International evaluation of an AI system for breast cancer screening.

Screening mammography aims to identify breast cancer at earlier stages of the disease, when treatment can be more successful1. Despite the existence of screening programmes worldwide, the interpretation of mammograms is affected by high rates of false positives and false negatives2. Here we present an artificial intelligence (AI) system that is capable of surpassing human experts in breast cancer prediction. To assess its performance in the clinical setting, we curated a large representative dataset from the UK and a large enriched dataset from the USA. We show an absolute reduction of 5.7% and 1.2% (USA and UK) in false positives and 9.4% and 2.7% in false negatives. We provide evidence of the ability of the system to generalize from the UK to the USA. In an independent study of six radiologists, the AI system outperformed all of the human readers: the area under the receiver operating characteristic curve (AUC-ROC) for the AI system was greater than the AUC-ROC for the average radiologist by an absolute margin of 11.5%. We ran a simulation in which the AI system participated in the double-reading process that is used in the UK, and found that the AI system maintained non-inferior performance and reduced the workload of the second reader by 88%. This robust assessment of the AI system paves the way for clinical trials to improve the accuracy and efficiency of breast cancer screening.

The nuclear matrix protein CIZ1 facilitates localization of Xist RNA to the inactive X-chromosome territory.

The nuclear matrix protein Cip1-interacting zinc finger protein 1 (CIZ1) promotes DNA replication in association with cyclins and has been linked to adult and pediatric cancers. Here we show that CIZ1 is highly enriched on the inactive X chromosome (Xi) in mouse and human female cells and is retained by interaction with the RNA-dependent nuclear matrix. CIZ1 is recruited to Xi in response to expression of X inactive-specific transcript (Xist) RNA during the earliest stages of X inactivation in embryonic stem cells and is dependent on the C-terminal nuclear matrix anchor domain of CIZ1 and the E repeats of Xist CIZ1-null mice, although viable, display fully penetrant female-specific lymphoproliferative disorder. Interestingly, in mouse embryonic fibroblast cells derived from CIZ1-null embryos, Xist RNA localization is disrupted, being highly dispersed through the nucleoplasm rather than focal. Focal localization is reinstated following re-expression of CIZ1. Focal localization of Xist RNA is also disrupted in activated B and T cells isolated from CIZ1-null animals, suggesting a possible explanation for female-specific lymphoproliferative disorder. Together, these findings suggest that CIZ1 has an essential role in anchoring Xist to the nuclear matrix in specific somatic lineages.

PLCζ is the physiological trigger of the Ca2+ oscillations that induce embryogenesis in mammals but conception can occur in its absence.

Activation of the egg by the sperm is the first, vital stage of embryogenesis. The sperm protein PLCζ has been proposed as the physiological agent that triggers the Ca2+ oscillations that normally initiate embryogenesis. Consistent with this, recombinant PLCζ induces Ca2+ oscillations in eggs and debilitating mutations in the PLCZ1 gene are associated with infertility in men. However, there has been no evidence that knockout of the gene encoding PLCζ abolishes the ability of sperm to induce Ca2+ oscillations in eggs. Here, we show that sperm derived from Plcz1-/- male mice fail to trigger Ca2+ oscillations in eggs, cause polyspermy and thus demonstrate that PLCζ is the physiological trigger of these Ca2+ oscillations. Remarkably, some eggs fertilized by PLCζ-null sperm can develop, albeit at greatly reduced efficiency, and after a significant time-delay. In addition, Plcz1-/- males are subfertile but not sterile, suggesting that in the absence of PLCζ, spontaneous egg activation can eventually occur via an alternative route. This is the first demonstration that in vivo fertilization without the normal physiological trigger of egg activation can result in offspring. PLCζ-null sperm now make it possible to resolve long-standing questions in fertilization biology, and to test the efficacy and safety of procedures used to treat human infertility.

Functional analysis of AEBP2, a PRC2 Polycomb protein, reveals a Trithorax phenotype in embryonic development and in ESCs.

