PSMB11 Orchestrates the Development of CD4 and CD8 Thymocytes via Regulation of Gene Expression in Cortical Thymic Epithelial Cells

This article is featured in In This Issue, p.627

Vertebrates express two main types of proteasomes: constitutive proteasomes and immunoproteasomes (1). These proteasomes regulate many basic cellular processes, including cell cycle progression and division, differentiation and development, morphogenesis, and response to stress via proteolysis of cellular proteins (2, 3). In addition, degradation of proteins by proteasomes generates peptides that are presented by MHC class I (MHC I) proteins (4, 5). More recently, Murata and colleagues (6) discovered that cortical thymic epithelial cells (cTECs) expressed a third type of proteasome that they named the thymoproteasome. Thymoproteasomes share two catalytic subunits with immunoproteasomes (PSMB9 and PSMB10) and contain a third catalytic chain unique to cTECs, PSMB11. PSMB11 is required for the generation of an adequate CD8 T cell protection (i.e., a number of self-tolerant CD8 T cells sufficient to effectively combat pathogens) (7). Indeed, Psmb11^−/− mice display defective development of CD8 thymocytes, and their postthymic CD8 T cells show impaired Ag responsiveness (8–10). Moreover, human Psmb11 polymorphisms affect CD8 T cell repertoire selection and may influence susceptibility to autoimmune diseases (10, 11).

Given that MHC I peptides (MIPs) regulate the selection of CD8 thymocytes, it was logical to posit that the key role of thymoproteasomes was to generate a unique MIP repertoire (8). Because thymoproteasomes have a lower chymotryptic activity than other types of proteasomes, it was initially suggested that they might generate peptides with hydrophilic C-termini and thereby a low MHC binding affinity (12). However, indirect approaches failed to support this idea; the cell surface density and t_1/2 of MHC I–peptide complexes are similar in wild-type (WT) and Psmb11^−/− cTECs (8, 9). To gain further insights into the impact of PSMB11 on the MIP repertoire, Sasaki et al. (13) sequenced ∼100 MIPs extracted from mouse embryonic fibroblasts transfected with Psmb8 or Psmb11 and treated with IFN-γ. MIPs from the two cell types showed similar MHC I binding affinity, hydrophobic C-termini, and anchor residues. However, analysis of 48 MIPs unique to Psmb8- or Psmb11-transfected cells revealed some discrepancies in residues located in the center of MIPs (P3 and P4). The mechanisms responsible for these differences and their potential significance remain elusive.

In this context, it is important to consider that the nonredundant functions of different types of proteasomes go well beyond MIP production and that a key role of proteasomal catalytic subunits is to regulate gene transcription via proteolytic destruction or activation of transcription factors (TFs) (1, 14). Indeed, replacement of constitutive proteasomes by immunoproteasomes has pleiotropic effects on cell functions, including proliferation, survival, and differentiation (3), and may even enhance organismal lifespan (15). In dendritic cells (DCs), for example, immunoproteasomes and constitutive proteasomes differentially regulate the proteolysis of numerous TFs that modulate the expression of thousands of genes. Furthermore, even when engineered to present optimal amounts of MIPs, immunoproteasome-deficient DCs are inefficient for T cell priming (3). These considerations prompted us to evaluate whether PSMB11 might have MIP-independent effects on cTEC biology. We report that PSMB11 regulates the expression of 850 genes in cTECs mainly by repressing WNT signaling. Deletion of Psmb11 perturbs cTEC differentiation, impinges on MIP-independent cell–cell interactions, and has pervasive effects not only on CD8 but also on CD4 thymocytes.

B6.B6CB-Psmb11^tm1Smta (Psmb11^−/−) mice were obtained from the Riken BioResource Center, and B6.129S2-H2^dlAb1-Ea/J (MHCII^−/−) and B6.129P2-B2m^tm1Unc/J (β2m^−/−) mice were purchased from The Jackson Laboratory. Psmb11^−/− mice used in this study were obtained by mating Psmb11^+/− parents, and unless stated otherwise, Psmb11^+/+ littermates were used as WT controls. Mice were maintained under specific pathogen-free conditions at the Institute for Research in Immunology and Cancer and at the Maisonneuve-Rosemont Hospital Research Center. All procedures were in accordance with the regulations of the Canadian Council on Animal Care and the University of Montreal.

