Citations

Abstract Benign breast disease (BBD) confers elevated risk for breast cancer, yet current tools lack precision in identifying lesions likely to progress. Immune mechanisms may influence progression, but the spatial and molecular features of immune responses in BBD versus invasive cancer remain poorly characterized. We applied Digital Spatial Proteomics (DSP) to evaluate immune-related protein expression across 1,540 regions of interest (ROIs) from breast tumors (n=58, Carolina Breast Cancer Study [CBCS]), a UNC BBD cohort (n=20), and a Mayo BBD cohort (n=177). Using Consensus Clustering in CBCS and UNC BBD, we identified two stable immune phenotypes: “High Immune,” enriched in tumors and high-risk benign lesions, and “Quiet Immune,” predominant in low-risk BBD. ROIs were classified by immune marker expression, and patients were labeled as “High” if ≥50% of their ROIs were High Immune. In CBCS, High Immune profiles were associated with ER-negative status (OR=2.92 [0.99-8.65] participant-level; OR=3.28 [1.34-8.04] ROI-level). In BBD, High Immune profiles trended toward higher odds of high-risk lesions (OR=1.00 [0.15-6.70] participant-level; OR=1.68 [0.57-4.97] ROI-level), though this pattern was not replicated in the Mayo cohort. BBD samples exhibited substantial spatial heterogeneity in immune expression (40% within-patient ROI agreement vs. 74% in tumors), potentially limiting biomarker reliability. Nonetheless, four immune proteins (CD44, CD80, CD14, CD66b) were significantly associated with future cancer development in the Mayo cohort, though these findings require validation due to within-person variability. Our findings highlight the challenges posed by spatial heterogeneity in assessing immune biomarkers in BBD and suggest that immune activation may emerge later in disease progression. These results raise important questions about the timing and consistency of immune responses in the natural history of breast cancer. Citation Format: A. Zhang, A. Cozzo, L. Olsson, E. Kirk, X. Gao, S. Nyante, B. Calhoun, R. Vierkant, M. Sherman, M. Troester. Immune Heterogeneity in Breast Cancer Natural History: Benign Breast Disease to Breast Cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS2-05-23.

Abstract Histones variant composition and post-translational modifications (PTMs) of core and linker histones jointly orchestrate chromatin architecture and tissue-specific gene regulation. However, capturing the full complexity of the “histone code” in native human tissues remains challenging. Here we present a human cardiac histone proteoform atlas—comprising intact histone variants with combinatorial PTMs—enabled by a streamlined top‑down proteomics workflow. In a single one‑dimensional reversed‑phase liquid chromatography (LC)-mass spectrometry (MS) run, we achieve baseline separation of all four core histone families (H2A, H2B, H3, and H4) together with the linker histone H1 directly from human myocardium. Notably, this intact-protein analysis preserves the connectivity of co-occurring PTMs to individual histone molecules to reveal the combinatorial histone code, enabling proteoform‑level quantification of histone‑variant composition and PTM stoichiometry at the chromatographic scale. Targeted MS/MS directly resolves co-occurring PTM combinations and distinguishes highly homologous histone isoforms. Using this approach, we assemble the first comprehensive cardiac histone proteoform atlas and uncover previously unobserved histone proteoforms. More broadly, this top-down proteomics workflow provides a robust framework for proteoform-resolved analysis of histone variants and PTMs in complex biological systems.

GGGGCC (G 4 C 2 ) repeat expansion in C9ORF72 is the most common genetic cause of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). Toxicity is thought to result from the accumulation of either repeat RNAs and/or dipeptide repeat proteins (DPRs) translated from repeat-containing transcripts through repeat-associated non-AUG (RAN) translation. To disentangle RNA from DPR toxicity, we mutated a CUG codon predominantly used to initiate DPR translation from all three reading frames. This mutation disrupted DPR synthesis while preserving the expression of repeat-containing RNAs. Despite the accumulation of RNA foci, behavioral deficits and pathological abnormalities, including p-TDP-43 inclusions, STING activation, motor neuron loss, neuroinflammation, and increased plasma neurofilament concentration, were alleviated in C9ORF72 mice. Base editing of the CUG codon also improved molecular phenotypes and survival in patient induced pluripotent stem cell–derived neurons, which highlights the potential of therapeutically targeting DPR production rather than repeat RNAs.

