Early Alzheimer's disease pathology in human cortex involves transient cell states.

Cellular perturbations underlying Alzheimer's disease (AD) are primarily studied in human postmortem samples and model organisms. Here, we generated a single-nucleus atlas from a rare cohort of cortical biopsies from living individuals with varying degrees of AD pathology. We next performed a systematic cross-disease and cross-species integrative analysis to identify a set of cell states that are specific to early AD pathology. These changes-which we refer to as the early cortical amyloid response-were prominent in neurons, wherein we identified a transitional hyperactive state preceding the loss of excitatory neurons, which we confirmed by acute slice physiology on independent biopsy specimens. Microglia overexpressing neuroinflammatory-related processes also expanded as AD pathology increased. Finally, both oligodendrocytes and pyramidal neurons upregulated genes associated with ß-amyloid production and processing during this early hyperactive phase. Our integrative analysis provides an organizing framework for targeting circuit dysfunction, neuroinflammation, and amyloid production early in AD pathogenesis.

Exposure of iPSC-derived human microglia to brain substrates enables the generation and manipulation of diverse transcriptional states in vitro.

Microglia, the macrophages of the brain parenchyma, are key players in neurodegenerative diseases such as Alzheimer's disease. These cells adopt distinct transcriptional subtypes known as states. Understanding state function, especially in human microglia, has been elusive owing to a lack of tools to model and manipulate these cells. Here, we developed a platform for modeling human microglia transcriptional states in vitro. We found that exposure of human stem-cell-differentiated microglia to synaptosomes, myelin debris, apoptotic neurons or synthetic amyloid-beta fibrils generated transcriptional diversity that mapped to gene signatures identified in human brain microglia, including disease-associated microglia, a state enriched in neurodegenerative diseases. Using a new lentiviral approach, we demonstrated that the transcription factor MITF drives a disease-associated transcriptional signature and a highly phagocytic state. Together, these tools enable the manipulation and functional interrogation of human microglial states in both homeostatic and disease-relevant contexts.

The molecular cytoarchitecture of the adult mouse brain.

The function of the mammalian brain relies upon the specification and spatial positioning of diversely specialized cell types. Yet, the molecular identities of the cell types and their positions within individual anatomical structures remain incompletely known. To construct a comprehensive atlas of cell types in each brain structure, we paired high-throughput single-nucleus RNA sequencing with Slide-seq1,2-a recently developed spatial transcriptomics method with near-cellular resolution-across the entire mouse brain. Integration of these datasets revealed the cell type composition of each neuroanatomical structure. Cell type diversity was found to be remarkably high in the midbrain, hindbrain and hypothalamus, with most clusters requiring a combination of at least three discrete gene expression markers to uniquely define them. Using these data, we developed a framework for genetically accessing each cell type, comprehensively characterized neuropeptide and neurotransmitter signalling, elucidated region-specific specializations in activity-regulated gene expression and ascertained the heritability enrichment of neurological and psychiatric phenotypes. These data, available as an online resource ( www.BrainCellData.org ), should find diverse applications across neuroscience, including the construction of new genetic tools and the prioritization of specific cell types and circuits in the study of brain diseases.

A predictive ensemble classifier for the gene expression diagnosis of ASD at ages 1 to 4 years.