Thymi were isolated and stromal cell enrichment was performed as previously described (16). Briefly, thymi were mechanically disrupted and digested with DNase I (Sigma-Aldrich) and Liberase (Roche). Single-cell suspensions were stained with biotinylated Ulex europaeus lectin 1 (UEA1; Vector Laboratories), PE-Cy7–conjugated streptavidin (BD Biosciences), and the following mAbs: Alexa Fluor 700 anti-CD45, PE anti–I-Ab, allophycocyanin-Cy7 anti-EpCAM, and Alexa Fluor 647 anti-Ly51 (BioLegend). Thymocytes were extracted by mechanical force. CCR7 staining on thymocytes was performed for 20 min at 37° with anti-CCR7 PE (4B12; eBioscience) or PE-Cy5 (4B12; BioLegend), followed by surface marker staining for 25 min with the following Abs: anti-CD4 AF700 (RM4-5; BD Biosciences), anti-CD8α V500 (53-6.7; BD Horizon) or BUV395 (53-6.7; BD Biosciences), anti-TCRβ PE/Dazzle594 (H57-597; BioLegend), anti-MHC I allophycocyanin (AF6-88.5.53; eBioscience) or PE (AF6-88.5; BD Biosciences), anti-CD25 allophycocyanin-Cy7 (PC61; BioLegend), CD1d–α-GalCer tetramers BV421 (National Institutes of Health), anti-TCRγδ BV421 (GL3; BioLegend), anti-CD44 BV650 (IM7; BioLegend), anti-CD69 PE-Cy7 (H1.2F3; BD Biosciences), anti-CD5 PE (53-7.3; BD Biosciences), anti-CD8β PE (H35-17.2; BD Biosciences), and anti-CD127 PE (A7R34; eBioscience). Staining with anti-CXCR4 PE (2B11; eBioscience) was performed for 40 min at 4° in staining buffer (PBS with 0.1% BSA). Cell viability and early apoptosis were assessed using 7-aminoactinomycin D (7-AAD) (BD Biosciences) and Annexin V Alexa Fluor 350 (Thermo Fisher). Flow cytometry was performed on a ZE5 (Bio-Rad) for thymocytes and on an LSR II (BD Biosciences) for stromal cells. Cell sorting was performed using a three-laser FACSAria (BD Biosciences), and all data were analyzed using the FACSDiva or FlowJo software.

Thymocytes were freshly isolated and stained with 500 μM CellROX Orange Reagent (Molecular Probes) for 45 min in RPMI 1640 (Thermo Fisher Scientific) with 10% FBS, followed by a wash step with PBS and staining for surface markers.

Frozen tissues were embedded in O.C.T. Compound (Sakura Finetek) and sliced into 8-μm-thick sections. Sections were fixed with 10% formalin (Chaptec) for 10 min and stained with rabbit anti-HSP70 (polyclonal; Novus Biologicals) for 1 h, followed by 30 min staining with anti-rabbit Alexa Fluor 594 (Molecular Probes) and DAPI at the end (Life Technologies). All staining steps were preceded by a wash step with PBS. Omission of primary Ab and replacement of the primary Ab by purified rabbit polyclonal isotype (BioLegend) were used as controls. Images were acquired with an LSM700 confocal microscope (Zeiss) with 40× magnification and analyzed using ImageJ.

The ImageStream^X Mark II Imaging Flow Cytometer (Amnis, Seattle, WA) was used to quantify the localization of β-catenin in cTECs. Briefly, freshly isolated thymic epithelial cells (TECs) were enriched using an LS Column and CD326 (EpCAM) MicroBeads (Miltenyi Biotec). After staining for extracellular markers, cells were stained for anti–β-catenin allophycocyanin Ab (Miltenyi Biotec) using a Foxp3 staining buffer kit (Miltenyi Biotec) following the manufacturer’s instructions. After washing, cells were resuspended in PBS with 0.5% BSA containing 0.1 μg/ml DAPI prior to acquisition of 50,000 total events on the ImageStream. All compensation and data analyses were performed postacquisition using the IDEAS 6.2 software package (Amnis). Each β-catenin⁺ cTEC was assessed for cytoplasmic or nuclear localization of β-catenin–allophycocyanin signal. For quantifying nuclear levels, the fluorescence image-based method relies on the spectral isolation of the β-catenin signal image from the DAPI-stained nuclear image. For quantifying cytoplasmic levels, we used the EpCAM allophycocyanin-Cy7 signal to define membrane localization and the DAPI-stained nuclear image; first, we eroded the EpCAM channel by a system mask of six pixels to eliminate the β-catenin signal associated with the cell membrane, and then we subtracted the nuclear β-catenin signal.

Thymic lobes from Psmb11^+/−, Psmb11^−/−, and β2m^−/− mice were embedded in 4% low-melt agarose and cut into 500-μm-thick thymic slices using a Leica VT1000 S vibrating blade microtome (17, 18). Slices were transferred to a 0.4-μm cell culture insert (BD Falcon) over 1.5 ml of complete RPMI 1640 media containing 10% FBS, 4 mM l-glutamine, and 1× penicillin/streptomycin (Wisent) in a six-well plate, and excess PBS was removed. Thymocytes from WT mice were harvested and incubated with Abs specific for murine TCRβ (H57-597), CD4 (GK1.5), and CD8α (53-6.7) (BioLegend) prior to cell sorting. Sorted thymocytes were labeled with 1 μM SNARF (Thermo Fisher Scientific), and 0.5–1.0 × 10⁶ cells were overlaid atop Psmb11^+/−, Psmb11^−/−, and β2m^−/− slices. All thymic slices were incubated at 37°C for 5 h before washing off the excess cells that did not migrate into the tissue and were further incubated for 16 h, allowing the cells to migrate to either cortex or medulla. Thymic slices were fixed for 5 min with 4% formaldehyde (Thermo Fisher Scientific) and passed through a sucrose gradient. The slices were then embedded in frozen section compound (VWR) over dry ice and kept at −80°C until they were cryosectioned (CM3050S; Leica). Ten-micrometer cryosections of each thymic slice were washed with PBS and stained with polyclonal anti-keratin 5 (BioLegend) Ab overnight, followed by 2 h staining with donkey anti-rabbit IgG Alexa Fluor 647 as secondary Ab in a humidified chamber at 4°C. The sections were subsequently mounted in ProLong Gold Antifade Mountant with DAPI (Thermo Fisher Scientific). Images were obtained with an Olympus FluoView1000 confocal microscope. Investigators were blinded to the sample identity in the outlining of medulla and cortex areas in thymic slice experiments. Quantification of the relative density of overlaid WT TCRβ^hi CD8⁺ CD4⁻ thymocytes between thymic cortex and medulla was performed by first counting the number of SNARF⁺ thymocytes using the ImageJ ComDet version 0.3.6.1 plugin (tissue margins with unspecific signal were automatically excluded from the analysis), then measuring the areas of cortex and medulla within each image based on DAPI, GFP reporter, and/or keratin 5 signals in the cortex or medulla by a custom-written MATLAB program (available upon request).

