Leveraging Directed Causal Discovery to Detect Latent Common Causes in Cause-Effect Pairs.

The discovery of causal relationships is a fundamental problem in science and medicine. In recent years, many elegant approaches to discovering causal relationships between two variables from observational data have been proposed. However, most of these deal only with purely directed causal relationships and cannot detect latent common causes. Here, we devise a general heuristic which takes a causal discovery algorithm that can only distinguish purely directed causal relations and modifies it to also detect latent common causes. We apply our method to two directed causal discovery algorithms, the information geometric causal inference (IGCI) of (Daniusis et al., 2010) and the kernel conditional deviance for causal inference of (Mitrovic et al., 2018), and extensively test on synthetic data-detecting latent common causes in additive, multiplicative and complex noise regimes-and on real data, where we are able to detect known common causes. In addition to detecting latent common causes, our experiments demonstrate that both the modified algorithms preserve the performance of the original in distinguishing directed causal relations.

Technology readiness levels for machine learning systems.

The development and deployment of machine learning systems can be executed easily with modern tools, but the process is typically rushed and means-to-an-end. Lack of diligence can lead to technical debt, scope creep and misaligned objectives, model misuse and failures, and expensive consequences. Engineering systems, on the other hand, follow well-defined processes and testing standards to streamline development for high-quality, reliable results. The extreme is spacecraft systems, with mission critical measures and robustness throughout the process. Drawing on experience in both spacecraft engineering and machine learning (research through product across domain areas), we've developed a proven systems engineering approach for machine learning and artificial intelligence: the Machine Learning Technology Readiness Levels framework defines a principled process to ensure robust, reliable, and responsible systems while being streamlined for machine learning workflows, including key distinctions from traditional software engineering, and a lingua franca for people across teams and organizations to work collaboratively on machine learning and artificial intelligence technologies. Here we describe the framework and elucidate with use-cases from physics research to computer vision apps to medical diagnostics.

Identification and Validation of CRISPR/Cas9 Off-Target Activity in Hematopoietic Stem and Progenitor Cells.

Targeted genome editing in hematopoietic stem and progenitor cells (HSPCs) using CRISPR/Cas9 can potentially provide a permanent cure for hematologic diseases. However, the utility of CRISPR/Cas9 systems for therapeutic genome editing can be compromised by their off-target effects. In this chapter, we outline the procedures for CRISPR/Cas9 off-target identification and validation in HSPCs. This method is broadly applicable to diverse CRISPR/Cas9 systems and cell types. Using this protocol, researchers can perform computational prediction and experimental identification of potential off-target sites followed by off-target activity quantification by next-generation sequencing.

Electroporation-Mediated Delivery of Cas9 Ribonucleoproteins Results in High Levels of Gene Editing in Primary Hepatocytes.

Adeno-associated virus vectors are the most used delivery method for liver-directed gene editing. Still, they are associated with significant disadvantages that can compromise the safety and efficacy of therapies. Here, we investigate the effects of electroporating CRISPR-Cas9 as mRNA and ribonucleoproteins (RNPs) into primary hepatocytes regarding on-target activity, specificity, and cell viability. We observed a transfection efficiency of >60% and on-target insertions/deletions (indels) of up to 95% in primary mouse hepatocytes electroporated with Cas9 RNPs targeting Hpd, the gene encoding hydroxyphenylpyruvate dioxygenase. In primary human hepatocytes, we observed on-target indels of 52.4% with Cas9 RNPs and >65% viability after electroporation. These results establish the impact of using electroporation to deliver Cas9 RNPs into primary hepatocytes as a highly efficient and potentially safe approach for therapeutic liver-directed gene editing and the production of liver disease models.

Targeted replacement of full-length CFTR in human airway stem cells by CRISPR-Cas9 for pan-mutation correction in the endogenous locus.