The Polycomb repressive complexes PRC1 and PRC2 are key mediators of heritable gene silencing in multicellular organisms. Here, we characterise AEBP2, a known PRC2 co-factor which, in vitro, has been shown to stimulate PRC2 activity. We show that AEBP2 localises specifically to PRC2 target loci, including the inactive X chromosome. Proteomic analysis confirms that AEBP2 associates exclusively with PRC2 complexes. However, analysis of embryos homozygous for a targeted mutation of Aebp2 unexpectedly revealed a Trithorax phenotype, normally linked to antagonism of Polycomb function. Consistent with this, we observe elevated levels of PRC2-mediated histone H3K27 methylation at target loci in Aebp2 mutant embryonic stem cells (ESCs). We further demonstrate that mutant ESCs assemble atypical hybrid PRC2 subcomplexes, potentially accounting for enhancement of Polycomb activity, and suggesting that AEBP2 normally plays a role in defining the mutually exclusive composition of PRC2 subcomplexes.

Rec8-containing cohesin maintains bivalents without turnover during the growing phase of mouse oocytes.

During female meiosis, bivalent chromosomes are thought to be held together from birth until ovulation by sister chromatid cohesion mediated by cohesin complexes whose ring structure depends on kleisin subunits, either Rec8 or Scc1. Because cohesion is established at DNA replication in the embryo, its maintenance for such a long time may require cohesin turnover. To address whether Rec8- or Scc1-containing cohesin holds bivalents together and whether it turns over, we created mice whose kleisin subunits can be cleaved by TEV protease. We show by microinjection experiments and confocal live-cell imaging that Rec8 cleavage triggers chiasmata resolution during meiosis I and sister centromere disjunction during meiosis II, while Scc1 cleavage triggers sister chromatid disjunction in the first embryonic mitosis, demonstrating a dramatic transition from Rec8- to Scc1-containing cohesin at fertilization. Crucially, activation of an ectopic Rec8 transgene during the growing phase of Rec8(TEV)(/TEV) oocytes does not prevent TEV-mediated bivalent destruction, implying little or no cohesin turnover for ≥2 wk during oocyte growth. We suggest that the inability of oocytes to regenerate cohesion may contribute to age-related meiosis I errors.

Rnf165/Ark2C enhances BMP-Smad signaling to mediate motor axon extension.

Little is known about extrinsic signals required for the advancement of motor neuron (MN) axons, which extend over long distances in the periphery to form precise connections with target muscles. Here we present that Rnf165 (Arkadia-like; Arkadia2; Ark2C) is expressed specifically in the nervous system and that its loss in mice causes motor innervation defects that originate during development and lead to wasting and death before weaning. The defects range from severe reduction of motor axon extension as observed in the dorsal forelimb to shortening of presynaptic branches of the phrenic nerve, as observed in the diaphragm. Molecular functional analysis showed that in the context of the spinal cord Ark2C enhances transcriptional responses of the Smad1/5/8 effectors, which are activated (phosphorylated) downstream of Bone Morphogenetic Protein (BMP) signals. Consistent with Ark2C-modulated BMP signaling influencing motor axons, motor pools in the spinal cord were found to harbor phosphorylated Smad1/5/8 (pSmad) and treatment of primary MN with BMP inhibitor diminished axon length. In addition, genetic reduction of BMP-Smad signaling in Ark2C (+/-) mice caused the emergence of Ark2C (-/-) -like dorsal forelimb innervation deficits confirming that enhancement of BMP-Smad responses by Ark2C mediates efficient innervation. Together the above data reveal an involvement of BMP-Smad signaling in motor axon advancement.

Disruption of a conserved region of Xist exon 1 impairs Xist RNA localisation and X-linked gene silencing during random and imprinted X chromosome inactivation.

