Article Text
PDF
Abstract
Background Tertiary Lymphoid Structures (TLS) are aggregates of immune cells that form at sites of chronic inflammation. Presence of tumor-associated TLS has recently been shown to associate with immune checkpoint inhibitor response independent of known biomarkers. This makes TLS induction a promising mechanism for novel immunotherapy target discovery. However, the gold standard TLS detection method by pathologists is both time and labor intensive, limiting TLS biology research at scale. Recent advances in AI and spatial profile technology, such as foundation models in computer vision and spatial transcriptomics, highlights the potential to identify TLS more efficiently and enables subsequent novel target discovery.
Methods We have built a spatial atlas of over 1000 non-small cell lung cancer (NSCLC) tumor tissues encompassing multiple paired modalities: subcellular resolution spatial transcriptomics from CosMx, 15-plex immunofluorescence images, hematoxylin-eosin stain images, and whole exome sequencing. We trained a multimodal foundation model on this dataset and then queried the model to propose pixelwise TLS segmentation masks based on predicted probabilities. Paired spatial transcriptomics data were then used to identify genes colocalized with TLS. Finally, we used the foundation model to perform in silico perturbation of genes to simulate their downstream biological impact.
Results We detect TLSes in approximately 40% of the lung tumors with high specificity. We found the presence of TLS is associated with elevated immune infiltration in tumors, and that the spatial gene expression profile differs between TLS-associated tumor microenvironment (TME) and TLS-free TME. Genes known to modulate lymph node organogenesis (e.g. lymphotoxin beta, LTB) are associated with TLS, along with gene sets related to lymphocyte activation and immune response. We also identified genes that promote TLS maturation. In silico gene perturbations using our model reveal potential targets that may promote TLS formation and impact immune cell infiltration through modulating cell-cell communication.
Conclusions TLS can be identified and segmented efficiently with a multimodal foundation model (figure 1). TLS associated genes are expressed in various tissue compartment and form a complex cellular communication network. These genes play a big role in activating anti-tumor immunity. In silico perturbation of these genes reveals potential targets that may promote anti-tumor response through TLS induction.
Ethics Approval Biospecimens were procured from 10 commercial vendors; each study was approved by an IRB in accordance with applicable country-specific regulatory requirements and are used under active or waived Informed Consent. Details available upon request.
Multimodal spatial atlas of NSCLC based TLS research
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.