Materials
EnVision detection system, peroxidase/diaminobenzidine, rabbit/mouse, cat # K5007, citrate, cat # S2369, and HER2 antibody, cat # A0485, for standard (Immunohistochemistry) IHC on tumour sections were obtained from Dako/Agilent. SuperFrost Plus microscope slides were from Daigger Scientific, Vernon Hills, IL, USA. Epitope retrieval solution (pH 6) was from Leica Biosystems, Wetzlar, Germany. Biotin conjugated Anti-HER2 Affibody was (ZHER2:477)2, from Affibody AB, Bromma, Sweden. Dispase, normal goat serum, QD (Qdot® 655 nm) streptavidin conjugate were from Thermo Fisher Scientific GmbH, Dreieich, Germany. ROTISOLV® high pressure liquid chromatography (HPLC) grade deionized water, acetone and ethanol, phosphate buffered saline (PBS) 10 × solution, electron microscopy grade glutaraldehyde (GA) 25% solution, D-saccharose, sodium chloride, glycine, biotin free and molecular biology grade bovine serum albumin fraction V (BSA), and sodium cacodylate trihydrate were from Carl Roth GmbH + Co. KG, Karlsruhe, Germany. Electron microscopy grade formaldehyde 16% solution was from Science Services GmbH, Munich, Germany. Histochoice Clearing Agent, collagenase IA, Tween, 0.01% poly-L-lysine (PLL) solution (mol wt 70,000-150,000), MAPTrix™ Reagent high MW, 0.5 mg/mL protein, sodium bicarbonate, sodium hydroxide, sodium tetraborate, sodium azide and boric acid were from Sigma-Aldrich Chemie GmbH, Munich, Germany. CELLVIEW cell culture dishes (35 mm) with 4 compartments and glass bottom were from Greiner Bio-One GmbH, Frickenhausen, Germany. Custom designed silicon microchips were purchased from DENSsolutions, Delft, Netherlands. The microchips had outer dimensions of 2.0 × 2.6 × 0.4 mm and each contained a central silicon nitride (SiN) membrane window with dimensions of 150 × 400 μm and a thickness of 50 nm. Trivial transfer multilayer graphene was purchased from, ACS Material LLC, Pasadena, CA, USA. NaCl2 crystals were from Plano GmbH, Wetzlar, Germany. All solutions, except those of HPLC grade, were filter-sterilized prior to use. If not indicated otherwise, all procedures were performed at room temperature (RT).
Graphene preparation
Poly-methyl-methacrylate (PMMA) covered multi-layer (3 to 5 layers) graphene was used. To remove the PMMA layer and detach the graphene from the polymer substrate, the composite was submerged into a NaCl2 saturated, deionized water solution at 45° angle and the floating graphene-PMMA stack was then scooped up with a NaCl2 crystal (Dahmke et al. 2017; Weatherup et al. 2016). After baking it in an oven at 100 °C for 20 min, the stack on salt crystal was immersed into acetone for 30 min to remove the PMMA, and was subsequently air-dried. After this point, the graphene on salt was cut with a razor blade into into pieces of ~ 2 × 2 mm as needed to cover the samples.
Preparation of SiN membrane microchips and cell culture dishes
To remove the protective resist layer from the silicon microchips they were incubated for 2-min in HPLC grade acetone, then rinsed for 2 min in HPLC grade ethanol, air-dried, and exposed for 5 min to plasma cleaning (ambient air plasma). Immediately afterwards, the microchips were incubated in 0.01% PLL for 5 min, rinsed twice with HPLC grade water, and incubated for 60 min, in 250 μl MAPTrix coating solution, corresponding to ~ 6 μg MAPTrix/cm2. MAPTrix contains components of the mussel adhesive protein, which, by absorption to the PLL coated surfaces turns them sticky, in order to robustly immobilize single, dispersed tumour cells. The MAPTrix 0.5 mg/mL stock solution was therefore diluted 1:8 in 0.1 M sodium bicarbonate buffer, pH 8.0, and the pH was raised by adding 1 M NaOH in a 1:50 ratio, to induce the absorption process, which was allowed to proceed for 1 h. Afterwards, and directly prior to the seeding of dissociated cells, the coated microchips were rinsed twice with HPLC grade water. Preparation of the glass bottom, four-compartment cell culture dishes was principally the same as for the microchips.