In XX female mammals a single X chromosome is inactivated early in embryonic development, a process that is required to equalise X-linked gene dosage relative to XY males. X inactivation is regulated by a cis-acting master switch, the Xist locus, the product of which is a large non-coding RNA that coats the chromosome from which it is transcribed, triggering recruitment of chromatin modifying factors that establish and maintain gene silencing chromosome wide. Chromosome coating and Xist RNA-mediated silencing remain poorly understood, both at the level of RNA sequence determinants and interacting factors. Here, we describe analysis of a novel targeted mutation, Xist(INV), designed to test the function of a conserved region located in exon 1 of Xist RNA during X inactivation in mouse. We show that Xist(INV) is a strong hypomorphic allele that is appropriately regulated but compromised in its ability to silence X-linked loci in cis. Inheritance of Xist(INV) on the paternal X chromosome results in embryonic lethality due to failure of imprinted X inactivation in extra-embryonic lineages. Female embryos inheriting Xist(INV) on the maternal X chromosome undergo extreme secondary non-random X inactivation, eliminating the majority of cells that express the Xist(INV) allele. Analysis of cells that express Xist(INV) RNA demonstrates reduced association of the mutant RNA to the X chromosome, suggesting that conserved sequences in the inverted region are important for Xist RNA localisation.

An integrated approach identifies the molecular underpinnings of murine anterior visceral endoderm migration.

The anterior visceral endoderm (AVE) differs from the surrounding visceral endoderm (VE) in its migratory behavior and ability to restrict primitive streak formation to the opposite side of the mouse embryo. To characterize the molecular bases for the unique properties of the AVE, we combined single-cell RNA sequencing of the VE prior to and during AVE migration with phosphoproteomics, high-resolution live-imaging, and short-term lineage labeling and intervention. This identified the transient nature of the AVE with attenuation of "anteriorizing" gene expression as cells migrate and the emergence of heterogeneities in transcriptional states relative to the AVE's position. Using cell communication analysis, we identified the requirement of semaphorin signaling for normal AVE migration. Lattice light-sheet microscopy showed that Sema6D mutants have abnormalities in basal projections and migration speed. These findings point to a tight coupling between transcriptional state and position of the AVE and identify molecular controllers of AVE migration.

Targeting the CXCR4-CXCL12 axis mobilizes autologous hematopoietic stem cells and prolongs islet allograft survival via programmed death ligand 1.

Antagonism of CXCR4 disrupts the interaction between the CXCR4 receptor on hematopoietic stem cells (HSCs) and the CXCL12 expressed by stromal cells in the bone marrow, which subsequently results in the shedding of HSCs to the periphery. Because of their profound immunomodulatory effects, HSCs have emerged as a promising therapeutic strategy for autoimmune disorders. We sought to investigate the immunomodulatory role of mobilized autologous HSCs, via target of the CXCR4-CXL12 axis, to promote engraftment of islet cell transplantation. Islets from BALB/c mice were transplanted beneath the kidney capsule of hyperglycemic C57BL/6 mice, and treatment of recipients with CXCR4 antagonist resulted in mobilization of HSCs and in prolongation of islet graft survival. Addition of rapamycin to anti-CXCR4 therapy further promoted HSC mobilization and islet allograft survival, inducing a robust and transferable host hyporesponsiveness, while administration of an ACK2 (anti-CD117) mAb halted CXCR4 antagonist-mediated HSC release and restored allograft rejection. Mobilized HSCs were shown to express high levels of the negative costimulatory molecule programmed death ligand 1 (PD-L1), and HSCs extracted from wild-type mice, but not from PD-L1 knockout mice, suppressed the in vitro alloimmune response. Moreover, HSC mobilization in PD-L1 knockout mice failed to prolong islet allograft survival. Targeting the CXCR4-CXCL12 axis thus mobilizes autologous HSCs and promotes long-term survival of islet allografts via a PD-L1-mediated mechanism.

Identification of a microRNA signature of renal ischemia reperfusion injury.

