Researchers at MUSC Hollings Cancer Center and international collaborators have developed a blood-based screening platform that could help shorten this diagnostic bottleneck. Known as Access to Diagnosis using Liquid Biopsy, or ADLiB, the system combines several genetic and biological signals from a blood sample with machine learning. Its purpose is not to replace tissue biopsy, which remains essential for confirming lymphoma, but to identify patients who should be prioritized for urgent biopsy and specialist evaluation.

The study, published in HemaSphere, focuses on one of the most difficult settings for lymphoma diagnosis: regions where tuberculosis is common. Tuberculosis can cause enlarged lymph nodes, fever, fatigue and weight loss, closely resembling lymphoma. Katherine Antel, M.D., Ph.D., a Hollings physician-scientist and the study’s lead author, previously practiced and conducted research in South Africa. She said that the delays she observed there were surprisingly similar to challenges faced by patients in South Carolina, particularly those living far from major medical centers or specialized pathology services.

“Lymphoma is a global problem,” Antel said. “When I came to South Carolina, I was surprised to see many of the same challenges here that I see in other parts of the world: late diagnosis and limited access to specialized pathology expertise.” The problem can begin when patients undergo fine-needle aspiration, a minimally invasive procedure that removes individual cells rather than a larger piece of tissue. Although useful for many conditions, this approach may not provide enough architectural information to distinguish lymphoma from benign disease or to classify its subtype accurately.

A false-negative result can be particularly dangerous. Patients may be told that no cancer was found, only to return later with more advanced illness. “To diagnose lymphoma, you really need tissue,” Antel said. “A fine-needle aspiration can come back falsely negative. Patients are reassured, and then the diagnosis is missed until they become much sicker.” Surgical biopsy remains the standard because lymphoma diagnosis depends not only on identifying abnormal cells, but also on understanding how those cells are organized and which molecular features they carry.

ADLiB takes a different approach by searching the bloodstream for fragments of cell-free DNA. These tiny pieces of genetic material are released when cells die, and most originate from normal tissues. A small fraction may come from tumor cells, creating what Antel described as “a needle in a haystack.” The platform measures the amount of circulating tumor DNA and looks for lymphoma-associated mutations and chromosome alterations. It also examines immune-cell receptor patterns, which can reflect abnormal immune-cell populations, while screening for infectious pathogens that may mimic lymphoma, including tuberculosis.

Machine learning enables the test to interpret these signals together rather than relying on a single genetic marker. That distinction is important because lymphoma is not one disease but a broad group of cancers with diverse biological features. A mutation that appears in one lymphoma subtype may be absent in another, while the quantity of tumor DNA in blood can vary depending on tumor location and burden. By combining multiple weak or incomplete clues, the algorithm generates an overall estimate of how likely a patient is to have lymphoma.

The researchers first tested whether the platform could detect lymphoma-related genetic changes in laboratory materials and patient samples with known diagnoses. They then evaluated ADLiB in 124 adults in Cape Town who had enlarged lymph nodes. Approximately three-quarters were ultimately diagnosed with lymphoma, while the remaining participants had tuberculosis, benign conditions or metastatic solid tumors. About one-third were living with HIV, a group with a substantially elevated risk of lymphoma. Patients with lymphoma generally had higher concentrations of circulating tumor DNA and more cancer-associated mutations than participants whose lymph-node enlargement had other causes.

When the machine-learning system combined the available genetic and biological information, it correctly identified patients with lymphoma 95% of the time and distinguished them from those with noncancerous causes of enlarged lymph nodes with 92% accuracy. The findings suggest that a relatively simple blood draw could help clinicians decide which patients need to move rapidly through the diagnostic pathway. This could be especially valuable when lymph nodes are located deep in the chest or abdomen, where obtaining tissue is more complicated, or when a patient lives in an area without immediate access to a lymphoma specialist.

Antel emphasized that ADLiB is currently a triage tool, not a stand-alone cancer diagnosis. A positive result would signal the need for expedited tissue biopsy and further testing, while a negative result would not automatically rule out lymphoma. Future studies will refine the algorithm, assess its ability to distinguish lymphoma subtypes and test it in larger and more diverse populations. If those investigations confirm its performance, ADLiB could help connect high-risk patients with the right specialists sooner, reducing diagnostic delays in both resource-limited regions and wealthier health systems where rural access and pathology capacity remain uneven.