Cystic fibrosis (CF) is a monogenic disease caused by impaired production and/or function of the CF transmembrane conductance regulator (CFTR) protein. Although we have previously shown correction of the most common pathogenic mutation, there are many other pathogenic mutations throughout the CF gene. An autologous airway stem cell therapy in which the CFTR cDNA is precisely inserted into the CFTR locus may enable the development of a durable cure for almost all CF patients, irrespective of the causal mutation. Here, we use CRISPR-Cas9 and two adeno-associated viruses (AAVs) carrying the two halves of the CFTR cDNA to sequentially insert the full CFTR cDNA along with a truncated CD19 (tCD19) enrichment tag in upper airway basal stem cells (UABCs) and human bronchial epithelial cells (HBECs). The modified cells were enriched to obtain 60%-80% tCD19+ UABCs and HBECs from 11 different CF donors with a variety of mutations. Differentiated epithelial monolayers cultured at air-liquid interface showed restored CFTR function that was >70% of the CFTR function in non-CF controls. Thus, our study enables the development of a therapy for almost all CF patients, including patients who cannot be treated using recently approved modulator therapies.

In vivo genome editing at the albumin locus to treat methylmalonic acidemia.

Methylmalonic acidemia (MMA) is a metabolic disorder most commonly caused by mutations in the methylmalonyl-CoA mutase (MMUT) gene. Although adeno-associated viral (AAV) gene therapy has been effective at correcting the disease phenotype in MMA mouse models, clinical translation may be impaired by loss of episomal transgene expression and magnified by the need to treat patients early in life. To achieve permanent correction, we developed a dual AAV strategy to express a codon-optimized MMUT transgene from Alb and tested various CRISPR-Cas9 genome-editing vectors in newly developed knockin mouse models of MMA. For one target site in intron 1 of Alb, we designed rescue cassettes expressing MMUT behind a 2A-peptide or an internal ribosomal entry site sequence. A second guide RNA targeted the initiator codon, and the donor cassette encompassed the proximal albumin promoter in the 5' homology arm. Although all editing approaches were therapeutic, targeting the start codon of albumin allowed the use of a donor cassette that also functioned as an episome and after homologous recombination, even without the expression of Cas9, as an integrant. Targeting the albumin locus using these strategies would be effective for other metabolic disorders where early treatment and permanent long-term correction are needed.

TNF-α synergises with IFN-γ to induce caspase-8-JAK1/2-STAT1-dependent death of intestinal epithelial cells.

Rewiring of host cytokine networks is a key feature of inflammatory bowel diseases (IBD) such as Crohn's disease (CD). Th1-type cytokines-IFN-γ and TNF-α-occupy critical nodes within these networks and both are associated with disruption of gut epithelial barrier function. This may be due to their ability to synergistically trigger the death of intestinal epithelial cells (IECs) via largely unknown mechanisms. In this study, through unbiased kinome RNAi and drug repurposing screens we identified JAK1/2 kinases as the principal and nonredundant drivers of the synergistic killing of human IECs by IFN-γ/TNF-α. Sensitivity to IFN-γ/TNF-α-mediated synergistic IEC death was retained in primary patient-derived intestinal organoids. Dependence on JAK1/2 was confirmed using genetic loss-of-function studies and JAK inhibitors (JAKinibs). Despite the presence of biochemical features consistent with canonical TNFR1-mediated apoptosis and necroptosis, IFN-γ/TNF-α-induced IEC death was independent of RIPK1/3, ZBP1, MLKL or caspase activity. Instead, it involved sustained activation of JAK1/2-STAT1 signalling, which required a nonenzymatic scaffold function of caspase-8 (CASP8). Further modelling in gut mucosal biopsies revealed an intercorrelated induction of the lethal CASP8-JAK1/2-STAT1 module during ex vivo stimulation of T cells. Functional studies in CD-derived organoids using inhibitors of apoptosis, necroptosis and JAKinibs confirmed the causative role of JAK1/2-STAT1 in cytokine-induced death of primary IECs. Collectively, we demonstrate that TNF-α synergises with IFN-γ to kill IECs via the CASP8-JAK1/2-STAT1 module independently of canonical TNFR1 and cell death signalling. This non-canonical cell death pathway may underpin immunopathology driven by IFN-γ/TNF-α in diverse autoinflammatory diseases such as IBD, and its inhibition may contribute to the therapeutic efficacy of anti-TNFs and JAKinibs.

The TRACE-Seq method tracks recombination alleles and identifies clonal reconstitution dynamics of gene targeted human hematopoietic stem cells.

