🧬 biology

Functional microbiome features distinguish clinical trajectories in low-grade cervical lesions

This study demonstrates that while taxonomic composition alone has limited value, integrating age with microbial ecological organization and inferred functional pathways significantly improves the prediction of clinical trajectories in women with low-grade cervical lesions.

Original authors: Milan Stosic, Mariano A. Molina, Dhananjay Mukhedkar, Sadaf Sakina Hassan, Anna Sahlin, Cristina Ruiz Ballester, Jiangrong Wang, Laila Sara Arroyo Mühr

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Milan Stosic, Mariano A. Molina, Dhananjay Mukhedkar, Sadaf Sakina Hassan, Anna Sahlin, Cristina Ruiz Ballester, Jiangrong Wang, Laila Sara Arroyo Mühr

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your cervix is a bustling city, and the tiny bacteria living there are the citizens. For a long time, doctors thought that if the city was run by the "good guys" (a specific type of bacteria called Lactobacillus), everything was fine. If the "bad guys" (other bacteria) took over, trouble was brewing. But a new study suggests the story is way more complicated than just "good guys vs. bad guys."

The researchers looked at 90 women who had a specific type of low-grade cervical change called LSIL. Think of LSIL as a small, fuzzy patch on a wall. Sometimes, the wall fixes itself (regression), and sometimes, the patch gets worse and turns into something more serious (progression). The big question was: Can we look at the bacterial city to predict which wall will fix itself and which will crumble?

The "Good Guys" Aren't All the Same

First, the team checked the overall population of the city. They looked at how many different types of bacteria were there and how they were mixed up. Surprisingly, they found no big difference between the women whose walls healed and those whose walls got worse. The total number of citizens and the general mix looked almost identical in both groups.

So, simply counting the bacteria or looking at the main "good guy" species wasn't enough to tell the future. It's like trying to predict a storm just by counting how many people are wearing raincoats; it doesn't tell you if a hurricane is coming.

The Secret is in the Neighborhoods (CSTs)

The researchers then zoomed in on how the bacteria organized themselves into "neighborhoods" called Community State Types (CSTs). They found that the women whose lesions healed were more likely to have a specific neighborhood called CST I-B.

Here is the twist: CST I-B is still a "good guy" neighborhood dominated by Lactobacillus, but it's not the exact same neighborhood as another common one called CST I-A.

CST I-A is like a quiet, orderly suburb where one type of Lactobacillus (L. crispatus) is the strict mayor.
CST I-B is a bit more lively and diverse. It has L. iners (a different type of Lactobacillus) and even some other bacteria like Prevotella bivia hanging out.

The study suggests that this slightly more chaotic, diverse neighborhood (CST I-B) was actually more common in the women whose lesions healed. It seems that having a slightly more complex mix of bacteria, rather than a super-strict, single-species city, might be the key to healing.

The Real Clue: What the Bacteria Are Doing

The most exciting part of the study is that the researchers didn't just look at who was living there, but what they were doing. They used a computer tool to guess the bacteria's "jobs" or metabolic functions.

They discovered that the bacteria in the healing group (CST I-B) were busy with different tasks than those in the other group:

The Healing Crew: Their bacteria were super active in carbohydrate metabolism and fermentation. Imagine them running a busy bakery, breaking down sugars and making energy.
The Other Crew: Their bacteria were less focused on that and more on building blocks for DNA (nucleotide biosynthesis).

The study suggests that this "bakery" activity—breaking down sugars and making fermentation products—might be creating a chemical environment that helps the body fight off the virus and heal the lesion. It's not just about who lives there, but what the neighborhood is cooking up.

Can We Predict the Future?

The researchers tried to build a crystal ball (a computer model) to predict who would heal and who wouldn't.

If they only looked at the woman's age, the crystal ball was basically guessing (about 57–61% accuracy).
If they only looked at the list of bacteria names, it was still just guessing (about 57–61% accuracy).
But, when they combined the age, the list of bacteria, AND the "job descriptions" (the metabolic functions), the crystal ball got much sharper. It correctly distinguished between healing and worsening about 76% of the time.

This suggests that knowing what the bacteria are doing adds a crucial piece of the puzzle that just knowing who they are misses.

What We Don't Know Yet

It's important to remember that this study suggests these patterns but doesn't prove them with absolute certainty. The researchers used a computer to infer (guess based on DNA) what the bacteria were doing, rather than measuring the chemicals directly in a lab. Also, the group of women was relatively small (90 people), so the results need to be tested on a much larger crowd to be sure.

The study also couldn't tell us exactly why the "bakery" activity helps. It's like seeing that the bakery is open and the city is safe, but not knowing exactly which cookie is the magic one.

The Bottom Line

The main takeaway is that the "good guy" bacteria aren't all the same. A slightly more diverse neighborhood where bacteria are busy fermenting sugars seems to be linked to healing. While looking at just the names of the bacteria doesn't tell us much, looking at their functional jobs combined with their names gives us a much better hint about whether a low-grade lesion will disappear on its own or need more attention. It's a step toward a smarter way to sort out who needs a watchful eye and who can just relax.