INTRODUCTION Recent work has identified unique cognitive profiles for early‐onset Alzheimer's disease (EOAD) relative to late‐onset Alzheimer's disease (LOAD), however, examination has been limited in determining whether the association between age and cognitive severity at presentation also differs across conditions. METHODS A series of linear spline regression models was conducted across baseline cognitive data from 325 EOAD and 314 LOAD participants, after accounting for education, sex, and apolipoprotein ε4 status. RESULTS Significant differences existed in the relationship between baseline age and cognitive performance between EOAD and LOAD samples for Processing Speed/Attention, Executive Functioning, and Episodic Immediate Memory. Younger participants from both EOAD and LOAD groups performed disproportionately worse on non‐amnestic cognitive domains, with this occurring to a greater extent in EOAD than LOAD. DISCUSSION In the age of disease‐modifying treatments, results highlight the importance of assessing for cognitive declines in individuals starting much earlier than age 65. Highlights Early‐onset Alzheimer's disease (EOAD) and late‐onset Alzheimer's disease (LOAD) participants each displayed cognitive impairments relative to same‐aged peers across most domains. Both groups displayed positive relationships between impairment among non‐amnestic cognitive domains and baseline age. This relationship displayed a significantly greater effect in EOAD than LOAD, with domains of Processing Speed/Attention and Executive Functioning skills being the most pronounced. Of those participants developing AD, age displayed a disproportionate impact on their symptom onset.

Abstract Background The accuracy and safety of generating medication orders by large language models (LLMs) must be demonstrated. Without standardization, performance evaluation is limited to time and resource-intensive clinician grading. This evaluation aimed to develop a standardized medication format that supports automated performance evaluation (MedMatch). Methods First, a survey of 40 medication prompts was given to clinicians to assess agreement in medication order communication. Second, a clinician panel developed a standardized medication format (MedMatch) for oral and intravenous medications. Third, a clinician-annotated dataset of medication prompts and standardized answers in the MedMatch format was developed for LLM testing. Finally, LLMs were retested with the same dataset, adjusted to exclude route information, to evaluate the appropriate categorization of medication route. Results The formal medication orders consistently showed low omission rates and high overlap for all entities, compared to the verbal and brief written communication types. Lexical overlap results demonstrated pattern norms amongst clinicians with entities appearing most commonly in positions 1-5 in the order of drug name, dose, unit, route, and frequency. In the second survey, the formal written group performed the highest with 78.3% of prompts considered appropriate as a computer-generated response. LLM accuracy on MedMatch order standardization was highest in oral solid (64.2-72.5%), intravenous intermittent (72.5-84.3%), and intravenous push (62.7-74.5%) categories. LLMs performed the worst at categorizing medication orders accurately into intravenous push (18-61%) and intravenous intermittent (51-100%) routes. Conclusions Standardized format for computer-based outputs may support automated performance analysis and enhance the clarity of medication communication.

Abstract Background Large language models (LLMs) are increasingly used in medication-related tasks despite limited evidence supporting their accuracy and safety. LLMs are vulnerable to adversarial attacks and may confabulate in response to fabricated information in the prompt. Given the substantial risk of harm from medication errors, the purpose of this study was to establish a benchmark task to evaluate LLM performance in distinguishing fabricated from real medications. Methods Two datasets (brand and generic), each consisting of 250 medication lists were developed, each including 4-6 medications and one fabricated medication (specifically a Pokémon character), with complete dosing information provided. Six LLMs were evaluated (GPT-5-Chat, GPT-4o-mini, DrugGPT, Gemma-3-27B-IT, Llama-3.3-70B-Instruct, and Qwen3-32B). For each medication list, models were queried for dosing information and disease indication across three experimental conditions: standard decoding, standard decoding with mitigation prompting, and deterministic decoding (temperature = 0). Each experiment was performed in triplicate. Outputs were deemed confabulations if the LLM failed to recognize and reject the fabricated medication. The primary outcome was the rate of confabulations. Exact paired-permutation tests were used to compare confabulation rates across LLMs and prompting approaches. Results Confabulation rates under standard conditions ranged from 2.7% to 99.6%. DrugGPT demonstrated the lowest baseline confabulation rates (2.7-6.4%), whereas other models demonstrated substantially higher rates (47.2-99.6%). Mitigation prompting significantly reduced confabulation rates compared to standard and deterministic decoding across all tasks and datasets (p < 0.001): GPT-5-Chat achieved confabulation rates of 0% with mitigation prompting from baseline rates of 66.4-88.5%. Deterministic decoding did not substantially impact confabulation rates. Conclusions LLMs performing medication-related tasks are susceptible to confabulation in response to invalid or fabricated inputs, raising important safety concerns for clinical use. While mitigation prompting reduced confabulation rates, this study underscores the importance of robust safeguards when deploying LLMs in medication-related decision-making.

