Subcellular localization of biomolecules and drug distribution by high-definition ion beam imaging

Abstract

Simultaneous visualization of the relationship between multiple biomolecules and their ligands or small molecules at the nanometer scale in cells will enable greater understanding of how biological processes operate. We present here high-definition multiplex ion beam imaging (HD-MIBI), a secondary ion mass spectrometry approach capable of high-parameter imaging in 3D of targeted biological entities and exogenously added structurally-unmodified small molecules. With this technology, the atomic constituents of the biomolecules themselves can be used in our system as the "tag" and we demonstrate measurements down to ~30 nm lateral resolution. We correlated the subcellular localization of the chemotherapy drug cisplatin simultaneously with five subnuclear structures. Cisplatin was preferentially enriched in nuclear speckles and excluded from closed-chromatin regions, indicative of a role for cisplatin in active regions of chromatin. Unexpectedly, cells surviving multi-drug treatment with cisplatin and the BET inhibitor JQ1 demonstrated near total cisplatin exclusion from the nucleus, suggesting that selective subcellular drug relocalization may modulate resistance to this important chemotherapeutic treatment. Multiplexed high-resolution imaging techniques, such as HD-MIBI, will enable studies of biomolecules and drug distributions in biologically relevant subcellular microenvironments by visualizing the processes themselves in concert, rather than inferring mechanism through surrogate analyses.

Fig. 1. Super-resolution visualization of nuclear structures using HD-MIBI.

a Workflow of super-resolution ion beam imaging (HD-MIBI). (1) Cells are treated with the drug of study, fixed, and stained with MoC-Abs. Endogenous elemental masses can also be detected. (2) Cells are rasterized by a cesium primary ion beam, and the secondary ions are collected by a magnetic sector mass spectrometer. (3) Spatial information is serially recorded in up to 8 channels simultaneously. (4) A composite image of the cell is reconstructed by combining and pseudo-coloring the total number of channels at each depth. (5) Serial acquisitions via HD-MIBI yields a depth profile that reveals the specific subcellular localization of endogenous and labeled targets. A 3D rendering of the depths are shown here. (6) Ion count extraction on a pixel-by-pixel basis for feature extraction allows the application of computational methods to identify subcellular interactions, such as nuclear neighborhoods. b Schematic of ¹⁹F-based MoC-Ab. The oligonucleotide includes a FITC fluorophore at the 3′ end and the modified base 2-F-Ac-C (green) in place of deoxycytidine. c Representative ^–e and ¹⁹F images of a HeLa cell stained with anti-nucleolin-¹⁹F/FITC. From left to right: (1) ^–e image of a HeLa cell; nucleoli are indicated with white arrows. (2) Image based on ¹⁹F signals originating from anti-nucleolin-¹⁹F/FITC. (3) Overlay of images showing the co-localization of ¹⁹F signal with the subnuclear structures predicted to be the nucleoli from the ^–e image. (4) An enlarged image of a nucleolus. n = 3. d Representative HD-MIBI image of a HeLa cell stained with anti-H3K27Ac-¹⁹F/FITC. Overlay of ion images for ¹⁹F (green) and ³¹P (blue). The image is the sum of 10 consecutive planes. n = 3. e HeLa cells were labeled for 24 h with IdU (¹²⁷I) and stained with anti-nucleolin-¹⁹F/FITC. Using HD-MIBI, 40 depths of a single cell were acquired with an increase of current every 10 scans. Images of newly synthesized DNA (top) and nucleolin (middle) and their overlay (bottom) show details of a region within the nucleus, at different currents. Details of DNA and nucleolus in these settings are increasing by lowering the current, at the cost of ion yield. Colored dots indicate the image used for the quantification shown in panel (f). n = 3. f Quantification of the relationship between HD-MIBI resolution and secondary ion counts. The lateral resolution of ¹²⁷I (present in IdU) at different currents from panel (e) was calculated using the 84–16% criterion. The R² for the relationship between the increase in resolution and decrease in ion yield was 0.91. Data are presented as mean values +/−SD. n = 3 line scans per cell per condition. See Fig. S6 for details.