Renal ischemia reperfusion injury (IRI) is associated with significant morbidity and mortality. Given the importance of microRNAs (miRNAs) in regulating gene expression, we examined expression profiles of miRNAs following renal IRI. Global miRNA expression profiling on samples prepared from the kidneys of C57BL/6 mice that underwent unilateral warm ischemia revealed nine miRNAs (miR-21, miR-20a, miR-146a, miR-199a-3p, miR-214, miR-192, miR-187, miR-805, and miR-194) that are differentially expressed following IRI when compared with sham controls. These miRNAs were also differently expressed following IRI in immunodeficient RAG-2/common gamma-chain double-knockout mice, suggesting that the changes in expression observed are not significantly influenced by lymphocyte infiltration and therefore define a lymphocyte-independent signature of renal IRI. In vitro studies revealed that miR-21 is expressed in proliferating tubular epithelial cells (TEC) and up-regulated by both cell-intrinsic and -extrinsic mechanisms resulting from ischemia and TGF-beta signaling, respectively. In vitro, knockdown of miR-21 in TEC resulted in increased cell death, whereas overexpression prevented cell death. However, overexpression of miR-21 alone was not sufficient to prevent TEC death following ischemia. Our findings therefore define a molecular fingerprint of renal injury and suggest miR-21 may play a role in protecting TEC from death.

Costimulation-dependent expression of microRNA-214 increases the ability of T cells to proliferate by targeting Pten.

T cell activation requires signaling through the TCR and costimulatory molecules, such as CD28. MicroRNAs (miRNAs) are small noncoding RNAs that regulate gene expression posttranscriptionally and are also known to be involved in lymphocyte development and function. In this paper, we set out to examine potential roles of miRNAs in T cell activation, using genome-wide expression profiling to identify miRNAs differentially regulated following T cell activation. One of the miRNAs upregulated after T cell activation, miR-214, was predicted to be capable of targeting Pten based on bioinformatics and reports suggesting that it targets Pten in ovarian tumor cells. Upregulation of miR-214 in T cells inversely correlated with levels of phosphatase and tensin homolog deleted on chromosome 10. In vivo, transcripts containing the 3' untranslated region of Pten, including the miR-214 target sequence, were negatively regulated after T cell activation, and forced expression of miR-214 in T cells led to increased proliferation after stimulation. Blocking CD28 signaling in vivo prevented miR-214 upregulation in alloreactive T cells. Stimulation of T cells through the TCR alone was not sufficient to result in upregulation of miR-214. Thus, costimulation-dependent upregulation of miR-214 promotes T cell activation by targeting the negative regulator Pten. Thus, the requirement for T cell costimulation is, in part, related to its ability to regulate expression of miRNAs that control T cell activation.

Prion-like domains drive CIZ1 assembly formation at the inactive X chromosome.

CIZ1 forms large assemblies at the inactive X chromosome (Xi) in female fibroblasts in an Xist lncRNA-dependent manner and is required for accurate maintenance of polycomb targets genome-wide. Here we address requirements for assembly formation and show that CIZ1 undergoes two direct interactions with Xist, via independent N- and C-terminal domains. Interaction with Xist, assembly at Xi, and complexity of self-assemblies formed in vitro are modulated by two alternatively spliced glutamine-rich prion-like domains (PLD1 and 2). PLD2 is dispensable for accumulation at existing CIZ1-Xi assemblies in wild-type cells but is required in CIZ1-null cells where targeting, assembly, and enrichment for H3K27me3 and H2AK119ub occur de novo. In contrast, PLD1 is required for both de novo assembly and accumulation at preexisting assemblies and, in vitro, drives formation of a stable fibrillar network. Together they impart affinity for RNA and a complex relationship with repeat E of Xist. These data show that alternative splicing of two PLDs modulates CIZ1's ability to build large RNA-protein assemblies.

ASPP2 maintains the integrity of mechanically stressed pseudostratified epithelia during morphogenesis.

During development, pseudostratified epithelia undergo large scale morphogenetic events associated with increased mechanical stress. Using a variety of genetic and imaging approaches, we uncover that in the mouse E6.5 epiblast, where apical tension is highest, ASPP2 safeguards tissue integrity. It achieves this by preventing the most apical daughter cells from delaminating apically following division events. In this context, ASPP2 maintains the integrity and organisation of the filamentous actin cytoskeleton at apical junctions. ASPP2 is also essential during gastrulation in the primitive streak, in somites and in the head fold region, suggesting that it is required across a wide range of pseudostratified epithelia during morphogenetic events that are accompanied by intense tissue remodelling. Finally, our study also suggests that the interaction between ASPP2 and PP1 is essential to the tumour suppressor function of ASPP2, which may be particularly relevant in the context of tissues that are subject to increased mechanical stress.