Autism Spectrum Disorder (ASD) diagnosis remains behavior-based and the median age of diagnosis is ~52 months, nearly 5 years after its first-trimester origin. Accurate and clinically-translatable early-age diagnostics do not exist due to ASD genetic and clinical heterogeneity. Here we collected clinical, diagnostic, and leukocyte RNA data from 240 ASD and typically developing (TD) toddlers (175 toddlers for training and 65 for test). To identify gene expression ASD diagnostic classifiers, we developed 42,840 models composed of 3570 gene expression feature selection sets and 12 classification methods. We found that 742 models had AUC-ROC ≥ 0.8 on both Training and Test sets. Weighted Bayesian model averaging of these 742 models yielded an ensemble classifier model with accurate performance in Training and Test gene expression datasets with ASD diagnostic classification AUC-ROC scores of 85-89% and AUC-PR scores of 84-92%. ASD toddlers with ensemble scores above and below the overall ASD ensemble mean of 0.723 (on a scale of 0 to 1) had similar diagnostic and psychometric scores, but those below this ASD ensemble mean had more prenatal risk events than TD toddlers. Ensemble model feature genes were involved in cell cycle, inflammation/immune response, transcriptional gene regulation, cytokine response, and PI3K-AKT, RAS and Wnt signaling pathways. We additionally collected targeted DNA sequencing smMIPs data on a subset of ASD risk genes from 217 of the 240 ASD and TD toddlers. This DNA sequencing found about the same percentage of SFARI Level 1 and 2 ASD risk gene mutations in TD (12 of 105) as in ASD (13 of 112) toddlers, and classification based only on the presence of mutation in these risk genes performed at a chance level of 49%. By contrast, the leukocyte ensemble gene expression classifier correctly diagnostically classified 88% of TD and ASD toddlers with ASD risk gene mutations. Our ensemble ASD gene expression classifier is diagnostically predictive and replicable across different toddler ages, races, and ethnicities; out-performs a risk gene mutation classifier; and has potential for clinical translation.

Pre-treatment clinical and gene expression patterns predict developmental change in early intervention in autism.

Early detection and intervention are believed to be key to facilitating better outcomes in children with autism, yet the impact of age at treatment start on the outcome is poorly understood. While clinical traits such as language ability have been shown to predict treatment outcome, whether or not and how information at the genomic level can predict treatment outcome is unknown. Leveraging a cohort of toddlers with autism who all received the same standardized intervention at a very young age and provided a blood sample, here we find that very early treatment engagement (i.e., <24 months) leads to greater gains while controlling for time in treatment. Pre-treatment clinical behavioral measures predict 21% of the variance in the rate of skill growth during early intervention. Pre-treatment blood leukocyte gene expression patterns also predict the rate of skill growth, accounting for 13% of the variance in treatment slopes. Results indicated that 295 genes can be prioritized as driving this effect. These treatment-relevant genes highly interact at the protein level, are enriched for differentially histone acetylated genes in autism postmortem cortical tissue, and are normatively highly expressed in a variety of subcortical and cortical areas important for social communication and language development. This work suggests that pre-treatment biological and clinical behavioral characteristics are important for predicting developmental change in the context of early intervention and that individualized pre-treatment biology related to histone acetylation may be key.

Multiple freeze-thaw cycles lead to a loss of consistency in poly(A)-enriched RNA sequencing.

BACKGROUND: Both RNA-Seq and sample freeze-thaw are ubiquitous. However, knowledge about the impact of freeze-thaw on downstream analyses is limited. The lack of common quality metrics that are sufficiently sensitive to freeze-thaw and RNA degradation, e.g. the RNA Integrity Score, makes such assessments challenging. RESULTS: Here we quantify the impact of repeated freeze-thaw cycles on the reliability of RNA-Seq by examining poly(A)-enriched and ribosomal RNA depleted RNA-seq from frozen leukocytes drawn from a toddler Autism cohort. To do so, we estimate the relative noise, or percentage of random counts, separating technical replicates. Using this approach we measured noise associated with RIN and freeze-thaw cycles. As expected, RIN does not fully capture sample degradation due to freeze-thaw. We further examined differential expression results and found that three freeze-thaws should extinguish the differential expression reproducibility of similar experiments. Freeze-thaw also resulted in a 3' shift in the read coverage distribution along the gene body of poly(A)-enriched samples compared to ribosomal RNA depleted samples, suggesting that library preparation may exacerbate freeze-thaw-induced sample degradation. CONCLUSION: The use of poly(A)-enrichment for RNA sequencing is pervasive in library preparation of frozen tissue, and thus, it is important during experimental design and data analysis to consider the impact of repeated freeze-thaw cycles on reproducibility.