For thymocytes, mice received two doses of 1 mg of BrdU (BD Biosciences) 2 h apart by i.p. injection, and BrdU incorporation was detected 24 h after the first injection. For TECs, mice received one daily i.p. injection of 1.5 mg of BrdU for 3 d, and BrdU uptake was detected 24 h after the last injection. BrdU was detected using an APC BrdU Flow Kit (BD Biosciences) following the manufacturer’s instructions.

Total RNA was isolated from three technical replicates (three mice per genotype) using TRIzol (Life Technologies). A minimum of 3 × 10⁴ TECs and 1 × 10⁵ thymocytes were used for each replicate, yielding an average of 50 and 75 ng of RNA, respectively. RNA-extracted samples were treated with DNAse I and purified using an RNeasy Mini Kit (QIAGEN) for TECs or purified using the RNeasy Micro Kit (Qiagen) for thymocytes, following the manufacturers’ instructions. Sample quality was assessed using 2100 Bioanalyzer RNA Pico chips (Agilent Technologies). Transcriptome libraries were generated from 50 ng of total RNA for TECs and 75 ng of total RNA for thymocytes using the KAPA stranded mRNA kit (KAPA Biosystems) following the manufacturer’s protocol. Paired-end sequencing was performed using an Illumina HiSeq200 sequencer for thymocytes (2 × 100 nt) and medullary TECs (mTECs) (2 × 75 nt) and by using an Illumina NextSeq 500 sequencer for cTECs (2 × 75 nt). Sequences were trimmed for sequencing adapters and low quality 3′ bases using Trimmomatic version 0.35 and aligned to the reference mouse genome version GRCm38 (or mm10) using STAR version 2.5.1b.

Genes whose expression was <1 fragments per kilobase of transcript per million mapped reads (FPKM) in all samples were removed. Differential gene expression between cell subsets from WT and Psmb11^−/− mice was determined using the R version 3.0.1 software (http://www.r-project.org/) with the DESeq2 package. DESeq2 applies a p value adjustment (Benjamini–Hochberg correction) to control for multiple comparisons. To select differentially expressed genes (DEGs), we filtered genes based on a familywise p-adjusted value of 0.1, meaning that we allowed only 10% of the independent genes with significant p values (p < 0.05) to be false positives. Thus, DEGs were extracted based on an adjusted p value < 0.1 and a minimal fold change of 1.4 for TECs and a fold change of 2 for thymocytes.

DEGs from each cell subset were submitted to DAVID 6.8 (http://david.ncifcrf.gov). Significantly enriched biological processes and KEGG pathways (p < 0.05 with Benjamini–Hochberg) were kept, and redundant categories were removed using REVIGO (http://revigo.irb.hr).

The Molecular Signatures DataBase v6.1 (Broad Institute) was used as source of TF target (TFT) gene sets (C3 collection). TFT gene sets (615 gene sets) were downloaded for analysis in the R statistical environment version 3.4.3. DEGs in Psmb11^−/− cTECs were separated into two groups: genes upregulated or downregulated in Psmb11^−/−. Mouse gene symbols were converted to human gene symbols using BiomaRt library for intersection with MSigDB data. Contingency tables were created using the number of genes differentially expressed and present (or not) in a given TFT gene set. Two-tailed Fisher test was performed on each table to retrieve odds ratio and p values, and the false discovery rate (FDR) was evaluated. Gene sets with an FDR <0.05 and odds ratio >2 were kept as significant. TF for which the source mRNA mean expression was <10 read counts across all conditions were flagged as not expressed and removed from further analyses. MSigDB TF names were manually mapped back to source gene names.

Enrichment analyses were also performed using the oPOSSUM tool for TF binding site motif enrichment version 3.0 (http://opossum.cisreg.ca/cgi-bin/oPOSSUM3/opossum_mouse_ssa). For oPOSSUM, genes upregulated in Psmb11^−/− were fed to the Mouse Single Site Analysis using default parameters and cutoffs. Both TF binding site searches were performed on the 2-kb upstream and downstream of the transcription start site of the DEGs. Only genes that were expressed at a minimum level (combined number of read counts >10 for all samples) were used as background. Supplemental Material 2 contains all results separated into two groups: z-score ≥5 and <5.

Statistical significance was determined using the GraphPad Prism software and calculated using a two-tailed unpaired Student t test unless stated otherwise: *p < 0.05, ** p < 0.01, and ***p < 0.001. All data were reported, including outliers.

RNA-sequencing (RNA-seq) data are available from the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE107534, GSE107535, and GSE107536.

