Predicting the Presence and Abundance of Bacterial Taxa in Environmental Communities through Flow Cytometric Fingerprinting

J. Heyse, Florian Schattenberg, Peter Rubbens, Susann Müller, Willem Waegeman, Nico Boon, Ruben Props

TL;DR

This paper presents a novel computational pipeline that uses flow cytometry (FCM) data to predict the presence and abundance of bacterial taxa in complex environmental communities. By integrating FCM with 16S rRNA gene sequencing, the method offers a faster, more scalable alternative to traditional sequencing-based microbiome analysis, enabling real-time monitoring of microbial dynamics.

What Problem Does This Paper Address?

Accurately monitoring microbial community composition is crucial for managing various biotechnological applications, from aquaculture to industrial fermentation. However, current gold-standard methods like cultivation and 16S rRNA gene sequencing are often time-consuming and labor-intensive, especially when high-throughput, temporally resolved data is needed. This bottleneck limits our ability to understand and respond to dynamic changes in microbial ecosystems. While flow cytometry offers a rapid, single-cell analysis method, its direct link to specific taxonomic identities in complex communities has been challenging to establish robustly. This paper aims to bridge this gap by developing a predictive model that translates FCM "fingerprints" into taxonomic information.

Methods

The core of this work is a two-stage computational pipeline designed to predict bacterial taxon presence and abundance from flow cytometry data.

  1. Training Stage:
    • Data Integration: The pipeline integrates flow cytometry data (measuring phenotypic properties like cell size and fluorescence) with 16S rRNA gene amplicon sequencing data (providing taxonomic profiles).
  2. Taxonomic Stratification: Cell sorting is used to physically separate different sub-communities within the FCM data space, allowing researchers to link specific FCM "regions" or "gates" to taxonomic groups identified by sequencing. This step is crucial for validating the connection between phenotype and taxonomy.
  3. Model Development: For each target taxon (Operational Taxonomic Unit or OTU), two machine learning models are trained:
  4. Presence/Absence Classifier: A random forest classifier predicts whether a taxon is present or absent in a sample.
  5. Abundance Regression Model: A regression ensemble (combining gradient boosting and support vector regression) predicts the relative abundance of the taxon.
  6. Pipeline Refinement: The predictions from the classifier are superimposed onto the regression model's output. Specifically, if a taxon is predicted to be absent by the classifier, its predicted abundance is set to zero. This step significantly reduces false positives.
  7. Deployment Stage:
    • Once trained, the pipeline can take new FCM data from a sampled environment and predict the presence and relative abundance of multiple target taxa simultaneously.

The novelty lies in directly training models on survey data from mixed communities, avoiding the potential inaccuracies of using axenic (pure culture) data, as the phenotypic characteristics of a bacterium can change when it's part of a community.

Datasets

  • Primary Dataset: The main dataset used for training and validation comes from an 18-day sampling campaign in a whiteleg shrimp ( Litopenaeus vannamei ) hatchery. This dataset includes both previously published flow cytometry and 16S rRNA gene amplicon sequencing data, supplemented with newly generated 16S rRNA gene amplicon data from sorted subcommunities.
  • Validation Datasets:
    • A three-strain cytometric mock community.
  • A dataset of insular reactor communities.

These validation datasets were used to assess the pipeline's applicability to other microbial systems beyond the initial aquaculture case study.

Key Insights and Ideas

  • FCM Fingerprints Encode Taxonomic Information: The study robustly demonstrates that specific regions within flow cytometric "fingerprints" correspond to distinct bacterial taxa. Cell sorting and subsequent sequencing confirm this link, showing that sorted subcommunities have significantly different taxonomic compositions.
  • Community Context Matters: Phenotypic properties measured by FCM can differ for a taxon in isolation versus within a mixed community. This highlights the importance of training predictive models on real community data rather than relying solely on axenic cultures.
  • Two-Stage Prediction is Powerful: Combining a presence/absence classifier with a regression model for abundance prediction significantly improves accuracy and reduces false positives, making the predictions more reliable.
  • Temporal Dynamics are Predictable: The developed pipeline can effectively predict the temporal dynamics of bacterial taxa in a community over time, offering a valuable tool for monitoring changes.
  • Phylogenetic Closeness and Cytometric Similarity: While not a perfect correlation, phylogenetically related taxa tend to occupy similar regions in the cytometric fingerprint, suggesting some conserved phenotypic traits within related groups.

