Traces of 98 Diseases in One Blood Tube: A Whole-Blood Transcriptome Atlas from Erzurum
In this article
Traces of 98 diseases in a single tube of blood: a whole-blood transcriptome atlas from Erzurum
In a joint study by Atatürk University and SciLifeLab in Sweden, researchers measured the activity of roughly 20,000 genes in 2-3 ml blood samples from 4,444 people in Erzurum, Turkey, and mapped the "RNA footprint" of 98 different health conditions in the blood. Published in Nature Communications, the Whole Blood Transcriptome Atlas (WBT Atlas) brings together, in a single resource, the imprint that many diseases, from cancer and sepsis to diabetes and Parkinson's disease, leave on the immune system and metabolism. When blood RNA and plasma proteins were combined, an artificial intelligence (AI) model correctly distinguished 28 health conditions in three out of four samples. We discussed the idea of using AI to distinguish multiple diseases from a single blood sample in an earlier article.
- What was done: Between 2022 and 2025, whole-blood RNA sequencing was performed in 4,444 people aged 0-99 at two hospitals in Erzurum; about 5,000 plasma proteins were also measured in 1,280 of them. The results have been released in a publicly accessible online atlas (phenomeatlas.org/wbt).
- Key finding: The strongest determinant of gene activity in the blood was not age or sex but the person's disease. Cancers shared a similar blood signature, and interferon responses were elevated in many cancer types.
- Artificial intelligence: Using only 10 genes per disease, the model distinguished 29 conditions with 64% accuracy; combining RNA and proteins raised accuracy to 75% across 28 conditions. In 10 independent cohorts (1,621 samples), the mean AUC was 0.79; with protein signatures in the UK Biobank, it was 0.73.
- In practice: This is a research resource, not a diagnostic test. It is not yet used for cancer screening or patient monitoring, but it provides a strong foundation for the concept of "many diseases from a single blood sample."
Why does this matter?
Blood circulates through every part of the body and carries traces of disease. Most blood tests that look for signs of cancer today analyze DNA fragments shed from tumors into the bloodstream (liquid biopsy) or proteins in the plasma. The whole-blood transcriptome opens a different window: it shows which genes the cells in the blood, particularly immune cells, are switching on at a given moment and to what extent. In other words, it reads the body's response to disease rather than the tumor itself.
This study matters for two reasons. First, it is the first disease-focused whole-blood transcriptome atlas built at this scale; according to the authors, no resource this comprehensive existed before. Second, it is based on samples collected in Erzurum, Turkey: the samples were collected at Atatürk University and Erzurum Regional Training and Research Hospital, RNA sequencing was performed in Istanbul, and more than 90 of the roughly 110 authors are physicians and scientists from institutions in Turkey. In the "omics" era that began with the Human Genome Project, it is especially gratifying for us to see Turkey take its place as a country that generates data.
- What is a first: According to the authors, there was previously no disease-focused whole-blood transcriptome atlas of this size bringing many diseases together under one roof.
- What is not a first: The protein map of blood had already been charted. In December 2025, Álvez and colleagues, from the same research network, published the "human pan-disease blood atlas" in Science, examining about 5,000 plasma proteins across 59 diseases in more than 8,000 people. The Human Protein Atlas, UK Biobank, TCGA, and GTEx are also core resources for tissues and diseases.
- What is new: RNA and protein were measured together in the same individuals, and the two were shown to complement each other: the 264 RNA signatures and 244 protein signatures shared not a single molecule.
The two corresponding authors are Prof. Dr. Hasan Türkez of the Atatürk University Faculty of Medicine, who is also a vice rector of the university, and Prof. Dr. Adil Mardinoğlu. An electrical and electronics engineering graduate of Istanbul University, Mardinoğlu leads systems biology groups at SciLifeLab (KTH Royal Institute of Technology) in Sweden and at King's College London in the UK, and is known for genome-scale metabolic models and multi-omics analyses. In 2019 he was appointed president of the Health Institutes of Türkiye (TÜSEB), with Türkez serving as its secretary general at the time.
On the Turkish side, more than 90 physicians and researchers from the Atatürk University faculties of medicine and dentistry and the Erzurum Regional Training and Research Hospital provided diagnoses and samples for patients with 98 different conditions. RNA sequencing was performed at SZA Omics (Acıbadem Mehmet Ali Aydınlar University) in Istanbul, protein measurements in Uppsala, Sweden, and the analyses in Stockholm. The atlas is part of the doctoral thesis of first author Mengzhen Li at KTH.
What is the whole-blood transcriptome?