Get SET Early to Identify and Treatment Refer Autism Spectrum Disorder at 1 Year and Discover Factors That Influence Early Diagnosis.

OBJECTIVES: To examine the impact of a new approach, Get SET Early, on the rates of early autism spectrum disorder (ASD) detection and factors that influence the screen-evaluate-treat chain. STUDY DESIGN: After attending Get SET Early training, 203 pediatricians administered 57 603 total screens using the Communication and Symbolic Behavior Scales Infant-Toddler Checklist at 12-, 18-, and 24-month well-baby examinations, and parents designated presence or absence of concern. For screen-positive toddlers, pediatricians specified if the child was being referred for evaluation, and if not, why not. RESULTS: Collapsed across ages, toddlers were evaluated and referred for treatment at a median age of 19 months, and those screened at 12 months (59.4% of sample) by 15 months. Pediatricians referred one-third of screen-positive toddlers for evaluation, citing lack of confidence in the accuracy of screen-positive results as the primary reason for nonreferral. If a parent expressed concerns, referral probability doubled, and the rate of an ASD diagnosis increased by 37%. Of 897 toddlers evaluated, almost one-half were diagnosed as ASD, translating into an ASD prevalence of 1%. CONCLUSIONS: The Get SET Early model was effective at detecting ASD and initiating very early treatment. Results also underscored the need for change in early identification approaches to formally operationalize and incorporate pediatrician judgment and level of parent concern into the process.

The ASD Living Biology: from cell proliferation to clinical phenotype.

Autism spectrum disorder (ASD) has captured the attention of scientists, clinicians and the lay public because of its uncertain origins and striking and unexplained clinical heterogeneity. Here we review genetic, genomic, cellular, postmortem, animal model, and cell model evidence that shows ASD begins in the womb. This evidence leads to a new theory that ASD is a multistage, progressive disorder of brain development, spanning nearly all of prenatal life. ASD can begin as early as the 1st and 2nd trimester with disruption of cell proliferation and differentiation. It continues with disruption of neural migration, laminar disorganization, altered neuron maturation and neurite outgrowth, disruption of synaptogenesis and reduced neural network functioning. Among the most commonly reported high-confidence ASD (hcASD) genes, 94% express during prenatal life and affect these fetal processes in neocortex, amygdala, hippocampus, striatum and cerebellum. A majority of hcASD genes are pleiotropic, and affect proliferation/differentiation and/or synapse development. Proliferation and subsequent fetal stages can also be disrupted by maternal immune activation in the 1st trimester. Commonly implicated pathways, PI3K/AKT and RAS/ERK, are also pleiotropic and affect multiple fetal processes from proliferation through synapse and neural functional development. In different ASD individuals, variation in how and when these pleiotropic pathways are dysregulated, will lead to different, even opposing effects, producing prenatal as well as later neural and clinical heterogeneity. Thus, the pathogenesis of ASD is not set at one point in time and does not reside in one process, but rather is a cascade of prenatal pathogenic processes in the vast majority of ASD toddlers. Despite this new knowledge and theory that ASD biology begins in the womb, current research methods have not provided individualized information: What are the fetal processes and early-age molecular and cellular differences that underlie ASD in each individual child? Without such individualized knowledge, rapid advances in biological-based diagnostic, prognostic, and precision medicine treatments cannot occur. Missing, therefore, is what we call ASD Living Biology. This is a conceptual and paradigm shift towards a focus on the abnormal prenatal processes underlying ASD within each living individual. The concept emphasizes the specific need for foundational knowledge of a living child's development from abnormal prenatal beginnings to early clinical stages. The ASD Living Biology paradigm seeks this knowledge by linking genetic and in vitro prenatal molecular, cellular and neural measurements with in vivo post-natal molecular, neural and clinical presentation and progression in each ASD child. We review the first such study, which confirms the multistage fetal nature of ASD and provides the first in vitro fetal-stage explanation for in vivo early brain overgrowth. Within-child ASD Living Biology is a novel research concept we coin here that advocates the integration of in vitro prenatal and in vivo early post-natal information to generate individualized and group-level explanations, clinically useful prognoses, and precision medicine approaches that are truly beneficial for the individual infant and toddler with ASD.

circTAIL-seq, a targeted method for deep analysis of RNA 3' tails, reveals transcript-specific differences by multiple metrics.

