New Model Captures Full RT-QuIC Kinetics in Prion Disease

Researchers at Sapienza University of Rome and the Istituto Superiore di Sanità have developed a statistical framework that analyzes the complete temporal trajectory of Real-Time Quaking-Induced Conversion (RT-QuIC) assays in cerebrospinal fluid. Published in Frontiers in Digital Health, the study demonstrates that reducing these continuous fluorescence curves to standard summary metrics discards significant heterogeneity in the reaction kinetics. By applying a three-way tensor clustering approach to 296 patient samples, the team identified distinct empirical profiles of prion seeding activity that are not visible through conventional diagnostic measures.

Prion diseases, including Creutzfeldt-Jakob disease, are invariably fatal neurodegenerative disorders characterized by rapid progression and clinical heterogeneity. Early diagnosis remains challenging because symptoms often overlap with other rapidly progressive dementias. The RT-QuIC assay has become a critical tool for antemortem detection, amplifying the pathogenic prion protein (PrPSc) with near-perfect specificity. However, routine clinical practice typically collapses the continuous fluorescence curve into just two parameters: the lag phase and the maximum fluorescence intensity. This reduction erases the complex temporal dynamics that may contain latent biological information regarding pathological burden and strain properties.

To address this information loss, the research team proposed an end-to-end statistical framework that captures the complete fluorescence distribution over time. The dataset comprised cerebrospinal fluid RT-QuIC trajectories from 301 patients referred to the National CJD Registry in Italy between 2014 and 2020 due to clinical suspicion of prion disease. After excluding five patients with consistently shorter recordings, 296 patient-level profiles were included in the tensor modeling and clustering analyses. Each patient’s data consisted of four technical replicates, which were retained in the tensor structure to characterize within-sample assay variability and derive a consensus fluorescence profile.

The analytical approach involved two main stages. First, the team fitted each patient-level consensus trajectory using a five-parameter growth-attenuation function. This phenomenological model included parameters for amplitude, growth rate, logistic midpoint, baseline level, and late-phase attenuation. Unlike standard sigmoidal models, this function accounts for the decay of fluorescence signal after the initial growth phase, a feature frequently observed in RT-QuIC trajectories but ignored by conventional summaries. Second, the researchers applied a three-way Tucker3 tensor clustering approach to the unreduced time courses across all technical replicates. This method simultaneously clusters patients based on similarities in their fluorescence curves while reducing the dimensionality of the variables.

The primary clustering solution, determined by the Calinski-Harabasz index, identified three broad empirical RT-QuIC response profiles. The largest cluster, comprising 206 patients, was characterized by low fluorescence values with only a slight increase over time, representing weak or almost absent responses. The second cluster, including 77 patients, showed an intermediate fluorescence response with a gradual increase followed by stabilization or slight decrease. The smallest cluster, consisting of 13 patients, exhibited a rapid and strong increase in fluorescence, indicating a high-response group. This three-group partition was retained as the primary descriptive solution due to its parsimony and high reproducibility.

To explore finer temporal heterogeneity, the team also examined an exploratory nine-cluster solution. This higher-resolution partition subdivided the broad response groups into more specific fluorescence patterns, distinguishing differences in the timing and intensity of the signal increase. For instance, the low-response group was split into clusters with nearly constant fluorescence and those with slight late increases. Although this nine-cluster solution was globally reproducible, its smaller subdivisions showed greater variability in cluster-specific reproducibility, leading the authors to classify it as an exploratory representation of heterogeneity rather than a definitive classification.

Resampling-based assessment confirmed the stability of the primary clustering solution. In 2,000 bootstrap repetitions involving 80% subsampling of the cohort, the three-group partition was consistently reproduced with a mean adjusted Rand index of 0.970. The largest cluster was almost perfectly reproduced, with a mean Jaccard similarity of 0.991. At the patient level, 99.0% of samples showed high assignment stability, defined as a probability of at least 0.80 of retaining the reference-cluster assignment. These results support the robustness of the identified kinetic phenotypes within this specific cohort.

The study also compared the curve-based clustering with a parameter-based clustering approach. Using K-means clustering on the five fitted model parameters, the team identified six groups of unequal size. The confusion matrix between the parameter-based and curve-based classifications revealed that the two approaches provide complementary information. Similar parameter values could correspond to different fluorescence curve shapes, particularly in the near-flat curve group, which included observations from almost all parameter-based clusters. This many-to-one relationship highlights the complexity of mapping fitted parameters to observable curve morphology and underscores the value of analyzing the full trajectory.

The authors emphasize that the identified profiles are cohort-specific empirical patterns of the RT-QuIC fluorescence reaction. The model is phenomenological rather than mechanistic, meaning it summarizes observable features of the assay trajectories without estimating microscopic rate constants for specific molecular processes such as nucleation or fragmentation. The late-phase attenuation parameter provides an additional descriptor not included in standard models, but its biological, clinical, and diagnostic interpretation requires independent validation. The differences described in the fitted trajectories should not be interpreted as evidence of specific aggregation mechanisms or prion strains.

