A Ghost of the Thymus
The thymus is where T cells learn to leave the body's own tissues alone. Medullary thymic epithelial cells (mTECs) accomplish this by expressing a staggering breadth of tissue-specific proteins, essentially staging a molecular parade of 'self' before newly minted T cells. Any T cell that reacts too strongly gets deleted. The process is called central tolerance, and it is one of the most elegant quality-control mechanisms in biology.
The problem is that it is imperfect. Some self-reactive T cells escape into the periphery, and the immune system needs a second line of defense. For years, a rare population of cells in the secondary lymphoid organs, defined by their expression of the Aire gene (Autoimmune Regulator, the same master transcription factor that orchestrates self-antigen display in the thymus), has been suspected of filling this role. These extrathymic Aire-expressing cells, or eTACs, could delete or silence autoreactive T cells in the lymph nodes. But their precise identity remained stubbornly unclear, with competing claims placing them variously among dendritic cells and innate lymphoid cells.
A study from Wang, Lareau, and colleagues at UCSF and Stanford now resolves that debate with unusual finality. Using single-cell multiomics across thousands of individual cells, the team not only confirms that eTACs are migratory dendritic cell-like populations, but discovers within them a wholly new cell type: the Janus cell, a population that co-expresses Aire and Rorc (encoding RORγt, a transcription factor previously associated with T helper 17 cells and innate lymphoid cells), and that bears a genomic signature almost indistinguishable from the mTECs it was never thought to resemble.
Mapping the Unmapped: A Multi-Omic Portrait
The experimental architecture here is worth dwelling on, because it is what makes the conclusions credible. Rare immune populations are notoriously difficult to characterize. eTACs represent a tiny fraction of lymph node cellularity, and previous studies had to rely on individual marker genes to assign identity, a strategy that is vulnerable to the very ambiguities this paper resolves.
The cosine similarity approach is the methodological backbone of the identity claims. Rather than asking 'does this cell express marker X?', the authors ask 'across the entire transcriptome, which reference population does this cell most resemble?' Applied to both RNA and chromatin data independently, the answer was unambiguous: both JCs and AmDCs cluster with migratory dendritic cells, not with innate lymphoid cells. The ILC3 hypothesis, which had gained traction based on Rorc co-expression, does not survive whole-transcriptome scrutiny. JCs share some transcripts with ILC3s, including Ccr6 and Tmem176a/b, but canonical ILC3 transcripts (Il17a, Il22, Il23r) are absent.
The chromatin data adds a layer that RNA alone cannot provide. Janus cells carry the highest chromatin accessibility of any cell type profiled in the lymph node, followed closely by migratory DC1/2s. A differential accessibility analysis comparing eTACs to all other lymph node populations found 12,571 peaks with greater accessibility in eTACs, against only 1,578 peaks with less. This is not a subtle difference. An open chromatin landscape of this breadth implies a cell primed to express an unusually wide range of genes, which is precisely what the transcriptional data shows.
A Reservoir of Self: Tissue-Specific Antigens in the Periphery
The functional logic of eTACs depends on their ability to display self-antigens to patrolling T cells. The transcriptional data makes clear that JCs are exceptionally well-equipped for this role. Among all profiled lymph node populations, JCs express the highest number of tissue-specific antigen (TSA) genes, a catalog of proteins normally restricted to peripheral organs. The TSA list used for comparison comprised 6,610 genes, 202 of which were significantly upregulated in JCs relative to other Aire-expressing populations.
The tissue distribution of these JC-enriched TSAs is telling. Neuronal-associated antigens and germ cell and placental antigens are particularly enriched. Genes like Gal (galanin, expressed in pancreas and nervous system), Prg2 (expressed in bone), and Pappa2 (expressed in placenta) are among the differentially upregulated transcripts. The placental enrichment connects directly to prior work showing that eTACs are required for normal maternal-fetal immune tolerance, though the precise mechanism remains an open question.
Both JCs and AmDCs also express a broad array of antigen processing and presentation genes spanning both MHC class I and class II pathways, alongside a suite of immunomodulatory transcripts including Socs2, Fas, Cd200, Ido1, Smad4, and Irf2bp2. The picture that emerges is of cells built not just to display self-antigens, but to do so in a context that actively suppresses rather than activates the T cells they encounter.