HER2 immunohistochemistry on FFPE tissue sections
Tissue sections from three cases were prepared of which two were HER2 positive and one was HER2 negative. HER2 positivity was defined by immunohistochemistry and scoring was done in accordance with the current HER2 scoring guidelines (Wolff et al. 2018). The two positive cases were diagnostically assigned with HER2 score 3+ while the negative case did not show any relevant staining (HER2 score 0). The tissue sections were first deparaffinized in xylene (3 × 5 min). The slides were then rehydrated in a decreasing ethanol series (100, 100, 96, 80%, 2 min each) before rinsing in distilled water twice. Heat-induced antigen retrieval was performed by incubation in Epitope Retrieval Solution (pH 6), citrate in a water bath at 95 °C for 40 min. Primary HER2 antibody (1:500 dilution) was added to the tissue section for 30 min. Detection was done using the EnVision detection system, peroxidase/diaminobenzidin, rabbit/mouse, according to the manufacturer’s instructions.
Preparation of single cell suspensions from FFPE tumour tissues
Single cell suspensions were prepared from 50 μm-thick tumour sections following the protocol described by Bolognesi et al. with some modifications (Bolognesi et al. 2016). Briefly, tumour samples were deparaffinized with Histochoice Clearing Agent at 65 °C followed by rehydration in an ethanol series (100, 70, 50%, 2 × 5 min each). Heat-induced antigen retrieval was performed by incubation in epitope retrieval solution (pH 6) at 80 °C for 60 min. After washing in PBS, tissue samples were enzymatically digested in 0.1% collagenase IA and 0.1% dispase at 37 °C for 30 min followed by mechanical dissociation with a syringe. Disintegrated cells were washed with PBS twice and then resuspended in PBS supplemented with 1% BSA and 0.05% Tween.
Processing of dissociated tumour cells for HER2 labelling
Biotin conjugated Anti-HER2 Affibody (HER2-AFF-B) stock solution (20 μM) was adjusted to a final concentration of 200 nM in PBS supplemented with 1% normal goat serum (GS), and 1% BSA (GS-BSA-PBS). Streptavidin-QD655 stock solution (1 μM) was diluted 1:2 in 40 mM borate buffer (sodium tetraborate, boric acid, pH 8.3), and then diluted to a final concentration of 20 nM by adding PBS supplemented with 1% BSA (BSA-PBS). The dissociated cell suspensions were processed in batches of 500 or 250 μl volumes, handled in 1.5 ml microcentrifuge tubes, and using centrifugation forces of 2500 g and centrifugation times of 1 min per 100 μl of cell suspension. To diminish unspecific HER2-AFF-B binding the cells were washed once, and then incubated for 10 min, at 37 °C, in GS-BSA-PBS. The cell pellets were resuspended in 250 μl HER2-AFF-B labelling solution, and incubated for 10 min, at 37 °C. After 3 washing cycles with PBS, the cell pellets were resuspended in ~ 75 μl PBS. The water-rinsed, MAPTrix-coated microchips were placed in new wells of a 96-well plate and the concentrated cell suspensions were pipetted onto the wet microchips. During the following 1 h incubation time the cells sunk onto the microchip and firmly adhered. The microchip samples were then rinsed once with 0.1 M cacodylate buffer (CB) supplemented with 0.1 M saccharose (S), pH 7.4 (CB-S), followed by fixation with 3% FA in CB for 10 min. After another rinse with CB-S, 3 rinses with PBS, a 2-min incubation in 0.1 M glycine (GLY) in PBS, pH 7.4 (GLY-PBS), and a rinse with PBS, the samples were placed in 75 μl strep-QD labelling solution, and incubated for 12 min. This two-step labelling protocol ensured HER2 labelling in a 1:1 stoichiometry (Peckys et al. 2015). After rinsing 3 × with BSA-PBS the samples were imaged with light microscopy (LM). To ensure a stable chemical fixation of the biological material for EM imaging, the cells were fixed with 2% GA in CB-S, similar to the FA fixation described above. The samples were stored in BSA-PBS, supplemented with 0.02% SA, at 4 °C, until electron microscopy was performed, usually within the next few days.
