Deep learning reveals cancer metastasis and therapeutic antibody targeting in whole body

Pan, C.; Schoppe, O.; Parra-Damas, A.; Cai, R.; Todorov, M. I.; Gondi, G.; von Neubeck, B.; Ghasemi, A.; Reimer, M. A.; Coronel, J.; Garvalov, B. K.; Menze, B.; Zeidler, R.; Erturk, A.

doi:10.1101/541862

bioRxiv · 10.1101/541862

Deep learning reveals cancer metastasis and therapeutic antibody targeting in whole body

Pan, C.·Schoppe, O.·Parra-Damas, A.·Cai, R.·Todorov, M. I.·Gondi, G.·von Neubeck, B.·Ghasemi, A.·Reimer, M. A.·Coronel, J.·Garvalov, B. K.·Menze, B.·Zeidler, R.·Erturk, A.

View original source ↗Export metadata ↓

Abstract

Reliable detection of disseminated tumor cells and of the biodistribution of tumor-targeting therapeutic antibodies within the entire body has long been needed to better understand and treat cancer metastasis. Here, we developed an integrated pipeline for automated quantification of cancer metastases and therapeutic antibody targeting, named DeepMACT. First, we enhanced the fluorescent signal of tumor cells more than 100-fold by applying the vDISCO method to image single cancer cells in intact transparent mice. Second, we developed deep learning algorithms for automated quantification of metastases with an accuracy matching human expert manual annotation. Deep learning-based quantifications in a model of spontaneous metastasis using human breast cancer cells allowed us to systematically analyze clinically relevant features such as size, shape, spatial distribution, and the degree to which metastases are targeted by a therapeutic monoclonal antibody in whole mice. DeepMACT can thus considerably improve the discovery of effective therapeutic strategies for metastatic cancer. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=137 SRC="FIGDIR/small/541862v1_ufig1.gif" ALT="Figure 1"> View larger version (49K): org.highwire.dtl.DTLVardef@ffe1b9org.highwire.dtl.DTLVardef@13c55eborg.highwire.dtl.DTLVardef@2ccb6corg.highwire.dtl.DTLVardef@df55e9_HPS_FORMAT_FIGEXP M_FIG C_FIG Supplementary Movies and deep learning algorithms of DeepMACT are available at http://discotechnologies.org/DeepMACT/

Explore related subjects

cancer biology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Pan, C., Schoppe, O., Parra-Damas, A., Cai, R., Todorov, M. I., Gondi, G., von Neubeck, B., Ghasemi, A., Reimer, M. A., Coronel, J., Garvalov, B. K., Menze, B., Zeidler, R., Erturk, A.. 2019-02-05. Deep learning reveals cancer metastasis and therapeutic antibody targeting in whole body. https://doi.org/10.1101/541862

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

The methionine cycle sustains the mesenchymal phenotype of tumor cells and drives PDA progression

Mesenchymal traits are associated with poor prognosis in Pancreatic Ductal Adenocarcinoma (PDA), but direct targeting of mesenchymal tumor cells remains a challenge. Metabolic profiling of tumor cells isolated from murine PDA and liver metastases (LM) revealed that LM rely more on the methionine cycle than does primary PDA. This metabolic shift was specifically associated with increased activation of the methionine cycle in mesenchymal tumor cells compared to differentiated tumor cells. We show that the targeting of the key enzyme of the methionine cycle S-adenosylhomocysteine hydrolase (SAHH) in mesenchymal cells lowers the levels of homocysteine (Hcy) and glutathione, resulting in reduced antioxidant capacity, deregulated intracellular calcium homeostasis, and ultimately loss of mesenchymal markers. Together, these events decrease cell proliferation and invasion, blunting the tumorigenic and metastatic potential of SAHH-deficient cells, eventually impeding tumor growth and progression of the metastatic disease. Clinically, the high expression of SAHH in patients with mesenchymal PDA is correlated with poor survival, positioning SAHH as a potential therapeutic target to impede the pro-metastatic, mesenchymal phenotype of PDA.

cancer biology↗

bioRxiv2026

Arrest-associated transcriptional programmes outweigh apoptosis-associated programmes in treatment-naive pancreatic ductal adenocarcinoma

Background: Pancreatic ductal adenocarcinoma (PDAC) is driven by mutant KRAS, and RAS inhibitors now extend survival in previously treated metastatic disease. Whether treatment shrinks a tumour or only halts its growth depends partly on whether cancer cells die or stop dividing, yet how untreated PDAC cells are balanced between these programmes is unknown. Aims: We aimed to characterise the balance between apoptosis-associated and arrest-associated transcriptional programmes in treatment-naive PDAC and to examine their relationships with tumour subtype, TGM2 expression, tissue organisation, survival and drug sensitivity. Methods and results: We scored 52,999 malignant cells from 18 untreated patients in the Hwang atlas using a transcriptional contrast between genes associated with apoptosis and genes associated with resistance to apoptosis or cell-cycle arrest. In every patient the contrast was shifted toward the arrest-associated side: using equality between the two sides as the dividing threshold, 86.3% of cells with an informative score fell on that side, although the proportion depends on the threshold chosen. The shift recurred in the independent Steele atlas (median 71.6% per sample, range 37.1% to 90.8%). In the discovery cohort, tumours with higher basal-like scores had fewer arrest-side cells, and TGM2 expression was modestly higher on the arrest-associated side in both cohorts. The contrast did not track cell-line drug sensitivity or survival in bulk tumours. Conclusion: Untreated PDAC cells show a reproducible transcriptional bias toward anti-apoptotic and cell-cycle-arrest programmes. Whether this bias predicts cellular responses to treatment remains to be tested.

cancer biology↗

bioRxiv2026

Quantitative live imaging reveals region-dependent medulloblastoma disseminationin the zebrafish brain

Medulloblastoma is the most common malignant pediatric brain tumor, and relapse is frequently driven by dissemination within the central nervous system. Yet the early dynamics of tumor spread remain difficult to resolve in vivo. Here, we orthotopically transplanted human medulloblastoma cells into larval zebrafish and followed individual xenografts by longitudinal high-resolution confocal imaging. ONS-76 showed particularly robust dissemination, with hindbrain tumors spreading into the developing cerebellum in 81.5% of larvae, more than fourfold the frequency observed for forebrain or midbrain tumors. Disseminating cells extended along brain-surface and leptomeningeal-like regions and frequently associated with cerebral vasculature. Quantitative analysis of tumor area, perimeter, compactness, convex hull area, and fractal dimension captured changes in growth and spatial complexity over time. Dasatinib reduced tumor expansion and dissemination-related features in a concentration-dependent manner, while higher doses revealed host toxicity. These findings establish a powerful in vivo model for resolving medulloblastoma dissemination and therapeutic response.

cancer biology↗