A Thymus in the Periphery: The mTEC Homology
The most arresting finding in this paper is not the discovery of Janus cells per se. It is what Janus cells look like when you compare them to the rest of the immune system's reference atlas.
When the authors scored each cell against ImmGen reference populations for transcriptional similarity, JCs lit up not only in the migratory DC cluster but also in a second, unexpected location: medullary thymic epithelial cells. The top differentially expressed genes defining JCs, when queried against the full ImmGen RNA-seq database, were enriched specifically in mTEC-high populations. An integrated atlas of published thymic epithelial single-cell data confirmed that many of these JC-specific genes track tightly with Aire expression during mTEC development. Critically, the majority of these shared genes are not themselves Aire-regulated in the thymus (only 3 of the top 50 differentially expressed genes, roughly 5.5%, appear on the Aire-dependent gene list). The homology reflects a shared regulatory program, not simply a shared output.
The chromatin data tells the same story from a different angle. Transcription factor binding-site accessibility analysis of the ATAC data showed that the open chromatin of JCs is most enriched for binding sites of RelB, NF-κB, and the less-characterized HIVEP family proteins, precisely the transcription factors known to drive Aire expression in mTECs. The transcripts encoding these factors, including Relb, Nfkb2, Hivep1, and Hivep3, are highly expressed in JCs and migratory DCs. When the same transcription factor binding-site enrichment analysis is applied to the ImmGen ATAC database, the mTEC population stands out as the closest match among all 89 reference populations.
This convergence between a hematopoietic peripheral cell and a stromal thymic cell is not what anyone would have predicted from first principles. The two populations arise from entirely different developmental lineages, occupy different anatomical compartments, and serve what were thought to be distinct phases of immune education. Yet at the level of chromatin architecture and transcriptional circuitry, they are running a recognizably similar program.
The Same Signal, Twice: RANK-RANKL in Thymus and Periphery
If eTACs and mTECs share a transcriptional program, they might also share the upstream signals that activate it. In the thymus, RANK-RANKL signaling through the noncanonical NF-κB pathway is required for Aire expression in mTECs. The scRNA-seq data shows that Tnfrsf11a (encoding RANK) is highly upregulated in eTACs, and that Tnfrsf11b (encoding OPG, a soluble RANKL antagonist that serves as a negative-feedback brake in cells receiving high RANK signaling) is expressed specifically in JCs among peripheral immune populations.
To test whether this signaling axis is functionally required, the authors treated Adig mice with an anti-RANKL blocking antibody or isotype control for three doses over six days. The result was unambiguous. Total eTAC numbers fell with a p < 0.0001 significance, and JCs, identified as GFP+ CCR7-negative CD11c-negative cells, nearly disappeared entirely (p < 0.0006). The same treatment in the thymus reproduced the well-established loss of Aire-expressing mTECs. The parallel is exact: the same cytokine signal, operating through the same transcription factor family, drives Aire expression in both the central and peripheral tolerance compartments.
This has a practical implication worth flagging. RANKL blockade is being explored in tumor immunology as a strategy to transiently reduce central tolerance and boost anti-tumor T cell responses. The data here suggest that such interventions will simultaneously deplete peripheral eTAC populations, potentially with consequences for autoimmunity that have not been fully accounted for in experimental designs.
Proof of Concept: Preventing Diabetes with Peripheral Tolerance
Transcriptional homology and chromatin accessibility are compelling, but the question that matters clinically is whether eTACs actually do what the model predicts. The thymic swap experiment answers that question with unusual directness.
The design is elegant in its logic. Non-obese diabetic (NOD) mice and Adig NOD mice were thymectomized and simultaneously given transplanted wild-type thymi under the kidney capsule. Two weeks later, both groups received bone marrow from 8.3 TCR-transgenic mice, whose CD8+ T cells carry a receptor specific for IGRP (islet-specific glucose-6-phosphatase related protein), a pancreatic self-antigen. In the Adig NOD recipients, IGRP is expressed under the Aire promoter in peripheral eTACs. In the WT NOD recipients, it is not expressed peripherally at all. Critically, the transplanted wild-type thymi in both groups lack IGRP, so any tolerance that develops in the Adig group must be mediated by peripheral eTACs alone.
