• Contact Us

Columbia University
Irving Medical Center
Neurological Institute

710 West 168th Street, 3rd floor
(212) 305-1818

TaubCONNECT Research Perspective:
April 2025

Modulation of CREB3L2-ATF4 Heterodimerization via Proteasome Inhibition and HRI Activation in Alzheimer's Disease Pathology

Krystal Herline-Killian, PhD		Ulrich Hengst, PhD

Widespread gene expression changes are a feature of many neurodegenerative disorders, including Alzheimer’s disease (AD). Identifying the mechanisms driving these changes could lead to the development of novel therapeutic targets to restore normal gene expression in AD. In our group, we have long been interested in the molecular pathways connecting exposure of neurons to oligomeric Aβ42 to the induction of AD-typical transcriptional changes. In a series of previous publications, we reported that a unique heterodimer of two bZIP transcription factors, CREB3L2 and ATF4, is present in the brains of individuals with AD. This heterodimer is associated with up to 50% of the gene expression changes observed in AD brains. Interestingly, its induction in neurons is sufficient to elicit AD-typical neuronal changes, such as the hyperphosphorylation and secretion of tau, and increased Aβ42 production over Aβ40. More importantly, we found that preventing the formation of this dimer rescues neurons from Aβ42-induced cell death. However, it remained puzzling why the CREB3L2-ATF4 heterodimer is formed specifically in the context of AD in response to oligomeric Aβ42, while most other cellular stressors seemingly do not trigger this signaling pathway.

Figure 5B. HRI activation interferes with CREB3L2-ATF4 signaling. Proximity labeling assay (PLA; red) for CREB3L2-ATF4 in cortical neurons treated with DMSO or the HRI activator BtdCPU and Aβ42 or vehicle for 8 h. Nuclei are in blue, microtubules in green. Scale bar, 5 µm.

As recently reported in Cell Death & Disease, we found that the formation of the CREB3L2-ATF4 requires the activation of the integrated stress response with simultaneous inhibition of the proteasome. The levels of stress response transcription factors, such as ATF4 and CREB3L2, are tightly controlled by translational mechanisms, especially by the integrated stress response (ISR). The ISR is a network of cellular signaling pathways activated by various stressors to restore cellular homeostasis. These pathways converge in the phosphorylation of eIF2ɑ, leading to the synthesis of stress response transcription factors. It is unclear which eIF2ɑ kinases (PERK, PKR, GCN2, HRI) are activated in response to Aβ42 in neurons and required for ATF4 and CREB3L2 synthesis and dimerization. Post-translational mechanisms, including ubiquitination and proteasomal degradation, also control ATF4 and CREB3L2 abundance. For example, CREB3L2 is rapidly ubiquitinated and degraded by the proteasome under non-stress conditions. The impact of translational activation downstream of eIF2ɑ and ubiquitin-dependent proteasomal degradation on the formation and abundance of the CREB3L2-ATF4 heterodimer in Aβ42-exposed neurons remained unexplored.

In this study, we explored the role of proteasome inhibition and the eIF2ɑ-kinase HRI in the formation of CREB3L2-ATF4 in neurons exposed to soluble oligomeric Aβ42. While HRI activation increased ATF4 expression, it decreased CREB3L2 and CREB3L2-ATF4 levels. Interestingly, proteasome inhibition, induced by Aβ42, led to an increase in both transcription factors in the nucleus. These findings suggest that CREB3L2 levels are typically maintained low due to rapid degradation, but proteasome inhibition in response to Aβ42 disrupts this balance, resulting in an increase in CREB3L2 and heterodimer levels. Moreover, HRI activation not only reduced CREB3L2 and heterodimer levels but also prevented the transcriptional dysregulation of a CREB3L2-ATF4 target, SNX3. Our results indicate that manipulating the HRI pathway during proteasome inhibition could potentially restore normal gene expression in the context of AD-related protein accumulation.

Ulrich Hengst, PhD
Professor of Pathology & Cell Biology (in the Taub Institute)
uh2112@cumc.columbia.edu

APOE and Alzheimer's Disease and Related Dementias Risk Among 12,221 Hispanics/Latinos

Sandra Barral, PhD		Giuseppe Tosto, MD, PhD

Despite the identification of over eighty genetic loci associated with Alzheimer’s disease and related dementias (ADRD), the Apolipoprotein E (APOE) ε4 allele remains the strongest and most consistently replicated genetic risk factor in the non-Hispanic White population. However, in populations with admixed genetic backgrounds, the risk conferred by the different APOE alleles is heterogeneous. These large differences in APOE-ε4 effects are likely to depend on both global genetic ancestry and local genetic ancestry, as well as gene–environment interactions. The strongest association between ADRD and APOE-ε4 has been observed in East Asians followed by Non-Hispanic Whites. Genetic studies examining the association between ADRD risk and APOE in African Americans and Caribbean-Hispanics showed that the risk is attenuated, especially when the ancestral background around the APOE locus is from the African haplotypes.