Subject of Research: People

Article Title: Access to diagnosis using liquid biopsy (ADLiB): Identifying lymphoma in a tuberculosis-endemic setting

News Publication Date: 29-May-2026

Web References: https://hollingscancercenter.musc.edu/ ; https://onlinelibrary.wiley.com/doi/10.1002/hem3.70394

References: HemaSphere. DOI: 10.1002/hem3.70394

Image Credits: Medical University of South Carolina; photo by Clif Rhodes

Keywords: Lymphoma, liquid biopsy, machine learning, circulating tumor DNA, cancer diagnosis, tuberculosis, cancer genetics, oncology, HIV, precision medicine

Traditionally, oncologists have relied heavily on tissue biopsies to study tumors, capturing static snapshots of cancer cells alongside their immediate healthy neighbors. These biopsies provide vital genetic information, but the invasive nature and temporal limitations restrict frequent monitoring. Moreover, tumor microenvironments are dynamic, continually evolving during disease progression and in response to treatments. The inability to observe these changes in real-time has hindered precision medicine efforts, often leaving clinicians to guess how tumors might adapt or resist therapies.

The Stanford and Mayo Clinic teams tackled this challenge by integrating cutting-edge machine learning algorithms with detailed molecular profiling. Central to their approach are two computational tools—CytoSPACE and Spatial EcoTyper—which piece together spatial maps of cellular neighborhoods within tumors based on gene expression patterns. CytoSPACE, developed just last year, accurately assigns individual cells to their precise coordinates within tumor tissue, akin to plotting every resident on a detailed city map. Spatial EcoTyper complements this by characterizing cell types, states, and interactions, revealing distinct spatial ecotypes—cellular communities with unique genetic and functional signatures.

By applying these methods to over 100 tumor samples spanning ten diverse cancer types, the researchers identified nine conserved spatial ecotypes within tumors, regardless of cancer origin. These cellular neighborhoods varied in location and composition, some residing at interfaces between tumor and healthy tissue, others deep within malignant masses. Crucially, certain ecotypes correlated strongly with patients’ responses to immunotherapy, suggesting these spatial patterns could serve as biomarkers guiding personalized treatment choices.

A critical insight emerged when the team observed that individual immune cells like CD8 T cells displayed context-dependent gene expression profiles influenced by their surrounding cellular “neighbors.” This discovery parallels human social behavior—distinct conversations and interactions occurring in different social settings despite the presence of common participants. In the tumor microenvironment, these spatially dependent gene expression programs define the functional states of cells, potentially tipping the balance between tumor suppression and progression.

Recognizing the clinical constraints of repeated biopsies, the scientists sought a less invasive method to capture these spatial relationships. Their solution lies in analyzing methylation patterns—chemical modifications on DNA that reflect gene expression activities—in circulating cell-free DNA found in patients’ blood. Tumors and surrounding cells release fragments of their DNA into the bloodstream upon cell death, carrying methylation “signatures” indicative of their origins and current states. Leveraging artificial intelligence, the team developed Liquid EcoTyper, a tool capable of reconstructing the tumor microenvironment from these methylation signals, effectively translating blood samples into dynamic maps of spatial ecotypes.

Validation studies demonstrated strong concordance between spatial ecotype profiles derived from tumor biopsies and those inferred via Liquid EcoTyper from patient blood samples. This breakthrough paves the way for longitudinal monitoring of tumor ecosystems through simple blood draws, enabling clinicians to track the evolving interactions within the tumor microenvironment as patients undergo therapy. Such real-time insights could alert physicians to emerging resistance mechanisms or shifts in immune landscapes, prompting timely adjustments to treatment regimens.

Beyond its diagnostic potential, this research challenges the tumor-centric paradigm that has dominated cancer biology. By highlighting the pivotal roles played by surrounding healthy cells—including fibroblasts, immune modulators, and vascular elements—in shaping tumor behavior, the study underscores the importance of targeting the tumor microenvironment alongside malignant cells. This holistic perspective promises novel therapeutic avenues that disrupt the nurturing niches fostering cancer growth.

