Spatial transcriptomics methods have been slow to move into clinical practice, but spatial proteomics are cheaper and more scalable, and could progress faster.
Cell atlas consortiums, such as the Human Cell Atlas (HCA), have generated immense amounts of data. Less than a decade ago, the HCA had the ambition to map every cell type in the human body across different developmental stages. Their single-cell-level transcriptomic data aimed to provide a foundation for understanding human health and disease. In the recently published collection of papers from the HCA, more than 60 million cells were networked and spatially mapped1. The next stage of the consortium will expand these existing datasets and networks, moving past transcriptomics to multiomics. To understand a disease phenotype, it is necessary to understand how these cells interact, where they are located, and the proteins they produce.
In contrast to dissociative technologies like single-cell RNA-seq, spatial ’omics technologies show the locations of RNAs, proteins and metabolites within a cell or tissue. While large-scale sequencing can be used for subtyping diseases, stratifying patients for clinical trials, and monitoring drug responses, spatial ’omics data can supplement current sequencing methods with additional resolution. In the last decade, spatial technologies have advanced in resolution and in number of measurable modalities, and can provide multiomics data (transcriptomics, proteomics and/or metabolomics) in a single assay.
Translating spatial ’omics data to improve human health has been slow compared to the speed of method development over the last several years. Until recently, the field has mainly been in a stage of data generation. But we may be approaching a tipping point. In the last few years, spatial transcriptomics has provided insights into tumor gene expression and the tumor microenvironment, and has identified prognostic or diagnostic biomarkers, although it is not yet used routinely in clinical practice. Part of the reason for this is scalability: spatial transcriptomics methods can still only capture the expression patterns of a relatively small number of genes per cell. Also, quantifying RNA expression provides limited information, as RNA expression does not always correlate directly with protein translation.
Spatial proteomics technologies are more scalable and provide data about cellular function. Protein-level data are often easier for pathologies to work with, as they are similar to existing diagnostic screens. Protein expression can also tell more about a cellular state during disease, especially when taken together with transcriptomics data. For example, imaging of both RNA transcripts and proteins is necessary for ligand–receptor analysis or investigation of cell–cell paracrine signaling pathways. Secreted proteins are often difficult to stain with immunofluorescence, and the mRNA transcripts could act as a surrogate in the analysis of spatial ligand–receptor interactions2. New methods can now map both proteins and RNAs in the same samples3,4. Nature Methods has chosen spatial proteomics as its Method of the Year for 2024 (ref. 5), demonstrating the promise of this technology.
Recently, a spatial proteomics technology, deep visual proteomics (DVP), was used to develop a therapeutic to toxic epidermal necrolysis6, a severe skin blistering condition that is a severe form of cutaneous adverse drug reaction. The main treatment for this often-fatal condition is stopping the trigger medication, followed by supportive care. DVP combines artificial intelligence (AI)-driven image analysis of cellular phenotypes with unbiased proteomics7. Using DVP, the protein expression signature of keratinocytes and immune cells in patient tissue sections was profiled at unprecedented depth, and the authors could more clearly understand the spatial proteome of epidermal detachment. They observed the Janus kinase–signal transducer and activator of transcription (JAK–STAT) pathway to be upregulated in toxic epidermal necrolysis. Treatment with known JAK inhibitors led to cutaneous re-epithelialization and recovery in seven patients with this condition.
Companies are taking note of the increased interest and higher resolution technologies. The authors of DVP and single-cell (sc)DVP8 have spun out an early-stage company, OmicVision, that uses DVP for biomarker identification and diagnostic tests, stratifying patients on the basis of tumor phenotypes and recruiting appropriate patients for clinical trials. They also examine whether existing therapeutics match tumor protein expression profiles.
Akoya Biosciences has developed a proprietary chemistry that enables whole-slide spatial multiomics at single-cell resolution. The combination of high-plex spatial proteomics and transcriptomics can be used for scalable, deep spatial phenotyping. They are similarly focusing on prediction of cancer immunotherapy response and biomarker discovery. Akoya collaborates with the clinical-stage oncology companies NeraCare and Acrivon, who are using their certified platforms and assays to develop personalized therapies and profile clinically relevant biomarkers in a variety of cancers, aiming to better match patients with treatments.
Companies are also quickly incorporating AI for spatial data analysis. Deep learning is easily integrated with hematoxylin and eosin staining to predict disease outcome9, and with hi-plex imaging mass cytometry to predict prognosis, genotype and therapy response in non-small cell lung cancer10. The spatial biology platform from Nucleai performs AI-guided analysis of imaged biopsies for cancer diagnostics and biomarker discovery. It has been used to predict the responses of patients with non-small cell lung cancer to immune checkpoints blockade therapies11 and in a clinical trial to aid in the identification of biomarkers differentiating responders and non-responders to immunotherapy12.
As with many emergent technologies, democratization is necessary. Further collaboration between the data-generating scientists and clinical pathologists is needed, as is the sharing of platforms and reagents. These efforts would not only boost collaborations and make the data available to a variety of researchers but also would standardize the techniques. Costs of instruments and reagents for spatial technologies are still high, hindering the routine incorporation of the technology at the point of care. Spatial proteomics methods might be cheaper than spatial transcriptomics approaches, where the cost per slide is high; however, cost remains a barrier to their integration into clinics.
With the rapid speed of technology development and data generation, it is clear that many disease-related insights will come from the combinations of transcriptomics and proteomics, and possibly other ’omics such as metabolomics and epigenomics. With more biotechnology companies now working on translating the knowledge into clinics, the technology can influence patient treatment and treatment outcomes.
