From cortex to medulla: Navigating the thymic landscape of T cell education

Main text

T cells are essential players in the body’s defense against pathogens and cancerous cells. To ensure a specific yet non-self-targeted response against a wide range of antigens, the development of T cells from hematopoietic precursors entails several meticulously controlled differentiation steps, which take place in the thymus with the support of thymic stromal and immune cells. To ensure a functional self-tolerant T cell receptor repertoire, thymic epithelial cells (TECs) in the cortical zone positively select T cells recognizing self-major histocompatibility complex (MHC) molecules, whereas medullary TECs and other antigen-presenting cells in the medulla promote the deletion or lineage diversion of T cells with strong reactivity to self-antigens.

Recent advances in single-cell technologies have facilitated high-resolution profiling of thymic cell types, revealing considerable diversity within the T lineage, resident immune cells, and the stromal compartment.^1,2,3 Additionally, changes in the cellular composition between fetal and early postnatal life highlighted the complexity and plasticity of this important immunological organ throughout its window of peak productivity.¹ Nevertheless, spatial tissue organization, which is critical for the directed differentiation of T cells, has not been analyzed systematically for this period of human development and thus remains poorly understood. Novel high-throughput methodologies for spatial profiling can now provide the necessary level of resolution to address this gap⁴ and offer a comprehensive and unbiased view of the cellular and molecular processes taking place in tissue compartments. However, variability in tissue morphology, sampling method, and spatial technology still represent a hurdle for data integration and quantitative comparisons.

To address this, we established the cortico-medullary axis (CMA),⁵ a morphology-derived common coordinate framework for the human thymus based on key tissue landmarks that can be consistently annotated in individual tissue sections. First, we compute the shortest distances from any location in the tissue to the cortex, medulla, and capsule/edge. Next, relative positioning is determined with respect to selected landmark pairs (cortico-edge and cortico-medullary). Finally, a weighted combination of these relative positions generates a directional, unit-free, non-linear metric, defining a continuous spatial axis—the CMA—spanning the thymic lobule (Figure 1A). CMA values are, therefore, directly comparable across thymic lobules, sections, and even spatial technologies, providing a foundation for integration and quantitative analyses of tissue samples (Figure 1B). This approach enables mapping of any cell or spot at a resolution far beyond the annotated thymic compartments, facilitating the exploration of intra-compartmental cellular and molecular variability while minimizing annotation errors that occur at tissue boundaries (Figure 1C). Additionally, we devised a binning strategy that groups continuous CMA values into biologically meaningful levels, simplifying visualization, quantification, and the characterization of local microenvironments.⁵

Figure 1.

Construction of a common coordinate framework for continuous and quantitative cell mapping

(A) Annotation of discrete morphological compartments in a tissue section serves as the basis for distance measurements. These are subsequently combined to establish an axis that describes the relative position of individual spots/cells within the thymic lobule. This offers increased resolution over discrete tissue annotations. (B) The axis framework can be applied to any tissue section in which the macroscopic compartments can be annotated. Due to its relative nature, this facilitates direct integration and comparison between tissue samples, even when these are derived from different conditions associated with altered morphology or profiled with differing spatial technologies. (C) Through cell annotation or spot deconvolution individual cells can be positioned on a continuous axis. This can help overcome situations where cells or spots cannot unambiguously be assigned to a discrete compartment.

To investigate the composition of the human thymus throughout early human development, we applied this novel framework to a collection of thymic single-cell RNA sequencing (scRNA_seq) and spatial data from 50 fetal and pediatric donors. CMA mapping was carried out on Visium spatial transcriptomics and IBEX 44-plex protein imaging data⁶ to derive CMA values for spots and segmented cells, respectively, which enabled the integration of samples from different donors despite substantial morphological differences. Cell type identification of thymic cells along the CMA was achieved through deconvolution or similarity mapping of the spatial data using the scRNA-seq dataset as reference.

