Multiresolution nondestructive 3D pathology of whole lymph nodes for breast cancer staging

Abstract

Significance: For breast cancer patients, the extent of regional lymph node (LN) metastasis influences the decision to remove all axillary LNs. Metastases are currently identified and classified with visual analysis of a few thin tissue sections with conventional histology that may underrepresent the extent of metastases.

Aim: We sought to enable nondestructive three-dimensional (3D) pathology of human axillary LNs and to develop a practical workflow for LN staging with our method. We also sought to evaluate whether 3D pathology improves staging accuracy in comparison to two-dimensional (2D) histology.

Approach: We developed a method to fluorescently stain and optically clear LN specimens for comprehensive imaging with multiresolution open-top light-sheet microscopy. We present an efficient imaging and data-processing workflow for rapid evaluation of H&E-like datasets in 3D, with low-resolution screening to identify potential metastases followed by high-resolution localized imaging to confirm malignancy.

Results: We simulate LN staging with 3D and 2D pathology datasets from 10 metastatic nodes, showing that 2D pathology consistently underestimates metastasis size, including instances in which 3D pathology would lead to upstaging of the metastasis with important implications on clinical treatment.

Conclusions: Our 3D pathology method may improve clinical management for breast cancer patients by improving staging accuracy of LN metastases.

Keywords: breast cancer; lymph node staging; open-top light-sheet microscopy; three-dimensional pathology.

Fig. 1

Sentinel LNs are resected during mastectomy or lumpectomy to facilitate evaluation of nodal metastases. However, sparse sampling with conventional histology may lead to underclassification of metastases in cases for which the maximum diameter of the tumor is not sampled on glass slides. A simplified diagram of a tangentially sectioned metastasis (pink) within surrounding LN tissue (gray) is shown (center). In this example, the 2D section under-represents the largest dimension of metastasis and could lead pathologists to underclassify the metastasis with implications on patient treatment.

Fig. 2

(a) En face visualization of an LN stained with our H&E fluorescent analog, cleared with CUBIC-R+(N), and imaged with multiresolution OTLS microscopy. Standard fluorescence intensity visualization is shown on the left at low resolution. H&E-like visualization is shown on the right, which is generated with an open-source false-coloring code that operates on two-channel fluorescence images of tissues stained with a nuclear and cytoplasmic fluorophore. (b) ROIs shown at high resolution. Standard fluorescence and H&E-like visualizations are shown on the left and right, respectively. Scale bars represent $10\mu \mathrm{m}$. Comparison of various techniques for [(c)–(f)] nuclear and [(g), (h)] cytoplasmic staining in LN tissue. Line profiles of staining intensity as a function of depth in the tissue are shown to the left of each vertical cross-section image. For nuclear staining, we show (c) our original H&E-analog staining protocol, (d) a SWITCH-mediated version of our H&E-analog staining protocol, (e) the original CUBIC protocol, and (f) our final protocol based on CUBIC-HV. At the bottom, en face views are shown at various tissue depths for the dataset in (f). For cytoplasmic staining, we compare (g) incubating a specimen with our cytoplasmic stain (AlexaFluor NHS ester) in PBS at pH 5, and (h) our final protocol, where the specimen is incubated with the same cytoplasmic stain in a PBS/THF mixture at pH 5. At the bottom, en face views are shown at various tissue depths for the dataset in (h). Scale bars represent $50\mu \mathrm{m}$ for vertical cross sections and $25\mu \mathrm{m}$ for en face views.

Fig. 3

(a) Side-view schematic of the multiresolution OTLS microscope used in this study. The collection arm is equipped with 5× and 20× objectives on a dual-objective turret for low- and high-resolution imaging, respectively. Specimens may be placed on the modular sample holder, which is attached to a motorized stage (not shown) that translates the specimen in XYZ during imaging. (b) Diagram of the focal region within the specimen. (c) To enable aberration-free imaging, the refractive index of the immersion media, sample holder, and sample must be precisely matched. (d) A single 3D image tile is acquired by stage-scanning the specimen in the $X$ direction. (e) Adjacent image tiles are collected in the lateral ($Y$) and vertical ($Z$) directions, which are assembled to enable the visualization of a large 3D volume. (f) Diagram of the pathology workflow used for staging axillary LNs from breast cancer patients. LN specimens are imaged at low resolution, false colored to mimic H&E histology, and are viewed by a pathologist in 3D to identify suspicious ROIs (i.e., possible metastases). These localized regions are subsequently imaged at high resolution in 3D, false-colored, and displayed to a pathologist for definitive diagnosis (tumor versus benign).

Fig. 4

(a) Step-by-step illustration of our previously reported workflow^, for OTLS image acquisition, postprocessing, and viewing of false-colored datasets in 3D. A series of volumetric image tiles were (1) acquired sequentially during imaging and (2) assembled to create a large volumetric image of the specimen. To create a “seamless” 3D dataset, the image tiles were then (3) fused into a single 3D image volume. For visual interpretation, the fused datasets could subsequently be (4) false-colored and saved as a stack of RGB TIFFs (or HDF5). TIFF stacks could then be loaded into Fiji for 3D visualization. (b) The workflow we report here enables H&E-like visualization of 3D image data immediately after imaging. After each 3D image tile is acquired (1), it is (2) immediately false-colored while the raw data are in RAM (before the next tile is imaged) and (3) saved at multiple levels of downsampling in a hierarchical data format (HDF5). After imaging is complete, the volumetric false-colored dataset (hierarchical RGB dataset) may be viewed in BigStitcher with an H&E-like appearance. (c) This new workflow reduces postprocessing time by $50\times$ compared to our previous workflow. The time required for each processing step is shown for the previous workflow (top) and optimized workflow (bottom). Here, the reported processing times are for a low-resolution dataset using a workstation equipped with an Intel Xeon processor, NVIDIA TITAN Xp graphics card with CUDA 10.2, and 128 GB of RAM. The time required for false coloring assumes that the depth direction ($z$ axis) is binned by $4\times$ in all cases.

Fig. 5

(a) Image atlas of regional LNs classified with our 3D pathology workflow. Low-resolution images of (a) benign node and (b) metastatic node. For both examples, the low-resolution datasets (left) were used to identify suspicious regions (center). Subsequent high-resolution imaging of those localized regions (right) revealed that the suspicious region in panel (a) is a benign vessel, as indicated by the flattened endothelial cells lining the vessel, and that the suspicious region in panel (b) contains cancer cells, as indicated by the enlarged and irregular-shaped nuclei (black arrows). Lymphocytes adjacent to metastatic cells are indicated by white arrows, exhibiting circular darkly stained nuclei that are densely packed (high nuclear-to-cytoplasm ratios). The maximum dimension of the metastatic nodes is used to classify them as individual tumor cells ($<200\mu \mathrm{m}$), micrometastases ($200\mu \mathrm{m}$ to 2 mm), or macrometastases ($>2\mathrm{mm}$). (c) Top: a micrometastasis ($<2\mathrm{mm}$) is observed in the simulated histology images. Bottom left: deep 3D imaging reveals that the tumor deposit is a macrometastasis ($>2\mathrm{mm}$), an upstaging that would lead to a more aggressive treatment plan (complete axillary LN dissection). Bottom right: high-resolution imaging of the interface between the metastasis and benign tissue. ROIs show clustered nuclei that are suggestive of tubule formation (top) and enlarged, irregularly shaped nuclei (middle) that are indicative of cancer. A high-resolution view of benign lymphocytes is also shown (bottom, white arrows). Scale bars of ROIs represent $10\mu \mathrm{m}$.