Computational pipeline uncovers subtype specific targets for cervical cancer

A new study presents a multi-layered strategy for identifying cell-surface targets that could help guide more precise antibody-drug conjugate (ADC) development for cervical cancer, with different targets showing distinct potential across tumor subtypes.

Cervical cancer is a major health burden among women of reproductive age, and although immunotherapies have improved outcomes for some patients with metastatic or refractory disease, substantial therapeutic challenges remain. One of these challenges is tumor heterogeneity: cervical cancer comprises biologically distinct subtypes, including squamous cell carcinoma (SCC) and adenocarcinoma, which arise from different tissues of origin.

A new study published in Computational Biomedicine, "Multi-Database Transcriptomic Screening to Identify Cervical Cancer Subtype-Selective Cell Surface Targets for Precision Medicine-based Application of Recombinant Immunotherapeutics," addresses this challenge by combining large-scale transcriptomic analysis with tissue-of-origin refinement, tumor subpopulation analysis, and laboratory validation.

The study developed a unified computational pipeline that reprocessed 304 primary cervical cancer tumors and 7,597 healthy tissue samples spanning 52 tissue types using consistent alignment and quantification parameters. The researchers screened a curated set of 259 high-confidence cell-surface protein genes to identify proteins that were preferentially expressed in cervical cancer and could potentially serve as targets for ADC-based therapies.

The initial global analysis identified 30 significantly overexpressed cell-surface candidates. Importantly, the researchers then moved beyond a simple tumor-versus-normal comparison.

Looking beyond a single definition of "normal"

The study highlights an important challenge in therapeutic target discovery: a protein that appears highly overexpressed when a tumor is compared with a broad collection of healthy tissues may not necessarily be selectively expressed relative to the specific normal tissue from which that tumor originated.

To address this issue, the researchers separately evaluated cervical cancer subtypes against normal ectocervical and endocervical tissues.

This additional layer of analysis substantially changed the interpretation of several candidate targets.

MSLN emerged as the most consistent pan-cervical cancer target, maintaining strong overexpression relative to both ectocervical and endocervical reference tissues.

TROP-2, meanwhile, demonstrated a more subtype-dependent profile. Although it showed the second-highest global fold change and was detected in all tumors analyzed, its expression was not selectively elevated relative to normal ectocervix. In contrast, it showed strong differential expression relative to normal endocervix, supporting its potential as an adenocarcinoma-selective target rather than a universal cervical cancer target.

The analysis also revealed an example of why average expression across an entire tumor cohort can conceal clinically relevant subgroups.

Finding a hidden high-expression subgroup

LIV-1 did not rank among the top 30 targets in the global analysis. Its cohort-wide fold change was only 0.97, which could have led it to being excluded from further consideration.

Instead, the researchers examined expression heterogeneity within the tumor cohort.

This analysis identified 14 high-expressing tumors-approximately 4.6% of the cohort-with a median expression of 122.7 TPM and a fold change of 3.07.

The finding suggests that LIV-1 may represent a potential precision medicine target for a biologically defined subgroup of cervical cancer patients, rather than a broadly expressed target across the disease.

According to the researchers, this illustrates why a single ranked list of differentially expressed genes may be insufficient for therapeutic target discovery in heterogeneous cancers.

From computational prediction to experimental testing

The researchers next evaluated whether transcriptomic predictions could be translated into functional activity.

Recombinant scFv-SNAP fusion proteins targeting MSLN, TROP-2, and LIV-1 were produced and characterized. When conjugated with the cytotoxic payload auristatin F, the resulting constructs demonstrated surface binding and dose-dependent cytotoxicity across cervical cancer cell lines.

For example, TROP-2-targeted constructs showed strong activity in squamous cell carcinoma-derived CaSki and SiHa cells, with IC50 values of 7.0 nM and 17.9 nM, respectively. The construct also demonstrated activity in the adenocarcinoma-derived HeLa cell line, although at a higher concentration.

MSLN-targeted constructs demonstrated surface binding ranging from 43.6% to 99.4% across the tested cervical cancer cell lines and showed selective nanomolar cytotoxicity.

The LIV-1 findings provided another important correspondence between computational and experimental analyses. Stronger cytotoxic activity was observed in CaSki and SiHa cells, while activity was substantially lower in HeLa and ME180 cells, consistent with the proposed existence of an LIV-1-high biological subgroup.

Toward more refined target selection

Together, the findings support a three-tier target prioritization framework:

Global surface-target screening → tissue-of-origin refinement → tumor subpopulation stratification

Each level provided different information.

MSLN was identified as a candidate with broad cervical cancer coverage, TROP-2 showed a more pronounced adenocarcinoma-selective profile, and LIV-1 emerged as an exploratory target for a smaller, biologically defined subgroup.

The study therefore suggests that precision target discovery may require more than simply identifying genes that are highly expressed in tumors. Understanding which normal tissue provides the appropriate biological reference, which cancer subtype expresses the target, and whether expression is concentrated in a specific patient subgroup may all be critical for defining a clinically meaningful therapeutic window.

The researchers emphasize that the current findings remain preclinical. The transcriptomic analyses are based on mRNA expression, while the experimental validation was performed in cell lines and therefore cannot fully reproduce the tumor–normal expression context or the tumor microenvironment found in patients.

Future studies incorporating patient-derived organoids, larger and more deeply resolved single-cell RNA sequencing datasets, surface proteomics, and patient-derived samples could help determine which cells express these targets, how frequently the identified subgroups occur in patients, and whether the proposed therapeutic windows can be reproduced in more physiologically relevant models.

Ultimately, the study demonstrates how computational target discovery and experimental validation can work together to refine therapeutic priorities in a heterogeneous cancer. Rather than treating cervical cancer as a single molecular entity, the approach provides a framework for identifying different surface targets for different biological subtypes and patient populations, potentially supporting the development of more precisely directed recombinant immunotherapeutics.