Research

Our projects

From the stability of the stem cell genome to new diagnostics at the bedside: an overview of what we are working on now.

01 · Genome integrity Ongoing

Jumping genes in blood stem cells

About 40% of the human genome consists of retrotransposons, so-called "jumping genes" long dismissed as junk DNA. A few families, such as L1HS and AluY, can still copy themselves into new places in the genome and disrupt genes. Blood stem cells have to keep their genome intact for a lifetime. We study how they keep these elements in check.

Methods
Whole-genome sequencing of single-cell-derived colonies, phylogenetics, knock-in mouse models, human HSC culture
Partners
Cambridge Stem Cell Institute (E. Laurenti, A.R. Green, S.J. Loughran, J. Nangalia)
Preprint
Deuel et al., bioRxiv 2025
mutations (SBS) →birth → age at samplingLSELLLINE-1SSINE / AluEEndogenous retrovirusone tip = one stem cell
Finding insertions in family trees of stem cells. We sequence the genomes of hundreds of single stem cells and use their mutations to build a family tree. PEAR-TREE then finds new retrotransposon insertions and places each one on its branch. Schematic.
OUTSIDE THE CELL CYTOPLASM THPO MPL JAK2 STAT5 · MAPK canonical signalling: survival, self-renewal (unchanged in mutant) pY HNRNPA1 P P L1 / ERV transcripts genome insertion blocked
A newly identified protective pathway. THPO acts through its receptor MPL and the kinase JAK2 to phosphorylate two tyrosines on the RNA-binding protein HNRNPA1, which then binds retrotransposon RNA differently and represses it.

How often do genes jump in blood stem cells?

We developed two methods: PEAR-TREE finds new insertions in whole-genome data from single stem cells, and PARTRIDGE detects new endogenous retrovirus insertions in bulk mouse stem cells with very high sensitivity. With them, we measured for the first time how often retrotransposition happens in healthy human and mouse blood formation. Compared with other tissues, the rate in stem cells is extremely low, but it is not zero.

3 / 368insertions in stem cells of a 63-year-old man
57 / 612insertions in stem cells of a 30-month-old mouse
~460 vs ~4,610estimated lifetime insertions across all stem cells, human vs mouse

Why so few?

Stem cells express a lot of retrotransposon RNA, yet very few new insertions appear. We found that thrombopoietin (THPO), a key stem cell cytokine, strongly represses retrotransposons while stem cells divide, which is when they are most vulnerable. In knock-in mice whose HNRNPA1 cannot be phosphorylated, this protection is lost and insertions rise, while canonical MPL/JAK/STAT signalling stays intact.

Next: leukaemic stem cells

Loss of frequently mutated genes such as DNMT3A, TET2 or TP53 de-represses retrotransposons in other cancers. We ask whether mutant haematopoietic and leukaemic stem cells depend on alternative brakes such as the THPO/MPL/HNRNPA1 axis to avoid lethal levels of insertional mutagenesis.

Team Jeremy Deuel, Teresa OlbrichCollaboration Elisa Laurenti, Jyoti Nangalia, Steve Loughran, Anthony R. Green (all Cambridge)

02 · Exhalomics Ongoing

Reading disease in a breath

Every breath carries hundreds of volatile metabolites. Secondary electrospray ionisation high-resolution mass spectrometry (SESI-HRMS) measures them in real time and without needles. We run two prospective clinical trials using this "exhalomics" method in patients with acute leukaemia.

This work received the SFK Candy Heberlein Award.

Method
SESI-HRMS, developed by Prof. M. Kohler (Pulmonology, USZ) and Prof. P. Sinues (University of Basel)
Partners
Prof. A. Egli, Institute of Medical Microbiology, UZH · Prof. N. Cabezas-Wallscheid, ETH Zurich
Data capture
Our own electronic case report form, approved by the ethics committee, with an iPad questionnaire for patients
Exhaled breath SESI-HRMS real-time mass spectrometry m/z Metabolic signature ? diagnosis, response
From breath to biomarker. A patient breathes into the SESI-HRMS instrument. The mass spectrum is compared across patients to find signatures of infection or of the leukaemia itself.
At each visit: breath analysis + bone marrowINDUCTION CYCLE 1CYCLE 2t1diagnosisbefore therapyt2~ day 14t3end ofcycle 1t4end ofcycle 2Breath profileslongitudinal exhalomics from diagnosis onPlasma metabolomicscorrelation with ETH ZurichRoutine diagnosticsflow cytometry, NGS, transcriptomeQuality of lifeFACT-Leu questionnaire on iPad
EXHALE study design. 120 patients with newly diagnosed or relapsed/refractory AML give breath samples at the same four timepoints as their routine bone marrow assessments.