Fig. 2. Multiparametric HD-MIBI using MoC-Abs identifies unique subcellular features.

a–c (Left) Representative HD-MIBI images of a HeLa cell stained with a anti-CENP-A-⁸¹Br/Cy3, b anti-H3K27Ac-¹²⁷Ir/Cy5, and c anti-SC35-biotin recognized by streptavidin-¹⁹⁷Au/FITC. Overlay of ion images for MoC-Ab (red) and phosphorus (³¹P; blue). (Right) A digital zoom of boxed area in the original image to show specificity of MoC-Ab signal. All images consist of the sums of 10 consecutive planes. n = 3. d Composite HD-MIBI image of nucleolin (cyan), H3K9me3 (magenta), H3K27Ac (green), and SC35 (red) in a HeLa cell. Cells were stained with anti-nucleolin-¹⁹F/FITC, anti-H3K9me3-⁸¹Br/Cy3, anti-H3K27Ac-¹²⁷Ir/Cy5, and anti-SC35-biotin (recognized by streptavidin-¹⁹⁷Au/FITC). The image consists of the sum of 10 consecutive planes. n = 3. e–g Multiple enlarged images from boxed regions in panel (d) showing e a nuclear speckle, f a nucleolus, and g heterochromatin. h The largest nucleus in panel (d) was masked using the phosphorus signal. The ion counts for each acquired marker (nucleolin, ¹⁹F; DNA, ³¹P; H3K9me3, ⁸¹Br; H3K27Ac, ¹²⁷I; and SC35, ¹⁹⁷Au) were extracted. i t-SNE maps of the nucleus in the center of the image shown in panel (d) were created using the ion counts for each of the 84,836 pixels from the mask described in (h). The ion count for each marker in each pixel was min–max normalized to scale the range to [0, 1]. Each point represents a pixel. Pixels grouped in distinct regions based on the expression of each marker. The color in each map represents the intensity of the indicated marker in each pixel. The areas marked A–C indicate three distinct groups of nucleolin-positive pixels identified manually in the nucleolin map. j Grouped pixels from the unsupervised viSNE map are differentially distributed in space as shown in HD-MIBI images of (1) total nucleolin (¹⁹F) from the HeLa cell shown in panel (d), (2) ¹⁹F signals from pixels within gate A, (3) gate B, (4) gate C, and (5) overlay of ¹⁹F signals from gates A (red), B (green), and C (blue).

Fig. 3. 3D nanoscale imaging of the nucleus through iterative HD-MIBI.

a Representative single-plane HD-MIBI images at different depths of a HeLa cell nucleolus. HeLa cells were stained with anti-nucleolin-¹⁹F/FITC, and 785 individual planes were acquired to obtain HD-MIBI images of a nucleolus from its appearance to its disappearance. Single planes every 100 depths show a distinctive molecular distribution of nucleolin in the 3D space. See Fig. S19 for images of each individual plane. Blue, red, and green arrows indicate x-axis, y-axis, and z-axis, respectively. b (Left) 3D surface reconstruction of images of nucleolin staining of a nucleolus shown in panel (a). (Right) Overviews of the same nucleolus along x-axis (blue arrow), y-axis (red arrow), and z-axis (green arrow) with the origin represented as a black dot. c Representative 3D surface reconstruction of nucleolus (green), centromeres (red), and nuclear speckles (cyan). HeLa cells were stained with anti-nucleolin-¹⁹F/FITC, anti-CENP-A-⁸¹Br/Cy3, and anti-SC35-biotin (recognized by streptavidin-¹⁹⁷Au/FITC). The image consists of the 3D reconstruction of a stack of 40 consecutive planes. d Representative 3D reconstruction of nucleolin (cyan), phosphorus (blue), H3K9me3 (magenta), H3K27Ac (green), and SC35 (red) in a HeLa cell stained with anti-nucleolin-¹⁹F/FITC, anti-H3K9me3-⁸¹Br/Cy3, anti-H3K27Ac-¹²⁷I/Cy5, and anti-SC35-biotin (recognized by streptavidin-¹⁹⁷Au/FITC). The image consists of the 3D reconstruction of a stack of 400 consecutive planes. Enlarged images from the boxed region show details of marker distribution. e (Left) Workflow of iterative HD-MIBI: (1) Five to ten depths at high current are acquired in a cell of interest at 25 × 25 μm to identify a ROI. (2) Iterative acquisition is performed by focusing the beam into the ROI at lower currents in a smaller area of 5 × 5 μm or 10 × 10 μm. (Right) (Top) Representative region of the IdU signal (¹²⁷I) in a nucleus of a HeLa cell labeled for 24 h with IdU (¹²⁷I). (Bottom) A 10 × 10 μm ROI was acquired for super-resolution imaging of chromatin folding. Enlarged image of the boxed region shows fine detail of IdU labeled chromatin. n = 2. f Quantification of the resolution of iterative HD-MIBI imaging of a nucleolus. From left to right: (1) ^–e image of a HeLa cell reveals subnuclear structures including the nucleoli chosen for iterative HD-MIBI. (2) A 3 × 3 μm ROI was acquired by iterative HD-MIBI for nucleolin (¹⁹F). n = 2. (3) An enlarged view from the boxed region in image 2 showing the nanoscale organization of nucleolin (¹⁹F). (4) Line scan along the red line in the boxed region in image 3 demonstrates identification of molecules spaced about 30 nm. (5) An enlarged view of the boxed region in image 3. See Fig. S21 for additional examples. g Iterative HD-MIBI 3D reconstruction. HeLa cells were stained with anti-FBL-⁸¹Br/Cy3, anti-nucleolin-¹²⁷Ir/Cy5, and anti-NPM1-biotin (detected with streptavidin-¹⁹⁷Au/FITC). Iterative HD-MIBI was performed on a site with three nucleolin. The ⁻e image confirmed that the ROI was acquired. Nucleolin (cyan), phosphorus (blue), FBL (green), and NPM1 (red) were used for 3D reconstruction from 40 consecutive planes. These images identify the granular component (GC, NPM1-positive), dense fibrillar component (FBL- and nucleolin-positive), and perinucleolar heterochromatin (PNC, phosphorus-high).