Flow cytometry and tetramer staining confirmed that thymic negative selection was equivalent between groups: IGRP-specific T cells left the transplanted thymi at comparable rates. The divergence happened in the periphery. In the Adig NOD mice, IGRP-specific CD8+ T cells were profoundly depleted from the spleen, cervical lymph nodes, and pancreatic lymph nodes, and surviving cells showed reduced tetramer avidity, a sign of functional inactivation. In the WT NOD controls, these cells accumulated unchecked.
The survival curves tell the story plainly. Every Adig NOD mouse in the thymic swap experiment remained diabetes-free throughout the observation period, a 100% protection rate, while control animals developed disease at the expected rate for the NOD background. The log-rank test comparing the two survival curves reached p < 0.0005. Self-antigen expression in peripheral eTACs, without any contribution from the thymus, was sufficient to prevent autoimmune diabetes in a model where autoreactive T cells were being continuously generated and exported.
What the Data Does and Does Not Settle
The methodological strengths here are real. Using cosine similarity against a comprehensive reference atlas, rather than a handful of marker genes, is the right way to assign cell identity in a field where lineage boundaries are contested. The reciprocal validation, scoring cells against ImmGen RNA and then independently against ImmGen ATAC, and getting the same answer both times, is reassuring. The REALTAR lineage-tracing mouse, which simultaneously reports active Aire expression and RORγt lineage history, is a genuinely clever tool for disentangling two populations that share a transcription factor.
A few questions remain open, and the authors are candid about them. The developmental origin of Janus cells is not established. RNA velocity analysis found no evidence of a developmental trajectory from JCs to AmDCs, and lineage tracing showed that most AmDCs do not arise from an RORγt-expressing precursor. But where JCs themselves come from, and what drives their unusual mixed phenotype, is not addressed. The sample sizes in the functional experiments are not reported in the abstract, and the thymic swap model, while elegant, involves surgical manipulation and irradiation that could introduce confounds not fully controlled for.
The reinterpretation of prior ILC3 tolerance data also deserves careful handling. Several published studies attributed tolerogenic phenotypes to MHCII-expressing ILC3s using RORγt-Cre/MHCII-flox systems. The REALTAR data shows that among MHCIIhi RORγt lineage-traced cells, over 50% are in fact Aire-expressing non-ILCs (GFP+, CD90-negative). This is a meaningful caveat for interpreting those prior experiments, though it does not necessarily invalidate their conclusions about commensal tolerance.
A Conserved Program for Self-Education
The broader significance of this work lies in what it suggests about the architecture of immune tolerance. Central and peripheral tolerance have long been treated as mechanistically distinct processes, separated by the thymic exit. What this paper proposes, with substantial evidence, is that a core transcriptional and epigenomic program for self-antigen display operates in both compartments, driven by the same upstream signals (RANK-RANKL, noncanonical NF-κB), organized around the same chromatin-regulatory logic, and producing cells with strikingly similar functional properties despite arising from entirely different developmental lineages.
The practical implications branch in several directions. In autoimmunity, understanding what maintains JC and AmDC populations could point toward strategies for restoring peripheral tolerance in patients where it has broken down. In maternal-fetal immunology, the enrichment of placental antigens in JCs offers a molecular handle on why eTAC loss disrupts pregnancy tolerance. In cancer immunology, the finding that RANKL blockade depletes eTACs alongside mTECs means that tolerance-breaking strategies targeting this axis will need to account for peripheral as well as central effects.
For those of us who work with rare immune populations by flow cytometry, the surface phenotype data here is immediately useful. JCs are CCR7-low and CD11c-low, which means they would be missed by standard dendritic cell gating strategies that require high CD11c expression. The REALTAR mouse provides a genetic solution, but for human work, the CD200 enrichment on JCs relative to AmDCs may offer a practical handle worth exploring in peripheral blood and lymph node samples.
The Janus cell is a fitting name. Like the deity, it faces two directions at once: toward the dendritic cell lineage that defines its genomic identity, and toward the thymic epithelium whose program it has, through some convergent evolutionary logic, come to share. Whether that convergence reflects a deeply conserved mechanism for self-representation or a more contingent co-option of available regulatory machinery is a question this paper opens rather than closes. It is the right kind of question to be left with.