Post-transcriptionally added RNA 3' nucleotide extensions, or tails, impose numerous regulatory effects on RNAs, including effects on RNA turnover and translation. However, efficient methods for in-depth tail profiling of a transcript of interest are still lacking, hindering available knowledge particularly of tail populations that are highly heterogeneous. Here, we developed a targeted approach, termed circTAIL-seq, to quantify both major and subtle differences of heterogeneous tail populations. As proof-of-principle, we show that circTAIL-seq quantifies the differences in tail qualities between two selected Trypanosoma brucei mitochondrial transcripts. The results demonstrate the power of the developed method in identification, discrimination, and quantification of different tail states that the population of one transcript can possess. We further show that circTAIL-seq can detect the tail characteristics for variants of transcripts that are not easily detectable by conventional approaches, such as degradation intermediates. Our findings are not only well supported by previous knowledge, but they also expand this knowledge and provide experimental evidence for previous hypotheses. In the future, this approach can be used to determine changes in tail qualities in response to environmental or internal stimuli, or upon silencing of genes of interest in mRNA-processing pathways. In summary, circTAIL-seq is an effective tool for comparing nonencoded RNA tails, especially when the tails are extremely variable or transcript of interest is low abundance.

Early Alzheimer's disease pathology in human cortex is associated with a transient phase of distinct cell states.

Cellular perturbations underlying Alzheimer's disease are primarily studied in human postmortem samples and model organisms. Here we generated a single-nucleus atlas from a rare cohort of cortical biopsies from living individuals with varying degrees of Alzheimer's disease pathology. We next performed a systematic cross-disease and cross-species integrative analysis to identify a set of cell states that are specific to early AD pathology. These changes-which we refer to as the Early Cortical Amyloid Response-were prominent in neurons, wherein we identified a transient state of hyperactivity preceding loss of excitatory neurons, which correlated with the selective loss of layer 1 inhibitory neurons. Microglia overexpressing neuroinflammatory-related processes also expanded as AD pathological burden increased. Lastly, both oligodendrocytes and pyramidal neurons upregulated genes associated with amyloid beta production and processing during this early hyperactive phase. Our integrative analysis provides an organizing framework for targeting circuit dysfunction, neuroinflammation, and amyloid production early in AD pathogenesis.

The cell type composition of the adult mouse brain revealed by single cell and spatial genomics.

The function of the mammalian brain relies upon the specification and spatial positioning of diversely specialized cell types. Yet, the molecular identities of the cell types, and their positions within individual anatomical structures, remain incompletely known. To construct a comprehensive atlas of cell types in each brain structure, we paired high-throughput single-nucleus RNA-seq with Slide-seq-a recently developed spatial transcriptomics method with near-cellular resolution-across the entire mouse brain. Integration of these datasets revealed the cell type composition of each neuroanatomical structure. Cell type diversity was found to be remarkably high in the midbrain, hindbrain, and hypothalamus, with most clusters requiring a combination of at least three discrete gene expression markers to uniquely define them. Using these data, we developed a framework for genetically accessing each cell type, comprehensively characterized neuropeptide and neurotransmitter signaling, elucidated region-specific specializations in activity-regulated gene expression, and ascertained the heritability enrichment of neurological and psychiatric phenotypes. These data, available as an online resource (BrainCellData.org) should find diverse applications across neuroscience, including the construction of new genetic tools, and the prioritization of specific cell types and circuits in the study of brain diseases.

Brain-region-specific changes in neurons and glia and dysregulation of dopamine signaling in Grin2a mutant mice.