To assess the overall impact of thymoproteasomes on the transcriptome of TECs, we performed RNA-seq on cTECs and mTECs from WT and Psmb11^−/− mice. Subsets of TECs (EpCAM⁺CD45⁻) were defined according to their cell surface phenotype: Ly51⁺UEA1^– for cTECs and Ly51^–UEA1⁺ for mTECs (Supplemental Fig. 1A). In cTECs, we found that among genes with an expression over one fragment per kilobase of transcript per million mapped reads (FPKM), 850 were DEGs with an adjusted p < 0.1 and a fold change >1.4 (Supplemental Fig. 1B, 1C, Supplemental Material 1). Of these, 680 and 170 DEGs were over- and underexpressed in Psmb11^−/− relative to WT cTECs, respectively (Fig. 1A). Hence, PSMB11 is predominantly a repressor of gene expression. Of note, Psmb11 deletion had no impact on the transcriptome of MHC class II (MHC II)^lo or MHC II^hi mTECs (mTEC^lo and mTEC^hi; Supplemental Fig. 1C). This means that in adult mice, transcriptional regulation by PSMB11 takes place exclusively in cTECs. Two inferences can be made from this. First, the 850 DEGs in cTECs do not result from random differences in the genetic background of the animals. Indeed, the probability that random genetic variations would affect only cTECs (but not mTEC^lo and mTEC^hi cells) is equal to (1/3)⁸⁵⁰ ≈ 0. Second, although adult mTECs originate from progenitors that express PSMB11 early in ontogeny (<1 wk of age) (19, 20), they keep no trace of this transient PSMB11 expression. Furthermore, analyses of in vivo BrdU incorporation and annexin V staining showed no differences between WT and PSMB11-deficient cTECs (Supplemental Fig. 1D–F). Hence, PSMB11 is not essential for the maintenance of cTEC homeostasis, and differential gene expression in PSMB11-deficient cTECs was not caused by cTEC-intrinsic changes in proliferation or survival.

Gene ontology (GO) term and KEGG pathway enrichment analyses revealed that a large proportion of the 850 PSMB11-regulated DEGs are involved in interactions between TECs and thymocytes (e.g., chemotaxis, cell–cell adhesion, and extracellular matrix [ECM] organization) (Fig. 1B). The TNF-induced genes Cxcl2, Cxcl9, and Ccl5, which serve as chemoattractants for hematopoietic progenitors to the thymus or for T cell emigration (21) were upregulated in cTECs from Psmb11^−/− mice (Fig. 1C). In WT thymi, these chemokines are primarily expressed in mTECs (22). In addition, the transcript levels of CCL25 and CXCL12, which are major chemokines secreted by cTECs (23), decreased in the absence of PSMB11. These results suggest that, relative to WT cTECs, the chemokine expression profile of Psmb11^−/− cTECs is more “mTEC-like.”

Positive selection is accompanied by periods of migration interrupted by brief migratory pauses (24). Acting in synergy with chemokines, the ECM components facilitate thymocyte migration and compartmentalization and also support thymocyte adhesion to TECs, with the thymic medulla containing a denser laminin network than the cortex (23, 25, 26). We found that several groups of ECM-related genes were upregulated in Psmb11^−/− cTECs (Fig. 1C). The first group includes the integrin subunit α m gene (Itgam) as well as fibronectin (Fndc1, Fndc7) and collagen (16 genes) components (Fig. 1C). A second group of DEGs includes laminin and laminin-associated genes: Lama1, Lamb2, Nid1, Ntn4. By analogy to the laminin-related protein Ntn1 (27), we surmise that the products of these genes may increase thymocyte adhesion in the cortex. Akin to laminins, the expression of HpsE, Vit, Parvb, and Mmp3 were elevated in cTECs lacking thymoproteasomes and could promote cell adhesion by degradation of the ECM. Moreover, several cadherin-encoding genes involved in cell adhesion and whose expression is stronger in WT mTECs than WT cTECs (28) were upregulated in cTECs lacking PSMB11 (Cdh13, Cdhr1, Pcdh17, Cdhr5). Lastly, the claudin-encoding transcripts Cldn3 and Cldn4, expressed only in mTEC^hi cells in WT thymi (29), were upregulated in Psmb11^−/− cTECs. Hence, in the absence of PSMB11, cTECs upregulate genes that orchestrate interactions between thymocytes and the thymic stroma (TECs and the ECM). Collective upregulation of these genes is likely to affect the migration pattern of thymocytes and their cross-talk with cTECs.

Lastly, we noted that several members of the TNF (Tnfrsf21, Tnfrsf8, Tnfsf10, Lif) and TGF (Tgfb2, Gdf7) families were overexpressed in PSMB11-deficient cTECs (Fig. 1C). These gene families may certainly affect thymocyte maturation; TNF-dependent activation of NF-κB is necessary for thymocyte survival and transition to mature stages (30), whereas TGF-β can induce expression of the CD8 lineage–specifying TF Runx3d (31). The Notch effector Dll4, which is instrumental for early thymocyte differentiation and T cell fate specification (32), was also downregulated in cTECs lacking PSMB11 (Supplemental Material 1). However, further studies will be required to determine whether these DEGs are functionally important for the Psmb11^−/− phenotype.