Mechanistically, cancer cells employ sophisticated strategies to co-opt their neighbors. They induce blood vessel formation (angiogenesis) to secure vital nutrients, reprogram immune cells to evade detection, and stimulate fibroblasts to generate collagen-rich scaffolds that shield tumors from immune attack. Each spatial ecotype represents a unique micro-ecosystem wherein these interactions unfold, orchestrating cancer progression and influencing therapeutic success. By mapping these ecotypes, Liquid EcoTyper offers a window into the molecular conversations driving malignancy.

The implications for immunotherapy are particularly profound. Immunotherapies, which unleash the immune system against cancer, often exhibit variable efficacy across patients with seemingly similar tumors. This variability may stem from differences in spatial ecotypes within tumors that modulate immune cell activation and infiltration. Liquid EcoTyper’s ability to noninvasively detect ecotype patterns associated with favorable immunotherapy responses heralds a new era of predictive and adaptive cancer care.

Looking ahead, the researchers envision integrating liquid biopsy-based spatial profiling into clinical workflows, furnishing oncologists with actionable intelligence throughout the treatment journey. Repeated assessments could refine individualized therapy plans, minimize unnecessary toxicities, and ultimately improve survival outcomes. Before widespread adoption, however, comprehensive clinical trials will be necessary to validate Liquid EcoTyper across broader patient populations and cancer subtypes.

The study represents a remarkable convergence of biomedical data science, genomics, and computational biology. It exemplifies how sophisticated algorithms can extract meaningful biological signals from high-dimensional data, transforming clinical diagnostics and deepening our understanding of tumor ecosystems. Fueled by collaborative efforts spanning prestigious institutions, the research underscores the growing importance of interdisciplinary approaches in tackling cancer’s complexity.

While significant challenges remain, such as standardizing assay protocols and ensuring robust interpretation in diverse clinical contexts, the demonstration of a blood test that noninvasively captures spatial ecotypes marks a paradigm shift in oncology. It opens avenues for exploring tumor ecology dynamically and noninvasively, moving beyond static snapshots toward continuous monitoring and precise intervention.

In essence, this innovative methodology reframes cancer not as a solitary invader but as a thriving community embedded within a complex cellular milieu. Understanding and disrupting these malignant neighborhoods may prove pivotal to defeating one of humanity’s deadliest foes. As the field moves forward, tools like Liquid EcoTyper promise to empower clinicians with unprecedented clarity into the tumor microenvironment’s spatial and functional architecture, heralding a future where cancer treatment is smarter, more adaptive, and profoundly personalized.

Subject of Research: Cells

Article Title: Non-invasive profiling of the tumour microenvironment with spatial ecotypes

News Publication Date: 6-May-2026

Web References:

• https://www.nature.com/articles/s41586-026-10452-4

References:

• Newman, A., et al. (2026). Non-invasive profiling of the tumour microenvironment with spatial ecotypes. Nature. DOI: 10.1038/s41586-026-10452-4

Keywords: Tumor cells, Cancer immunotherapy

Colorectal cancer remains one of the deadliest malignancies worldwide, ranking as the second leading cause of cancer-related deaths. One of the major hurdles in improving patient outcomes lies in early diagnosis. Colonoscopy, though highly effective, is costly and uncomfortable, deterring many from regular screening and consequently delaying detection until advanced stages when treatment options are limited and prognosis worsens. The UNIGE team’s novel approach directly addresses this challenge by providing a simpler route to identifying cancer presence through the microbiome’s intricate composition.

Central to this breakthrough is the recognition that not all bacteria within the gut microbiota contribute equally to disease development. Previous research established a link between microbial communities and colorectal cancer but treating bacterial species as uniform entities glossed over critical differences. Remarkably, strains within a single species can diverge functionally — some may promote carcinogenesis, while others remain benign. By developing a framework to identify these organisms at the subspecies level, the researchers have pinpointed more relevant microbial actors with finer resolution, bringing clarity to a previously murky biological landscape.

The team, led by Professor Mirko Trajkovski of the Department of Cell Physiology and Metabolism at UNIGE, innovated by moving beyond traditional taxonomic classifications. They explained that subspecies-level analysis strikes a balance: it resolves bacterial groups sufficiently to capture functional diversity but remains sufficiently consistent across individuals and populations to yield meaningful, reproducible insights. This middle ground overcomes the variability that complicates attempts to use strain-level data clinically, which often suffers from extreme heterogeneity.