Fig. 4. Subcellular localization of cisplatin distribution.

a Representative HD-MIBI images of TYK-nu ovarian cancer cells treated with DMSO (top row) or 15 μM cisplatin for 72 h (middle and bottom rows) and stained with anti-nucleolin-¹⁹F/FITC, anti-H3K9me3-⁸¹Br/Cy3, anti-H3K27Ac-¹²⁷I/Cy5, and anti-SC35-biotin (detected with streptavidin-¹⁹⁷Au/FITC). We simultaneously acquired images of nucleolin (¹⁹F), DNA (³¹P), H3K9me3 (⁸¹Br), H3K27Ac (¹²⁷I), cisplatin (¹⁹⁴Pt), and SC35 (¹⁹⁷Au), in addition to ^–e. (First column) Cisplatin (¹⁹⁴Pt) distribution within the cell. (Second to sixth columns) Overlay of ion images for cisplatin (¹⁹⁴Pt; white) and a second marker. The brightness intensities for H3K9me3 and H3K27ac were lowered for cisplatin-treated images, compared to untreated images, to avoid overexposure of the cisplatin-treated images. Images consist of the sums of 100 consecutive planes. Scale bars, 5 μm. n = 2. b A schematic for the identification of nuclear neighborhoods. (1) HD-MIBI acquisition is performed on single TYK-nu cells treated with cisplatin and labeled with MoC-Abs as described in panel (a). (2) A sliding window of 3 × 3 × 10 pixels is used for feature extraction for all channels imaged. (3) Unsupervised clustering is performed on extracted features to identify nuclear neighborhoods. (4) Identified clusters are recolored back onto the cell to visualize nuclear neighborhoods spatially. c A heatmap of the 10 distinctive neighborhoods identified based on the five indicated nuclear markers. The intensity of each marker is denoted by the scale bar on the left. The identified clusters of interactions, termed nuclear neighborhoods, are colored and numbered. d Identified nuclear neighborhoods are used to recreate the cell, resulting in distinctive structures that resemble known features in the nucleus. Two representative TYK-nu nuclei are shown. e Violin plots showing cisplatin counts for each nuclear neighborhood identified in panel (c) for the two nuclei shown in panel (d).

Fig. 5. An analytical framework to identify nuclear neighborhood interactions in multiplexed super-resolved data.

