Optimized model architectures for deep learning on genomic data

The success of deep learning in various applications depends on task-specific architecture design choices, including the types, hyperparameters, and number of layers. In computational biology, there is no consensus on the optimal architecture design, and decisions are often made using insights from more well-established fields such as computer vision. These may not consider the domain-specific characteristics of genome sequences, potentially limiting performance. Here, we present GenomeNet-Architect, a neural architecture design framework that automatically optimizes deep learning models for genome sequence data. It optimizes the overall layout of the architecture, with a search space specifically designed for genomics. Additionally, it optimizes hyperparameters of individual layers and the model training procedure. On a viral classification task, GenomeNet-Architect reduced the read-level misclassification rate by 19%, with 67% faster inference and 83% fewer parameters, and achieved similar contig-level accuracy with ~100 times fewer parameters compared to the best-performing deep learning baselines.

Generation of complex bone marrow organoids from human induced pluripotent stem cells

The human bone marrow (BM) niche sustains hematopoiesis throughout life. We present a method for generating complex BM-like organoids (BMOs) from human induced pluripotent stem cells (iPSCs). BMOs consist of key cell types that self-organize into spatially defined three-dimensional structures mimicking cellular, structural and molecular characteristics of the hematopoietic microenvironment. Functional properties of BMOs include the presence of an in vivo-like vascular network, the presence of multipotent mesenchymal stem/progenitor cells, the support of neutrophil differentiation and responsiveness to inflammatory stimuli. Single-cell RNA sequencing revealed a heterocellular composition including the presence of a hematopoietic stem/progenitor (HSPC) cluster expressing genes of fetal HSCs. BMO-derived HSPCs also exhibited lymphoid potential and a subset demonstrated transient engraftment potential upon xenotransplantation in mice. We show that the BMOs could enable the modeling of hematopoietic developmental aspects and inborn errors of hematopoiesis, as shown for human VPS45 deficiency. Thus, iPSC-derived BMOs serve as a physiologically relevant in vitro model of the human BM microenvironment to study hematopoietic development and BM diseases.

Drug repurposing screen identifies lonafarnib as respiratory syncytial virus fusion protein inhibitor

Respiratory syncytial virus (RSV) is a common cause of acute lower respiratory tract infection in infants, older adults and the immunocompromised. Effective directly acting antivirals are not yet available for clinical use. To address this, we screen the ReFRAME drug-repurposing library consisting of 12,000 small molecules against RSV. We identify 21 primary candidates including RSV F and N protein inhibitors, five HSP90 and four IMPDH inhibitors. We select lonafarnib, a licensed farnesyltransferase inhibitor, and phase III candidate for hepatitis delta virus (HDV) therapy, for further follow-up. Dose-response analyses and plaque assays confirm the antiviral activity (IC50: 10-118 nM). Passaging of RSV with lonafarnib selects for phenotypic resistance and fixation of mutations in the RSV fusion protein (T335I and T400A). Lentiviral pseudotypes programmed with variant RSV fusion proteins confirm that lonafarnib inhibits RSV cell entry and that these mutations confer lonafarnib resistance. Surface plasmon resonance reveals RSV fusion protein binding of lonafarnib and co-crystallography identifies the lonafarnib binding site within RSV F. Oral administration of lonafarnib dose-dependently reduces RSV virus load in a murine infection model using female mice. Collectively, this work provides an overview of RSV drug repurposing candidates and establishes lonafarnib as a bona fide fusion protein inhibitor.

