Cancer is not one enemy, but many
Inside every tumor lies not one disease but many — a mosaic of cell populations, each carrying its own evolutionary history. A team at the University of Texas has built a computational tool called TUSCAN that reads the chromosomal scars common to nearly all cancers, overlaying genetic information onto tissue images to reveal where tumor cells live and how they have diverged over time. Rather than chasing unreliable molecular markers that shift from patient to patient, TUSCAN grounds its analysis in copy number variations — large-scale DNA duplications and deletions that mark the landscape of malignancy with unusual consistency. The work is a step toward treating cancer not as a single adversary, but as the fractured, evolving population it truly is.
- Traditional gene-marker strategies kept failing researchers because no single molecular flag reliably identifies cancer across different patients or tumor types, leaving detection strategies that worked in one case collapse in the next.
- TUSCAN sidesteps this instability by targeting copy number variations — chromosomal gains and losses present in the vast majority of solid tumors — giving the tool a near-universal foothold that marker-based methods lack.
- Tested against five competing tools across six cancer datasets, TUSCAN outperformed on nearly every measure, and was the only method to correctly separate distinct tumor foci in a breast cancer sample rather than merging them into a single undifferentiated mass.
- The tool went further than detection: in one breast cancer specimen it reconstructed six distinct tumor subclones, traced their evolutionary lineage, and identified an aggressive clone that had silenced immune defenses and gained the ability to invade surrounding tissue.
- Limitations remain — single-cell resolution is beyond its reach, blood cancers with few CNVs fall outside its scope — but for the solid tumors responsible for most cancer deaths, TUSCAN is now freely available to the research community.
Cancer is not a single disease confined to a single location. Inside any tumor lies a landscape of different cell populations, each with its own genetic signature, each responding differently to treatment. For decades, researchers have struggled to map where cancerous cells actually reside within a tissue sample — and to read the evolutionary story written in their genomes.
Traditional approaches hunted for specific gene markers, but these proved unreliable. They shift from patient to patient and tumor type to tumor type. In aggressive cancers like triple-negative breast cancer, the genetic variability is so extreme that no single marker works consistently. A team at the University of Texas took a different path, building a tool called TUSCAN — TUmor Segmentation and Classification ANalysis in spatial transcriptomics — that instead targets copy number variations, the large-scale DNA duplications and deletions carried by nearly all tumor cells. TUSCAN reconstructs a tissue's copy number profile from gene expression data and overlays it onto standard microscopic images to pinpoint exactly where cancer lives.
The method works in three stages: identifying normal tissue as a baseline using gene expression patterns and microscopic color characteristics; reconstructing the chromosomal gain-and-loss landscape for every spot in the sample by comparing it to that baseline; and clustering the tissue into tumor and non-tumor regions accordingly. Tested against five existing tools across six datasets spanning breast, prostate, pancreatic, and ovarian cancers, TUSCAN achieved the highest accuracy on nearly every benchmark. It was the only method to exceed a key performance threshold on a prostate cancer sample, and it successfully separated distinct tumor foci in a breast cancer specimen where a leading competitor merged them into one.
Beyond detection, TUSCAN revealed something deeper. In a detailed breast cancer case study, the researchers used its copy number profiles to reconstruct the tumor's evolutionary history — identifying six distinct subclones, each occupying a specific region of tissue. One clone showed hallmarks of aggressive behavior: elevated expression of genes enabling invasion and suppressed immune signaling suggesting it had evolved to evade the body's defenses. Another, occupying a region of early-stage carcinoma, appeared to be an ancestral population from which the more aggressive clones had diverged. A phylogenetic tree of these clones let the researchers trace the tumor's developmental trajectory and the chromosomal events that drove it.
Limitations exist: TUSCAN cannot achieve single-cell resolution, cannot be applied to cancers where copy number variations are rare, and requires full transcriptome data. But for the solid tumors that account for the vast majority of cancer deaths, it offers a more universal and reliable way to see where disease lives and how it has evolved. The code is freely available. The harder work — using these maps of tumor diversity to design treatments that account for cancer's fractured, many-faced nature — now begins.
Cancer is not a single disease confined to a single location. Inside any tumor lies a landscape of different cell populations, each with its own genetic signature, each responding differently to treatment. For decades, researchers have struggled with a fundamental problem: how to map where the cancerous cells actually are within a tissue sample, and how to understand the evolutionary story written in their genomes.
Traditional approaches relied on hunting for specific gene markers—molecular flags that supposedly identify cancer cells. The trouble is that these markers are unreliable. They shift from patient to patient, from tumor type to tumor type. In highly aggressive cancers like triple-negative breast cancer or melanoma, the genetic variability is so extreme that no single marker works consistently. Researchers would spend months designing a detection strategy only to find it failed when applied to a different patient's tissue.
A team led by researchers at the University of Texas and supported by the National Institutes of Health took a different approach. Instead of chasing elusive markers, they looked for something more fundamental: copy number variations, or CNVs. These are large-scale duplications and deletions of DNA segments that appear in the vast majority of cancers. Nearly all tumor cells carry these chromosomal scars. The researchers built a computational tool called TUSCAN—TUmor Segmentation and Classification ANalysis in spatial transcriptomics—that reconstructs a tumor's copy number profile from gene expression data and then overlays that information onto standard tissue images to pinpoint exactly where cancer cells live.
