Comprehensive genomic profiling: the goal and the gap. Precision oncology depends on identifying the specific genetic alterations driving each patient's cancer so that targeted treatments can be matched to them. Comprehensive genomic profiling (CGP) tests, which analyze hundreds of cancer-relevant genes simultaneously, aim to maximize this identification. However, most current CGP tests rely exclusively on DNA from tumor tissue - a single data type that misses important biological information.
What DNA alone misses. Three specific limitations stand out. First, standard DNA panels have insufficient probe coverage to reliably detect copy number variations (deletions and amplifications of chromosomal regions). Second, DNA-based tests miss up to 20% of gene fusions - a particularly important class of cancer driver events - that are best detected by analyzing messenger RNA (mRNA). Third, tissue biopsies capture only a static snapshot of the tumor at a single moment, missing how the cancer evolves during treatment.
Liquid biopsy: monitoring cancer in real time. Tumors shed DNA fragments into the bloodstream. This circulating tumor DNA (ctDNA) can be detected and measured from a simple blood draw - a "liquid biopsy" - providing a minimally invasive window into tumor burden, treatment response, and emerging resistance mutations. Currently, liquid biopsy and tissue-based CGP are typically performed as separate, expensive tests.
An integrated solution from Vietnam. Researchers at Gene Solutions in Vietnam developed and validated an approach that simultaneously extracts and sequences both DNA and mRNA from tumor tissue while also monitoring ctDNA from blood. This paper reports the analytical performance and early clinical evidence for this integrated strategy.
Component 1: High-density probe DNA sequencing. DNA extracted from formalin-fixed paraffin-embedded (FFPE) tumor tissue and matched white blood cells is sequenced using a 504-gene panel. The key innovation here is the high-density probe (HDP) design, which concentrates more DNA capture probes in genomic regions prone to copy number changes, improving the sensitivity to detect amplifications, deletions, and large structural rearrangements.
Component 2: Whole transcriptome mRNA sequencing. Total RNA is simultaneously extracted from the same tissue sample and sequenced across all 19,435 protein-coding genes. This broad coverage enables detection of gene fusions that DNA sequencing misses (including those with complex or cryptic breakpoints), captures MET exon 14 skipping mutations, and generates gene expression profiles used for tissue-of-origin prediction.
Component 3: Hybrid liquid biopsy ctDNA monitoring. Blood plasma samples are analyzed using two complementary approaches. The first uses a panel of 700 lung cancer-specific hotspot mutations to find tumor-agnostic alterations (including emerging resistance mutations). The second uses ultra-deep sequencing to track 5-8 personalized mutations identified from the patient's own FFPE profile, providing individualized tumor burden monitoring across multiple blood draws over time.
Clinical validation cohorts. The study used 604 archived tissue samples of 12 cancer types for the DNA and mRNA components. The liquid biopsy component was evaluated in a separate cohort of 55 stage-IV lung cancer patients with 147 plasma samples collected longitudinally before and after treatment initiation with targeted therapies or immunotherapy.
Chromosome-level structural variants: a clear difference. In two glioma tissue samples, the high-density probe (HDP) panel identified chromosome 1p/19q co-deletion and chromosome 10 loss that the standard probe design (STD) failed to detect. These chromosome-level changes are clinically important - the 1p/19q co-deletion defines a specific glioma subtype with a favorable prognosis and specific treatment options.
Gene-level copy number accuracy is dramatically improved. When quantifying how many copies of a specific gene are present, the HDP panel achieved a copy number estimation error of only 0.1-0.4 copies, compared to errors of up to 1.5 copies with the standard panel. For gene deletions, the HDP panel detected both heterozygous and homozygous co-deletion of MTAP-CDKN2A in reference samples, while the standard panel missed both.
Better detection of large genomic rearrangements in BRCA genes. Large insertions or deletions spanning one or more exons (LGRs) are particularly hard to detect with low probe coverage. For BRCA1/2 genes - critical for hereditary cancer risk and for predicting response to PARP inhibitors - the HDP panel achieved 100% sensitivity versus 81.8% for the standard panel. In a lung cancer cohort of 49 samples, 12.2% had homozygous loss of MTAP-CDKN2A, matching published prevalence estimates.
HRD status in ovarian cancer. By combining BRCA mutation detection with a whole-genome instability score from shallow whole-genome sequencing, the assay determined homologous recombination deficiency (HRD) status in 169 ovarian cancer samples. 55% were HRD-positive, consistent with the published literature - validating that the assay can reliably identify patients who would benefit from PARP inhibitor therapy.
20% more fusions detected by mRNA. In reference samples with known fusion alterations, mRNA sequencing achieved 100% sensitivity, while DNA sequencing captured only 80%. Some fusions were not covered by the targeted DNA probe design, and others arose from complex genomic rearrangements that create fusion transcripts without leaving a signature at the DNA level that probe-based sequencing can reliably capture.
The FFPE RNA quality problem and its solution. A significant practical challenge emerged: 35% of real-world clinical FFPE samples had poor mRNA quality (DV200 below 30%), often failing quality control entirely. This is a common problem in resource-limited settings where tissue storage conditions may not be optimal. The solution is to combine both methods - when mRNA quality is insufficient, DNA sequencing fills the gap, and when DNA probes miss a fusion, mRNA catches it.