As mentioned above, the expression profile of chemokine and claudin genes in Psmb11^−/− cTECs presented obvious similarities to that of WT mTECs (Figs. 1C, 2A). Furthermore, we found that four genes coding for prototypical cTEC markers Enpep, Ly75, Ctsl, and Prss16 (33) were expressed at lower levels in Psmb11^−/− than WT cTECs (Fig. 2A). We therefore asked whether the presence or absence of PSMB11 influenced the global divergence between cTECs and mTECs. To assess the overall impact of PSMB11, we compared for the 850 DEGs, the difference in expression between MHC II^lo mTECs (mTEC^lo), and WT versus Psmb11^−/− cTECs. We compared cTECs to mTEC^lo because, relative to MHC II^hi mTECs, mTEC^lo are ontogenically closer to cTECs (34). We found that the divergence between WT mTEC^lo and Psmb11^−/− cTECs was significantly smaller than the divergence between WT mTEC^lo and WT cTECs (p < 2.2 × 10⁻¹⁶; Fig. 2B, 2C). This effect was also observed, albeit to a lesser extent, when we considered the entire TEC transcriptome [excluding promiscuously expressed genes (35)] (Supplemental Fig. 1F). Hence, loss of PSMB11 renders cTECs more similar to mTEC^lo.

Because proteasomes regulate gene expression via proteolytic destruction or activation of TFs (1, 14), we asked whether specific TF binding motifs were enriched in cTEC DEGs. Using TF gene sets from the Molecular Signatures Database, we performed enrichment analyses on the DEGs over- and underexpressed in Psmb11^−/− cTECs, respectively (see 2Materials and Methods). No TF motifs were significantly enriched in the DEGs downregulated in Psmb11^−/− cTECs, consistent with the predominantly repressive impact of PSMB11 on gene expression (Fig. 1). In contrast, several TF motifs were enriched in the DEGs overexpressed in Psmb11^−/− cTECs (Fig. 2D, Supplemental Material 2). The most conspicuous enrichment involved effectors of the WNT pathway (36): TCF3 (p = 6.3 × 10⁻²⁵), JUN (p = 1 × 10⁻¹⁴), and LEF1 (p = 7 × 10⁻⁷). In line with this, we also found a major enrichment for motifs recognized by modulators of the WNT pathway such as NFATs (p = 3.5 × 10⁻¹⁰), ETS2 (p = 7.4 × 10⁻⁹), FOXA1-3 (p = 3.1 × 10⁻⁵), and RUNX1 (p = 9 × 10⁻⁶). Most hits, including TCF3, JUN, and NFAT, were also significantly enriched using other TF binding site databases (Supplemental Material 2). Furthermore, four mTEC-specific TFs (34) (Ehf, Fezf2, Pou2f3, and Spib) were found to be upregulated in Psmb11^−/− cTECs (Supplemental Material 1) and to harbor binding motifs for TCF3 and other WNT effectors (Supplemental Material 2). These data demonstrate that genes overexpressed in Psmb11^−/− cTECs display strong enrichment for selected TF binding motifs. Moreover, they strongly suggest that enhanced WNT signaling is instrumental in the acquisition of mTEC features by Psmb11^−/− cTECs. WNT signaling is normally repressed in cTECs (in contrast to thymocytes), and upon overactivation of WNT signaling, cTECs acquire features of mTECs (36, 37). The key switch in the canonical WNT pathway is the cytoplasmic protein β-catenin, whose abundance is regulated by proteasomal degradation (36). Consistent with our TF analyses, flow cytometry analyses revealed a marked increase in the abundance of cytoplasmic and nuclear β-catenin (CTNNB1) in Psmb11^−/− cTECs (Fig. 2E, 2F, Supplemental Fig. 1H). This indicates that the PSMB11 proteasomal subunit increases β-catenin degradation and therefore inhibits WNT signaling in cTECs. We conclude from the above results that PSMB11-mediated control of the WNT pathway is necessary for the maintenance of cTEC hallmarks in the thymus.

To assess the impact of Psmb11 deletion on αβ T cell development, we analyzed the semimature (SM), mature 1 (M1), and mature 2 (M2) subsets of postselection thymocytes defined by Hogquist and colleagues (38) (Fig. 3A). We found that the previously reported decrease in CD8 thymocytes in Psmb11^−/− mice (6, 39) was conspicuous at all three stages of maturation (Fig. 3B) and must therefore affect early events in induction of the CD8 lineage fate.

Unexpectedly, although the percentage of total CD4 single-positive (SP) thymocytes was similar between the two genotypes (Supplemental Fig. 2A), 4M2 cells were significantly more abundant in Psmb11^−/− relative to WT mice (Fig. 3C). Expansion of the 4M2 subset in these mice was MHC II–dependent and not due to an erroneous commitment of CD8 T cells to the CD4 lineage. Indeed, in MHC II–deficient mice, the presence or absence of PSMB11 did not affect the number of 4M2 cells (Supplemental Fig. 2B). We conclude that PSMB11 influences the development of both CD4 and CD8 thymocytes.

In an effort to understand the mechanisms responsible for the altered cellularity in the CD4 and CD8 cell subsets in Psmb11^−/− mice, we analyzed their proliferation and apoptosis rate. To examine the proliferation ability, we performed in vivo BrdU labeling in young adult mice. BrdU was pulsed twice at 2-h intervals and was chased for 24 h to allow incorporation into newly synthesized DNA. The percentage of BrdU⁺ cells in Psmb11^−/− and WT mice was similar for all CD4 and CD8 subsets except for 8SM, which contained more BrdU⁺ cells in Psmb11^−/− thymi (Fig. 3D, Supplemental Fig. 2C). The early apoptosis status was assessed using annexin V staining. We found that a significantly higher percentage of 8M1 and 8M2 thymocytes were apoptotic in the absence of PSMB11 (Fig. 3E, Supplemental Fig. 2D). Our results indicate that proliferation of CD8 thymocytes selected in the absence of PSMB11 is unimpaired, but these cells are more susceptible to apoptosis. The increased proportion of BrdU⁺ cells in Psmb11^−/− 8SM likely represents a compensatory reaction to the decreased number of CD8 thymocytes. The mechanism responsible for the increased number of 4M2 in Psmb11^−/− thymi was not elucidated because it could be ascribed neither to enhanced proliferation nor to decreased apoptosis.

