Stromal cell analysis and sorting We identified the four LNSC populations among CD45− cells using the aforementioned PDPN and CD31 markers, as well as HLA-DRhigh CD45low cells among CD45+ cells. The latter were also of interest as: (1) they had been included in previous LNSC studies because of their lower expression of CD45 and radioresistance; and (2) they are enriched in autoimmune regulator (AIRE)+ dendritic cells [16, 17]. In addition, human cells were stained for HLA-DR (MHC-II), HLA-A,B,C (MHC-I) and PD-L1, and mouse cells for I-Ag7 (MHC-II; detected with the cross-reacting anti-I-Ak antibody), H2-Kd (MHC-I) and PD-L1 (ESM Table 2). After staining, cells were acquired using a Fortessa cell analyser (BD, Franklin Lakes, NJ, USA) and FCS Express 6 (DeNovo Software, Glendale, CA, USA) was used for data analysis. Mean fluorescence intensity (MFI) values were normalised in each sample against CD45+ MHC-II− cells as control and represent a fold change. If the number of cells was sufficient, cell populations were sorted into TRI Reagent LS (Sigma, St Louis, MO, USA) using the BD Influx cell sorter and stored at −80°C.

Targeted analysis of gene expression in LNSCs from human PLNs using Biomark Total RNA was isolated from cells sorted in TRI Reagent LS, using a modified protocol of chloroform extraction followed by purification using an RNeasy Micro Kit (Qiagen, Hilden, Germany). cDNA was synthesised from 200 ng of total RNA for each sample using iScript Reverse Transcription Supermix (Bio-Rad, Hercules, CA, USA). All Delta Gene primers (Fluidigm, South San Francisco, CA, USA) were purchased pre-validated for assay performance (ESM Table 3). Samples were pre-amplified using a pool of all primers (minus endogenous controls; 50 nmol/l final) and 18 cycles, as per the Biomark protocol. Samples (16 in triplicates) were loaded onto Biomark 48 × 48 IFC chips (Fluidigm) and assayed against the 48 Eva Green-based assays (primers at 5 μmol/l final) listed in ESM Table 4. Target gene expression was calculated using the comparative method for relative quantification after normalisation to expression of the housekeeping HPRT1 gene, the expression of which was the most homogeneous across multiple samples compared with other housekeeping genes. An insufficient number of sorted cells, poor RNA quality (assessed using BioAnalyzer PicoChip; Agilent Technology, Waldbronn, Germany) or failed amplification were criteria for sample exclusion in the gene expression analysis. For comparison of relative gene expression between human and mouse LNSC subsets (our data vs Immunological Genome Project [ImmGen] RNA-Seq data [www.immgen.org]), we normalised gene expression to 100% in subsets in which it was most highly expressed in each set of data independently. Biomark data were deposited at: https://data.mendeley.com/datasets/d9rdzdmvyf/1 [18].