MET exon 14 skipping: mRNA wins again. MET exon 14 skipping mutations are important therapeutic targets in lung cancer, and they are notoriously difficult to detect by DNA sequencing alone. The mRNA sequencing component detected these variants with over 99% sensitivity, confirming that the transcriptome approach adds meaningful clinical value for this specific class of actionable alteration.
Novel fusion variants as a bonus. Whole transcriptome mRNA sequencing also identified novel fusion variants not anticipated by targeted DNA panels, providing additional research value and the potential to identify patients who might benefit from emerging targeted therapies targeting novel fusion partners.
The problem of cancer of unknown primary. Approximately 5% of metastatic cancer cases are classified as "cancer of unknown primary" (CUP) - the original organ where the cancer started cannot be determined by standard diagnostic methods. CUP patients have poor outcomes in part because they cannot receive site-specific treatments. Predicting the tumor's tissue of origin (TOO) from its gene expression pattern could unlock targeted treatment options.
OriCUP: an ensemble machine learning model trained on 9,889 tumors. The researchers trained their tissue-of-origin prediction model, OriCUP, on gene expression profiles from The Cancer Genome Atlas covering 32 cancer types. Three gene sets were tested (90 genes, 2,000 genes, and 739 genes), and an ensemble of multiple machine learning algorithms was selected for each set based on 10-fold cross-validation performance. The 2,000-gene set with a Liblinear Support Vector Classifier performed best.
Validated performance: 87.7% for primary, 81.4% for metastatic tumors. On an independent validation dataset of 731 samples - combining public data and the lab's own cases, including both primary and metastatic tumors - OriCUP achieved 87.7% accuracy for primary tumors and 81.4% for metastatic tumors. Metastatic tumor prediction is inherently harder because gene expression patterns change as tumors spread and evolve.
Outperforming a competing model. OriCUP's 81.4% accuracy for metastatic tumors was approximately 10 percentage points higher than the CUP-AI-Dx model (71.2%) tested on the same validation set. The improvement was most pronounced for lung and gastrointestinal cancers - two common cancer types where precise tissue-of-origin assignment has the most treatment implications.
Finding more actionable mutations through blood testing. In 55 stage-IV lung cancer patients, FFPE profiling identified 69 actionable mutations (88.5% of all found). The liquid biopsy hotspot panel added 9 additional tumor-agnostic mutations (11.5%) not found in tissue - including EGFR T790M resistance mutations. This means blood testing identified clinically important targets that tissue biopsy alone missed, likely because resistance mutations can emerge in metastatic deposits not sampled during biopsy.
Personalized ctDNA tracking: most patients show detectable ctDNA at baseline. From 227 tumor-derived mutations selected for personalized tracking across 55 patients, 114 mutations (50.2%) were detectable in plasma at baseline in 42 patients (76.4%). Adding tumor-agnostic hotspot mutations raised ctDNA detectability to 87.3% of patients - meaning the vast majority of patients could be monitored using blood draws rather than repeat biopsies.
ctDNA decrease predicts survival with remarkable accuracy. In 40 patients with serial blood samples before and after treatment, those whose ctDNA decreased by more than 50% from baseline ("molecular responders") had dramatically better outcomes than non-responders. The 12-month progression-free survival was 95.5% for molecular responders versus 31.7% for non-responders - a hazard ratio of 9.42 (p less than 0.0001). Among molecular responders, 84% achieved clinical complete or partial response; among molecular non-responders, 93% had confirmed progressive disease.
Ultra-sensitive detection at 0.01% tumor fraction. By combining mutation-based and non-mutation-based signals (copy number changes and DNA fragment length alterations), the assay achieved a limit of detection of 0.01% tumor fraction in circulating DNA - equivalent to detecting one cancer cell's worth of DNA among 10,000 normal cells' worth. This sensitivity is critical for early detection of residual disease before it becomes clinically apparent.
Three complementary data types, one integrated test. This study demonstrates that combining high-density probe DNA sequencing, whole transcriptome mRNA sequencing, and hybrid liquid biopsy ctDNA monitoring provides clinically actionable information beyond any single method. The integrated approach detects copy number variations with greater precision, captures fusions missed by DNA alone, predicts tumor origin for CUP patients, and monitors treatment response in real time through blood testing.
Addressing barriers to access in Vietnam and similar settings. The study is notable for being conducted in Vietnam, where FFPE sample quality challenges and limited sequencing infrastructure represent real-world constraints. The finding that combining DNA and mRNA compensates for frequent RNA quality failures is directly relevant to clinical oncology in resource-limited healthcare systems where such sample quality issues are common.
Important limitations acknowledged. The study has not yet demonstrated clinical utility in terms of improved patient outcomes through prospective trials - it validates technical performance and shows correlations with clinical response. Sample sizes, particularly for the ctDNA cohort (n=40 with serial testing), are limited. The high cost of the integrated approach remains a barrier to widespread adoption, though the parallel testing strategy is more efficient than running separate tests.
The path forward. Future randomized clinical trials will need to test whether treatment decisions guided by this integrated profiling approach actually improve patient survival compared to standard care. Reducing sequencing costs and expanding access to bioinformatics infrastructure will also be essential to make this approach viable at scale, particularly in the developing-country healthcare systems where such innovations could have the greatest impact on unmet clinical need.