T cell lineage choice and differentiation in the absence of the RNase III enzyme Dicer.

The ribonuclease III enzyme Dicer is essential for the processing of micro-RNAs (miRNAs) and small interfering RNAs (siRNAs) from double-stranded RNA precursors. miRNAs and siRNAs regulate chromatin structure, gene transcription, mRNA stability, and translation in a wide range of organisms. To provide a model system to explore the role of Dicer-generated RNAs in the differentiation of mammalian cells in vivo, we have generated a conditional Dicer allele. Deletion of Dicer at an early stage of T cell development compromised the survival of alphabeta lineage cells, whereas the numbers of gammadelta-expressing thymocytes were not affected. In developing thymocytes, Dicer was not required for the maintenance of transcriptional silencing at pericentromeric satellite sequences (constitutive heterochromatin), the maintenance of DNA methylation and X chromosome inactivation in female cells (facultative heterochromatin), and the stable shutdown of a developmentally regulated gene (developmentally regulated gene silencing). Most remarkably, given that one third of mammalian mRNAs are putative miRNA targets, Dicer seems to be dispensable for CD4/8 lineage commitment, a process in which epigenetic regulation of lineage choice has been well documented. Thus, although Dicer seems to be critical for the development of the early embryo, it may have limited impact on the implementation of some lineage-specific gene expression programs.

Immunomodulatory function of bone marrow-derived mesenchymal stem cells in experimental autoimmune type 1 diabetes.

Human clinical trials in type 1 diabetes (T1D) patients using mesenchymal stem cells (MSC) are presently underway without prior validation in a mouse model for the disease. In response to this void, we characterized bone marrow-derived murine MSC for their ability to modulate immune responses in the context of T1D, as represented in NOD mice. In comparison to NOD mice, BALB/c-MSC mice were found to express higher levels of the negative costimulatory molecule PD-L1 and to promote a shift toward Th2-like responses in treated NOD mice. In addition, transfer of MSC from resistant strains (i.e., nonobese resistant mice or BALB/c), but not from NOD mice, delayed the onset of diabetes when administered to prediabetic NOD mice. The number of BALB/c-MSC trafficking to the pancreatic lymph nodes of NOD mice was higher than in NOD mice provided autologous NOD-MSC. Administration of BALB/c-MSC temporarily resulted in reversal of hyperglycemia in 90% of NOD mice (p = 0.002). Transfer of autologous NOD-MSC imparted no such therapeutic benefit. We also noted soft tissue and visceral tumors in NOD-MSC-treated mice, which were uniquely observed in this setting (i.e., no tumors were present with BALB/c- or nonobese resistant mice-MSC transfer). The importance of this observation remains to be explored in humans, as inbred mice such as NOD may be more susceptible to tumor formation. These data provide important preclinical data supporting the basis for further development of allogeneic MSC-based therapies for T1D and, potentially, for other autoimmune disorders.

Loss of sister kinetochore co-orientation and peri-centromeric cohesin protection after meiosis I depends on cleavage of centromeric REC8.

Protection of peri-centromeric (periCEN) REC8 cohesin from Separase and sister kinetochore (KT) attachment to microtubules emanating from the same spindle pole (co-orientation) ensures that sister chromatids remain associated after meiosis I. Both features are lost during meiosis II, resulting in sister chromatid disjunction and the production of haploid gametes. By transferring spindle-chromosome complexes (SCCs) between meiosis I and II in mouse oocytes, we discovered that both sister KT co-orientation and periCEN cohesin protection depend on the SCC, and not the cytoplasm. Moreover, the catalytic activity of Separase at meiosis I is necessary not only for converting KTs from a co- to a bi-oriented state but also for deprotection of periCEN cohesion, and cleavage of REC8 may be the key event. Crucially, selective cleavage of REC8 in the vicinity of KTs is sufficient to destroy co-orientation in univalent chromosomes, albeit not in bivalents where resolution of chiasmata may also be required.