a Representative HD-MIBI image of cisplatin-treated TYK-nu cells. TYK-nu ovarian cancer cells were treated with 5 μM cisplatin for 24 h and stained with anti-nucleolin-¹⁹F/FITC, anti-H3K9me3-⁸¹Br/Cy3, anti-H3K27Ac-¹²⁷I/Cy5, and anti-SC35-biotin (detected with streptavidin-¹⁹⁷Au/FITC). Images of nucleolin (¹⁹F), DNA (³¹P), H3K9me3 (⁸¹Br), H3K27Ac (¹²⁷I), cisplatin (¹⁹⁴Pt), and SC35 (¹⁹⁷Au) were simultaneously acquired. A 100-μm² ROI in the nucleus was acquired by iterative HD-MIBI. Displayed are nucleolin (cyan), phosphorus (blue), H3K9me3 (magenta), H3K27Ac (green), cisplatin (gray), and SC35 (red). Denoising was performed using a k-nearest neighbor approach. An unfiltered image is shown in Fig. S24. Enlarged images from boxed regions exemplify the nuclear organization diversity. Images consist of the sums of 20 consecutive planes. b t-SNE map colored by the 11 identified nuclear neighborhoods. The t-SNE map is derived from 100,000 voxels of dimension (x, y, z) = (10, 10, 5) pixels (20,000 voxels were randomly sampled across 5 different cells). Each point represents a voxel. Voxels grouped in distinct regions based on the expression of each marker. The 11 hierarchical clusters were identified by unsupervised hierarchically clustering, followed by manual annotation. c An expression profile of the mean marker expression in each of the 11 distinctive nuclear neighborhoods. The scale intensity of each marker is denoted by the color bar on the top right (Z-score, normalized to each row). d Annotation of each neighborhood, based on their mean marker expression profile from panel (c). e Identified nuclear neighborhoods described in panel (d) are used to recreate the cell, resulting in distinctive structures that resemble known features in the nucleus shown in panel (a). f Pairwise interaction frequency calculations. A permutation test was implemented to quantify the frequency of pairwise neighborhood interactions. If two points (e.g., A and B) were within 5 pixels of each other, they were defined as interacting. The mean of 1000 shuffled interaction frequencies was taken as the expected interaction frequency. (Left) The neighborhood interaction frequency map was calculated using the log2 enrichment of the real over expected number of pairwise interaction frequencies. (Right) The same interaction frequency was represented here in graph form, where size-proportional nodes represent neighborhoods, while width-proportional edges represent log2 enrichment of the real over expected interactions between neighborhoods. Only statistically significant edges (one-sided test, p < 0.05, see “Methods” for details) are plotted. The directionality of the edges is indicated by the color of the node it originates from. g, h Nuclear neighborhoods are colored as shown in the legend of panel (d) to show g chromatin-specific neighborhoods, and h cisplatin-enriched neighborhoods. Enlarged images from boxed regions exemplify different neighborhood interactions: accessible euchromatin (purple box), heterochromatin near the nucleolus (green box), an inactive/active boundary of cisplatin-containing nuclear speckles (orange box), and cisplatin-euchromatin adducts within the nucleolus (brown box). i Differentially spliced transcripts between control and treatment conditions were identified (p-adj < 0.05, from left to right, n = 3170, 3842, 1012, and 1034). The splicing score was calculated, where 0 represents no change, >0 represents increased splicing and <0 represents decreased splicing in the treatment compared to control.

Fig. 6. HD-MIBI based identification of drug relocalization in resistant cells upon multi-drug treatment.

a 5000 sampled voxels from 33 FOVs were combined for analysis. (Left) Dimensional reduction with t-SNE, followed by hierarchical clustering, identified seven distinct cellular neighborhoods (left panel). (Middle) These neighborhoods were then annotated based on their median expression profiles (upper right panel), as well as visual confirmation. n = 8 cells for DMSO, n = 6 cells for JQ1, n = 10 cells for cisplatin, and n = 9 cells for JQ1 + cisplatin examined in 1 experiment. b The secondary electron image for each cell is overlaid with the locations of identified neighborhoods. Two representative FOVs are shown for each treatment. Scale bars, 3 μm. c The raw secondary electron and cisplatin images for all FOVs treated with cisplatin or JQ1 + cisplatin, together with a representative FOV from the DMSO control, are shown. Scale bars, 3 μm. d Neighborhood interaction frequencies between each neighborhood represented here in graph form. Size-proportional nodes represent neighborhoods, while width-proportional edges represent log2 enrichment of the real over expected interactions between neighborhoods. Only statistically significant edges (one-sided, p < 0.05, see “Methods” for details) are plotted. The directionality of the edges is indicated by the color of the node it originates from. e Violin and box plots showing the proportion of cisplatin positive voxels for each neighborhood, comparing cisplatin to JQ1 + cisplatin conditions (* indicates p < 0.05 for a two-sided Wilcoxon test; n.s. not significant). The center of the box corresponds to the median. The minima and maxima bound of box correspond to the 25th and 75th percentiles, respectively. The whiskers extend from the minima and maxima bounds of box to the largest value no further than 1.5 times the inter-quartile range. The outliers are shown as dots. See Figure S34B for a representation of the variability on each single cell. Cisplatin (Cis). n = 8 cells for DMSO, n = 6 cells for JQ1, n = 10 cells for cisplatin, and n = 9 cells for JQ1 + cisplatin examined in 1 experiment.