REDEFINE: detecting fungal infections early 2024 – 2027

Invasive fungal infections of the lung, such as aspergillosis, are dangerous for patients receiving intensive induction therapy for acute leukaemia. They are hard to diagnose, and late diagnosis has serious consequences. REDEFINE tests whether breath analysis can detect them earlier. We built the bioinformatic analysis from scratch. Results are expected in 2027.

EXHALE: tracking the leukaemia itself since Oct 2026

EXHALE looks for metabolic signatures of the leukaemia in breath. We take breath samples before, during and after treatment to predict response and to find new volatile biomarkers linked to outcome. Plasma and leukaemia-cell metabolomics at ETH Zurich serve as external validation. The study is single-centre at USZ and planned to run for three years.

Team Jeremy Deuel, Kevin HoferCollaboration Adrian Egli (UZH), Kohler group (USZ), Nina Cabezas-Wallscheid (ETH Zurich)

03 · Digitalisation & AI Ongoing

Digital haematology

Haematology diagnosis means combining microscopy, flow cytometry, cytogenetics and sequencing. Our web application Cinderella brings all of these together and has been in daily diagnostic use at USZ for more than five years. It covers AI-assisted microscopy, integrated flow cytometry reporting, molecular genetics and NGS, clinical annotation and laboratory statistics.

In clinical use
Cinderella (integrated diagnostics), diagnostic RNA-seq, patient-specific ddPCR MRD assays
Coming up
Unified AML risk stratification from ~500 cases · human vs AI blast counting · open-source RNA-seq platform
Teaching
BlastenBlaster, a cell-recognition game for haematology training
blast continuous maturation score mature dashed box = called blast
Every cell, on a continuum. Our classifiers segment each cell of a bone marrow smear and place it on a continuous maturation scale. A slide with thousands of cells is annotated within seconds, live in the browser. Schematic.
0%0%20%20%40%40%60%60%20% AML thresholdcontext-free classifierhuman observerssparse slide:stricterblast-rich slide:more generousblast fraction, classifierblasts counted by humansschematic
Humans count blasts in context. Experts are more generous counting blasts on blast-rich slides than on sparse ones. Because a 20% blast cut-off still defines AML in many cases (WHO-5, ICC-2022), AI models must take the whole slide into account, not only single cells. Schematic.
Patient samplebone marrow / bloodRNA extractionRNA sequencingwhole-transcriptomereadsFusion & variant callersincl. in-house callersWeb reportfor haematologists,no coding neededMRD assay designpatient-specificddPCRIGH::DUX4FLT3-ITDPAX5 dup.+ standard fusion panelguides transplant & post-transplant therapy
Diagnostic RNA-seq, end to end. The first platform of its kind in routine diagnostics in Switzerland, validated for acute lymphoblastic leukaemia and other blood cancers. In-house callers detect fusions that are hard to find, such as IGH::DUX4, FLT3-ITDs and intragenic PAX5 duplications. From the detected fusion, the platform designs patient-specific ddPCR assays for minimal residual disease. We have validated a dozen so far, and they guide decisions on transplantation and post-transplant therapy.

Team Jeremy Deuel, Sebastian HahnCollaboration Stefan Balabanov, Max Rieger (both USZ)

Past projects

Infectious diseases Completed 2023

COVID-19 in young adults

In March 2020, one of the first and fastest COVID-19 outbreaks in Switzerland hit the army base in Airolo. An ad-hoc research group on site produced some of the first descriptions of transmission, immunity and long-term sequelae in young adults, together with Dr Roland Züst (Spiez Laboratory). Soldiers infected before social distancing, masks and hygiene measures were introduced often fell ill: 29% of 354 developed COVID-19. In a spatially separated group of 154 soldiers, where the virus only arrived after these measures, 15% of those tested were infected, but nobody developed COVID-19. A lower viral inoculum seems to make the disease milder and transmission slower (Clin Infect Dis 2021). Among 199 recruits, predicted maximal aerobic capacity (VO2max) fell by a median of 0.9 ml/min/kg in COVID-19 convalescents, but not in asymptomatically infected or naive recruits.