A genetically valid animal model could transform our understanding of schizophrenia (SCZ) disease mechanisms. Rare heterozygous loss-of-function (LoF) mutations in GRIN2A, encoding a subunit of the NMDA receptor, greatly increase the risk of SCZ. By transcriptomic, proteomic, and behavioral analyses, we report that heterozygous Grin2a mutant mice show (1) large-scale gene expression changes across multiple brain regions and in neuronal (excitatory and inhibitory) and non-neuronal cells (astrocytes and oligodendrocytes), (2) evidence of hypoactivity in the prefrontal cortex (PFC) and hyperactivity in the hippocampus and striatum, (3) an elevated dopamine signaling in the striatum and hypersensitivity to amphetamine-induced hyperlocomotion (AIH), (4) altered cholesterol biosynthesis in astrocytes, (5) a reduction in glutamatergic receptor signaling proteins in the synapse, and (6) an aberrant locomotor pattern opposite of that induced by antipsychotic drugs. These findings reveal potential pathophysiologic mechanisms, provide support for both the "hypo-glutamate" and "hyper-dopamine" hypotheses of SCZ, and underscore the utility of Grin2a-deficient mice as a genetic model of SCZ.

Comparative proteomics analysis of Arabidopsis thaliana response to light-emitting diode of narrow wavelength 450 nm, 595 nm, and 650 nm.

Incident light is a central modulator of plant growth and development. However, there are still open questions surrounding wavelength-specific plant proteomic responses. Here we applied tandem mass tag based quantitative proteomics technology to acquire an in-depth view of proteome changes in Arabidopsis thaliana response to narrow wavelength blue (B; 450 nm), amber (A; 595 nm), and red (R; 650 nm) light treatments. A total of 16,707 proteins were identified with 9120 proteins quantified across all three light treatments in three biological replicates. This enabled examination of changes in the abundance for proteins with low abundance and important regulatory roles including transcription factors and hormone signaling. Importantly, 18% (1631 proteins) of the A. thaliana proteome is differentially abundant in response to narrow wavelength lights, and changes in proteome correlate well with different morphologies exhibited by plants. To showcase the usefulness of this resource, data were placed in the context of more than thirty published datasets, providing orthogonal validation and further insights into light-specific biological pathways, including Systemic Acquired Resistance and Shade Avoidance Syndrome. This high-resolution resource for A. thaliana provides baseline data and a tool for defining molecular mechanisms that control fundamental aspects of plant response to changing light conditions, with implications in plant development and adaptation. SIGNIFICANCE: Understanding of molecular mechanisms involved in wavelength-specific response of plant is question of widespread interest both to basic researchers and to those interested in applying such knowledge to the engineering of novel proteins, as well as targeted lighting systems. Here we sought to generate a high-resolution proteomic profile of plant leaves, based on exposure to specific narrow-wavelength lights. Although changes in plant physiology in response to light spectral composition is well documented, there is limited knowledge on the roles of specific light wavelengths and their impact. Most previous studies have utilized relatively broad wavebands in their experiments. Such multi-wavelengths lights trigger diverse and complex signaling networks that pose major challenges in inference of wavelength-specific molecular processes that underly the plant response. Moreover, most studies have compared the effect of blue and red wavelengths comparing with FL, as control. As FL light consists the mixed spectra composition of both red and blue as well as numerous other wavelengths, comparing undeniably results in inconsistent and overlapping responses that will hamper effects to elucidate the plant response to specific wavelengths [1, 2]. Monitoring plant proteome response to specific wavelengths and further contrasting the changes with one another, rather than comparing plants proteome to FL, is thus necessary to gain detailed insights on underlying biological pathways and their consequences in plant physiology. Here, we employed narrow wavelength LED lights in our design to eliminate a potential overlap in molecular responses by ensuring non-overlapping wavelengths in the light treatments. We further applied TMT-labeling technology to gain a high-resolution view on the proteome changes. Our proteomics data provides an in-depth coverage suitable for system-wide analyses, providing deep insights on plant molecular response particularly because of the tremendous increase in the coverage of identified proteins which outreach the other biological data.