MVA-based vaccine candidates encoding the native or prefusion-stabilized SARS-CoV-2 spike reveal differential immunogenicity in humans

In response to the COVID-19 pandemic, multiple vaccines were developed using platforms such as viral vectors and mRNA technology. Here, we report humoral and cellular immunogenicity data from human phase 1 clinical trials investigating two recombinant Modified Vaccinia virus Ankara vaccine candidates, MVA-SARS-2-S and MVA-SARS-2-ST, encoding the native and the prefusion-stabilized SARS-CoV-2 spike protein, respectively. MVA-SARS-2-ST was more immunogenic than MVA-SARS-2-S, but both were less immunogenic compared to licensed mRNA- and ChAd-based vaccines in SARS-CoV-2 naïve individuals. In heterologous vaccination, previous MVA-SARS-2-S vaccination enhanced T cell functionality and MVA-SARS-2-ST boosted the frequency of T cells and S1-specific IgG levels when used as a third vaccination. While the vaccine candidate containing the prefusion-stabilized spike elicited predominantly S1-specific responses, immunity to the candidate with the native spike was skewed towards S2-specific responses. These data demonstrate how the spike antigen conformation, using the same viral vector, directly affects vaccine immunogenicity in humans.

Epstein-Barr virus-driven B cell lymphoma mediated by a direct LMP1-TRAF6 complex

Epstein-Barr virus (EBV) latent membrane protein 1 (LMP1) drives viral B cell transformation and oncogenesis. LMP1's transforming activity depends on its C-terminal activation region 2 (CTAR2), which induces NF-κB and JNK by engaging TNF receptor-associated factor 6 (TRAF6). The mechanism of TRAF6 recruitment to LMP1 and its role in LMP1 signalling remains elusive. Here we demonstrate that TRAF6 interacts directly with a viral TRAF6 binding motif within CTAR2. Functional and NMR studies supported by molecular modeling provide insight into the architecture of the LMP1-TRAF6 complex, which differs from that of CD40-TRAF6. The direct recruitment of TRAF6 to LMP1 is essential for NF-κB activation by CTAR2 and the survival of LMP1-driven lymphoma. Disruption of the LMP1-TRAF6 complex by inhibitory peptides interferes with the survival of EBV-transformed B cells. In this work, we identify LMP1-TRAF6 as a critical virus-host interface and validate this interaction as a potential therapeutic target in EBV-associated cancer.

A type VII-secreted lipase toxin with reverse domain arrangement

The type VII protein secretion system (T7SS) is found in many Gram-positive bacteria and in pathogenic mycobacteria. All T7SS substrate proteins described to date share a common helical domain architecture at the N-terminus that typically interacts with other helical partner proteins, forming a composite signal sequence for targeting to the T7SS. The C-terminal domains are functionally diverse and in Gram-positive bacteria such as Staphylococcus aureus often specify toxic anti-bacterial activity. Here we describe the first example of a class of T7 substrate, TslA, that has a reverse domain organisation. TslA is widely found across Bacillota including Staphylococcus, Enterococcus and Listeria. We show that the S. aureus TslA N-terminal domain is a phospholipase A with anti-staphylococcal activity that is neutralised by the immunity lipoprotein TilA. Two small helical partner proteins, TlaA1 and TlaA2 are essential for T7-dependent secretion of TslA and at least one of these interacts with the TslA C-terminal domain to form a helical stack. Cryo-EM analysis of purified TslA complexes indicate that they share structural similarity with canonical T7 substrates. Our findings suggest that the T7SS has the capacity to recognise a secretion signal present at either end of a substrate.

Probabilities of developing HIV-1 bNAb sequence features in uninfected and chronically infected individuals

HIV-1 broadly neutralizing antibodies (bNAbs) are able to suppress viremia and prevent infection. Their induction by vaccination is therefore a major goal. However, in contrast to antibodies that neutralize other pathogens, HIV-1-specific bNAbs frequently carry uncommon molecular characteristics that might prevent their induction. Here, we perform unbiased sequence analyses of B cell receptor repertoires from 57 uninfected and 46 chronically HIV-1- or HCV-infected individuals and learn probabilistic models to predict the likelihood of bNAb development. We formally show that lower probabilities for bNAbs are predictive of higher HIV-1 neutralization activity. Moreover, ranking bNAbs by their probabilities allows to identify highly potent antibodies with superior generation probabilities as preferential targets for vaccination approaches. Importantly, we find equal bNAb probabilities across infected and uninfected individuals. This implies that chronic infection is not a prerequisite for the generation of bNAbs, fostering the hope that HIV-1 vaccines can induce bNAb development in uninfected people.