The follow-up study LoCoMo, run with Prof. Patricia Schlagenhauf (EBPI, University of Zurich) and funded by the Swiss Armed Forces, examined 501 unvaccinated service members. More than 180 days after COVID-19, 177 previously infected participants had a higher BMI, higher cholesterol and LDL, and a lower aerobic threshold than 251 never-infected controls. This points towards a metabolic pattern, while other psychosocial, eye and sperm outcomes did not differ.

LoCoMo Study
Deuel et al., Lancet Infect Dis 2022
Airolo outbreak
Bielecki … Deuel, Clin Infect Dis 2021
Crameri … Deuel, Eurosurveillance 2020
Assessments
Vital signs, blood tests, immunology, cardiopulmonary exercise testing, lung function, FeNO, OCT angiography, sensory and psychological testing, spermiogram
A. Clin Infect Dis 2021: social distancing and the course of COVID-19Airolo army base, March 2020: soldiers infected before vs after social distancing, masks and hygiene measures.
Army base Airolo, March 2020. 354 soldiers infected before social distancing, masking and hygiene: 29% developed COVID-19 (102) and 62% were infected (113 of 181 tested), consistent with a high viral inoculum, more symptomatic cases and a higher transmission rate. 154 spatially separated soldiers infected afterwards: 0% developed COVID-19 and 15% were infected (13 of 88), consistent with a low viral inoculum, fewer symptomatic cases and a lower transmission rate. B. Eurosurveillance 2020: recruits losing more than 10% of predicted VO₂maxShare of recruits, before vs about 45 days after the Airolo outbreak (paired data)SARS-CoV-2 naive0% (0 of 36)Asymptomatically infected1.9% (1 of 55)COVID-19 convalescents18.8% (9 of 48)0%20%Median change in VO₂max: +0.1, 0 and −0.9 ml/min/kg respectively (p = 0.005).
C. Lancet Infect Dis 2022 (LoCoMo): more than 180 days after COVID-19Never-infected controls (blue, n = 251) vs non-recent COVID-19 (red, n = 177); boxes show median and IQR. * p < 0.05, ** p < 0.01. Lipids in mmol/L.
Box plots with individual data points comparing never-infected controls and participants more than 180 days after COVID-19: BMI (median 23.2 vs 24.0 kg/m2, p < 0.05), aerobic threshold as % of VO2max (41 vs 39, p < 0.05), LDL (2.2 vs 2.4 mmol/L, p < 0.01) and total cholesterol (3.9 vs 4.2 mmol/L, p < 0.01); horizontal lines mark reference limits.

Fatigue was also slightly higher after COVID-19 (Chalder score 12 vs 11, p = 0.027); other psychosocial scores, eye findings and sperm quality did not differ.

Main findings of the three papers. A: Bielecki … Deuel, Clin Infect Dis 2021. B: Crameri … Deuel, Eurosurveillance 2020, Table (199 recruits; 139 with paired data). C: Deuel et al., Lancet Infect Dis 2022; never-infected controls (n = 251) vs non-recent COVID-19 (mean 340 days after diagnosis, n = 177). Observational data, not proof of causation.
0 of 154soldiers infected after social distancing developed COVID-19, compared with 29% of 354 infected before it
−0.9ml/min/kg median change in predicted VO2max in 48 COVID-19 convalescent recruits (naive +0.1, asymptomatic 0; p = 0.005). 9 of 48 (18.8%) lost more than 10%, vs 0 of 36 naive.
180 daysafter COVID-19, 177 young adults had a higher BMI (24.0 vs 23.2 kg/m2) and LDL (2.4 vs 2.2 mmol/L) than 251 controls, and a lower aerobic threshold (39% vs 41%)
Metabolicpattern: higher BMI, cholesterol (4.2 vs 3.9 mmol/L) and LDL with lower endurance. Fatigue score slightly higher (12 vs 11); no differences in eye findings or sperm quality.

Team Jeremy Deuel, Michel Bielecki, Giovanni A. CrameriCollaboration Patricia Schlagenhauf (UZH), Roland Züst (Labor Spiez), Sandrine Zweifel (USZ), Swiss Armed Forces