The whole-blood transcriptome is the measurement of all RNA molecules produced by the genes active at a given moment in the cells of the blood. Most of the RNA in blood comes from immune cells such as neutrophils and lymphocytes. The whole-blood transcriptome therefore provides a snapshot of the body's immune and metabolic response to infection, inflammation, cancer, or metabolic disease.
We described the role of RNA sequencing of tumor tissue in cancer treatment in our guide to RNA sequencing; here, the same technology is applied to blood. If DNA is likened to a cookbook containing every recipe a cell carries, RNA shows which recipes are being cooked at that moment. RNA in blood comes from four sources: immune cells (the largest share), red blood cells (hemoglobin RNA), platelets, and circulating extracellular RNA. Because hemoglobin RNA can make up more than half of all RNA and masks other signals, it was removed in the laboratory; as a result, hemoglobin genes accounted for less than 0.85% of total expression.
| Ways of reading disease from blood | What is measured? | What does it reflect? | Current status |
|---|---|---|---|
| Tumor markers (CEA, CA 19-9, CA 125) | Individual proteins | Substances secreted by specific tumors | Used for monitoring; inadequate for screening |
| Cell-free DNA and methylation (e.g., Galleri) | DNA fragments released from tumors into the blood | The tumor itself, directly | In use as a multi-cancer early detection test; its role in screening is debated |
| Plasma proteomics (Olink, about 5,000 proteins) | Circulating proteins | End products released by organs and cells | Research; Science 2025 pan-disease atlas |
| Whole-blood transcriptome (this study) | RNA from about 20,000 genes | The immune and metabolic response to disease | Research; open atlas of 98 conditions |
We covered tumor markers in detail in our articles on CEA, CA 19-9, and CA 125, and multi-cancer blood tests in our article on Galleri.
How was the study conducted?
Figure 1 · Building the Whole Blood Transcriptome Atlas
From blood collected in Erzurum to international validation.
| Feature | WBT Atlas |
|---|---|
| Participants | 4,444 people; aged 0-99 (mean 42.3); 48.7% male |
| Health conditions | 98 conditions in 15 major groups: cancers, cardiovascular, neurologic, psychiatric, autoimmune, infectious, and childhood diseases; plus healthy individuals and "long-lived" people over 80 |
| Setting and period | Atatürk University Health Research and Application Center and Erzurum Regional Training and Research Hospital; January 2022 - September 2025 |
| RNA measurement | PAXgene tubes, hemoglobin RNA depletion, sequencing on NovaSeq 6000 (Istanbul) |
| Protein measurement | 1,280 people, 28 conditions; about 5,000 proteins with Olink Explore HT (Uppsala, Sweden) |
| External validation | 10 independent RNA cohorts (1,621 samples; sepsis, bipolar disorder, Parkinson's disease, rheumatoid arthritis, type 2 diabetes) and UK Biobank protein data (9,998 samples; breast cancer, lung cancer, coronary artery disease, Parkinson's disease, type 2 diabetes) |
What did the study find?
1. Disease changes the blood the most
The researchers calculated how much of the variation in blood gene activity was explained by age, sex, body mass index, and disease. For most of the highly variable genes, disease accounted for the largest share, with age in second place. This is a fundamental finding showing that blood truly carries information about disease.
2. Cancers resemble one another, while sepsis and COPD stand apart
When diseases were grouped by the similarity of their blood RNA profiles, those with similar mechanisms clustered together: cancers grouped with other cancers, and neurologic diseases with other neurologic diseases. Interferon gamma and interferon alpha responses were elevated in many cancer types, suggesting that the immune system reflects the presence of a tumor in the blood as well. As expected, leukemia had the most distinct and characteristic blood signature, because in leukemia the disease itself resides in the blood cells.
Sepsis and COPD, by contrast, stood clearly apart from all other conditions. Both showed reduced gene modules associated with B, T, and NK cells and lower metabolic activity, a picture of immunosuppression and energy depletion. In rheumatoid arthritis and ankylosing spondylitis, TNF-alpha and NF-κB inflammatory pathways were markedly upregulated.
3. Signs of aging in the blood: thresholds at ages 25 and 75
The researchers defined a "blood aging signature" consisting of the 32 genes most strongly associated with age. These genes were concentrated in pathways of age-related inflammation (inflammaging) and immunosenescence, such as IL-2, TNF-alpha, mTOR, and complement, and declined steadily with age. A model built on these 32 genes estimated age in 100 independent healthy individuals with a mean error of 8.2 years; for a 33-gene aging clock developed in Korea, the error was 13.8 years. This also contributes to the discussion of chronological versus biological age.