Key Results

  • Presence/Absence Classification: Achieved high accuracies, ranging from 78% to 98% for the top 50 OTUs, with an average accuracy of 88.13% ± 4.78%.
  • Relative Abundance Prediction: The final pipeline (after classifier-regression integration) yielded an average R² of 0.35 ± 0.24 across the top 50 OTUs. This represents a significant improvement over raw regression predictions, particularly in reducing false positives.
  • Temporal Prediction: The models successfully predicted the temporal presence and abundance of taxa in time-series data, capturing overall dynamics well, especially for taxa with good model performance (e.g., OTU 1 with R² = 0.81).
  • Validation Success: The pipeline showed promising performance on external datasets, including a mock community (average R² of 0.89 ± 0.03) and reactor communities (average R² of 0.33 ± 0.27), indicating broad applicability.
  • Co-occurrence Impact: The study found that co-occurrence patterns between taxa had a minor influence on model performance, suggesting the models are not solely relying on these relationships.

Limitations and Future Work

  • Model Performance Variability: There is substantial variation in model performance across different OTUs, with some taxa exhibiting very low prediction accuracy. This suggests that for certain taxa, it may be inherently difficult to build performant models, possibly due to indistinguishable cytometric fingerprints or high phenotypic plasticity.
  • Snapshot Sample Caution: While good for monitoring trends over time, the models should be used with more caution when interpreting single snapshot samples, as their accuracy can be lower.
  • Data Requirements: Model performance is sensitive to the amount of training data; performance significantly drops with smaller datasets, indicating that more data is often needed for robust predictions.
  • Sorting Gate Precision: The study used relatively large sorting gates, which allowed for verification of general region association but not precise identification of exact cell populations. Future work could benefit from more refined sorting gates.
  • Natural Environments: The pipeline's applicability to natural environments (like lakes and oceans) has not yet been investigated and warrants further study.
  • Future Improvements: The authors suggest several avenues for improvement, including increasing training dataset sizes, optimizing acquisition settings, incorporating additional fluorescence detectors or stains, and potentially using more precise sorting gates for validation.

Summary updated

Abstract

Microbiome management research and applications rely on temporally resolved measurements of community composition. Current technologies to assess community composition make use of either cultivation or sequencing of genomic material, which can become time-consuming and/or laborious in case high-throughput measurements are required. Here, using data from a shrimp hatchery as an economically relevant case study, we combined 16S rRNA gene amplicon sequencing and flow cytometry data to develop a computational workflow that allows the prediction of taxon abundances based on flow cytometry measurements. The first stage of our pipeline consists of a classifier to predict the presence or absence of the taxon of interest, with yielded an average accuracy of 88.13% ± 4.78% across the top 50 operational taxonomic units (OTUs) of our data set. In the second stage, this classifier was combined with a regression model to predict the relative abundances of the taxon of interest, which yielded an average <i>R</i><sup>2</sup> of 0.35 ± 0.24 across the top 50 OTUs of our data set. Application of the models to flow cytometry time series data showed that the generated models can predict the temporal dynamics of a large fraction of the investigated taxa. Using cell sorting, we validated that the model correctly associates taxa to regions in the cytometric fingerprint, where they are detected using 16S rRNA gene amplicon sequencing. Finally, we applied the approach of our pipeline to two other data sets of microbial ecosystems. This pipeline represents an addition to the expanding toolbox for flow cytometry-based monitoring of bacterial communities and complements the current plating- and marker gene-based methods. <b>IMPORTANCE</b> Monitoring of microbial community composition is crucial for both microbiome management research and applications. Existing technologies, such as plating and amplicon sequencing, can become laborious and expensive when high-throughput measurements are required. In recent years, flow cytometry-based measurements of community diversity have been shown to correlate well with those derived from 16S rRNA gene amplicon sequencing in several aquatic ecosystems, suggesting that there is a link between the taxonomic community composition and phenotypic properties as derived through flow cytometry. Here, we further integrated 16S rRNA gene amplicon sequencing and flow cytometry survey data in order to construct models that enable the prediction of both the presence and the abundances of individual bacterial taxa in mixed communities using flow cytometric fingerprinting. The developed pipeline holds great potential to be integrated into routine monitoring schemes and early warning systems for biotechnological applications.
BibTeX
@inproceedings{heyse2021predicting,
  author = {J. Heyse and Florian Schattenberg and Peter Rubbens and Susann Müller and Willem Waegeman and Nico Boon and Ruben Props},
  title = {{Predicting the Presence and Abundance of Bacterial Taxa in Environmental Communities through Flow Cytometric Fingerprinting}},
  booktitle = {mSystems},
  year = {2021},
  doi = {10.1128/msystems.00551-21},
}
Figure