Single-cell genomic profiling of human dopamine neurons identifies a population that selectively degenerates in Parkinson's disease.

The loss of dopamine (DA) neurons within the substantia nigra pars compacta (SNpc) is a defining pathological hallmark of Parkinson's disease (PD). Nevertheless, the molecular features associated with DA neuron vulnerability have not yet been fully identified. Here, we developed a protocol to enrich and transcriptionally profile DA neurons from patients with PD and matched controls, sampling a total of 387,483 nuclei, including 22,048 DA neuron profiles. We identified ten populations and spatially localized each within the SNpc using Slide-seq. A single subtype, marked by the expression of the gene AGTR1 and spatially confined to the ventral tier of SNpc, was highly susceptible to loss in PD and showed the strongest upregulation of targets of TP53 and NR2F2, nominating molecular processes associated with degeneration. This same vulnerable population was specifically enriched for the heritable risk associated with PD, highlighting the importance of cell-intrinsic processes in determining the differential vulnerability of DA neurons to PD-associated degeneration.

Atypical genomic cortical patterning in autism with poor early language outcome.

Cortical regionalization develops via genomic patterning along anterior-posterior (A-P) and dorsal-ventral (D-V) gradients. Here, we find that normative A-P and D-V genomic patterning of cortical surface area (SA) and thickness (CT), present in typically developing and autistic toddlers with good early language outcome, is absent in autistic toddlers with poor early language outcome. Autistic toddlers with poor early language outcome are instead specifically characterized by a secondary and independent genomic patterning effect on CT. Genes involved in these effects can be traced back to midgestational A-P and D-V gene expression gradients and different prenatal cell types (e.g., progenitor cells and excitatory neurons), are functionally important for vocal learning and human-specific evolution, and are prominent in prenatal coexpression networks enriched for high-penetrance autism risk genes. Autism with poor early language outcome may be explained by atypical genomic cortical patterning starting in prenatal development, which may detrimentally affect later regional functional specialization and circuit formation.

Prenatal Origins of ASD: The When, What, and How of ASD Development.

Autism spectrum disorder (ASD) is a largely heritable, multistage prenatal disorder that impacts a child's ability to perceive and react to social information. Most ASD risk genes are expressed prenatally in many ASD-relevant brain regions and fall into two categories: broadly expressed regulatory genes that are expressed in the brain and other organs, and brain-specific genes. In trimesters one to three (Epoch-1), one set of broadly expressed (the majority) and brain-specific risk genes disrupts cell proliferation, neurogenesis, migration, and cell fate, while in trimester three and early postnatally (Epoch-2) another set (the majority being brain specific) disrupts neurite outgrowth, synaptogenesis, and the 'wiring' of the cortex. A proposed model is that upstream, highly interconnected regulatory ASD gene mutations disrupt transcriptional programs or signaling pathways resulting in dysregulation of downstream processes such as proliferation, neurogenesis, synaptogenesis, and neural activity. Dysregulation of signaling pathways is correlated with ASD social symptom severity. Since the majority of ASD risk genes are broadly expressed, many ASD individuals may benefit by being treated as having a broader medical disorder. An important future direction is the noninvasive study of ASD cell biology.

Gene Function Discovery for Kinetoplastid Pathogens.

We propose to integrate the existing and new experimental data with computational tools to model interaction networks for the most prominent kinetoplastid pathogens. These interaction networks will vastly expand the functional annotation of the kinetoplastid genomes, which in turn are critical for identifying new routes of disease intervention.

From Genotype to Phenotype: Augmenting Deep Learning with Networks and Systems Biology.

Cells, as complex systems, consist of diverse interacting biomolecules arranged in dynamic hierarchical modules. Recent advances in deep learning methods now allow one to encode this rich existing knowledge in the architecture of the learning procedure, thus providing the models with the knowledge that is absent in the training data. By encoding biological networks in the architecture, one can develop flexible deep models that propagate information through the molecular networks to successfully classify cell states. Moreover, this flexibility in the architecture can be harnessed to model the hierarchical structure of real biological systems, efficiently converting gene-level data to pathway-level information with an ultimate impact on cell phenotype. Furthermore, such models could require fewer training samples, are more generalizable across diverse biological contexts, and can make predictions that are more consistent with the current understanding on the inner-working of biological systems.

