Confronting tumor heterogeneity directly, rather than averaging past it, has become one of oncology drug development’s central problems, and the evidence for why is old enough to predate the spatial biology toolkit now used to address it. A landmark study of kidney cancer found that most somatic mutations in a tumor were not detectable in every region sampled from that same tumor, which means a single biopsy is, structurally, a survey of one neighborhood rather than the whole city.
Key takeaways
|
Why heterogeneity defeats drugs
A drug can be genuinely effective against the cells it was designed to hit and still fail in the clinic, if a tumor also contains a substantial population of cells the drug does not affect. Intratumor heterogeneity, meaning the coexistence of multiple malignant subpopulations with different drug sensitivities within the same tumor, is a documented and specific mechanism of treatment failure rather than a general caveat about cancer being complicated.
The evidence for this is not new, and it did not originate from spatial biology. A landmark 2012 multiregion sequencing study performed exome sequencing, chromosome aberration analysis, and ploidy profiling on multiple spatially separated samples from primary kidney tumors and their metastatic sites. Phylogenetic reconstruction revealed branched evolutionary growth, with 63% to 69% of all somatic mutations not detectable across every sampled tumor region. The authors noted directly that this heterogeneity may explain difficulties validating oncology biomarkers because of sampling bias, may contribute to Darwinian selection of preexisting drug-resistant clones, and may predict therapeutic resistance outright.
That finding matters for a specific reason: it means the sampling problem exists independent of any measurement technology’s limitations. Even a perfect sequencing method applied to one biopsy is still only describing one region of a tumor that, on the evidence above, may have evolved along different branches in different places. Spatial biology does not create this problem. It is the first set of tools built to actually see it, at scale, without requiring a separate biopsy from every suspected region.
Spatial vs. bulk views of heterogeneity
Three approaches to measuring a tumor answer three different questions, and conflating them is a common source of confusion in how heterogeneity gets discussed.
Approach | What it reports | What it cannot tell you |
Bulk sequencing of a homogenized sample | A single average value across whatever mixture of regions and cell types was included in the sample | Whether that average reflects a uniform tumor or a mixture of very different subpopulations |
Single biopsy, sequenced in depth | Detailed molecular profile of one specific tumor region | Whether other regions of the same tumor resemble that one or diverge from it |
Spatially resolved profiling | Where in the tissue specific mutations, expression patterns or cell states are located | Full whole-genome depth at every point, since spatial platforms generally trade some depth or coverage for spatial resolution |
Table 1. Three measurement approaches and the specific question each is built to answer. A drug development decision that assumes one approach answers all three questions is a decision built on an assumption the data does not support.
The practical consequence is that bulk and single-biopsy data are not simply lower-resolution versions of spatial data; they answer a genuinely different question and can actively mislead about heterogeneity specifically, since both methods are structurally unable to report whether the value they measured is representative of the whole tumor or specific to the sample taken.
Intratumoral spatial architecture
Spatial platforms have moved the heterogeneity question from whether it exists, which the 2012 multiregion study already established, to specifically where it sits and what it does mechanistically. Two recent examples illustrate the shift.
A 2025 review of cancer therapy resistance from a spatial omics perspective describes a re-analysis of colorectal cancer samples, profiled on a widely used sequencing-based spatial transcriptomics platform, from patients treated with oxaliplatin-based chemotherapy. The re-analysis found that cancer-associated fibroblasts expressing a specific marker were spatially proximate to oxaliplatin-resistant malignant cells and interacted with them directly through collagen, implicating that specific spatial relationship in mediating chemoresistance. That is a considerably more actionable finding than "resistant cells exist somewhere in this tumor": it names which stromal cell type sits next to the resistant cells and by what mechanism they appear to interact.