To gain a global understanding of the commitment of thymocytes to the CD4 and CD8 lineages in the absence of PSMB11, we carried out RNA-seq on discrete thymocyte subsets and used the DESeq2 package in the R software to extract DEGs between the WT and Psmb11^−/− thymocytes. DEGs were defined as genes with an adjusted p < 0.1 and an expression fold change >2.

The most striking finding was a 20–80-fold upregulation of Hspa1a and Hspa1b transcripts, which code for the stress-responsive HSP70 chaperone in all CD4 and CD8 thymocyte subsets from Psmb11^−/− mice (Fig. 4A). To validate this observation at the protein level, we performed immunofluorescence staining for HSP70 on thymic slices from 4-wk-old mice. We compared the HSP70 intensity in the medulla and cortex between Psmb11^−/− and WT thymi (Supplemental Fig. 2E), and we detected a significant increase in the HSP70 levels in the medulla but not in the cortex of Psmb11^−/− thymus sections (Fig. 4B, 4C). Because the majority of the cells in the medulla are SP thymocytes, these results demonstrate increased cellular stress in CD4 and CD8 thymocytes selected in the absence of PSMB11.

The expression of HSP70 chaperones increases in response to various stressful stimuli, especially those causing oxidative stress or endoplasmic reticulum stress (40). We did not find signs of endoplasmic reticulum stress in the DEG dataset according to biomarker gene expression (41). However, thymocytes from Psmb11^−/− mice showed upregulation of several oxidant-induced transcripts, namely Fos, Jun, Hsph1, Hspa1a, Hspa1b, Gadd45, and Hmox1 (heme oxygenase) (42) (Fig. 4A, 4D). Additionally, several genes involved in the apoptotic process in response to oxidative stress were also upregulated in the absence of PSMB11: Rhob, Pmaip1, and Nr4a2 (43–45) (Fig. 4D). Most of these markers of oxidative stress were upregulated in both CD4 and CD8 SP thymocytes. Finally, flow cytometry analyses revealed increased amounts of reactive oxygen species (ROS) in all subsets of CD4 and CD8 SP thymocytes from Psmb11^−/− relative to WT mice (Fig. 4E, Supplemental Fig. 2F). We conclude that SP thymocytes undergo oxidative stress in Psmb11^−/− thymi.

We next performed more detailed analyses of RNA-seq data from CD8 thymocytes. We identified 284 DEGs in CD8 thymocytes from WT versus Psmb11^−/− mice, with limited overlap between thymocyte subsets (Supplemental Fig. 3A, Supplemental Material 3). GO term analysis of the DEGs in each CD8 thymocyte subset revealed an enrichment in cell adhesion and surface receptor signaling processes (Fig. 5A), coherent with the differential expression of cell adhesion and cytokine genes in cTECs (Fig. 1C). In addition, CD8 thymocytes from Psmb11^−/− thymi showed aberrant expression of genes involved in lineage commitment (Supplemental Material 3, Supplemental Table I). For example, 8SM and 8M1 thymocytes displayed a lag in the downregulation of Zbtb7b, whereas both 8M1 and 8M2 subsets upregulated Cd4 and transcripts restricted to regulatory T cells, namely Tigit and Foxp3 (Supplemental Fig. 3B, Supplemental Table I). This altered transcriptional profile supports the idea that CD8 T cells receive stronger MHC I–TCR signals in the PSMB11-deficient cortex (9). This inference aligns with the higher expression of negative selection–associated genes in Psmb11^−/− cTECs (Fig. 1C), including Pvr (CD155, which has high affinity for TIGIT) (46). The expression level of Ctla4 and of genes involved in apoptosis (Faslg, Pdcd1, Gadd45b) was also elevated in 8M1 and 8M2 thymocytes from mice lacking PSMB11, in accordance with the increased percentage of annexin V⁺ cells in the CD8 lineage (Fig. 3E, Supplemental Table I). The fact that CD8 thymocytes skew to some extent their expression profile to that of unconventional T cell lineages is consistent with reports that CD8 selection is perturbed in Psmb11^−/− thymi (10, 39, 47) and suggests that this might be linked to strong TCR signals.

