Cancer treatment by immune checkpoint blockade (ICB) can bring long-lasting clinical benefits, but only a fraction of patients respond to treatment. To predict ICB response, we developed TIDE, a computational method to model two primary mechanisms of tumor immune evasion: the induction of T cell dysfunction in tumors with high infiltration of cytotoxic T lymphocytes (CTL) and the prevention of T cell infiltration in tumors with low CTL level. We identified signatures of T cell dysfunction from large tumor cohorts by testing how the expression of each gene in tumors interacts with the CTL infiltration level to influence patient survival. We also modeled factors that exclude T cell infiltration into tumors using expression signatures from immunosuppressive cells. Using this framework and pre-treatment RNA-Seq or NanoString tumor expression profiles, TIDE predicted the outcome of melanoma patients treated with first-line anti-PD1 or anti-CTLA4 more accurately than other biomarkers such as PD-L1 level and mutation load. TIDE also revealed new candidate ICB resistance regulators, such as SERPINB9, demonstrating utility for immunotherapy research.

Signaling between programmed cell death protein 1 (PD-1) and the PD-1 ligands (PD-L1, PD-L2) is essential for malignant Hodgkin Reed-Sternberg (HRS) cells to evade antitumor immunity in classical Hodgkin lymphoma (cHL). Copy number alterations of 9p24.1/()() contribute to robust PD-L1 and PD-L2 expression by HRS cells. PD-L1 is also expressed by nonmalignant tumor-associated macrophages (TAMs), but the relationships among PD-L1 HRS cells, PD-L1 TAMs, and PD-1 T cells remain undefined. We used multiplex immunofluorescence and digital image analysis to examine the topography of PD-L1 and PD-1 cells in the tumor microenvironment (TME) of cHL. We find that the majority of PD-L1 in the TME is expressed by the abundant PD-L1 TAMs, which physically colocalize with PD-L1 HRS cells in a microenvironmental niche. PD-L1 TAMs are enriched for contacts with T cells, and PD-L1 HRS cells are enriched for contacts with CD4 T cells, a subset of which are PD-1 Our data define a unique topology of cHL in which PD-L1 TAMs surround HRS cells and implicate CD4 T cells as a target of PD-1 blockade.

We propose a new method for determining the target genes of transcriptional enhancers in specific cells and tissues. It combines global trends across many samples and sample-specific information, and considers the joint effect of multiple enhancers. Our method outperforms existing methods when predicting the target genes of enhancers in unseen samples, as evaluated by independent experimental data. Requiring few types of input data, we are able to apply our method to reconstruct the enhancer-target networks in 935 samples of human primary cells, tissues and cell lines, which constitute by far the largest set of enhancer-target networks. The similarity of these networks from different samples closely follows their cell and tissue lineages. We discover three major co-regulation modes of enhancers and find defense-related genes often simultaneously regulated by multiple enhancers bound by different transcription factors. We also identify differentially methylated enhancers in hepatocellular carcinoma (HCC) and experimentally confirm their altered regulation of HCC-related genes.

In classical Hodgkin lymphoma (cHL), the host antitumor immune response is ineffective. Hodgkin Reed-Sternberg (HRS) cells have multifaceted mechanisms to evade the immune system, including 9p24.1 genetic alterations, overexpression of PD-1 ligands, and associated T-cell exhaustion and additional structural bases of aberrant antigen presentation. The clinical success of PD-1 blockade in cHL suggests that the tumor microenvironment (TME) contains reversibly exhausted T effector cells (Teffs). However, durable responses are observed in patients with β2-microglobulin/major histocompatibility complex (MHC) class I loss on HRS cells, raising the possibility of non-CD8 T cell-mediated mechanisms of efficacy of PD-1 blockade. These observations highlight the need for a detailed analysis of the cHL TME. Using a customized time-of-flight mass cytometry panel, we simultaneously assessed cell suspensions from diagnostic cHL biopsies and control reactive lymph node/tonsil (RLNT) samples. Precise phenotyping of immune cell subsets revealed salient differences between cHLs and RLNTs. The TME in cHL is CD4 T-cell rich, with frequent loss of MHC class I expression on HRS cells. In cHLs, we found concomitant expansion of T helper 1 (Th1)-polarized Teffs and regulatory T cells (Tregs). The cHL Th1 Tregs expressed little or no PD-1, whereas the Th1 Teffs were PD-1 The differential PD-1 expression and likely functional Th1-polarized CD4 Tregs and exhausted Teffs may represent complementary mechanisms of immunosuppression in cHL.

Tumor-infiltrating B cells are an important component in the microenvironment with unclear anti-tumor impacts. We enhanced our previous computational algorithm TRUST to extract the B cell immunoglobulin (Ig) hypervariable regions from bulk tumor RNA-seq data. TRUST assembled over 30 million complementarity-determining region 3 (CDR3s) of the B cell heavy chain (IgH) from The Cancer Genome Atlas (TCGA). Widespread B cell clonal expansions and Ig subclass switch events were observed in diverse human cancers. Prevalent somatic copy number alterations in MICA and MICB genes related to antibody-dependent cell mediated cytotoxicity (ADCC) were identified in tumors with elevated B cell activity. IgG3-1 subclass switch interacts with the B cell receptor affinity maturation and defects in the ADCC pathway. Comprehensive pan-cancer analyses of tumor-infiltrating B cell receptor repertoires identified novel tumor immune evasion mechanisms through genetic alterations. The IgH sequences identified here are potentially useful resources for future development of immunotherapies.