Our analyses revealed that the canonical T cell differentiation trajectory is fully established by the second trimester of fetal development, despite substantial ongoing changes in the macroscopic architecture of the thymus.⁵ In line with this, the majority of stromal and resident immune cells were consistently mapped to the same lobular regions throughout fetal and postnatal life, and many cytokine gradients remained constant within this developmental window. This suggests assembly of the structural framework supporting T cell development within the first weeks of thymus organogenesis, consistent with the detection of efficient output of mature αβ T cells from 12 to 14 post-conception weeks onward.¹

We also identified several remarkable differences between fetal and postnatal thymus. The fetal capsular and sub-capsular regions were noticeably enriched for diverse stromal and immune cell types, including certain fibroblasts and macrophage subtypes, which may be indicative of heightened cellular activity in this zone in early thymic development. After birth, these cells shifted to the medulla, accompanied by changes in the spatial distribution of several cytokines, which likely reflect structural remodeling of the organ in late gestation. Furthermore, we identified the TEC progenitor niche exclusively in the highly vascularized capsular region of the fetal thymus, while the postnatal thymus harbored an additional niche in the cortico-medullary junction and perivascular regions of the medulla, which indicates a role of vascular proximity in TEC progenitor development. By assessing proximity to Hassall’s corpuscles independently of the medullary depth (CMA), we identified these structures as additional cellular organization centers in the postnatal thymic medulla and uncovered novel highly differentiated TEC subtypes with distinct localization.

To enhance our spatial analyses and harness the power of our multimodal RNA/protein single-cell data, we employed a hyper-clustering approach to map thymocyte stages independent of discrete annotations in combination with trajectory analysis. This uncovered distinct maturation and migration kinetics for CD4 and CD8 T lineages, which were accompanied by differences in chemokine receptor expression patterns, demonstrating the usefulness of the CMA for tracing dynamic processes in static data.

Overall, CMA mapping yielded robust and highly similar results in two completely different data types—RNA sequencing-based Visium spot data and single-cell-resolution IBEX protein imaging—underscoring the utility of our axis framework for cross-sample and cross-technology comparisons. These observations lend credibility not only to the CMA as a tool for reproducible spatial mapping but also to the biological findings obtained through this approach.

Importantly, the CMA framework offers significant potential for future research into thymic development by enabling the easy integration of novel with existing data. Likewise, complementary spatial approaches, such as targeted gene expression profiling, protein imaging with dedicated antibody panels, or spatial metabolomics, can be integrated later on to facilitate cross-technology analyses. Finally, the CMA is not limited to two-dimensional (2D) tissue sections but could be extended to 3D data (Figure 2A), such as CT scans,⁷ which are otherwise difficult to visualize and analyze quantitatively. These novel imaging modalities could train machine learning or artificial intelligence models to better estimate errors in 2D data caused by lacking spatial information in all three dimensions (Figure 2B). This may allow prediction of 3D positioning based solely on 2D histological landmarks, which could eliminate the need for tissue annotation altogether (Figure 2C).

Figure 2.

Future perspective: Continuous 3D spatial modeling

(A) 3D isotropic imaging across the organ scale allows the generation of corresponding morphological annotations and derivation of the CMA model in 3D. These imaging data can then be virtually sectioned to produce 2D training datasets. (B) 2D virtual sections, annotations, and the 2D CMA will be used to fine-tune the 2D-to-3D CMA model using advanced artificial intelligence (AI) or machine learning (ML) models. (C) The trained model could allow a direct estimation of a 3D CMA from 2D histological sections.

In conclusion, our study provides a framework for understanding thymic architecture and its role in T cell development. This approach not only advances basic immunological research but also holds potential for translational applications, such as cell and tissue engineering efforts, through better comprehension of the thymic structure and the composition of relevant cytokine niches. Beyond these analyses, the CMA could prove highly informative in the disease context, e.g., for Down syndrome, where a holistic view of the associated tissue changes might help unravel the cause of aberrant T cell development. Beyond the thymus, our axis approach (termed OrganAxis) has already been used to model the development of the human reproductive,⁸ skeletal,⁹ and cardiac¹⁰ systems within a continuous spatial framework, demonstrating its global applicability.

Acknowledgments

L.B. and T.T. are funded by the Fund for Scientific Research Flanders (G075421N, G0A1725N, G0ARH25N, and 12D9523N) and the Concerted Research Action from the Ghent University Research Fund (BOF24-GOA-035). N.Y. and S.A.T. are funded in whole, or in part, by the Wellcome Trust (226795/Z/22/Z). S.A.T. is funded by the CIFAR Macmillan Multi-scale Human Program. This work was supported by the Engineering and Physical Sciences Research Council (grant number EP/Y02978X/2). The author has applied a CC BY public copyright to any Author Accepted Manuscript version arising from this submission.

Declaration of interests

S.A.T. is a scientific advisory board member of ForeSite Labs, OMass Therapeutics, Qiagen, and Xaira Therapeutics; a co-founder and equity holder of TransitionBio and Ensocell Therapeutics; a non-executive director of 10× Genomics; and a part-time employee of GlaxoSmithKline.