Abstract Background Mitochondrial dysfunction in Alzheimer's Disease (AD) is characterized by impaired energy production, oxidative stress, and disrupted calcium homeostasis, which collectively contribute to neuronal damage, synaptic dysfunction, and the progression of AD pathology. A significant knowledge gap in studying mitochondrial dysfunction in Alzheimer's Disease (AD) lies in understanding the precise molecular mechanisms linking mitochondrial metabolic alterations to neuronal damage, particularly how sex‐specific and cell‐type‐specific mitochondrial processes contribute to the progression of AD pathology. Method This study integrates multiple datasets, post‐mortem brain RNAseq from ROSMAP, Mayo Clinic brain bank and Mount Sinai Brain Bank cohort data available via AD Knowledge Portal hosted by Sage Bionetworks), Human Protein Atlas (HPA), MitoCarta3, genome‐scale human metabolic reconstruction, mitochondriome analysis, and brain & blood metabolomics data (data generated by the Alzheimer's Disease Metabolomics Consortium (ADMC)) to identify mitochondrial genes associated with disrupted metabolism in AD. Additionally, biodomain analysis was used to identify key processes associated with pathology. Result A total of 62 common genes were identified across these approaches, highlighting their potential significance in mitochondrial processes. We also developed the framework to perform metabolite‐gene correlation analyses revealing key mitochondrial‐metabolic interactions (eg. homocysteine and MTHFR gene; acetyl‐CoA and PDHA1 gene). Integrated in silico metabolic analysis identified reaction fluxes, uncovering sex‐specific differences in mitochondrial transport and tricarboxylic acid (TCA) cycle subsystems, with notable alterations in females. Biodomain enrichment analysis focused on mitochondrial processes, identifying disruptions in pathways critical for energy production and redox homeostasis. Additionally, we analyzed datasets related to metformin and complex I inhibitors, linking mitochondrial complex I modulation with metabolic alterations in AD. This analysis identified druggable targets and potential therapeutic candidates for mitochondrial metabolic pathways. Conclusion Our findings integrate multiple data sources to provide a systems‐level understanding of mitochondrial metabolism in AD. Sex‐specific differences associated with mitochondrial dysfunction in AD were identified. We identify important mitochondrial transport proteins and metabolites that could potentially be used as druggable targets in AD.

Abstract Background Early mechanisms of neuronal damage in Alzheimer's Disease (AD) include mitochondrial dysfunction and abnormal energy homeostasis. AD‐associated mitochondrial signaling was shown to occur upstream of amyloid precursor protein processing, Ab production, apolipoprotein E (APOE) expression, and tau pathology. However, whether alterations in mitochondrial function are sufficient to instigate AD remains unclear. Mitochondrial complex I (mtCI) is a rate‐limiting enzyme in the oxidative phosphorylation (OXPHOS) machinery. Changes in mtCI activity influence ATP and ROS production, cellular energy and redox homeostasis. We examined how reduced activity of mtCI due to a knockout of one of its accessory subunits Ndufs4 in mice affects transcriptional networks in the brain. We also determined whether reduced activity of mtCI abrogates neuroprotective properties of a small molecule CP2. Method RNA‐sequencing of the brain tissue was performed using Ndufs4 ‐/‐ and control mice ( n = 3–5 per group). The co‐expression modules of differentially expressed genes were compared to the AMP‐AD patient cohort data. Key mechanisms, including mitophagy signaling, were confirmed using ELISA and western blot analyses; the effect on mtCI activity was confirmed using functional assays. Results In Ndufs4 ‐/‐ mice, a residual activity of mtCI was ∼50% of the control. This loss of function led to a global disruption of mitochondrial homeostasis, energy metabolism, synaptic function, significant increase in mitophagy and a reduction in biogenesis. Transcriptomic signatures in brain tissue of male and female Ndufs4 ‐/‐ mice overlapped with changer reported for AD patients and AD mouse models with Ab and p ‐tau pathologies. These changes were partially rescued by a neuroprotective mitochondria‐targeting small molecule CP2. Consistent with studies in AD mice, CP2 treatment in Ndufs4 ‐/‐ mice augmented the expression of genes associated with mitochondrial biogenesis and turnover, synaptic activity, autophagy, redox balance, and reduced expression of genes related to inflammation. Sex‐specific differences included better adaptation to a reduced activity of mtCI and to CP2 treatment in female mice. Conclusion Reduced mtCI activity was sufficient to induce the AD‐like transcriptomic changes in brain. Residual mtCI function was sufficient to mediate neuroprotective effect of CP2, emphasizing a key role of mitochondria in AD development and treatment.