Several limitations constrain the interpretation of these findings. The data were derived from a single cohort in Italy, and RT-QuIC kinetics are highly sensitive to instrumentation, shaking cycles, and recombinant substrate batches. Therefore, these clusters must be treated as a structured hypothesis space rather than an absolute diagnostic taxonomy until validated on external, independent datasets. Clinical, biomarker, genetic, and survival variables were not included in the clustering analyses, so the identified groups cannot be linked to specific clinical phenotypes or molecular subtypes in this study.

Future research should assess associations between curve-derived parameters and clinical phenotype, PRNP genotype, molecular subtype, biomarkers, and survival. A two-cohort design, with an independent discovery cohort for model development and a separate validation cohort for assessing cluster reproducibility, is recommended. By moving beyond binary diagnostic readouts toward a quantitative characterization of the complete amplification process, this framework provides a foundation for exploring RT-QuIC kinetics as a potential tool for patient stratification and longitudinal monitoring of pathological seeding activity.

The RT-QuIC assay detects prion seeding activity by amplifying the pathogenic prion protein, PrPSc, which templates the conversion of recombinant prion protein into amyloid fibrils under cyclic shaking. This process is tracked in real time through Thioflavin-T fluorescence, producing a continuous curve that reflects the reaction’s kinetics. Second-generation RT-QuIC assays have demonstrated a specificity approaching 100% and a sensitivity of 90 to 96% in cerebrospinal fluid, performance that has led to the assay’s inclusion in formal diagnostic criteria for probable Creutzfeldt-Jakob disease. Despite these high diagnostic metrics, the underlying measurements are quantitative, yet their conventional clinical use is essentially categorical, determining whether a sample is classified as positive or negative rather than describing how the reaction evolves over time.

The observed fluorescence trajectories exhibit striking heterogeneity across patients and even across technical replicates. Some curves rise in a classical sigmoid shape and plateau, while others amplify late, peak weakly and transiently, or decay along an unstable baseline. Curves with similar threshold-defined lag phases or maximum fluorescence values can differ markedly in the steepness and timing of their rise, baseline behavior, and the extent of late signal attenuation. These variations are features of the observed assay trajectory and may contain reproducible information beyond conventional summary metrics, although their biological and clinical determinants remain to be established.

The five-parameter growth-attenuation function used in this study is phenomenological rather than mechanistic. It is intended to summarize observable features of the RT-QuIC fluorescence trajectories without estimating microscopic rate constants for primary nucleation, elongation, fragmentation, or aggregate maturation. The model includes a growth phase representing the initial increase in fluorescence, followed by an exponential decay component. The late-phase attenuation parameter, denoted as d, summarizes the observed curve shape and should not be interpreted as a direct measure of a single biological mechanism. Possible contributors to this attenuation include fluorescence-related, optical, aggregation-related, and assay-specific processes, which cannot be distinguished in the present analysis.

Establishing a mechanistic correspondence between the fitted parameters and specific molecular processes would require experiments conducted under systematically varied reaction conditions. In mechanistic aggregation models, macroscopic kinetic features can arise from coupled processes, meaning a single phenomenological parameter may reflect the combined contribution of several molecular events. Therefore, the fitted parameters in this study serve as empirical descriptors of curve morphology rather than direct estimates of molecular rate constants. The model accommodates different curve families, including monotonic logistic growth, monotonic decay, and rise-and-fall trajectories, providing a compact empirical fingerprint of the assay response.

Quantitative characterization of the complete RT-QuIC trajectory could potentially transform the assay into a tool for characterizing the dynamics of pathological seeding activity. In individuals at risk for genetic prion disease, cerebrospinal fluid prion seeding activity detected by RT-QuIC has been shown to precede symptom onset by one to several years. A quantitative and reproducible measure may allow changes in RT-QuIC kinetics to be assessed longitudinally, potentially providing information on disease evolution that cannot be captured by repeated binary measurements. In future disease-modifying trials, such measures could serve as exploratory pharmacodynamic or progression biomarkers, allowing longitudinal assessment of treatment effects on pathological seeding activity before irreversible neuronal damage occurs.

Emerging evidence from alpha-synuclein seed amplification assays supports the concept that quantitative kinetic measures can provide information related to disease stage, clinical severity, and longitudinal progression. Although these applications remain to be demonstrated and validated for prion diseases, they highlight the potential value of moving beyond a binary diagnostic readout. The current framework provides a quantitative basis for testing whether RT-QuIC curve morphology can capture clinically meaningful changes, offering a step toward a more detailed characterization of pathological seeding activity for diagnosis and patient stratification.

Subject of Research: Statistical modeling of RT-QuIC fluorescence kinetics in prion disease

Article Title: Statistical learning for RT-QuIC kinetics phenotyping in prion disease

Article References: D'Andrea, G., Poleggi, A., Sbriccoli, M., Moracci, G., Porreca, F., Costanzo, M., Gigante, G., & Vichi, M. (2026). Statistical learning for RT-QuIC kinetics phenotyping in prion disease. Frontiers in Digital Health, 8, 1940768. https://doi.org/10.3389/fdgth.2026.1940768

Image Credits: AI Generated

DOI: 10.3389/fdgth.2026.1940768

Keywords: prion disease, RT-QuIC, statistical modeling, tensor clustering, kinetics, neurodegeneration, diagnostics, Creutzfeldt-Jakob disease, Statistical, learning, phenotyping, prion