Figure. Meta-analysis and forest plot for APOE-ε4 (upper panel) and APOE-ε2 (lower panel) alleles. The plot details the beta coefficients (BETAs), standard errors (SEs), odd ratios (ORs), 95% confidence intervals (CIs), and the individual study weights according to their contributions to the pooled estimates. The horizontal lines represent the study’s 95% CIs, with each end of the line representing the boundaries of the CI. A black dot represents the point estimate of the study, and it also provides a visual representation of the size of the study (the largest dot corresponds to a larger sample size). The dotted vertical lines are drawn at the value of the overall common effect. The diamond below the studies represents the overall pooled effect.

In the current study, we investigated the association between APOE genotype and ADRD in 12,221 Hispanics/Latinos from four different populations: Caribbean-Hispanics, Mexican Americans, Mexicans, and Peruvians/Bolivians. As recently reported in Alzheimer’s & Dementia, our results showed a significant association between APOE-ε4 allele dosage and ADRD across the cohorts. The genetic effect of at least one copy of the ε4 allele on disease risk was higher in Peruvians/Bolivians (OR=6.13, 95%CI= 2.71–13.83) compared to Mexicans (OR=4.31, 95%CI=1.58-11.74), Mexican-Americans (OR=3.06, 95%CI=2.04-4.59) or Caribbean Hispanics (OR=2.22, 95%CI=1.99-2.48). Although not statistically significant, we also observed a protective effect for the APOE-ε2 allele on ADRD risk across cohorts.

When stratified by global ancestry, samples with predominantly Native-American ancestry and APOE-ε4 heterozygous carriers showed a significant association between APOE-ε4 allele and ADRD. Results restricted to predominant European ancestry also found a significantly higher risk among APOE-ε4 carriers. In sex stratified analysis, the association between the ε4 allele and ADRD in the men strata (OR=2.20, 95%CI=1.85-2.62) was found significant yet weaker than in women (OR=2.84, 95%CI=2.51-3.22).

Survival analysis in Caribbean-Hispanics cohorts showed statistically significant differences in the number of conversions between ε4 allele carriers vs. non-carriers (HR=3.65, 95%CI=1.62-8.23 for homozygous ε4-carriers). The effect of the ε4 allele was further corroborated by plasma ADRD biomarkers analyses. Despite the heterogeneity of the platforms employed, all cohorts showed a significant increase in plasma levels of p-tau181 or p-tau217 among APOE-ε4 carriers. Our results showed that APOE-ε4 allele confers a heterogeneous risk for ADRD across Hispanic/Latino sample, the largest to date. None of the analysis cohorts replicated the effect of local ancestral background around the APOE region in determining ADRD risk, as previously reported by less-powered studies.

Different genetic architectures among ethnic groups may influence how genetic factors contribute to ADRD risk. To gain further insight into ADRD pathogenesis, future work should aim to reproduce our findings with larger, and more diverse admixed populations, more specifically those with Native American background.

Sandra Barral, PhD
Associate Professor of Neurogenetics (in Neurology, the Taub Institute, and the Gertrude H. Sergievsky Center) at CUIMC
smb2174@cumc.columbia.edu

Giuseppe Tosto, MD, PhD
Assistant Professor of Neurological Sciences (in Neurology, the Taub Institute, and the Gertrude H. Sergievsky Center)
smb2174@cumc.columbia.edu

A Human Brain Map of Mitochondrial Respiratory Capacity and Diversity

Eugene Mosharov, PhD		Martin Picard, PhD

The brain is the most energy-hungry organ in the body, consuming 20-25% of the body’s energy budget. The brain’s high burn rate powers synapses and makes new proteins required for its complex functioning—from encoding our memories to sustaining the oscillations required for consciousness. To power its activities, the brain cell contains thousands of mitochondria which populate the cell bodies, the axons and dendrites of neurons, as well as the cytoplasm of the dozens of glial cell types that keep the brain alive. Mitochondrial oxidative phosphorylation (OxPhos) powers brain activity while mitochondrial defects are linked to neurodegenerative and neuropsychiatric disorders, underscoring the need to define the brain’s molecular energetic landscape.

As recently reported in Nature, we’ve made the first attempt to bridge a scale gap between the subcellular mitochondrial biology and the macro-level anatomical structures and neuroimaging models that are the focus of cognitive neuroscience. This project and the resulting MitoBrainMap v1.0 are exceptional for three main reasons.