A perturbed gene network containing PI3K-AKT, RAS-ERK and WNT-ß-catenin pathways in leukocytes is linked to ASD genetics and symptom severity.

Hundreds of genes are implicated in autism spectrum disorder (ASD), but the mechanisms through which they contribute to ASD pathophysiology remain elusive. Here we analyzed leukocyte transcriptomics from 1- to 4-year-old male toddlers with ASD or typical development from the general population. We discovered a perturbed gene network that includes highly expressed genes during fetal brain development. This network is dysregulated in human induced pluripotent stem cell-derived neuron models of ASD. High-confidence ASD risk genes emerge as upstream regulators of the network, and many risk genes may impact the network by modulating RAS-ERK, PI3K-AKT and WNT-ß-catenin signaling pathways. We found that the degree of dysregulation in this network correlated with the severity of ASD symptoms in the toddlers. These results demonstrate how the heterogeneous genetics of ASD may dysregulate a core network to influence brain development at prenatal and very early postnatal ages and, thereby, the severity of later ASD symptoms.

Evaluation of the Diagnostic Stability of the Early Autism Spectrum Disorder Phenotype in the General Population Starting at 12 Months.

Importance: Universal early screening for autism spectrum disorder (ASD) in primary care is becoming increasingly common and is believed to be a pivotal step toward early treatment. However, the diagnostic stability of ASD in large cohorts from the general population, particularly in those younger than 18 months, is unknown. Changes in the phenotypic expression of ASD across early development compared with toddlers with other delays are also unknown. Objectives: To examine the diagnostic stability of ASD in a large cohort of toddlers starting at 12 months of age and to compare this stability with that of toddlers with other disorders, such as developmental delay. Design, Setting, and Participants: In this prospective cohort study performed from January 1, 2006, to December 31, 2018, a total of 2241 toddlers were referred from the general population through a universal screening program in primary care or community referral. Eligible toddlers received their first diagnostic evaluation between 12 and 36 months of age and had at least 1 subsequent evaluation. Exposures: Diagnosis was denoted after each evaluation visit as ASD, ASD features, language delay, developmental delay, other developmental issue, typical sibling of an ASD proband, or typical development. Main Outcomes and Measures: Diagnostic stability coefficients were calculated within 2-month age bands, and logistic regression models were used to explore the associations of sex, age, diagnosis at first visit, and interval between first and last diagnosis with stability. Toddlers with a non-ASD diagnosis at their first visit diagnosed with ASD at their last were designated as having late-identified ASD. Results: Among the 1269 toddlers included in the study (918 [72.3%] male; median age at first evaluation, 17.6 months [interquartile range, 14.0-24.4 months]; median age at final evaluation, 36.2 months [interquartile range, 33.4-40.9 months]), the overall diagnostic stability for ASD was 0.84 (95% CI, 0.80-0.87), which was higher than any other diagnostic group. Only 7 toddlers (1.8%) initially considered to have ASD transitioned into a final diagnosis of typical development. Diagnostic stability of ASD within the youngest age band (12-13 months) was lowest at 0.50 (95% CI, 0.32-0.69) but increased to 0.79 by 14 months and 0.83 by 16 months (age bands of 12 vs 14 and 16 months; odds ratio, 4.25; 95% CI, 1.59-11.74). A total of 105 toddlers (23.8%) were not designated as having ASD at their first visit but were identified at a later visit. Conclusions and Relevance: The findings suggest that an ASD diagnosis becomes stable starting at 14 months of age and overall is more stable than other diagnostic categories, including language or developmental delay. After a toddler is identified as having ASD, there may be a low chance that he or she will test within typical levels at 3 years of age. This finding opens the opportunity to test the impact of very early-age treatment of ASD.