Comprehensive analyses of CD8 thymocyte–positive selection unveiled two phases of surface CD8 expression (48). In phase 1, TCR signaling intensity determines the extent of CD8 loss; stronger TCR signaling accelerates the decline in CD8 surface expression at the intermediate maturation stage (CD4⁺CD69⁺CCR7⁺). Phase 2 of positive selection corresponds to cytokine signaling, which fully restores CD8 expression. To further evaluate the hypothesis that CD8 thymocytes receive stronger TCR signals in Psmb11^−/− thymi, we examined Nr4a1 (encoding NUR77) and CD5, whose expression correlates with the strength of TCR signaling (9). Consistent with previous analyses on total CD8 T cells (9), we observed that 8SM, 8M1, and 8M2 thymocytes displayed increased expression of Nr4a1 transcripts and cell surface CD5 levels in the absence of PSMB11 (Fig. 5B, 5C, Supplemental Fig. 3C). In addition, we found that surface CD8β expression did not reach normal levels in the 8M1 and 8M2 subsets of Psmb11^−/− mice, suggesting a defect in phase 2 of positive selection (Fig. 5D, Supplemental Fig. 3C). IL-7 and TGF-β are two cytokines that are instrumental in phase 2 (31). Flow cytometry analyses revealed that 8M1 and 8M2 thymocytes from Psmb11^−/− thymi had slightly but significantly lower levels of surface CD127 (Fig. 5E, Supplemental Fig. 3C). In addition, a 3.5-fold downregulation in the expression of Itgae, which is specifically induced by TGF-β signaling (31), was observed in the 8SM subset from Psmb11^−/− mice (Supplemental Table I). Overall, these data suggest that in Psmb11^−/− thymi, CD8 thymocytes receive strong TCR signals but show evidence of impaired cytokine signaling thereafter.

Lack of PSMB11 does not perturb the cortical and medullary compartmentalization in the thymus (6). However, several observations led us to postulate that migration of CD8 lineage committed thymocytes from the cortex to the medulla might be perturbed in the absence of PSMB11. First, the largest families of DEGs in Psmb11^−/− cTECs are implicated in chemokine signaling and cell adhesion (Fig. 1B, 1C), and cTECs acquire features of mTECs in the absence of PSMB11 (Fig. 2A–C). Second, several genes regulating thymocyte migration were modulated by PSMB11 in CD8 thymocytes (Fig. 5A), including CXCR4, which was upregulated in Psmb11^−/− 8SM and 8M1 (Supplemental Fig. 3D, Supplemental Table I). Third, cytokine-dependent restoration of surface CD8 during phase 2 of positive selection, which takes place in the medulla, is deficient in Psmb11^−/− thymi (Fig. 5E). To determine the impact of the gene expression changes in thymus of Psmb11-deficient mice, TCRβ^hi CD8 SP thymocytes from WT mice were overlaid on thymic slices from Psmb11^−/− or Psmb11^+/− littermates. As anticipated, CD8 SP thymocytes predominantly localized to the medulla on thymic slices from Psmb11^+/− mice. However, this preferential localization was abrogated on slices from Psmb11^−/− thymi (Fig. 6). Limitations in cell numbers precluded us from performing these experiments with Psmb11^−/− CD8 thymocytes. Nonetheless, our data show that CD8 thymocyte localization is dysregulated when cTECs do not express Psmb11. From this, we infer that migration of CD8 thymocytes from the cortex to the medulla must be defective in Psmb11^−/− thymi and that prolonged retention of CD8 thymocytes in the Psmb11^−/− cortex ought to impair phase 2 of positive selection.

Theoretically, retention of CD8 thymocytes in the cortex could depend on lack of positively selecting MIPs and/or on MIP-unrelated changes in the transcriptome of cTECs. To determine whether lack of positively selecting MIPs could explain the mislocalization of thymocytes, we overlaid TCRß^hi CD8 SP thymocytes from WT mice on β2-microglobulin^−/− (B2m^−/−) thymic slices. In contrast to Psmb11^−/− slices, CD8 thymocytes largely localized to the medulla of B2m-deficient thymic slices lacking MIP presentation (Fig. 6). This demonstrates that the PSMB11-mediated gene expression in cTECs is essential for the proper migration of SP thymocytes to the medullary compartment, whereas the MIP repertoire is dispensable for this process.

A total of 582 DEGs were identified between WT and Psmb11^−/− CD4 thymocyte subsets (Supplemental Fig. 3E, Supplemental Material 3). At all three stages of maturation, these DEGs were primarily involved in biosynthetic and gene expression processes (Fig. 7A). The most enriched GO term, heterocycle biosynthetic process, contained a large set of zinc finger protein–encoding genes that were repressed in the 4M2 subset of Psmb11^−/− mice (Supplemental Fig. 3F). To our knowledge, the function of these zinc finger genes in T cell development is unknown. Nonetheless, considering the occurrence of oxidative stress in PSMB11-deficient SP thymocytes, a salient finding was the downregulation of 10 mitochondrial (Mt) genes in the 4M2 subset (Fig. 7B). Indeed, mitochondria are hypersensitive to oxidative stress, and the suppression of mitochondrial gene expression facilitates protection from cellular injury induced by ROS (49, 50). Moreover, mitochondria become the prime target of the heat shock proteins–mediated defense against ROS (49, 51). In line with this, we found that, unlike SP8 thymocytes, CD4 thymocytes upregulated a group of genes encoding cochaperones of HSP70, suppressors of apoptosis (Areg, Serpine1, Ppp1r15l) and molecules involved in the shut-off of protein synthesis in response to stress (Ppp1r15a) (Supplemental Table I). We reason that the conspicuous modulation of gene expression in 4SM, 4M1, and 4M2 subsets represents an adaptive response to survive the major cellular stress induced by Psmb11^−/− cTECs (Fig. 4).

The defect in CD8 lineage T cell development in the absence of PSMB11 has, to date, been attributed to differences in the peptide repertoire presented by MHC I molecules (9, 13). The main conclusion from this work is that PSMB11 has dramatic effects on the transcriptome of cTECs as well as on that of both CD4 and CD8 thymocytes. By definition, the effects of PSMB11 on the transcriptome of CD4 and CD8 thymocytes is cell-extrinsic. Because the transcriptome of mTECs was not affected by the lack of PSMB11, we assume that the differential gene expression in Psmb11^−/− cTECs is a cell-autonomous process driven by differential proteolysis of TFs. Nonetheless, it is formally possible that some cTEC DEGs result from perturbed cross-talk between Psmb11^−/− cTECs and thymocytes.

