Predicting trajectories of acute illness using RNA velocity of whole blood

Aubrey J. Cunnington, Claire Dunican, Clare Wilson, Dominic Habgood-Coote, Suzanna Patterson, Mahdad Noursadeghi, Raymond M Moseki, Cari Stek, Robert J. Wilkinson, Philipp Agyeman, Coco R. Beudeker, Giske Biesbroek, Ulrich von Both, Karen Brengel‐Pesce, Enitan D. Carrol, Lachlan Coin, Giselle D’Souza, Tisham De, Marieke Emonts, Katy Fidler, Colin G. Fink, Michiel van der Flier, Ioanna Georgaki, Laura Kolberg, Mojca Kolnik, Taco W. Kuijpers, Federico Martinón‐Torres, Marine Mommert-Tripon, Samuel J. Nichols, Stéphane Paulus, Marko Pokorn, Andrew J. Pollard, Irene Rivero‐Calle, Aleksandra Rudzate, Luregn J. Schlapbach, Nina A. Schweintzger, Ching‐Fen Shen, Shrijana Shrestha, Chantal Tan, Μαρία Τσολιά, Effua Usuf, Fabian van der Velden, Clementien Vermont, Marie Voice, Shunmay Yeung, Dace Zavadska, Werner Zenz, Victoria Wright, Michael Levin, Jethro Herberg, Rachel Lai, Graeme Meintjes, Christopher Chiu, Mauricio Barahona, Myrsini KaforouResearch Square 2025 · DOI

Introduction: Moving from Reactive to Proactive Diagnostics

In acute clinical care, the ability to anticipate a patient's trajectory is often the difference between a successful intervention and a critical failure. Current diagnostic tools typically provide a "static snapshot"—a measurement of how sick a patient is at the exact moment of the blood draw. This limitation often leads to "cautious over-treatment," where stable patients are hospitalized unnecessarily, or "missed windows," where patients who appear stable are on the verge of rapid deterioration.

To address this, researchers have developed a novel method called VeloCD (RNA velocity-based transition predictions of future disease states). By adapting RNA velocity—a technique originally designed for single-cell analysis—to bulk whole-blood samples, it is now possible to predict the future clinical status of individuals with acute illness from a single blood draw.

The Method: Harnessing the "Velocity" of Gene Expression

The core innovation of VeloCD lies in its use of the ratio between nascent (unspliced) and mature (spliced) mRNA.

  • Biological Basis: When a gene is activated, it first produces unspliced pre-mRNA containing introns. This is subsequently processed into spliced mature mRNA.
  • The Vector: A high ratio of unspliced to spliced transcripts indicates that gene expression is ramping up (positive velocity), while a low ratio suggests it is being dialed back (negative velocity). By comparing these abundances, the system calculates a "velocity vector."
  • VeloCD Framework: The researchers introduced a gene selection algorithm called DISCO (Discordance-concordance analysis) to identify genes where the velocity signal aligns with observed clinical changes. This allows for the calculation of transition probabilities toward specific reference states, such as "Severe Illness" or "Recovery."

Validation Across Diverse Clinical Scenarios

The study validated VeloCD using three distinct datasets, demonstrating its versatility across different pathogens and patient populations:

  1. Human Challenge Models (CHIMs): Healthy volunteers were intentionally inoculated with Influenza A or SARS-CoV-2. VeloCD was able to predict which individuals would eventually test PCR-positive using samples taken before the virus was detectable by standard diagnostic tests.
  2. TB-IRIS Prediction: In HIV-TB coinfected patients, the tool predicted the onset of Immune Reconstitution Inflammatory Syndrome (TB-IRIS)—a dangerous inflammatory overreaction—weeks before clinical symptoms appeared.
  3. Pediatric Risk Stratification (PERFORM Study): In a multinational study of 399 febrile children presenting to emergency departments, VeloCD successfully differentiated between mild cases and those requiring intensive care.