Dynamics and durability of HIV-1 neutralization are determined by viral replication

Human immunodeficiency virus type 1 (HIV-1)-neutralizing antibodies (nAbs) that prevent infection are the main goal of HIV vaccine discovery. But as no nAb-eliciting vaccines are yet available, only data from HIV-1 neutralizers-persons with HIV-1 who naturally develop broad and potent nAbs-can inform about the dynamics and durability of nAb responses in humans, knowledge which is crucial for the design of future HIV-1 vaccine regimens. To address this, we assessed HIV-1-neutralizing immunoglobulin G (IgG) from 2,354 persons with HIV-1 on or off antiretroviral therapy (ART). Infection with non-clade B viruses, CD4+ T cell counts <200 µl-1, being off ART and a longer time off ART were independent predictors of a more potent and broad neutralization. In longitudinal analyses, we found nAb half-lives of 9.3 and 16.9 years in individuals with no- or low-level viremia, respectively, and 4.0 years in persons who newly initiated ART. Finally, in a potent HIV-1 neutralizer, we identified lower fractions of serum nAbs and of nAb-encoding memory B cells after ART initiation, suggesting that a decreasing neutralizing serum activity after antigen withdrawal is due to lower levels of nAbs. These results collectively show that HIV-1-neutralizing responses can persist for several years, even at low antigen levels, suggesting that an HIV-1 vaccine may elicit a durable nAb response.

A self-supervised deep learning method for data-efficient training in genomics

Deep learning in bioinformatics is often limited to problems where extensive amounts of labeled data are available for supervised classification. By exploiting unlabeled data, self-supervised learning techniques can improve the performance of machine learning models in the presence of limited labeled data. Although many self-supervised learning methods have been suggested before, they have failed to exploit the unique characteristics of genomic data. Therefore, we introduce Self-GenomeNet, a self-supervised learning technique that is custom-tailored for genomic data. Self-GenomeNet leverages reverse-complement sequences and effectively learns short- and long-term dependencies by predicting targets of different lengths. Self-GenomeNet performs better than other self-supervised methods in data-scarce genomic tasks and outperforms standard supervised training with ~10 times fewer labeled training data. Furthermore, the learned representations generalize well to new datasets and tasks. These findings suggest that Self-GenomeNet is well suited for large-scale, unlabeled genomic datasets and could substantially improve the performance of genomic models.

Nutritional and host environments determine community ecology and keystone species in a synthetic gut bacterial community

A challenging task to understand health and disease-related microbiome signatures is to move beyond descriptive community-level profiling towards disentangling microbial interaction networks. Using a synthetic gut bacterial community, we aimed to study the role of individual members in community assembly, identify putative keystone species and test their influence across different environments. Single-species dropout experiments reveal that bacterial strain relationships strongly vary not only in different regions of the murine gut, but also across several standard culture media. Mechanisms involved in environment-dependent keystone functions in vitro include exclusive access to polysaccharides as well as bacteriocin production. Further, Bacteroides caecimuris and Blautia coccoides are found to play keystone roles in gnotobiotic mice by impacting community composition, the metabolic landscape and inflammatory responses. In summary, the presented study highlights the strong interdependency between bacterial community ecology and the biotic and abiotic environment. These results question the concept of universally valid keystone species in the gastrointestinal ecosystem and underline the context-dependency of both, keystone functions and bacterial interaction networks.

A protective measles virus-derived vaccine inducing long-lasting immune responses against influenza A virus H7N9

A novel Influenza A virus (subtype H7N9) emerged in spring 2013 and caused considerable mortality in zoonotically infected patients. To be prepared for potential pandemics, broadly effective and safe vaccines are crucial. Recombinant measles virus (MeV) encoding antigens of foreign pathogens constitutes a promising vector platform to generate novel vaccines. To characterize the efficacy of H7N9 antigens in a prototypic vaccine platform technology, we generated MeVs encoding either neuraminidase (N9) or hemagglutinin (H7). Moraten vaccine strain-derived vaccine candidates were rescued; they replicated with efficiency comparable to that of the measles vaccine, robustly expressed H7 and N9, and were genetically stable over 10 passages. Immunization of MeV-susceptible mice triggered the production of antibodies against H7 and N9, including hemagglutination-inhibiting and neutralizing antibodies induced by MV_vac2-H7(P) and neuraminidase-inhibiting antibodies by MV_vac2-N9(P). Vaccinated mice also developed long-lasting H7- and N9-specific T cells. Both MV_vac2-H7(P) and MV_vac2-N9(P)-vaccinated mice were protected from lethal H7N9 challenge.

