside the ferritin protein. Because one ferritin molecule can bind up to 4,500 iron atoms the PNN-clad neurons gain an enormous iron storage capacity by having in average a 12% higher ferritin H content than the PNN-less neurons. Furthermore, ferritin H exhibits a ferroxidase activity, which facilitates a faster uptake of iron into the ferritin molecule. In general, we conclude from the higher amount of ferritin H that in contrast to PNN-less neurons, PNN-clad neurons are better prepared to accomplish the storage of greater amounts of potentially toxic iron ions. The combination of both, the reduction of the oxidative stress via PNN in the microenvironment of the neuron and an adopted iron metabolism of the neuron itself, supports the hypothesis of reduced vulnerability against degenerative processes. This can be observed for the PNN-clad neurons in diseases as Alzheimer's disease (23).
The concurrence of an increased ferritin H content and the occurrence of PNNs were also demonstrated by Western blot analyses (data not presented in this article). Thus, these results support the findings obtained by SBC. However, the two techniques cannot directly be compared, because Western blotting has its limitation for tissue-based single-cell analysis. It requires homogenized tissue that excludes conclusions for single neurons in preserved tissue sections.
A technique that uniquely enables the visualization and mapping of presently up to 100 different protein species in situ in a single cell is multi-epitope-ligand-cartography/toponome imaging system (MELC/TIS; for review, see Ref.24). This fluorescence imaging technology is based on the application of incubation-imaging-bleaching cycles on dye-labeled cell components (e.g., proteins). The colored mosaic structures showing the protein organization, the toponome, can finally be used for the systematic mapping of protein systems comparing healthy and diseased populations. MELC/TIS expands the analytical capabilities by breaking the spectral limit of traditional multichannel fluorescence microscopy utilizing repeated bleaching. These steps are necessary for the exploration of large molecular networks. However, the MELC/TIS technology is based on a binary mode that differentiates between the presence and absence of a tag depending on a fluorescence signal threshold. This technique is outstanding for cellular multiprotein mapping. Although on is able to differentiate between two concentration levels of minor difference by thresholding, it is not appropriate for quantifying fluorescence intensities. However, when the target-proteins yield sufficient fluorescence intensities, the utilization of incubation-imaging-bleaching cycles for SBC is an option to enhance the multichannel capabilities. Yet, at low protein concentrations, the remaining background fluorescence after bleaching will increase the noise level, thus, hampering a reliable quantification.
The present study introduces SBC as a powerful method to quantify the protein expression of single cells in immuno- and lectinhistochemically stained tissue sections on microscopic slides. In our study, we analyzed cells in the parietal cortex in rat brain slices and distinguished between neurons, which are surrounded by a specialized net of extracellular matrix (a perineuronal net, PNN), and neurons, which do not possess a PNN. With the iCys cell recognition software, we were able to identify and enumerate more than 6,000 single neurons without a PNN and ∼300 neurons with a PNN in an automated analysis. Hence, the proportion of WFA-reactive neurons in the parietal cortex of a rat brain is about 5%.
The comparison of the protein concentrations of ferritin H in PNN-positive and PNN-negative neurons of the parietal cortex showed that PNN-clad neurons have in average a 12% higher protein concentration of ferritin H than neurons without a PNN (confidence level > 99%).
Slide-based LSC is applicable not only for fast and repeatable identification and discrimination of single cells, but also for quasi-quantitative measurement of intracellular protein concentrations in multicolor-labeled tissue. The direct quantification of protein concentrations is also feasible when using primary fluorescence-labeled antibodies with the prerequisite of a careful instrument calibration and standardized preanalytics (25).