The proportions of immune cells in the blood also changed with age: neutrophils increased, while "naive" CD4 T cells, which have not yet encountered any pathogen, gradually declined. Interestingly, the largest changes clustered around ages 25 and 75, with the period in between remaining relatively stable. Long-lived individuals over 80 had high proportions of neutrophils and plasmablasts and low proportions of naive T and B cells, consistent with the known pattern of immunosenescence.
Estimating age from blood: mean error
Years; mean difference between predicted and actual age in 100 independent healthy individuals. Shorter bar is better.
Compared using the same machine learning method and data. Explained variance (R²) was 0.31 for the WBT signature and -0.14 for the Korean clock.
4. RNA and protein tell different stories
When both RNA and protein were measured in the same individuals, the overlap between genes and proteins altered in disease was very small (Jaccard index 0 to 0.13). According to the authors, RNA reflects regulatory processes within cells and the early cellular state of disease, whereas proteins reflect end products and secreted substances. In sepsis, for example, protein signatures (IL1RL1, IL15, PTX3, CSF3) pointed to intense inflammation, while RNA signatures (PPARG, CYP1B1, OLAH) revealed the underlying immune dysregulation and metabolic stress. Using the two together therefore provides more information.
How well did AI distinguish between diseases?
The researchers selected 29 health conditions with more than 40 samples each; models were trained on 70% of the data and tested on the remaining 30%. In distinguishing each disease from healthy individuals, even the top 10 genes alone achieved high performance (AUC 0.89-1.0 on internal test data). The harder question, however, was identifying which of the 29 conditions a person had by looking at their blood. Using 10 genes per disease, this question was answered correctly 64% of the time.
Combining RNA and protein to distinguish diseases
Multiclass model distinguishing 28 health conditions; F1 score on test data (1.00 = perfect). Colored bar: combined model.
Accuracy was 55%, 67%, and 75%, respectively. The top 10 RNAs and top 10 proteins for each disease were used. Across 29 conditions, the RNA-only model had an accuracy of 64%.
The results largely held up in independent datasets. This validation is important, because AI models often perform well on the data they were trained on but poorly on data from other hospitals.
Validation in independent datasets
Mean AUC (0.50 = chance, 1.00 = perfect). Colored bars: signatures selected by the study; gray bars: comparison models built with all molecules.
RNA validation was performed in sepsis, bipolar disorder, Parkinson's disease, rheumatoid arthritis, and type 2 diabetes; protein validation in breast cancer, lung cancer, coronary artery disease, Parkinson's disease, and type 2 diabetes.
Several notable signatures also emerged for cancer: PTP4A2, associated with tumor progression and spread, in lung cancer; CXCL12, which regulates the tumor microenvironment, in gastric cancer; and HLA-DRA, reflecting an immunologically "inflamed" tumor environment, in breast cancer. Two of the five conditions in which protein signatures were tested in the UK Biobank were breast and lung cancer. However, the study does not report in detail, disease by disease, whether cancer patients were sampled at diagnosis or during treatment; these findings therefore cannot be used directly for cancer screening.
This atlas is not a diagnostic or screening test. The model gave an incorrect result in about one-third of cases across the 29 conditions, and it has not yet been tested for whether it detects disease early in people who appear healthy. Proven cancer screening methods (colonoscopy, mammography, HPV testing, and low-dose lung CT in eligible individuals) remain the standard.
What are the limitations of this study?
- Single region, hospital-based: All samples came from two hospitals in Erzurum. Although the authors describe the cohort as "unselected," participants were people presenting to the hospital; generalizability to other geographic regions and ethnic backgrounds was only partly tested through external validation.
- Unbalanced groups: Some diseases had small sample sizes. The authors used statistical methods that are robust to this, but results in small groups are less certain.
- Disease-level analysis: For example, "lung cancer" was treated as a single group; subtypes, stages, and treatment status were not examined separately.
- Confounders: Apart from age, sex, and body mass index, factors such as lifestyle and medication use were not fully accounted for.
- Cross-sectional design: Only one sample was taken from each person; whether the blood signature appears before disease onset, and thus its usefulness for early detection, is unknown.
- Limited external validation: RNA models were independently tested in 5 diseases and protein models in 5 diseases; there is no external validation for the remaining 90-plus conditions.
What could change in the future?
The authors state that the atlas will grow as part of the "Global 1M Phenome" initiative, which aims to include 1 million people. In the future, multilayered tests that examine tumor DNA, plasma proteins, and immune-cell RNA from a single blood sample may be used together for early detection of disease and for monitoring treatment response. Proteogenomic and AI-assisted approaches are advancing rapidly in this direction. Blood-based colon cancer screening and early detection panels for pancreatic cancer are the examples in this field that are closest to clinical use today.