In-depth virological and immunological characterization of HIV-1 cure after CCR5Δ32/Δ32 allogeneic hematopoietic stem cell transplantation

Despite scientific evidence originating from two patients published to date that CCR5Δ32/Δ32 hematopoietic stem cell transplantation (HSCT) can cure human immunodeficiency virus type 1 (HIV-1), the knowledge of immunological and virological correlates of cure is limited. Here we characterize a case of long-term HIV-1 remission of a 53-year-old male who was carefully monitored for more than 9 years after allogeneic CCR5Δ32/Δ32 HSCT performed for acute myeloid leukemia. Despite sporadic traces of HIV-1 DNA detected by droplet digital PCR and in situ hybridization assays in peripheral T cell subsets and tissue-derived samples, repeated ex vivo quantitative and in vivo outgrowth assays in humanized mice did not reveal replication-competent virus. Low levels of immune activation and waning HIV-1-specific humoral and cellular immune responses indicated a lack of ongoing antigen production. Four years after analytical treatment interruption, the absence of a viral rebound and the lack of immunological correlates of HIV-1 antigen persistence are strong evidence for HIV-1 cure after CCR5Δ32/Δ32 HSCT.

Molecular consequences of SARS-CoV-2 liver tropism

Extrapulmonary manifestations of COVID-19 have gained attention due to their links to clinical outcomes and their potential long-term sequelae¹. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) displays tropism towards several organs, including the heart and kidney. Whether it also directly affects the liver has been debated^2,3. Here we provide clinical, histopathological, molecular and bioinformatic evidence for the hepatic tropism of SARS-CoV-2. We find that liver injury, indicated by a high frequency of abnormal liver function tests, is a common clinical feature of COVID-19 in two independent cohorts of patients with COVID-19 requiring hospitalization. Using autopsy samples obtained from a third patient cohort, we provide multiple levels of evidence for SARS-CoV-2 liver tropism, including viral RNA detection in 69% of autopsy liver specimens, and successful isolation of infectious SARS-CoV-2 from liver tissue postmortem. Furthermore, we identify transcription-, proteomic- and transcription factor-based activity profiles in hepatic autopsy samples, revealing similarities to the signatures associated with multiple other viral infections of the human liver. Together, we provide a comprehensive multimodal analysis of SARS-CoV-2 liver tropism, which increases our understanding of the molecular consequences of severe COVID-19 and could be useful for the identification of organ-specific pharmacological targets.

The landscape of viral associations in human cancers

Here, as part of the Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, for which whole-genome and-for a subset-whole-transcriptome sequencing data from 2,658 cancers across 38 tumor types was aggregated, we systematically investigated potential viral pathogens using a consensus approach that integrated three independent pipelines. Viruses were detected in 382 genome and 68 transcriptome datasets. We found a high prevalence of known tumor-associated viruses such as Epstein-Barr virus (EBV), hepatitis B virus (HBV) and human papilloma virus (HPV; for example, HPV16 or HPV18). The study revealed significant exclusivity of HPV and driver mutations in head-and-neck cancer and the association of HPV with APOBEC mutational signatures, which suggests that impaired antiviral defense is a driving force in cervical, bladder and head-and-neck carcinoma. For HBV, HPV16, HPV18 and adeno-associated virus-2 (AAV2), viral integration was associated with local variations in genomic copy numbers. Integrations at the TERT promoter were associated with high telomerase expression evidently activating this tumor-driving process. High levels of endogenous retrovirus (ERV1) expression were linked to a worse survival outcome in patients with kidney cancer.