Factorized embeddings learns rich and biologically meaningful embedding spaces using factorized tensor decomposition.

MOTIVATION: The recent development of sequencing technologies revolutionized our understanding of the inner workings of the cell as well as the way disease is treated. A single RNA sequencing (RNA-Seq) experiment, however, measures tens of thousands of parameters simultaneously. While the results are information rich, data analysis provides a challenge. Dimensionality reduction methods help with this task by extracting patterns from the data by compressing it into compact vector representations. RESULTS: We present the factorized embeddings (FE) model, a self-supervised deep learning algorithm that learns simultaneously, by tensor factorization, gene and sample representation spaces. We ran the model on RNA-Seq data from two large-scale cohorts and observed that the sample representation captures information on single gene and global gene expression patterns. Moreover, we found that the gene representation space was organized such that tissue-specific genes, highly correlated genes as well as genes participating in the same GO terms were grouped. Finally, we compared the vector representation of samples learned by the FE model to other similar models on 49 regression tasks. We report that the representations trained with FE rank first or second in all of the tasks, surpassing, sometimes by a considerable margin, other representations. AVAILABILITY AND IMPLEMENTATION: A toy example in the form of a Jupyter Notebook as well as the code and trained embeddings for this project can be found at: https://github.com/TrofimovAssya/FactorizedEmbeddings. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Differential Features of AIRE-Induced and AIRE-Independent Promiscuous Gene Expression in Thymic Epithelial Cells.

Establishment of self-tolerance in the thymus depends on promiscuous expression of tissue-restricted Ags (TRA) by thymic epithelial cells (TEC). This promiscuous gene expression (pGE) is regulated in part by the autoimmune regulator (AIRE). To evaluate the commonalities and discrepancies between AIRE-dependent and -independent pGE, we analyzed the transcriptome of the three main TEC subsets in wild-type and Aire knockout mice. We found that the impact of AIRE-dependent pGE is not limited to generation of TRA. AIRE decreases, via non-cell autonomous mechanisms, the expression of genes coding for positive regulators of cell proliferation, and it thereby reduces the number of cortical TEC. In mature medullary TEC, AIRE-driven pGE upregulates non-TRA coding genes that enhance cell-cell interactions (e.g., claudins, integrins, and selectins) and are probably of prime relevance to tolerance induction. We also found that AIRE-dependent and -independent TRA present several distinctive features. In particular, relative to AIRE-induced TRA, AIRE-independent TRA are more numerous and show greater splicing complexity. Furthermore, we report that AIRE-dependent versus -independent TRA project nonredundant representations of peripheral tissues in the thymus.

Immunoproteasomes shape the transcriptome and regulate the function of dendritic cells.

By regulating protein degradation, constitutive proteasomes (CPs) control practically all cellular functions. In addition to CPs, vertebrates express immunoproteasomes (IPs). The major nonredundant role ascribed to IPs is their enhanced ability to generate antigenic peptides. We report that CPs and IPs differentially regulate the expression of >8000 transcripts in maturing mouse dendritic cells (DCs) via regulation of signaling pathways such as IFN regulatory factors, STATs, and NF-κB. IPs regulate the transcription of many mRNAs and maturation of a few of them. Moreover, even when engineered to present optimal amounts of antigenic peptide, IP-deficient DCs are inefficient for in vivo T cell priming. Our study shows that the role of IPs in DCs is not limited to Ag processing and reveals a major nonredundant role for IPs in transcription regulation. The dramatic effect of IPs on the transcriptional landscape could explain the various immune and nonimmune phenotypes observed in vertebrates with IP deficiency or mutations.

Rejection of leukemic cells requires antigen-specific T cells with high functional avidity.

