Background

Microbiomes play a central role in maintaining soil health and supporting plant development, influencing nutrient cycling, disease resistance, and overall ecosystem stability [1,2]. In both agricultural and natural systems, diverse microbial communities dominated by bacteria in soils and plant tissues are key drivers of biogeochemical processes such as carbon sequestration and nitrogen transformation [3]. Understanding the composition and functions of microbial communities provides a foundation for the development of new technologies, a guide for ecological management decisions, and the improvement of agricultural practices. However, accurate and consistent microbiome profiling remains a technical challenge, especially when dealing with complex environmental samples such as soil samples [4].

For bacterial community profiling, amplicon sequencing of the 16S rRNA gene remains a widely used, cost-effective approach for estimating community composition and diversity across environmental gradients and experimental manipulations [5]. In soils and plant compartments (rhizosphere, endosphere, and phyllosphere), where communities are both diverse and uneven, the ability to resolve taxa at finer ranks (e.g., species and, where possible, strain) improves the interpretability of ecological patterns and the portability of findings across studies [6,7]. Methodological advances have progressively increased taxonomic resolution and reproducibility in amplicon workflows. Early pipelines grouped reads into operational taxonomic units (OTUs) at fixed similarity thresholds, which simplified analysis but conflated biological and technical variation. However, these methods often fall short in delivering species-level resolution and are prone to variability across pipelines and datasets [8,9].

The shift toward exact sequence-based approaches (e.g., amplicon sequence variants, ASVs) reduced clustering artifacts for short-read data, improving comparability across experiments [10]. More recently, long-read sequencing has enabled recovery of near full-length 16S rRNA genes, potentially improving species-level assignment, disambiguating closely related taxa, and stabilizing ecological inferences in complex communities [11,12]. However, leveraging full-length reads requires algorithms that can accommodate read length–specific error structures, model multi-mapping to reference databases, and estimate abundances without overinflating diversity [13]. Emu, a taxonomic profiling method based on expectation-maximization (EM) to resolve ambiguous long-read mappings, yields compositional estimates that aim to be both accurate and robust for long-read 16S rRNA datasets [14].

Despite these advances, environmental microbiome studies still face challenges related to reproducibility and scale [5,15]. Reported community differences can be sensitive to choices in primer sets, reference databases, quality filters, and classification parameters, complicating meta-analysis and the accumulation of knowledge [8]. Soil and plant microbiomes also impose heavy computational demands due to complex experimental designs, due to high richness and the need for deeper sequencing to capture rare taxa, which can strain local computing resources [16]. High-performance computing (HPC) environments address scalability but are often perceived as inaccessible to new users and can themselves be sources of variability when software stacks, dependencies, or job-scheduling constraints differ across clusters [17,18]. For long-read 16S rRNA specifically, the lack of shared, domain-tailored, and fully documented workflows that integrate Emu with transparent quality control and benchmarking makes it difficult to evaluate performance across soil and plant compartments and to reproduce results across independent laboratories.

Consequently, there is a practical and conceptual gap: we lack a standardized, open, and reproducible long-read 16S rRNA pipeline that (i) is expressly designed for soil and plant microbiome questions, (ii) operationalizes best practices for quality control and reference-based inference with Emu, and (iii) scales predictably on HPC while producing portable, versioned outputs for downstream ecological analysis. Existing tutorials often target short-read OTU/ASV workflows or provide minimal guidance on how to tune long read–specific steps (e.g., length/quality screening, handling mapping, and database curation) under realistic environmental complexity [5,8,19]. Moreover, reproducibility guidance typically emphasizes containerization or environment capture but stops short of demonstrating that end-to-end results remain stable across different HPC clusters, settings, or modest updates to reference databases, factors that commonly change across institutional environments [15]. For interdisciplinary audiences, these limitations hinder the translation of microbiome insights into real practice interventions or ecological theory because conclusions may be contingent on opaque computational choices.

To address this need, a reproducible Emu-based workflow is presented for soil and plant microbiome profiling, designed for HPC execution while remaining accessible to users with varying computational backgrounds. The workflow is organized into four steps: long-read 16S rRNA gene input validation, quality control suited to full-length reads, reference-informed taxonomic assignment with Emu, and standardized outputs for ecological analyses, each accompanied by versioned configurations and human-readable reports. Reproducibility is supported through declarative configuration files so that identical inputs produce consistent outputs across different clusters, and scalability is demonstrated.