Multimodal Imaging Focus of Three HuBMAP Collaborations
Multimodal Imaging Focus of Three HuBMAP Collaborations
HuBMAP SciTech Webinar, January 10, 2022
Three HuBMAP collaborations presented their work in integrating disparate imaging methods at the Jan. 10 HubMAP webinar. A Northwestern University RTI/Vanderbilt University TMC collaboration presented on their work in unifying mass spectroscopy (MS) lipid/metabolite and protein imaging to provide a proteoform atlas. The University of Florida and California Institute of Technology TMCs are combining RNA-based gene expression with antibody-based protein imaging in lymphatic tissues. And the Columbia University/Pennsylvania State University TTD is working with the Stanford University RTI to develop workflows and software to create 3D registration of multiple imaging modalities in single liver cells.
Proteoform Atlas for Imaging Mass Spectrometry
The Northwestern RTI and the Vanderbilt TMC are working to pair the Vanderbilt group’s expertise in lipid and metabolite imaging through MS with the Northwestern team’s multiplex imaging of proteins using MS, reported Vanderbilt’s Madeline Colley.
Sample preparation for lipid imaging is relatively simple, consisting of little more than bringing it to room temperature prior to spraying with a 1,5-diaminonaphthalene (DAN) matrix. A clean protein MS signature, on the other hand, requires a more elaborate preparation, involving a series of washes to remove lipids and salts, a 2′,6′-dihydroxyacetophenone (DHA) matrix, and then rehydration. The instrumentation necessary poses different m/z ratios (500-1,400 for lipids, with a maximum of 4,000; 790-24,000 for proteins. Spectral resolution is more comparable, about 40,000 for lipids and 60,000 for proteins. But perhaps the biggest mismatch and challenge to workflow are the vastly different acquisition times — roughly 30 minutes for lipids, 15 hours for proteins.
The consolidation poses significant challenges. Tissue wash optimization for protein imaging includes working around the salt and lipid removal steps required for a clear signal. Data must be collected within one month of sectioning because of faster degradation of proteins and subsequent signal loss compared with lipids and metabolites. Perhaps the biggest issue is space charging, by which different ionic environments cause inconsistent charge separation. This results in peptide peaks shifting from pixel to pixel — though inclusion of internal protein standards in the sample allows calibration of this effect.
To date the collaborators have scored a number of successes in overcoming these obstacles, according to Jeannie Camarillo of Northwestern. The team is optimizing parallel workflows on sequential sections using the well-characterized rat brain, followed by application to human kidney samples. Spatiomolecular mapping has been excellent, with Nano-DESI’s liquid-based extraction, then ionization compared with MALDI’s direct, laser-based ionization posing challenges as well but the combination offering the ability to detect different proteoforms. They have succeeded in imaging proteins in both rat and human samples using this workflow, with complex spectra offering many proteoform species in the latter. Proteins so investigated to date include histone H4, acyl-CoA-binding protein, histone H1.4, thymosin beta-4, S100-A6, polyubiquitin-B, and the 10 kDa heat shock protein. The results have allowed clear imaging separation between glomerular and cortical cells, with autofluorescence providing registration.
Future directions for the work include matrix optimization for higher molecular weight proteins, instrument optimization, dataset calibration and confirmation of proteoforms via immunofluorescence. Informatics goals include top-down automatic generation of theoretical isotope distributions and use of those theoretical distributions for machine-learning pattern recognition of protein species.
Lymphatic Tissue and in Situ Transcriptomic Profiling
A collaboration of the Florida and Cal Tech TMCs is imaging gene activity in the human spleen via mRNA species and protein expression via “bar code” antibody binding. Clive Wasserfall of UF presented the work, which will investigate splenic functions including in utero hematopoiesis, postnatal blood filtration, function as a secondary immune organ and cleanup of senescent erythrocytes.
Using the hilum as a reference point, the collaborators maintained registration and orientation of samples within the spleen. The initial strategy was to investigate 50 T-cell associated genes via seqFISH RNA visualization and antibody-based CODEX protein visualization. Optimization of processing and signal to noise, dependent on RNA species integrity, are an in initial focus.
To date the team has employed seqFISH probes for 194 genes (each represented by greater than 15 probes), enabling those 194 genes to be bar-coded via two CODEX color channels. This has enabled single-cell imaging providing identification of cell type. Wasserfall included an image of a cell labeled with GZMB, CD40LG and FGFBP2 probes, identifying it as likely to be a CD8 cytotoxic T-lymphocyte (CTL). Cell types identified in the images included littoral cells, B cells, CTLs and macrophages, with excellent anatomic separation. A possible insight offered is the possibility that the spleen employs class 1 HLA proteins, which are present in all nucleated cells and platelets but not erythrocytes, to carry out its function in clearing senescent red blood cells.
Wasserfall concluded the presentation with the observation that the combined method had enabled spatially resolved protein and mRNA visualization and cell-type identification. The two methods have proved complementary in approaching anatomical structure, cell type and biomarker (ASCT+B) annotation (see https://hubmapconsortium.github.io/ccf-releases/v1.1/docs/asct-b/spleen.html for more detail). Future goals include expanding to additional probes and cell types and sequential slides for CODEX and seqFISH, allowing registration of the signals.
Cross-Platform Cell Segmentation and 3D Reconstruction
Hua Tian of the Columbia/Penn State TTD and Jay Tarolli (Ionpath Inc.) of the Stanford RTI presented on the status of their collaboration to integrate different methods for 3D visualization of single liver cells.
Multimodal imaging represents complex workflows with different methods requiring very different sample preparation, Tian said. Further, useful methods such as hematoxylin-eosin (H&E) staining, DESI MSI, and mRNAscope in situ hybridization provide 3D omics integration at the tissue level, not matching the cellular 3D resolution of methods such as water gas cluster ion beam secondary ion mass spectrometry [(H₂O)ₙ-GCIB-SIMS] or C60-cluster primary ion bombardment secondary ion mass spectrometry [C60-SIMS]. Combining these methods in single, 3D images represents the potential of understanding the metabolic states of different cell types and their roles in those cells’ functions.
Tian demonstrated alignment of (H₂O)ₙ-GCIB-SIMS and C60-SIMS signals, using Deepcell segmentation. The collaborators have used mouse liver tissue as a model system to establish workflows, with the aim of application to human tissues. A high-content analysis (HCA) map provides a useful visualization of the image data, comparing lipid/metabolite and protein signals in a 2D matrix that discriminates both activity and cell type. The images can also be reconstructed in 3D, with the signals color-coded.
Tarolli presented on the group’s software platform development using open-source ImagingSIMS to process imaging MS data and generating 2D and 3D visualizations. A particular aim of the project is to improve 3D renderings, currently implemented outside the ImagingSIMS platform but buildable into it. Future development will focus on automation of peak annotation, alignment and mass data extraction, as well as improved analytics including principal component analysis (PCA), clustering, t-distributed stochastic neighbor embedding (t-SNE) and orthogonal projections to latent structures discriminant analysis (OPLS-DA). For more information, contact Brent Stockwell at firstname.lastname@example.org or Hua Tian at email@example.com.