- All samples in the study came from Erzurum, Turkey; RNA sequencing was performed at the SZA Omics laboratory, affiliated with Acıbadem Mehmet Ali Aydınlar University, in Istanbul. The author list includes physicians from dozens of clinical departments at Atatürk University.
- Disease-level results are publicly available at phenomeatlas.org/wbt. Individual-level data are not openly available because of participant consent and legal regulations, but they can be shared on request for noncommercial research.
- Whole-blood transcriptome analysis is not currently offered as a clinical test, either in Turkey or elsewhere in the world; caution is warranted if paid "disease screening" based on this type of analysis is offered.
Frequently asked questions
- Can 98 diseases be diagnosed with a single blood test?
- No, not yet. This study is a research resource that maps the RNA traces of 98 health conditions in the blood. The AI model correctly distinguished 29 conditions about two-thirds of the time and, when RNA and proteins were combined, 28 conditions about three-quarters of the time; this is not sufficient for a clinical test.
- What is the whole-blood transcriptome?
- It is the measurement of RNA produced by the roughly 20,000 genes active at a given moment in the cells of the blood, particularly immune cells. It provides a snapshot of the body's immune and metabolic response to disease.
- Can this test detect cancer early?
- That is unknown. The study found distinct signatures in the blood of patients with cancer, but it did not test whether cancer could be detected early in people who appear healthy. Proven methods such as colonoscopy, mammography, and HPV testing should be used for cancer screening.
- Why is the study important for Turkey?
- All samples were collected in Erzurum, and RNA sequencing was performed in Istanbul. An omics dataset of this scale from Turkey forming the basis of an international atlas is significant because it positions the country as a data generator in biomedical research.
Glossary
- Transcriptome
- The complete set of RNA molecules produced in a cell or tissue at a given moment.
- Proteomics
- The simultaneous measurement of a large number of proteins in blood or tissue.
- Cell-free DNA
- DNA fragments shed into the blood from cells, including tumor cells; the basis of liquid biopsy.
- AUC
- A measure of a test's ability to distinguish patients from healthy individuals; 0.5 indicates chance level and 1.0 indicates perfect discrimination.
- F1 score
- A measure that combines a model's ability to capture true positives with its ability to avoid false alarms.
- Immunosenescence
- Age-related changes in the number and function of immune cells.
DROZDOGAN Academy Commentary
Blood is like the body's "black box." This atlas reads the immune records inside that box at an unprecedented breadth, and it does so from Turkey.
⚡ Why it matters
An open resource that examines 98 health conditions with the same method, making it possible to compare shared and distinct immune and metabolic signatures across diseases. It shows in the same individuals that RNA and protein complement each other.
✅ Strengths
Large sample size, standardized sample collection, successful removal of hemoglobin RNA, matched protein data, external validation in independent cohorts and the UK Biobank, and a publicly accessible data portal.
⚠️ Limitations
Hospital-based sampling from a single region, unbalanced group sizes, no separation by disease subtype or treatment status, cross-sectional design, and an error rate of about one-third in multiclass classification.
🩺 In practice
It does not change patient management today. For physicians, it is a valuable research tool for exploring the immune footprints of diseases in the blood and generating new biomarker ideas.
❓ Open questions
Do blood RNA signatures change before disease onset? Can treatment response or recurrence be monitored with these signatures? Could a single blood test combining cell-free DNA, protein, and RNA detect cancer earlier than current screening methods?
Bottom line: The WBT Atlas is an important building block on the path toward "many diseases from a single blood sample." Reaching clinical use will require prospective, multicenter studies in healthy individuals.
- This article was prepared in October 2026.
- Data are taken from the full text of the article published as an "Article in Press" in Nature Communications; the final edited version may contain minor changes.
- Sample sizes and classification performance by cancer type are reported in the article's supplementary tables and are therefore not detailed here.
- The diagram is derived from verified statements in the text and contains no new data. The Academy commentary is our own.
References
- Li M, Tatar A, Green E, ... Turkez H, Mardinoglu A. Whole blood transcriptomics atlas across human diseases. Nat Commun. 2026 (Article in Press). doi:10.1038/s41467-026-78440-w
- WBT Atlas data portal. phenomeatlas.org/wbt
- Bueno Álvez M, et al. A human pan-disease blood atlas of the circulating proteome. Science. 2025. proteinatlas.org
- KTH Royal Institute of Technology. Mengzhen Li, doctoral thesis: Multi-disease characterization and classification through transcriptomics-based multi-omics analysis (June 2026). kth.se
- Global 1M Phenome Initiative. mphenome.org
Was this article helpful?
Your feedback helps us improve our content.
Reader comments
Share your comment
Your name and comment are published only after administrator approval. Your email address is never shown to visitors.
Be the first to comment.