Key Results and Insights

The performance of VeloCD across these cohorts suggests a powerful new dimension for transcriptomic analysis:

  • High Predictive Accuracy: VeloCD achieved an AUROC of 0.89 for predicting influenza infection up to three days in advance. For SARS-CoV-2, AUROCs ranged between 0.72 and 0.73.
  • Long-Range Forecasting: The ability to predict TB-IRIS weeks in advance suggests that the molecular "decision" to develop complications occurs long before physical symptoms manifest.
  • Early Warning for Moderate Cases: In the pediatric study, patients admitted to general wards who had higher VeloCD transition probabilities were significantly more likely to eventually require surgical intervention or intensive care.
  • Cross-Viral Utility: Gene signatures identified for one virus (Influenza) showed moderate utility for another (SARS-CoV-2), suggesting that the velocity of the interferon response is a conserved feature of the human viral defense.

Future Directions and Limitations

While VeloCD represents a significant leap forward, several hurdles remain before clinical implementation. The current method relies on RNA-Seq, which is relatively slow and expensive for emergency settings. Future work will focus on translating these "velocity signatures" to faster, point-of-care platforms like qPCR or LAMP.

Additionally, larger, independent prospective cohorts are needed to refine the reference "fate maps" and account for the confounding effects of medical treatments. By adding a temporal dimension to blood transcriptomics, VeloCD offers a glimpse into the future of personalized, proactive medicine, allowing clinicians to treat the trajectory of an illness rather than just its current state.

Summary updated

Abstract

<title>Abstract</title> Transcriptomic analyses reveal the status of cells, tissues, or organisms, across states of health and disease. RNA velocity adds a temporal dimension to single cell analyses, predicting future transcriptomic and phenotypic states, based on current spliced and unspliced mRNA of each cell. We hypothesized that RNA velocity could be adapted to predict future clinical status of individuals with acute illness using their whole-blood transcriptome. We developed a method for quantitative prediction of transitions in clinical state from a single time-point sample, which we call VeloCD. This predicted transcriptomic trajectories and future infection status in influenza A and SARS-CoV-2 human challenge studies. In HIV-TB coinfected individuals, it predicted the onset of immune reconstitution inflammatory syndrome. In a multinational observational study of acutely unwell febrile children, VeloCD predicted those with greatest medical care requirements. Our results demonstrate a novel application of RNA velocity to predict the trajectory of acute illness.
BibTeX
@inproceedings{cunnington2025predicting,
  author = {Aubrey J. Cunnington and Claire Dunican and Clare Wilson and Dominic Habgood-Coote and Suzanna Patterson and Mahdad Noursadeghi and Raymond M Moseki and Cari Stek and Robert J. Wilkinson and Philipp Agyeman and Coco R. Beudeker and Giske Biesbroek and Ulrich von Both and Karen Brengel‐Pesce and Enitan D. Carrol and Lachlan Coin and Giselle D’Souza and Tisham De and Marieke Emonts and Katy Fidler and Colin G. Fink and Michiel van der Flier and Ioanna Georgaki and Laura Kolberg and Mojca Kolnik and Taco W. Kuijpers and Federico Martinón‐Torres and Marine Mommert-Tripon and Samuel J. Nichols and Stéphane Paulus and Marko Pokorn and Andrew J. Pollard and Irene Rivero‐Calle and Aleksandra Rudzate and Luregn J. Schlapbach and Nina A. Schweintzger and Ching‐Fen Shen and Shrijana Shrestha and Chantal Tan and Μαρία Τσολιά and Effua Usuf and Fabian van der Velden and Clementien Vermont and Marie Voice and Shunmay Yeung and Dace Zavadska and Werner Zenz and Victoria Wright and Michael Levin and Jethro Herberg and Rachel Lai and Graeme Meintjes and Christopher Chiu and Mauricio Barahona and Myrsini Kaforou},
  title = {{Predicting trajectories of acute illness using RNA velocity of whole blood}},
  booktitle = {Research Square},
  year = {2025},
  doi = {10.21203/rs.3.rs-5764288/v1},
}
Figure