Targeted DNA correction of disease-causing mutations in hematopoietic stem and progenitor cells (HSPCs) may enable the treatment of genetic diseases of the blood and immune system. It is now possible to correct mutations at high frequencies in HSPCs by combining CRISPR/Cas9 with homologous DNA donors. Because of the precision of gene correction, these approaches preclude clonal tracking of gene-targeted HSPCs. Here, we describe Tracking Recombination Alleles in Clonal Engraftment using sequencing (TRACE-Seq), a methodology that utilizes barcoded AAV6 donor template libraries, carrying in-frame silent mutations or semi-randomized nucleotides outside the coding region, to track the in vivo lineage contribution of gene-targeted HSPC clones. By targeting the HBB gene with an AAV6 donor template library consisting of ~20,000 possible unique exon 1 in-frame silent mutations, we track the hematopoietic reconstitution of HBB targeted myeloid-skewed, lymphoid-skewed, and balanced multi-lineage repopulating human HSPC clones in mice. We anticipate this methodology could potentially be used for HSPC clonal tracking of Cas9 RNP and AAV6-mediated gene targeting outcomes in translational and basic research settings.

Genome editing of donor-derived T-cells to generate allogenic chimeric antigen receptor-modified T cells: Optimizing αß T cell-depleted haploidentical hematopoietic stem cell transplantation.

Allogeneic hematopoietic stem cell transplantation is an effective therapy for high-risk leukemias. In children, graft manipulation based on the selective removal of aß T cells and B cells has been shown to reduce the risk of acute and chronic graft-versus-host disease, thus allowing the use of haploidentical donors which expands the population of recipients in whom allogeneic hematopoietic stem cell transplantation can be used. Leukemic relapse, however, remains a challenge. T cells expressing chimeric antigen receptors can potently eliminate leukemia, including those in the central nervous system. We hypothesized that by engineering the donor aß T cells that are removed from the graft by genome editing to express a CD19-specific chimeric antigen receptor, while simultaneously inactivating the T-cell receptor, we could create a therapy that enhances the anti-leukemic efficacy of the stem cell transplant without increasing the risk of graft-versus-host disease. Using genome editing with Cas9 ribonucleoprotein and adeno-associated virus serotype 6, we integrated a CD19-specific chimeric antigen receptor inframe into the TRAC locus. More than 90% of cells lost T-cell receptor expression, while >75% expressed the chimeric antigen receptor. The initial product was further purified with less than 0.05% T-cell receptorpositive cells remaining. In vitro, the chimeric antigen receptor T cells efficiently eliminated target cells and produced high cytokine levels when challenged with CD19+ leukemia cells. In vivo, the gene-modified T cells eliminated leukemia without causing graft-versus-host disease in a xenograft model. Gene editing was highly specific with no evidence of off-target effects. These data support the concept that the addition of aß T-cell-derived, genome-edited T cells expressing CD19-specific chimeric antigen receptors could enhance the anti-leukemic efficacy of aß T-celldepleted haploidentical hematopoietic stem cell transplantation without increasing the risk of graft-versus-host disease.

Tools for experimental and computational analyses of off-target editing by programmable nucleases.

Genome editing using programmable nucleases is revolutionizing life science and medicine. Off-target editing by these nucleases remains a considerable concern, especially in therapeutic applications. Here we review tools developed for identifying potential off-target editing sites and compare the ability of these tools to properly analyze off-target effects. Recent advances in both in silico and experimental tools for off-target analysis have generated remarkably concordant results for sites with high off-target editing activity. However, no single tool is able to accurately predict low-frequency off-target editing, presenting a bottleneck in therapeutic genome editing, because even a small number of cells with off-target editing can be detrimental. Therefore, we recommend that at least one in silico tool and one experimental tool should be used together to identify potential off-target sites, and amplicon-based next-generation sequencing (NGS) should be used as the gold standard assay for assessing the true off-target effects at these candidate sites. Future work to improve off-target analysis includes expanding the true off-target editing dataset to evaluate new experimental techniques and to train machine learning algorithms; performing analysis using the particular genome of the cells in question rather than the reference genome; and applying novel NGS techniques to improve the sensitivity of amplicon-based off-target editing quantification.

Publisher Correction: Improving the accuracy of medical diagnosis with causal machine learning.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Improving the accuracy of medical diagnosis with causal machine learning.