Light microscopy
Labelled- and formaldehyde-fixed cells on microchips were imaged in BSA-PBS using an inverted light microscope (DMI6000B, Leica, Germany). For this purpose, the microchips were placed upside-down in a glass bottom dish. Direct interference contrast (DIC) images were acquired from each sample yielding information about membrane topography, and fluorescence images were acquired for the detection of HER2-bound strep-QDs, using a filter cube “A” with a 340–380 nm excitation and a > 420 nm emission window. Images were recorded with 20 x and 40 x objectives. Cells of the control groups, immobilized on the glass bottom of the CELLVIEW dishes, were imaged similarly.
Graphene-coating
Directly prior to STEM, the labelled cells on the microchips were covered with graphene sheets as described earlier (Dahmke et al. 2017). While viewing the sample using a binocular microscope, the graphene-NaCl crystal, held with tweezers, was slowly submerged under an angle of 20–45 ° with respect to the liquid surface in ~ 100 ml of HPLC-grade water in a clean glass beaker. Upon contact with water, the graphene sheet detached (within seconds) from the NaCl crystal, and floated on the water surface. The sheet was then carefully scooped up by submerging a microchip sample held by curved tweezers. The surface of the microchip was subsequently air-dried for a few minutes (Dahmke et al. 2017). During this process the graphene sheet sank onto the cells and the supporting SiN membrane, thus tightly wrapping and enclosing the cells so that the enclosed cells were maintained in liquid state.
Liquid-phase STEM
The graphene-coated sample was imaged with dark field STEM (ARM 200, JEOL, Japan) to observe the individual QD-labelled HER2 positions. The used electron beam energy was 200 keV. Firstly, overview STEM images were acquired of the entire SiN window that served as orientation purpose. By comparing the STEM overview images to the fluorescence- and DIC images of the same sample, it was possible to relocate cells during electron microscopy, and to choose cells with sufficient HER2 labelling for the high-resolution STEM. High-resolution images of selected areas of cells were subsequently acquired for the measurement of the individual positions of QD-labels. The pixel dwell-time was 14 μs, and magnifications of 60,000–120,000 × were used, corresponding to pixel sizes between 1.66 and 0.66 nm. The image size was 2048 × 2048 pixels, yielding a scanning area of 7.0–2.8 μm2 per image, the recording time per image was approx. 1 min. The applied electron doses were in the range of 16–63 e−/Å2, well below the radiation damage limit of these samples (Dahmke et al. 2017).
Statistical analysis
The label positions of QDs in STEM images were automatically detected using a plugin written in ImageJ (NIH) as described elsewhere (Peckys et al. 2015). In short, an image was first checked for artefacts such as larger debris particles, and some images were excluded for the analysis. When the local label density was too high (i.e. > 2.000 labels/μm2) for automatic detection, as was the case in a few membrane regions, data of these regions were omitted as well. A macro allowed automatic processing of a series of images, resulting in a data file with x/y-coordinates for each image. The measured label positions were then statistically analysed for all data files in a group using software of local design programmed in c++ in an Unix environment on OSX. The statistical analysis was based on calculating the pair correlation function g(r) (Stoyan and Stoyan 1996):
$$ g(r)=\frac{1}{\pi {p}^2 ry(r)}\sum \limits_{i=1}^N\sum \limits_{j=i+1}^Nk\left(r-\left|{x}_i-{x}_j\right|\right). $$
(1)
Here, r is the pair distance, ρ the labelling density, and N the number of labels. The covariance function (Stoyan et al. 1993) γ and the kernel (Fiksel 1988) k account for a correction at the edge of the image, and a smoothing of g(r), see also (Peckys et al. 2015). The centre-to-centre distance between the pair of labels at positions i and j is obtained from the modulus |xi - xj| of the two-dimensional (x, y) position vector x. Pair distances smaller than 10 nm were excluded to avoid counting overlapping QDs, thus assuming all labels were located in one horizontal plane, which neglects curvature of the cell surface. The resulting g(r) histogram exhibited a bin width in r of 2.5 nm. The data of the images of one group were averaged by weighting the data of each image by its corresponding particle density. The time for a complete analysis of a series of selected images from a patient (images with too high label densities had to be excluded) was in the range of 20–30 min.