Figure. Creating MitoBrainMap-1. (A) An image of the top surface of a human brain slab after a square 3x3 mm grid was milled to the depth of 3 mm. Lower image shows the voxels collected for biochemical assays. (B) The respiratory capacity of mitochondria in each voxel was assessed using biochemical techniques, and the values are represented by color coding. Voxels with brighter colors contain mitochondria that are more highly specialized for energy transformation than mitochondria in voxels with darker colors. (C) A map of mitochondrial respiratory capacity across the brain, predicted from brain-imaging data of various modalities, including functional data (reflecting activity associated with changes in blood flow) and various types of structural data.

The first advance was a new method to physically “voxelize” the frozen tissue. Using a healthy donor brain provided by the Quantitative Brain Biology (Brain QUANT, Psychiatry), directed by Maura Boldrini and maintained by Gorazd Rosoklija and Andrew Dwork, we systematically partition a single frozen section into 703 small voxels, comparable in size (3x3x3mm) to neuroimaging resolution.

The second advance was to molecularly define the bioenergetic landscape of mitochondrial phenotypes and diversity across brain regions, including white matter and different portions of the cortical and sub-cortical gray matter regions. This was performed in collaboration with two other teams. Philip L. De Jager, Vilas Menon and Ya Zhang from the Taub Institute and the Center for Translational and Computational Neuroimmunology, who performed single-nucleus RNA sequencing on four brain samples with Anna Monzel in Picard’s laboratory. And Orian Shirihai’s team including Linsey Stiles and Corey Osto at the UCLA Metabolism Theme, who deployed a frozen tissue respirometry technique they pioneered a few years back. It was the first time these assays were performed on hundreds of samples from a human brain. Such tour de force approach revealed qualitative differences in mitochondria density and OxPhos capacity between both regions and cell types in the brain. Quite surprisingly, we also found that higher OxPhos capacity is found in brain areas that appeared later in evolution, such as the cortex.

The third advance came from a computational model, extending the data from a single brain slice to the whole brain. Michel Thiebaut de Shotten, neuroanatomist and Research Director at CNRS in Bordeaux, transposed the mitochondrial data from this brain section to the “standard” brain in neuroimaging research. This allowed us to visualize and interpret the biochemical data in relation to widely available neuroimaging modalities and to build a computational model to predict molecular markers of mitochondrial density and OxPhos capacity from standard structural and functional MRI neuroimaging parameters. The final step consisted in expanding the prediction model to the scale of the whole brain. This led to the first probabilistic map of mitochondrial biology, meaning that even though we never measured mitochondrial features in some deep part of the temporal lobe, our model can predict the amount and types of mitochondria found there based on MRI data.

The brain voxelization method and the brain mitochondrial findings reveal the landscape of human brain mitochondrial energetics and provide a new tool to understand how brain energy relates to anatomy, development, behavior, neurodegeneration, well-being, and perhaps even biggest questions around the mind and consciousness. Further research is now needed to test, validate and apply this neuroimaging-based mitochondrial profiling approach in various biological and clinical contexts.

Eugene Mosharov, PhD
Research Scientist in the Department of Psychiatry
em706@cumc.columbia.edu

Martin Picard, PhD
Science of Health Associate Professor of Behavioral Medicine (in Psychiatry, Neurology and the Robert N. Butler Columbia Aging Center)
mp3484@cumc.columbia.edu

Preclinical Alzheimer's Disease Shows Alterations in Circulating Neuronal-Derived Extracellular Vesicle MicroRNAs in a Multiethnic Cohort

Paolo Reho, PhD		Badri N. Vardarajan, PhD, MS

In a recent study, conducted in collaboration with colleagues at the Mailman School of Public Health and Columbia University Irving Medical Center, we explored the potential of blood-based biomarkers—specifically microRNAs (miRNAs) found in neuronal-derived extracellular vesicles (NDEVs)—to signal early changes in the brain. These vesicles cross the blood–brain barrier and carry neuron-specific cargo, making them a compelling target for non-invasive diagnostics. We analyzed blood samples from individuals with clinical and preclinical AD, as well as healthy controls, to determine whether changes in NDEV miRNAs could reveal early-stage disease biology.

Figure 1. Study workflow. Schematic representation of the study workflow. Abbreviations: AD, Alzheimer's disease; EVs, extracellular vesicles; miRNA, microRNA. Figure created with BioRender.com.

As recently reported in Alzheimer’s & Dementia, we identified fourteen NDEV miRNAs associated with AD risk, with more pronounced transcriptional alterations observed in individuals with preclinical AD. Functional analysis revealed that these miRNAs target genes involved in key neurodegenerative pathways, including those related to tau (MAPT), alpha-synuclein (SNCA), and cytochrome c (CYCS). These findings support the potential of NDEV miRNAs as early biomarkers for AD and provide insight into the molecular processes that may precede the onset of clinical symptoms.

Badri N. Vardarajan, PhD, MS
Associate Professor of Neurological Science (in Neurology and the Gertrude H. Sergievsky Center and the Taub Institute) at the CUIMC
bnv2103@cumc.columbia.edu