PSMB11 modulates the expression of 850 genes in cTECs, 582 in CD4 thymocytes, and 284 in CD8 thymocytes. Although further studies will be required to understand the mechanisms responsible for such widespread changes in gene expression and to precisely evaluate the role of specific DEGs in thymocyte development, several major themes emerge from our analyses. A salient finding is that, in the absence of PSMB11, cTECs lose part of their distinctness relative to mTECs; they downregulate several cTEC genes and upregulate a set of mTEC genes. Based on flow cytometry and TF binding site analyses, we found that PSMB11 preserved the integrity of the cTEC-specific transcriptome by inhibiting WNT signaling via degradation of β-catenin. This conclusion is consistent with the observation that constitutive WNT activation in TECs disrupts their function and skews cTECs toward an mTEC phenotype (37). The sole difference between immunoproteasomes (found in mTECs) and thymoproteasomes (present in cTECs) is the nature of their chymotryptic catalytic subunit: PSMB8 in immunoproteasomes and PSMB11 in thymoproteasomes. Of note, thymoproteasomes are replaced by immunoproteasomes in Psmb11^−/− cTECs (8). From this, we infer that thymoproteasomes and immunoproteasomes are instrumental in establishing the unique differentiation program of cTECs versus mTECs. PSMB11-regulated DEGs in cTECs regulate, in large part, cell–cell adhesion, ECM organization, and thymocyte chemotaxis. Consistent with this, in thymic slice cultures, the proportion of CD8 thymocytes that accumulated in the thymic cortex was increased by ∼7-fold in PSMB11-deficient thymi. Importantly, the medullary localization of CD8 thymocytes on B2m-deficient thymic slices was not affected, which underscores the importance of the MIP-independent role of PSMB11 in thymocyte development. Retention in the cortex should decrease migration of CD8 thymocytes from the cortex to the medulla in Psmb11^−/− thymi, a crucial step for the positive selection and maturation of CD8 T cells (52, 53). Concretely, we posit that PSMB11 deficiency leads to a prolongation of phase 1 (in the cortex) and deficient phase 2 (in the medulla) in positive selection of CD8 thymocytes.

CD8 thymocytes are affected most severely by the absence of PSMB11 as compared with CD4 cells. They suffer massive cellular oxidative stress, acquire features of unconventional T cell lineages, and undergo apoptosis. It is impossible to untangle the extent to which the failure of CD8 thymocytes to develop in Psmb11^−/− thymi is contingent upon alterations in the MIP repertoire (13, 47) versus perturbed translational regulation in cTECs. Nevertheless, the extent of differential gene expression between WT and Psmb11^−/− cTECs strongly suggests a key role for MIP-independent effects of PSMB11 on T cell development. This assertion is supported by analogous observations in DCs. Immunoproteasome-deficient DCs show an altered MIP repertoire and a perturbed gene expression profile and are poor APCs (3, 5). It was initially postulated that deficient Ag presentation by immunoproteasome-deficient DCs depended on their altered MIP repertoire (4). However, it was subsequently shown that differential gene expression was sufficient to explain defective Ag presentation by immunoproteasome-deficient DCs (3).

The current paradigm holds that PSMB11 deficiency impinges only on CD8 T cell development. However, our data demonstrate that a lack of PSMB11 causes intense oxidative stress in CD4 thymocytes and has a dramatic effect on their transcriptome. On close examination, some hints that PSMB11 might affect CD4 T cells could be found in previous reports; Nitta et al. (8, 11) have reported a significant increase in the number of CD4 T cells in the lymph nodes of Psmb11^−/− mice, and a recent study showed that one out of three genetic variants of Psmb11 had a subtle yet significant impact on the total number of CD4 thymocytes. Nonetheless, we observed that compared with CD8 thymocytes, CD4 thymocytes can compensate for the stress induced by PSMB11 deletion via 1) upregulation of chaperones and suppressors of apoptosis together with 2) repression of protein synthesis and mitochondrial function. Thymic medullary cytokines are important for the development of CD8 thymocytes (54) but are dispensable for the development of conventional CD4 thymocytes (55). This might explain why CD4 thymocytes fare better than CD8 thymocytes in Psmb11^−/− thymi. The major variations in gene expression affecting CD4 thymocytes developing in a Psmb11^−/− thymus provide a strong rationale for in-depth functional analyses of this cell subset.

We are grateful to Danièle Gagné, Gaël Dulude, Martine Dupuis (flow cytometry and cell sorting), Christian Charbonneau (immunofluorescence), Micheline Fortin and Julie Hinsinger (histology and immunofluorescence), Raphaelle Lambert and Jennifer Huber (genomics), and Marilaine Fournier (image analysis) for technical assistance, as well as to the bioinformatics and animal facility personnel at the Institute for Research in Immunology and Cancer and the Maisonneuve-Rosemont Hospital Research Center. Special thanks to Marie-Hélène Lacombe (McGill University Health Center) for expert assistance with the ImageStream^X Mark II Imaging Flow Cytometer, and to Jalila Chagraoui for helpful discussion.

The authors have no financial conflicts of interest.