In a context where injection of antigen (Ag)-specific T cells probably represents the future of leukemia immunotherapy, identification of optimal target Ags is crucial. We therefore sought to discover a reliable marker for selection of the most potent Ags. To this end, (1) we immunized mice against 8 individual Ags: 4 minor histocompatibility Ags (miHAs) and 4 leukemia-associated Ags (LAAs) that were overexpressed on leukemic relative to normal thymocytes; (2) we assessed their ability to reject EL4 leukemic cells; and (3) we correlated the properties of our Ags (and their cognate T cells) with their ability to induce protective antileukemic responses. Overall, individual miHAs instigated more potent antileukemic responses than LAAs. Three features had no influence on the ability of primed T cells to reject leukemic cells: (1) MHC-peptide affinity; (2) the stability of MHC-peptide complexes; and (3) epitope density at the surface of leukemic cells, as assessed using mass spectrometry. The cardinal feature of successful Ags is that they were recognized by high-avidity CD8 T cells that proliferated extensively in vivo. Our work suggests that in vitro evaluation of functional avidity represents the best criterion for selection of Ags, which should be prioritized in clinical trials of leukemia immunotherapy.

Transposable elements regulate thymus development and function.

Transposable elements (TEs) are repetitive sequences representing ~45% of the human and mouse genomes and are highly expressed by medullary thymic epithelial cells (mTECs). In this study, we investigated the role of TEs on T-cell development in the thymus. We performed multiomic analyses of TEs in human and mouse thymic cells to elucidate their role in T-cell development. We report that TE expression in the human thymus is high and shows extensive age- and cell lineage-related variations. TE expression correlates with multiple transcription factors in all cell types of the human thymus. Two cell types express particularly broad TE repertoires: mTECs and plasmacytoid dendritic cells (pDCs). In mTECs, transcriptomic data suggest that TEs interact with transcription factors essential for mTEC development and function (e.g., PAX1 and REL), and immunopeptidomic data showed that TEs generate MHC-I-associated peptides implicated in thymocyte education. Notably, AIRE, FEZF2, and CHD4 regulate small yet non-redundant sets of TEs in murine mTECs. Human thymic pDCs homogenously express large numbers of TEs that likely form dsRNA, which can activate innate immune receptors, potentially explaining why thymic pDCs constitutively secrete IFN ɑ/ß. This study highlights the diversity of interactions between TEs and the adaptive immune system. TEs are genetic parasites, and the two thymic cell types most affected by TEs (mTEcs and pDCs) are essential to establishing central T-cell tolerance. Therefore, we propose that orchestrating TE expression in thymic cells is critical to prevent autoimmunity in vertebrates.

Two types of human TCR differentially regulate reactivity to self and non-self antigens.

Based on analyses of TCR sequences from over 1,000 individuals, we report that the TCR repertoire is composed of two ontogenically and functionally distinct types of TCRs. Their production is regulated by variations in thymic output and terminal deoxynucleotidyl transferase (TDT) activity. Neonatal TCRs derived from TDT-negative progenitors persist throughout life, are highly shared among subjects, and are reported as disease-associated. Thus, 10%-30% of most frequent cord blood TCRs are associated with common pathogens and autoantigens. TDT-dependent TCRs present distinct structural features and are less shared among subjects. TDT-dependent TCRs are produced in maximal numbers during infancy when thymic output and TDT activity reach a summit, are more abundant in subjects with AIRE mutations, and seem to play a dominant role in graft-versus-host disease. Factors decreasing thymic output (age, male sex) negatively impact TCR diversity. Males compensate for their lower repertoire diversity via hyperexpansion of selected TCR clonotypes.

UM171-Expanded Cord Blood Transplants Support Robust T Cell Reconstitution with Low Rates of Severe Infections.

Rapid T cell reconstitution following hematopoietic stem cell transplantation (HSCT) is essential for protection against infections and has been associated with lower incidence of chronic graft-versus-host disease (cGVHD), relapse, and transplant-related mortality (TRM). While cord blood (CB) transplants are associated with lower rates of cGVHD and relapse, their low stem cell content results in slower immune reconstitution and higher risk of graft failure, severe infections, and TRM. Recently, results of a phase I/II trial revealed that single UM171-expanded CB transplant allowed the use of smaller CB units without compromising engraftment (www.clinicaltrials.gov, NCT02668315). We assessed T cell reconstitution in patients who underwent transplantation with UM171-expanded CB grafts and retrospectively compared it to that of patients receiving unmanipulated CB transplants. While median T cell dose infused was at least 2 to 3 times lower than that of unmanipulated CB, numbers and phenotype of T cells at 3, 6, and 12 months post-transplant were similar between the 2 cohorts. T cell receptor sequencing analyses revealed that UM171 patients had greater T cell diversity and higher numbers of clonotypes at 12 months post-transplant. This was associated with higher counts of naive T cells and recent thymic emigrants, suggesting active thymopoiesis and correlating with the demonstration that UM171 expands common lymphoid progenitors in vitro. UM171 patients also showed rapid virus-specific T cell reactivity and significantly reduced incidence of severe infections. These results suggest that UM171 patients benefit from rapid T cell reconstitution, which likely contributes to the absence of moderate/severe cGVHD, infection-related mortality, and late TRM observed in this cohort.