Characterization of a common progenitor pool of the epicardium and myocardium.

The mammalian heart is derived from multiple cell lineages; however, our understanding of when and how the diverse cardiac cell types arise is limited. We mapped the origin of the embryonic mouse heart at single-cell resolution using a combination of transcriptomic, imaging, and genetic lineage labeling approaches. This mapping provided a transcriptional and anatomic definition of cardiac progenitor types. Furthermore, it revealed a cardiac progenitor pool that is anatomically and transcriptionally distinct from currently known cardiac progenitors. Besides contributing to cardiomyocytes, these cells also represent the earliest progenitor of the epicardium, a source of trophic factors and cells during cardiac development and injury. This study provides detailed insights into the formation of early cardiac cell types, with particular relevance to the development of cell-based cardiac regenerative therapies.

MCPH1 inhibits Condensin II during interphase by regulating its SMC2-Kleisin interface.

Dramatic change in chromosomal DNA morphology between interphase and mitosis is a defining features of the eukaryotic cell cycle. Two types of enzymes, namely cohesin and condensin confer the topology of chromosomal DNA by extruding DNA loops. While condensin normally configures chromosomes exclusively during mitosis, cohesin does so during interphase. The processivity of cohesin's loop extrusion during interphase is limited by a regulatory factor called WAPL, which induces cohesin to dissociate from chromosomes via a mechanism that requires dissociation of its kleisin from the neck of SMC3. We show here that a related mechanism may be responsible for blocking condensin II from acting during interphase. Cells derived from patients affected by microcephaly caused by mutations in the MCPH1 gene undergo premature chromosome condensation. We show that deletion of Mcph1 in mouse embryonic stem cells unleashes an activity of condensin II that triggers formation of compact chromosomes in G1 and G2 phases, accompanied by enhanced mixing of A and B chromatin compartments, and this occurs even in the absence of CDK1 activity. Crucially, inhibition of condensin II by MCPH1 depends on the binding of a short linear motif within MCPH1 to condensin II's NCAPG2 subunit. MCPH1's ability to block condensin II's association with chromatin is abrogated by the fusion of SMC2 with NCAPH2, hence may work by a mechanism similar to cohesin. Remarkably, in the absence of both WAPL and MCPH1, cohesin and condensin II transform chromosomal DNAs of G2 cells into chromosomes with a solenoidal axis.

Developing a reporting guideline for artificial intelligence-centred diagnostic test accuracy studies: the STARD-AI protocol.

INTRODUCTION: Standards for Reporting of Diagnostic Accuracy Study (STARD) was developed to improve the completeness and transparency of reporting in studies investigating diagnostic test accuracy. However, its current form, STARD 2015 does not address the issues and challenges raised by artificial intelligence (AI)-centred interventions. As such, we propose an AI-specific version of the STARD checklist (STARD-AI), which focuses on the reporting of AI diagnostic test accuracy studies. This paper describes the methods that will be used to develop STARD-AI. METHODS AND ANALYSIS: The development of the STARD-AI checklist can be distilled into six stages. (1) A project organisation phase has been undertaken, during which a Project Team and a Steering Committee were established; (2) An item generation process has been completed following a literature review, a patient and public involvement and engagement exercise and an online scoping survey of international experts; (3) A three-round modified Delphi consensus methodology is underway, which will culminate in a teleconference consensus meeting of experts; (4) Thereafter, the Project Team will draft the initial STARD-AI checklist and the accompanying documents; (5) A piloting phase among expert users will be undertaken to identify items which are either unclear or missing. This process, consisting of surveys and semistructured interviews, will contribute towards the explanation and elaboration document and (6) On finalisation of the manuscripts, the group's efforts turn towards an organised dissemination and implementation strategy to maximise end-user adoption. ETHICS AND DISSEMINATION: Ethical approval has been granted by the Joint Research Compliance Office at Imperial College London (reference number: 19IC5679). A dissemination strategy will be aimed towards five groups of stakeholders: (1) academia, (2) policy, (3) guidelines and regulation, (4) industry and (5) public and non-specific stakeholders. We anticipate that dissemination will take place in Q3 of 2021.