Technical Summary: Functional Microbiome Features Distinguish Clinical Trajectories in Low-Grade Cervical Lesions

Problem Statement
Despite the efficacy of HPV vaccination and screening, a significant challenge in cervical cancer prevention remains the inability to predict the clinical trajectory of low-grade squamous intraepithelial lesions (LSIL). While high-risk HPV infection is the primary driver of carcinogenesis, the majority of LSIL cases regress spontaneously, whereas a minority progress to high-grade lesions (HSIL+) or invasive cancer. Current screening strategies, relying on cytology and HPV genotyping, often fail to distinguish between these divergent outcomes, leading to unnecessary follow-ups, patient anxiety, and overtreatment. Although the cervicovaginal microbiome is increasingly linked to HPV persistence and disease progression, existing studies have largely been cross-sectional and focused on taxonomic composition. It remains unclear whether microbial taxonomic, ecological, and functional features can differentiate LSIL lesions that regress from those that progress, particularly within Lactobacillus-dominated communities where biological heterogeneity is suspected.

Methodology
This retrospective case-control study utilized a cohort of 90 women diagnosed with LSIL and known clinical outcomes (45 regression, 45 progression) derived from the Stockholm Cervical Screening Programme (2013–2023).

Data Generation: Cervicovaginal samples underwent 16S rRNA gene amplicon sequencing (V3–V4 regions) on an Illumina NextSeq 2000 platform.
Bioinformatics: Data were processed using the nf-core/ampliseq pipeline and DADA2 for Amplicon Sequence Variant (ASV) inference. Taxonomic classification was performed against the SILVA database.
Ecological Analysis: Alpha diversity (Shannon index) and beta diversity (Bray–Curtis PCoA) were calculated. Samples were classified into Community State Types (CSTs) based on Ravel et al. and France et al., with further exploration of CST I subtypes (I-A vs. I-B).
Functional Inference: Microbial functional potential was inferred using PICRUSt2, mapping ASVs to MetaCyc pathways. Differential abundance analysis and functional module aggregation (e.g., carbohydrate metabolism, fermentation) were conducted.
Statistical Modeling: Logistic regression models were constructed to predict clinical outcomes (regression vs. progression). Models tested individual features (age, taxa, CST) and integrated models combining age, microbiome taxa, and predicted functional pathway scores. Performance was evaluated using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC).

Key Results

Taxonomic Composition: Global microbiome diversity (alpha and beta) and genus-level composition did not significantly differ between regression and progression groups.
Community Structure: Microbial profiles clustered primarily by CST. CST I-B was significantly more frequent in the regression group compared to other CSTs ( $p=0.0467$ ).
Subtype Heterogeneity: Within Lactobacillus-dominated communities (CST I), subtype I-B exhibited higher alpha diversity and distinct compositional features compared to I-A. Specifically, CST I-B was enriched with Lactobacillus iners and Prevotella bivia, whereas CST I-A was dominated by L. crispatus.
Functional Profiles: While global functional differences between outcome groups were not significant, distinct functional profiles emerged within CST I subtypes. CST I-B showed enrichment in carbohydrate metabolism and fermentation pathways (specifically lactose and galactose degradation) and a relative depletion of nucleotide biosynthesis pathways compared to CST I-A.
Predictive Performance: Models based solely on age or microbiome taxa showed limited discriminatory power (AUC $\approx$ 0.57–0.61). However, an integrated model incorporating age, microbiome taxa, and predicted functional pathway features significantly improved discrimination between regression and progression (AUC = 0.76, 95% CI 0.66–0.86). This integrated model significantly outperformed the age-only model ( $p=0.015$ ) and the taxa-only model ( $p=0.006$ ). Exploratory analysis restricted to CST I samples suggested even higher discrimination (AUC > 0.90), though this requires validation in larger cohorts.

Significance and Claims
The authors claim that while broad taxonomic composition alone has limited value in distinguishing LSIL outcomes, microbial community organization and inferred functional states provide critical additional information. Specifically, the study demonstrates that:

Functional Heterogeneity Matters: Lactobacillus-dominated communities are not biologically equivalent; CST I-B represents a functionally distinct ecosystem characterized by specific metabolic activities (carbohydrate metabolism/fermentation) associated with lesion regression.
Ecological Context is Crucial: The presence of specific taxa (e.g., P. bivia) does not inherently predict progression; rather, their impact appears dependent on the broader ecological and functional context of the community.
Improved Risk Stratification: Integrating functional pathway features with taxonomic and clinical data modestly but significantly improves the prediction of lesion progression compared to traditional metrics.

The paper concludes that microbiome-derived functional features hold potential as candidate biomarkers for cervical lesion behavior. It supports the utility of integrative predictive modeling frameworks that combine multidimensional microbiome features to refine risk stratification, potentially reducing unnecessary interventions. The authors maintain a modest stance, acknowledging limitations such as the small cohort size, the use of inferred (rather than measured) functional data, and the lack of longitudinal sampling, emphasizing that these findings require validation in larger, prospective studies.