Widespread and tissue-specific expression of endogenous retroelements in human somatic tissues.

BACKGROUND: Endogenous retroelements (EREs) constitute about 42% of the human genome and have been implicated in common human diseases such as autoimmunity and cancer. The dominant paradigm holds that EREs are expressed in embryonic stem cells (ESCs) and germline cells but are repressed in differentiated somatic cells. Despite evidence that some EREs can be expressed at the RNA and protein levels in specific contexts, a system-level evaluation of their expression in human tissues is lacking. METHODS: Using RNA sequencing data, we analyzed ERE expression in 32 human tissues and cell types, including medullary thymic epithelial cells (mTECs). A tissue specificity index was computed to identify tissue-restricted ERE families. We also analyzed the transcriptome of mTECs in wild-type and autoimmune regulator (AIRE)-deficient mice. Finally, we developed a proteogenomic workflow combining RNA sequencing and mass spectrometry (MS) in order to evaluate whether EREs might be translated and generate MHC I-associated peptides (MAP) in B-lymphoblastoid cell lines (B-LCL) from 16 individuals. RESULTS: We report that all human tissues express EREs, but the breadth and magnitude of ERE expression are very heterogeneous from one tissue to another. ERE expression was particularly high in two MHC I-deficient tissues (ESCs and testis) and one MHC I-expressing tissue, mTECs. In mutant mice, we report that the exceptional expression of EREs in mTECs was AIRE-independent. MS analyses identified 103 non-redundant ERE-derived MAPs (ereMAPs) in B-LCLs. These ereMAPs preferentially derived from sense translation of intronic EREs. Notably, detailed analyses of their amino acid composition revealed that ERE-derived MAPs presented homology to viral MAPs. CONCLUSIONS: This study shows that ERE expression in somatic tissues is more pervasive and heterogeneous than anticipated. The high and diversified expression of EREs in mTECs and their ability to generate MAPs suggest that EREs may play an important role in the establishment of self-tolerance. The viral-like properties of ERE-derived MAPs suggest that those not expressed in mTECs can be highly immunogenic.

Expression of immunoproteasome genes is regulated by cell-intrinsic and -extrinsic factors in human cancers.

Based on transcriptomic analyses of thousands of samples from The Cancer Genome Atlas, we report that expression of constitutive proteasome (CP) genes (PSMB5, PSMB6, PSMB7) and immunoproteasome (IP) genes (PSMB8, PSMB9, PSMB10) is increased in most cancer types. In breast cancer, expression of IP genes was determined by the abundance of tumor infiltrating lymphocytes and high expression of IP genes was associated with longer survival. In contrast, IP upregulation in acute myeloid leukemia (AML) was a cell-intrinsic feature that was not associated with longer survival. Expression of IP genes in AML was IFN-independent, correlated with the methylation status of IP genes, and was particularly high in AML with an M5 phenotype and/or MLL rearrangement. Notably, PSMB8 inhibition led to accumulation of polyubiquitinated proteins and cell death in IPhigh but not IPlow AML cells. Co-clustering analysis revealed that genes correlated with IP subunits in non-M5 AMLs were primarily implicated in immune processes. However, in M5 AML, IP genes were primarily co-regulated with genes involved in cell metabolism and proliferation, mitochondrial activity and stress responses. We conclude that M5 AML cells can upregulate IP genes in a cell-intrinsic manner in order to resist cell stress.