We introduce the TRUST4 open-source algorithm for reconstruction of immune receptor repertoires in αβ/γδ T cells and B cells from RNA-sequencing (RNA-seq) data. Compared with competing methods, TRUST4 supports both FASTQ and BAM format and is faster and more sensitive in assembling longereven full-length-receptor repertoires. TRUST4 can also call repertoire sequences from single-cell RNA-seq (scRNA-seq) data without V(D)J enrichment, and is compatible with both SMART-seq and 5′ 10x Genomics platforms.Both T and B cells can generate diverse receptor (TCR and BCR, respectively) repertoires, through somatic V(D)J recombination, to recognize various external antigens or tumor neoantigens. Following antigen recognition, BCRs also undergo somatic hypermutations (SHMs) to further improve antigen-binding affinity. Repertoire sequencing has been increasingly adopted in infectious disease 1 , allergy 2 , autoimmune 3 , tumor immuology 4 and cancer immunotherapy 5 studies, but it is an expensive assay and consumes valuable tissue samples. Alternatively, RNA-seq data contain expressed TCR and BCR sequences in tissues or peripheral blood mononuclear cells (PBMC). However, because repertoire sequences from V(D)J recombination and SHM are different from the germline, they are often eliminated in the read-mapping step.Previously we developed the TRUST algorithm 6-8 , utilized to de novo assemble immune receptor repertories directly from tissue or blood RNA-seq data. When applied to The Cancer Genome Atlas (TCGA) tumor RNA-seq data, TRUST revealed profound biological insights into the repertoires of tumor-infiltrating T cells 6 and B cells 8 , as well as their associated tumor immunity. Although less sensitive than TCR-seq and BCR-seq, TRUST is able to identify the abundantly expressed and potentially more clonally expanded TCRs/BCRs in the RNA-seq data that are more likely to be involved in antigen binding 9 . Recent years have also seen other computational methods introduced for immune repertoire construction from RNA-seq data, including V'DJer 10 , MiXCR 11 , CATT 12 and ImRep 13 . These methods focus on reconstruction of complementary-determining region 3 (CDR3), with limited ability to assemble full-length V(D)J receptor sequences, although CDR1 and CDR2 on the V sequence still contribute considerably to antigen recognition and binding. For example, five out of six mutations predicted in a recent study to influence antibody affinity in the acidic tumor environment are located in CDR1 and CDR2 (ref. 14 ), and four out of nine positions contributing most to 4A8 antibody binding to the SARS-CoV-2 spike protein are in CDR1 and CDR2 (ref. 15 ). Therefore, algorithms that can infer full-length immune receptor repertoires can facilitate better receptor-antigen interaction modeling.

Direct conversion of somatic cells into neurons holds great promise for regenerative medicine. However, as neuronal conversion is relatively inefficient on human cells compared to mouse cells, it has been unclear what might be key barriers to reprogramming in human cells. We recently elucidated an RNA program mediated by the polypyrimidine tract binding protein PTB to convert mouse embryonic fibroblasts (MEFs) into functional neurons. On human adult fibroblasts (HAFs), however, we unexpectedly find that invoke of the documented PTB-REST-miR-124 loop only generates immature neurons. We now report that the functionality requires sequential inactivation of PTB and the PTB paralog nPTB in HAFs. Inactivation of nPTB triggers another self-enforcing loop essential for neuronal maturation, which comprises nPTB, the transcription factor BRN2, and miR-9. These findings suggest two separate gatekeepers to control neuronal conversion and maturation and consecutively overcoming these gatekeepers enables deterministic reprogramming of HAFs into functional neurons.

Recently, several experimental techniques have emerged for probing RNA structures based on high-throughput sequencing. However, most secondary structure prediction tools that incorporate probing data are designed and optimized for particular types of experiments. For example, RNAstructure-Fold is optimized for SHAPE data, while SeqFold is optimized for PARS data. Here, we report a new RNA secondary structure prediction method, restrained MaxExpect (RME), which can incorporate multiple types of experimental probing data and is based on a free energy model and an MEA (maximizing expected accuracy) algorithm. We first demonstrated that RME substantially improved secondary structure prediction with perfect restraints (base pair information of known structures). Next, we collected structure-probing data from diverse experiments (e.g. SHAPE, PARS and DMS-seq) and transformed them into a unified set of pairing probabilities with a posterior probabilistic model. By using the probability scores as restraints in RME, we compared its secondary structure prediction performance with two other well-known tools, RNAstructure-Fold (based on a free energy minimization algorithm) and SeqFold (based on a sampling algorithm). For SHAPE data, RME and RNAstructure-Fold performed better than SeqFold, because they markedly altered the energy model with the experimental restraints. For high-throughput data (e.g. PARS and DMS-seq) with lower probing efficiency, the secondary structure prediction performances of the tested tools were comparable, with performance improvements for only a portion of the tested RNAs. However, when the effects of tertiary structure and protein interactions were removed, RME showed the highest prediction accuracy in the DMS-accessible regions by incorporating in vivo DMS-seq data.