To achieve this, the researchers amassed and processed vast datasets encompassing the human gut microbiome. Matija Trickovic, the study’s first author and a PhD student under Trajkovski, tackled this bioinformatic challenge by creating the first exhaustive catalogue of human gut microbiota subspecies. Utilizing cutting-edge machine learning techniques, they developed computational methods capable of efficiently parsing extensive microbiome data to identify subspecies signatures correlated with colorectal cancer presence.

Combining this subspecies catalogue with clinical data from patients, the research team trained predictive models capable of diagnosing colorectal cancer solely based on the bacteria found in stool samples. The results were extraordinary: the diagnostic model detected 90% of colorectal cancer cases, approaching the 94% sensitivity conventionally achieved by colonoscopy. Notably, this performance outstripped that of all existing non-invasive detection methods. This validation confirms the power of subspecies microbiota analysis in medical diagnostics.

The implications extend well beyond screening. By integrating additional clinical information, the model’s predictive power is anticipated to improve further, potentially matching or even surpassing colonoscopy accuracy. Such a tool could be deployed in routine screenings worldwide, reserving invasive procedures for patients at high risk. This paradigm shift would not only enhance early cancer detection but also reduce healthcare costs and improve patient compliance due to the non-invasive nature of stool sampling.

Currently, UNIGE is collaborating with Geneva University Hospitals (HUG) to initiate clinical trials aimed at refining the detection capabilities of the model. A key objective is to determine the earliest cancer stages and specific lesions identifiable through microbiome signatures, paving the way for personalized medical interventions. These trials represent a significant step towards clinical translation and widespread use.

Beyond colorectal cancer, this subspecies-based microbiome analysis opens a vast frontier in understanding the gut’s impact on human health. Different subspecies of the same bacterial species can exert opposing physiological roles, influencing disease pathways from metabolic disorders to immune dysfunction. Capturing this nuanced microbial diversity provides a powerful lens through which researchers may unravel complex host-microbiome interactions that underlie numerous diseases.

The technological foundation of this breakthrough lies in the synergy between bioinformatics, machine learning, and microbiology. High-throughput sequencing generates massive data describing microbiome composition, but extracting biologically relevant information demands sophisticated algorithms capable of detecting subtle patterns. The UNIGE team’s approach exemplifies how interdisciplinary innovation can harness big data to address unmet medical needs effectively.

Furthermore, the non-invasive nature of stool sample analysis aligns with patient-centered care principles, potentially increasing participation in cancer screening programs. Regular microbiota profiling could enable longitudinal monitoring of gut health and early disease detection, fundamentally changing preventive medicine. The technique offers scalability and accessibility, especially in resource-limited settings where colonoscopy infrastructure is scarce.

Professor Trajkovski emphasizes that this research ushers in a new era for microbiome studies—not merely cataloging species but deciphering the functional and pathological implications hidden at finer taxonomic levels. This breakthrough exemplifies how sub-microscopic variations in microbial populations shape human health and disease, challenging scientists to rethink current diagnostic and therapeutic strategies.

The success of this study also exemplifies how machine learning can transform biological research. By training algorithms on catalogued microbial data aligned with clinical outcomes, the researchers created predictive tools that learn and improve over time. This dynamic capacity positions microbiome analysis as a cornerstone technology for next-generation diagnostics across a wide spectrum of diseases.

As this subspecies-focused diagnostic technology matures, it holds promise for integration with other omics data—such as metabolomics and genomics—to build multifaceted disease prediction platforms. The potential to detect subtle shifts in microbial communities before clinical symptoms arise could vastly improve early intervention and patient prognosis.

In conclusion, the UNIGE team has redefined the frontiers of cancer diagnostics by unveiling a microbiome-based, machine learning-powered tool for early colorectal cancer detection. This innovative method not only rivals established colonoscopy standards but also heralds a future where non-invasive, microbiota-informed diagnostics augment the fight against cancer and other diseases. As research progresses, it promises to democratize screening, reduce the burden of invasive procedures, and deepen our understanding of the microscopic ecosystems within us that ultimately influence our health.

Subject of Research: People

Article Title: Subspecies of the human gut microbiota carry implicit information for in-depth microbiome research

News Publication Date: 13-Aug-2025

Web References: 10.1016/j.chom.2025.07.015

Keywords: colorectal cancer, gut microbiota, subspecies, machine learning, non-invasive diagnostics, microbiome, early cancer detection, bioinformatics, stool sample screening, personalized medicine, microbiota catalog, cancer biomarkers