Abstract Background Alzheimer's disease (AD) affects all brain cells and has complex genomic and immunological alterations. Previous research discovered missense AD risk or protective variants in microglial genes ABI3 and PLCG2, respectively. Expression levels of these genes are altered in AD and can influence microglial function. This study aims to uncover protective, and risk microglial molecular signatures associated with these variants to determine their role in microglial subtypes and states in AD by single cell expression and functional studies. Method We generated microglia‐enriched snRNAseq data from donors harboring either AD protective PLCG2 or AD risk ABI3 missense mutations, or neither mutation. After standard snRNAseq QC, we performed differential expressed gene (DEG) analysis of all microglial cells between variant carriers and non‐carriers using MAST. We defined protective signature as genes that are both down in ABI3 and up in PLCG2 mutation‐carriers. In contrast, risk signature genes are up in ABI3 and down in PLCG2 mutation‐carriers. We investigated the conservation of protective and risk signatures across multiple datasets, including scRNAseq data from iPSC‐derived microglia cells carrying PLCG2 protective variant, snRNAseq data from AD‐resilient donors, and data from AD and other diagnostic groups across multiple brain regions sourced from external datasets. Result We obtained snRNAseq profiles of 35,000 microglia from AD variant carriers. Our DEG analysis among all microglia cells provided 227 microglial protective and 293 risk signature genes defined by these variants. Using integrated analysis of multiple internal and external datasets, we further narrowed down these signatures. We determined that these high‐confidence protective signature genes are downregulated in early AD, upregulated in late AD brains and positively‐correlated with protective variant load in in vitro models. In contrast, risk signature genes are upregulated in early AD, downregulated in late AD and resilient donors. Risk signature expression is decreased with protective PLCG2 variant load, and altered with Aβ treatment in in vitro models. Conclusion Our study uncovers microglia specific protective and risk signatures associated with AD using sn/scRNAseq datasets from multiple sources and models. These findings nominate novel immune targets and pathways with implications for microglial function in health and disease, and ultimately therapeutic potential.

Abstract Background Alzheimer's disease (AD) affects all brain cells and has complex genomic and immunological alterations. Previous research discovered missense AD risk or protective variants in microglial genes ABI3 and PLCG2 , respectively. Expression levels of these genes are altered in AD and can influence microglial function. This study aims to uncover protective, and risk microglial molecular signatures associated with these variants to determine their role in microglial subtypes and states in AD by single cell expression and functional studies. Method We generated microglia‐enriched snRNAseq data from donors harboring either AD protective PLCG2 or AD risk ABI3 missense mutations, or neither mutation. After standard snRNAseq QC, we performed differential expressed gene (DEG) analysis of all microglial cells between variant carriers and non‐carriers using MAST. We defined protective signature as genes that are both down in ABI3 and up in PLCG2 mutation‐carriers. In contrast, risk signature genes are up in ABI3 and down in PLCG2 mutation‐carriers. We investigated the conservation of protective and risk signatures across multiple datasets, including scRNAseq data from iPSC‐derived microglia cells carrying PLCG2 protective variant, snRNAseq data from AD‐resilient donors, and data from AD and other diagnostic groups across multiple brain regions sourced from external datasets. Result We obtained snRNAseq profiles of 35,000 microglia from AD variant carriers. Our DEG analysis among all microglia cells provided 227 microglial protective and 293 risk signature genes defined by these variants. Using integrated analysis of multiple internal and external datasets, we further narrowed down these signatures. We determined that these high‐confidence protective signature genes are downregulated in early AD, upregulated in late AD brains and positively‐correlated with protective variant load in in vitro models. In contrast, risk signature genes are upregulated in early AD, downregulated in late AD and resilient donors. Risk signature expression is decreased with protective PLCG2 variant load, and altered with Aβ treatment in in vitro models. Conclusion Our study uncovers microglia specific protective and risk signatures associated with AD using sn/scRNAseq datasets from multiple sources and models. These findings nominate novel immune targets and pathways with implications for microglial function in health and disease, and ultimately therapeutic potential.