The method works in three steps. First, TUSCAN examines a tissue sample's gene expression patterns and its microscopic appearance to identify spots of normal, healthy tissue that can serve as a baseline. It does this by combining two signals: gene expression variance and the color characteristics of the tissue under a microscope. Lighter-staining regions and areas with consistent gene expression patterns tend to be normal tissue. Second, the tool reconstructs the copy number landscape for every spot in the sample by comparing gene expression levels to that normal baseline, smoothing the signal across adjacent genes to reveal the large-scale chromosomal gains and losses that define cancer. Third, it uses clustering algorithms to partition the tissue into tumor and non-tumor regions based on the copy number patterns it has inferred.
The researchers tested TUSCAN against five existing methods on six different datasets spanning breast, prostate, pancreatic, and ovarian cancers. The results were decisive. TUSCAN achieved the highest accuracy scores across nearly every dataset. On a prostate cancer sample, TUSCAN was the only method to exceed a performance threshold that other tools failed to reach. It also demonstrated a striking ability to distinguish tumor tissue from necrotic (dead) tissue—a distinction that stumped some competing approaches. When applied to a breast cancer sample with multiple tumor foci, TUSCAN successfully separated and outlined each distinct region, while a leading competitor merged them into a single blob.
Beyond simple tumor detection, TUSCAN revealed something deeper. In a detailed case study of a breast cancer specimen, the researchers used the tool's copy number profiles to reconstruct the tumor's evolutionary history. They identified six distinct tumor subclones, each occupying a specific region of the tissue. One clone showed hallmarks of aggressive behavior: enriched expression of genes involved in epithelial-mesenchymal transition, a process that grants cancer cells the ability to migrate and invade. This same clone showed suppressed immune signaling, suggesting it had evolved mechanisms to evade the body's defenses. Another clone, which occupied the region of carcinoma in situ—early-stage cancer confined to the duct—appeared to be an ancestral population from which the more aggressive clones had diverged. By reconstructing a phylogenetic tree of these clones, the researchers could trace the tumor's developmental trajectory and identify the specific chromosomal events that drove its evolution.
The tool is not without limitations. Because spatial transcriptomics measures gene expression in small tissue spots that contain multiple cells, TUSCAN cannot achieve single-cell resolution. The tumor regions it identifies may contain some normal cells mixed in. It also cannot be applied to cancers where copy number variations are rare, such as certain blood cancers. And it requires access to the full transcriptome, making it incompatible with imaging-based methods that measure only hundreds of genes. Still, for the cancers where it applies—the solid tumors that account for the vast majority of cancer deaths—TUSCAN offers a more universal, more reliable way to see where disease lives and how it has evolved. The code is freely available, and the researchers have made their method accessible to the broader research community. What comes next is the harder work: using these maps of tumor diversity to design treatments that account for the fact that cancer is not one enemy, but many.
Citações Notáveis
The cluster with the lightest color on the histology image is most likely to represent normal tissue, because tumor cells exhibit larger, more intensely stained nuclei due to increased nuclear-cytoplasmic ratio— TUSCAN methodology
Clone 1 exhibits significant enrichment in genes associated with epithelial-mesenchymal transition, coagulation, and UV response, suggesting enhanced migratory and invasive capabilities and increased metastatic potential— Breast cancer case study findings
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that you're using copy number variations instead of gene markers? Aren't markers more direct?
Markers are direct, but they're also fragile. A marker gene that works perfectly in one patient's breast cancer might be silent in another's. With CNVs, you're reading something more fundamental—the actual chromosomal architecture that cancer cells have rewritten. It's universal across cancer types because almost all tumors accumulate these large-scale DNA changes.
So you're trading specificity for robustness?
Exactly. You lose some precision at the single-cell level, but you gain the ability to work across different cancers and different patients without redesigning your approach each time. It's a pragmatic trade.
The case study showed six distinct clones in one tissue sample. How does knowing that help a clinician treat the patient?
It tells you that the tumor isn't monolithic. One clone might be sensitive to a drug that another clone resists. If you only sequence the bulk tumor, you miss that heterogeneity. With spatial mapping, you can see which regions are most aggressive, which have immune evasion signatures, which might be vulnerable to specific therapies.
But you said the method can't achieve single-cell resolution. Doesn't that limit what you can actually do with the information?
It does. You're seeing populations of cells, not individuals. But that's actually closer to how tumors behave clinically—as regional ecosystems, not isolated cells. And it's a practical trade-off. Single-cell methods destroy spatial information entirely. This preserves it.
What happens when you apply this to a cancer type where CNVs are rare?
The method fails. It's honest about that limitation. For blood cancers like acute myeloid leukemia, where copy number changes are uncommon, you'd need a different approach. TUSCAN isn't universal—it's universal within the domain where CNVs are the dominant driver of cancer biology.