Generating functional protein variants with variational autoencoders.

The vast expansion of protein sequence databases provides an opportunity for new protein design approaches which seek to learn the sequence-function relationship directly from natural sequence variation. Deep generative models trained on protein sequence data have been shown to learn biologically meaningful representations helpful for a variety of downstream tasks, but their potential for direct use in the design of novel proteins remains largely unexplored. Here we show that variational autoencoders trained on a dataset of almost 70000 luciferase-like oxidoreductases can be used to generate novel, functional variants of the luxA bacterial luciferase. We propose separate VAE models to work with aligned sequence input (MSA VAE) and raw sequence input (AR-VAE), and offer evidence that while both are able to reproduce patterns of amino acid usage characteristic of the family, the MSA VAE is better able to capture long-distance dependencies reflecting the influence of 3D structure. To confirm the practical utility of the models, we used them to generate variants of luxA whose luminescence activity was validated experimentally. We further showed that conditional variants of both models could be used to increase the solubility of luxA without disrupting function. Altogether 6/12 of the variants generated using the unconditional AR-VAE and 9/11 generated using the unconditional MSA VAE retained measurable luminescence, together with all 23 of the less distant variants generated by conditional versions of the models; the most distant functional variant contained 35 differences relative to the nearest training set sequence. These results demonstrate the feasibility of using deep generative models to explore the space of possible protein sequences and generate useful variants, providing a method complementary to rational design and directed evolution approaches.

On-target activity predictions enable improved CRISPR-dCas9 screens in bacteria.

The ability to block gene expression in bacteria with the catalytically inactive mutant of Cas9, known as dCas9, is quickly becoming a standard methodology to probe gene function, perform high-throughput screens, and engineer cells for desired purposes. Yet, we still lack a good understanding of the design rules that determine on-target activity for dCas9. Taking advantage of high-throughput screening data, we fit a model to predict the ability of dCas9 to block the RNA polymerase based on the target sequence, and validate its performance on independently generated datasets. We further design a novel genome wide guide RNA library for E. coli MG1655, EcoWG1, using our model to choose guides with high activity while avoiding guides which might be toxic or have off-target effects. A screen performed using the EcoWG1 library during growth in rich medium improved upon previously published screens, demonstrating that very good performances can be attained using only a small number of well designed guides. Being able to design effective, smaller libraries will help make CRISPRi screens even easier to perform and more cost-effective. Our model and materials are available to the community through crispr.pasteur.fr and Addgene.

Identification of the cleavage sites leading to the shed forms of human and mouse anti-aging and cognition-enhancing protein Klotho.

Klotho is an age-extending, cognition-enhancing protein found to be down-regulated in aged mammals when age-related diseases start to appear. Low levels of Klotho occur in neurodegenerative diseases, kidney disease and many cancers. Many normal and pathologic processes involve the proteolytic shedding of membrane proteins. Transmembrane (TM) Klotho contains two homologous domains, KL1 and KL2 with homology to glycosidases. After shedding by ADAM 10 and 17, a shed Klotho isoform is released into serum and urine by the kidney, and into the CSF by the choroid plexus. We previously reported that human Klotho contains two major cleavage sites. However, the exact cleavage site responsible for the cleavage between the KL1 and KL2 domains remains unknown for the human Klotho, and both sites are unknown for mouse Klotho. In this study, we aimed to identify the cleavage sites leading to the shed forms of human and mouse Klotho. Mutations in the region close to the TM domain of mouse Klotho result in the reduced shedding of the 130 kD (KL1+KL2) and 70 kD (KL1) fragments, suggesting that the cleavage site lies within the mutated region. We further identified the cleavage sites responsible for the cleavage between KL1 and KL2 of human and mouse Klotho. Moreover, mutated Klotho proteins have similar subcellular localization patterns as wild type Klotho. Finally, in an FGF23 functional assay, all Klotho mutants with a nine amino acid deletion can also function as an FGFR1 co-receptor for FGF23 signaling, however, the signaling activity was greatly reduced. The study provides new and important information on Klotho shedding, and paves the way for studies aimed to distinguish between the distinct roles of the various isoforms of Klotho.

Neuroepithelial cysts of the middle ear.

Neuroepithelial cysts are lesions of the central nervous system that have previously been reported in cerebral parenchyma, in the spinal cord, and within the ventricles in association with the choroid plexus. We describe 2 cases of neuroepithelial cysts of the middle ear. One was diagnosed after surgery for a retraction pocket and chronic otitis media complicated by an intraoperative cerebrospinal fluid (CSF) leak. The other produced bilateral spontaneous CSF otorrhea and mimicked the presentation of arachnoid granulations. Both cases were verified with immunohistochemical stains. Neuroepithelial cysts, although rare, should be considered in the differential diagnosis of spontaneous CSF otorrhea or an epitympanic mass.

Genomic structure, cochlear expression, and mutation screening of KCNK6, a candidate gene for DFNA4.

KCNK6 encodes a tandem pore domain potassium channel, TWIK-2, that maps to chromosome 19. Both STS and linkage maps established KCNK6 as a positional candidate gene for DFNA4, a form of autosomal dominant nonsyndromic hereditary hearing loss. Identification and characterization of Kcnk6 expression within the mammalian cochlea established the gene as a functional candidate for DFNA4. Identification of Twik-2 expression in the mouse cochlea was initially established via RT-PCR assay of cochlear RNA. Subsequent immunoblot analysis of cochlear homogenate yielded a distinct 35-kDa band corresponding to the calculated molecular weight of the mouse Twik-2. Immunohistochemical studies localized Twik-2 expression in the cochlea predominantly within the stria vascularis. This vascular tissue borders the cochlear duct and is a critical regulator of potassium concentration in the endolymph. Genomic structure of TWIK-2 was subsequently determined and shown to consist of three coding exons with splice acceptor and donor sites in accordance with the consensus GT-AG rule. Two separate DFNA4 families were screened for KCNK6 sequence alterations. No mutations were found, thus excluding TWIK-2 as the DFNA4 candidate disease gene. Nevertheless, expression of Twik-2 within the stria vascularis suggests a potential role for this protein as one of the terminal components of the potassium ion-recycling pathway that contributes toward its reabsorption into the endolymph.