Machine learning promises to revolutionize clinical decision making and diagnosis. In medical diagnosis a doctor aims to explain a patient's symptoms by determining the diseases causing them. However, existing machine learning approaches to diagnosis are purely associative, identifying diseases that are strongly correlated with a patients symptoms. We show that this inability to disentangle correlation from causation can result in sub-optimal or dangerous diagnoses. To overcome this, we reformulate diagnosis as a counterfactual inference task and derive counterfactual diagnostic algorithms. We compare our counterfactual algorithms to the standard associative algorithm and 44 doctors using a test set of clinical vignettes. While the associative algorithm achieves an accuracy placing in the top 48% of doctors in our cohort, our counterfactual algorithm places in the top 25% of doctors, achieving expert clinical accuracy. Our results show that causal reasoning is a vital missing ingredient for applying machine learning to medical diagnosis.

Metabolic engineering generates a transgene-free safety switch for cell therapy.

Safeguard mechanisms can ameliorate the potential risks associated with cell therapies but currently rely on the introduction of transgenes. This limits their application owing to immunogenicity or transgene silencing. We aimed to create a control mechanism for human cells that is not mediated by a transgene. Using genome editing methods, we disrupt uridine monophosphate synthetase (UMPS) in the pyrimidine de novo synthesis pathway in cell lines, pluripotent cells and primary human T cells. We show that this makes proliferation dependent on external uridine and enables us to control cell growth by modulating the uridine supply, both in vitro and in vivo after transplantation in xenograft models. Additionally, disrupting this pathway creates resistance to 5-fluoroorotic acid, which enables positive selection of UMPS-knockout cells. We envision that this approach will add an additional level of safety to cell therapies and therefore enable the development of approaches with higher risks, especially those that are intended for limited treatment durations.

Pitfalls in Single Clone CRISPR-Cas9 Mutagenesis to Fine-map Regulatory Intervals.

The majority of genetic variants affecting complex traits map to regulatory regions of genes, and typically lie in credible intervals of 100 or more SNPs. Fine mapping of the causal variant(s) at a locus depends on assays that are able to discriminate the effects of polymorphisms or mutations on gene expression. Here, we evaluated a moderate-throughput CRISPR-Cas9 mutagenesis approach, based on replicated measurement of transcript abundance in single-cell clones, by deleting candidate regulatory SNPs, affecting four genes known to be affected by large-effect expression Quantitative Trait Loci (eQTL) in leukocytes, and using Fluidigm qRT-PCR to monitor gene expression in HL60 pro-myeloid human cells. We concluded that there were multiple constraints that rendered the approach generally infeasible for fine mapping. These included the non-targetability of many regulatory SNPs, clonal variability of single-cell derivatives, and expense. Power calculations based on the measured variance attributable to major sources of experimental error indicated that typical eQTL explaining 10% of the variation in expression of a gene would usually require at least eight biological replicates of each clone. Scanning across credible intervals with this approach is not recommended.

AAV-CRISPR Gene Editing Is Negated by Pre-existing Immunity to Cas9.

Adeno-associated viral (AAV) vectors are a leading candidate for the delivery of CRISPR-Cas9 for therapeutic genome editing in vivo. However, AAV-based delivery involves persistent expression of the Cas9 nuclease, a bacterial protein. Recent studies indicate a high prevalence of neutralizing antibodies and T cells specific to the commonly used Cas9 orthologs from Streptococcus pyogenes (SpCas9) and Staphylococcus aureus (SaCas9) in humans. We tested in a mouse model whether pre-existing immunity to SaCas9 would pose a barrier to liver genome editing with AAV packaging CRISPR-Cas9. Although efficient genome editing occurred in mouse liver with pre-existing SaCas9 immunity, this was accompanied by an increased proportion of CD8+ T cells in the liver. This cytotoxic T cell response was characterized by hepatocyte apoptosis, loss of recombinant AAV genomes, and complete elimination of genome-edited cells, and was followed by compensatory liver regeneration. Our results raise important efficacy and safety concerns for CRISPR-Cas9-based in vivo genome editing in the liver.

Site-Specific Post-translational Surface Modification of Adeno-Associated Virus Vectors Using Leucine Zippers.