When architecture, not just genetics, drives resistance The same review describes chromothriptic medulloblastomas, a pediatric brain tumor subtype marked by a specific pattern of catastrophic chromosome shattering, exhibiting pronounced spatial intratumor heterogeneity. These tumors showed increased proliferative and stemness signatures alongside reduced immune infiltration, and the analysis integrated fluorescence in situ hybridization validation with matched single-cell DNA and RNA sequencing data on adjacent sections. The authors state that distinct spatial clonal architectures can drive therapy resistance and relapse. Read together, these two examples make the same point from different angles. Heterogeneity is not only a matter of which mutations are present somewhere in a tumor; it is a matter of which cells sit next to which other cells, and that positional relationship itself can be the mechanism driving resistance, not merely a correlate of it. |
Implications for target and dose
Heterogeneity has direct, practical consequences for two decisions a drug development program has to make regardless of tumor type: which target to pursue, and what dose or regimen to design around.
A 2021 study of neuroblastoma, a childhood tumor responsible for a substantial share of pediatric cancer deaths, combined transcriptomic and genomic profiling with ultra-deep targeted sequencing of multiregional biopsies from 10 patients. The researchers found high spatial and temporal heterogeneity in somatic mutations and copy-number alterations, reflected at the transcriptomic level, and specifically noted that mutations in druggable target genes, including ALK and FGFR1, were heterogeneous at diagnosis, at relapse, or both. The authors state this directly raises the issue of whether current target prioritization and molecular risk stratification procedures, based on single biopsies, are sufficiently reliable for therapy decisions.
That is not an abstract heterogeneity concern. ALK and FGFR1 are both genes with approved, targeted therapies in oncology, meaning the heterogeneity described here is heterogeneity in exactly the genes a clinician might use to select a specific drug. If a single diagnostic biopsy does not reliably represent a tumor’s ALK or FGFR1 status throughout, a target selection decision made from that biopsy carries a specific, named risk of missing resistant or divergent clones elsewhere in the same tumor.
Dosing carries a related but distinct implication. A dose calibrated to be effective against the dominant, most heavily sampled subpopulation may be systematically under-dosed against a minority subpopulation the sampling missed, particularly if that minority population is the one carrying a resistance-conferring alteration. Spatial data does not solve dosing by itself, but it does make visible which subpopulations exist and roughly how much of the tumor they occupy, information a single-biopsy workflow structurally cannot provide.
Designing around heterogeneity
Treating heterogeneity as a property to characterize before a therapy is chosen, rather than an explanation discovered after a treatment fails, changes several design decisions.
- Sample more than one region where feasible. Multiregion sampling, even without full spatial profiling, directly addresses the sampling-bias problem the 2012 kidney cancer study established, and remains informative even where a full spatial platform is not available.
- Ask whether a candidate target is heterogeneous, not only whether it is present. The neuroblastoma findings above argue for checking a target’s spatial and temporal consistency specifically, rather than confirming its presence in a single diagnostic sample and treating that as sufficient.
- Look for spatial relationships that could mediate resistance, not only resistant cell populations themselves. The oxaliplatin-CAF example shows that a resistance mechanism can depend on which cells sit adjacent to which others, a question spatial data is specifically built to answer and bulk or single-region data cannot.
- Treat a combination or sequencing strategy as a response to known heterogeneity, not only to anticipated resistance. If spatial or multiregion data show a tumor already contains a resistant subpopulation before treatment begins, that is different information than resistance emerging under treatment pressure, and it argues for a different upfront strategy.
The tumor microenvironment itself, meaning the immune, stromal, and vascular context surrounding these heterogeneous tumor cell populations, is the subject of its own dedicated guide within this cluster, since interactions with that surrounding environment are a major additional driver of the resistance patterns described above. That guide, Spatial Biology in Oncology: Decoding the Tumor Microenvironment, develops the microenvironment side of this argument in full.
For the broader tissue-context argument this spoke builds on, including the clinical-success evidence for spatially informed target discovery, see Spatial biology for target discovery: Mapping disease at tissue resolution, and for where disease heterogeneity sits within the full spatial biology pipeline, Spatial biology in drug discovery: From target discovery to translational medicine.
This article was produced under Drug Discovery News’s editorial policies.