Adeno-associated virus (AAV) is widely favored as a gene therapy vector, tested in over 200 clinical trials internationally. To improve targeted delivery a variety of genetic capsid modifications, such as insertion of targeting proteins/peptides into the capsid shell, have been explored with some success but larger insertions often have unpredictable deleterious impacts on capsid formation and gene delivery. Here, we demonstrate a modular platform for the integration of exogenous peptides and proteins onto the AAV capsid post-translationally while preserving vector functionality. We decorated the AAV capsid with leucine-zipper coiled-coil binding motifs that exhibit specific noncovalent heterodimerization. AAV capsids successfully display hexahistidine tagged-peptides using this approach, as demonstrated through nickel column affinity. This protein display platform may facilitate the incorporation of biological moieties on the AAV surface, expanding possibilities for vector enhancement and engineering.

High-Efficiency, Selection-free Gene Repair in Airway Stem Cells from Cystic Fibrosis Patients Rescues CFTR Function in Differentiated Epithelia.

Cystic fibrosis (CF) is a monogenic disorder caused by mutations in the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) gene. Mortality in CF patients is mostly due to respiratory sequelae. Challenges with gene delivery have limited attempts to treat CF using in vivo gene therapy, and low correction levels have hindered ex vivo gene therapy efforts. We have used Cas9 and adeno-associated virus 6 to correct the ΔF508 mutation in readily accessible upper-airway basal stem cells (UABCs) obtained from CF patients. On average, we achieved 30%-50% allelic correction in UABCs and bronchial epithelial cells (HBECs) from 10 CF patients and observed 20%-50% CFTR function relative to non-CF controls in differentiated epithelia. Furthermore, we successfully embedded the corrected UABCs on an FDA-approved porcine small intestinal submucosal membrane (pSIS), and they retained differentiation capacity. This study supports further development of genetically corrected autologous airway stem cell transplant as a treatment for CF.

Therapeutically relevant engraftment of a CRISPR-Cas9-edited HSC-enriched population with HbF reactivation in nonhuman primates.

Reactivation of fetal hemoglobin (HbF) is being pursued as a treatment strategy for hemoglobinopathies. Here, we evaluated the therapeutic potential of hematopoietic stem and progenitor cells (HSPCs) edited with the CRISPR-Cas9 nuclease platform to recapitulate naturally occurring mutations identified in individuals who express increased amounts of HbF, a condition known as hereditary persistence of HbF. CRISPR-Cas9 treatment and transplantation of HSPCs purified on the basis of surface expression of the CD34 receptor in a nonhuman primate (NHP) autologous transplantation model resulted in up to 30% engraftment of gene-edited cells for >1 year. Edited cells effectively and stably reactivated HbF, as evidenced by up to 18% HbF-expressing erythrocytes in peripheral blood. Similar results were obtained by editing highly enriched stem cells, defined by the markers CD34+CD90+CD45RA-, allowing for a 10-fold reduction in the number of transplanted target cells, thus considerably reducing the need for editing reagents. The frequency of engrafted, gene-edited cells persisting in vivo using this approach may be sufficient to ameliorate the phenotype for a number of genetic diseases.

Highly efficient editing of the ß-globin gene in patient-derived hematopoietic stem and progenitor cells to treat sickle cell disease.

Sickle cell disease (SCD) is a monogenic disorder that affects millions worldwide. Allogeneic hematopoietic stem cell transplantation is the only available cure. Here, we demonstrate the use of CRISPR/Cas9 and a short single-stranded oligonucleotide template to correct the sickle mutation in the ß-globin gene in hematopoietic stem and progenitor cells (HSPCs) from peripheral blood or bone marrow of patients with SCD, with 24.5 ± 7.6% efficiency without selection. Erythrocytes derived from gene-edited cells showed a marked reduction of sickle cells, with the level of normal hemoglobin (HbA) increased to 25.3 ± 13.9%. Gene-corrected SCD HSPCs retained the ability to engraft when transplanted into non-obese diabetic (NOD)-SCID-gamma (NSG) mice with detectable levels of gene correction 16-19 weeks post-transplantation. We show that, by using a high-fidelity SpyCas9 that maintained the same level of on-target gene modification, the off-target effects including chromosomal rearrangements were significantly reduced. Taken together, our results demonstrate efficient gene correction of the sickle mutation in both peripheral blood and bone marrow-derived SCD HSPCs, a significant reduction in sickling of red blood cells, engraftment of gene-edited SCD HSPCs in vivo and the importance of reducing off-target effects; all are essential for moving genome editing based SCD treatment into clinical practice.