Lauren Michelle Lui

Agriculture is both humanity’s largest infrastructure and the leading driver of environmental degradation. Many early-stage biotechnologies could mitigate this impact, but their deployability has not been systematically evaluated. We analyze deployability of 77 biotechnologies across eight domains (crops, soils, livestock, watersheds, aquaculture, health and security, food loss/waste and frontiers) by applying a Balanced Readiness Level assessment spanning technological, market, regulatory, acceptance and organizational dimensions. We find that market viability and social acceptance, not technological readiness, represent the primary constraints. High-readiness technologies tend to be ‘drop-in’, contained, transient and perceived as natural. These characteristics indicate strategies to elevate overall deployability, such as addressing genetically modified organism containment. Assessing the scalability and environmental impacts of biotechnologies for prioritization requires additional quantitative analysis, as shown in a Fermi analysis of microbial fertilizers. We combine our findings into a near- to long-term roadmap for living infrastructure: predictively designed multicomponent biological communities that function as systems at the ecosystem level to enhance food security and environmental resilience.

Phylum Omnitrophota (formerly candidate division OP3) is represented in public databases almost entirely by metagenome-assembled genomes; GTDB R232 contains two complete Omnitrophota assemblies. We present 176 complete and 53 high-quality Omnitrophota genomes from Oxford Nanopore metagenomes of Fennoscandian deep groundwater and the Baltic Sea water column, an 88-fold expansion of the complete Omnitrophota count. The 229 genomes resolve to 202 distinct species at 95 % ANI; 162 of these have no conspecific match in the 714 NCBI HQ Omnitrophota MAGs. Phylogenomically, 171 of the 176 complete genomes fall in class Gorgyraeia — which contains the cultured episymbiont Velamenicoccus archaeovorus — and 5 in Omnitrophia , with multiple Gorgyraeia orders and families represented. Phylum Omnitrophota hosts many very long proteins, with the longest in our corpus reaching 147,155 AA; the long end of the length distribution is concentrated on Gorgyraeia contigs across multiple Gorgyraeia orders. 24–28 % of GTDB-Tk-classified Omnitrophota contigs in the deep-groundwater and Baltic samples host at least one ≥ 10-kAA protein. Across the 916-protein long-protein domain-architecture catalog, 94 % carry transmembrane helices or a signal peptide; the four complete-genome proteins above 100,000 amino acids are all in inner-membrane-anchored architectures, the 147,155-AA protein with 147 TM helices. The 176 complete genomes share a uniform metabolic profile across the dominant orders: intact bacterial peptidoglycan biosynthesis alongside fragmentary TCA, incomplete electron transport, absent aerobic terminal oxidase, and partial cofactor and amino-acid biosynthesis. The profile matches the cultured V. archaeovorus phenotype and is consistent with a host-dependent episymbiotic lifestyle. Hypervariable-region calling across the 229 chromosomes returns 1,909 candidate loci, distributed across 223 of them; ribosomal-protein and EF-Tu / EF-G content sits inside called HVRs on 150 of those 223 (67 %), recovering across the collection the housekeeping-cargo integrations documented in Nielsen (2026b). All genomes, the OrthoFinder supermatrix and its ML tree, the 916-protein giant-protein domain-architecture catalog, and per-step scripts are released as a community resource at Zenodo (DOI [TBD]).

The archaeal phylum Nanobdellota (formerly Nanoarchaeota) was previously represented by four complete genomes. We present 208 complete Nanobdellota genomes from Oxford Nanopore metagenomes of the Baltic Sea water column and Fennoscandian groundwater (69-201 m below sea level), rotated to the ORC1/Cdc6 replication origin -- a 52-fold expansion of complete-genome representation. Across the ar53 supermatrix and a Nanobdellota-tuned 71-marker supermatrix on 1,239 taxa, the named GTDB orders within Nanobdellota are recovered as monophyletic clades, including the three orders that dominate our environmental sampling: Woesearchaeales, Pacearchaeales, and the GTDB placeholder order SCGC-AAA011-G17. This is consistent with the existing GTDB R232 order-level circumscription. We retire the SCGC-AAA011-G17 placeholder name, replacing it with a complete-genome-anchored SeqCode nomenclatural chain (Maxwellarchaeales ord. nov., Maxwellarchaeaceae fam. nov., Maxwellarchaeum gen. nov., and Maxwellarchaeum balticum sp. nov.) without altering the order-level circumscription. Pacearchaeales and Maxwellarchaeales retain no central or energy metabolism beyond Form III RuBisCO, PEP synthase, and ferredoxin; Woesearchaeales retains partial glycolysis and a V/A-type ATPase. A 4,262-tip phylogeny of rbcL (the RuBisCO large-subunit gene) identifies nine candidate archaea-to-Patescibacteriota Form III RuBisCO transfer events -- including one to a Baltic Minisyncoccia -- versus two reciprocal candidates, consistent with archaea-to-CPR being the more frequently identified direction in our data. All 256 Nanobdellota genomes (208 complete + 48 high-quality non-circular), the ar71 marker set with its 1,239-taxon ML tree, 154 Nanobdellota-trained HMMs for KEGG-ortholog detection in DPANN proteomes (94 ROBUST), and the 4,262-tip rbcL reference tree are released as a community resource, alongside the full analysis archive -- alignments, intermediate trees, structural predictions, and per-step scripts --at Zenodo (DOI 10.5281/zenodo.20174424; see Using the resource).

Pelagibacter, the largest genus within the SAR11 clade, is the most abundant bacterium in the ocean, yet the vast majority of its species-level diversity remains uncharacterized at the genomic level. Here we present 135 complete Pelagibacter genomes -- the largest such collection assembled to date -- comprising 75 from Oxford Nanopore metagenomes of the San Francisco Estuary (SFE), 31 from a deeply sequenced station within the same transect, and 29 from public databases. These genomes define 52 species at 95% ANI, of which 44 (85%) are taxonomically novel. An expanded phylogeny incorporating 89 additional high-quality NCBI genomes confirms that our collection captures the phylogenetic backbone of the genus, with genomes from Hawaii, Namibia, and the Sargasso Sea nesting within SFE clades. The pangenome is open (14,862 singletons, 62%), driven by two distinct mechanisms. First, a universal hypervariable region (HVR) at a conserved chromosomal position (7-15% from dnaA) is present in all 135 genomes, anchored by tRNA genes at both boundaries (Phe/His and Arg). The HVR carries genome-specific surface polysaccharide biosynthesis genes with a GC age gradient -- highest GC at the tRNA boundaries, lowest in the center -- consistent with a two-ended phage insertion model. Only this HVR is positionally conserved across the genus; the three other hypervariable regions previously described in a single reference genome are not. Second, scattered genomic islands throughout the chromosome contribute the remaining singleton content, including chimeric islands with genes from four bacterial phyla. Biosynthetic pathway reconstruction reveals auxotrophies that are phylogenetically structured, not uniform: biotin, reduced sulfur, and glycine are genus-wide dependencies, while isoleucine, pantothenate, histidine, and glyoxylate cycle capacity vary across lineages with significant phylogenetic clustering. Structural annotation with ESMFold and Foldseek resolved 3,125 hypothetical proteins; 1,222 remain uncharacterized by any method, including a 47-amino-acid protein conserved in two-thirds of all genomes within a fixed operonic context -- independently predicted by two gene callers yet matching nothing in any database. A controlled depth comparison at one station demonstrates that standard metagenome sequencing systematically underestimates Pelagibacter diversity, with three species recovered only at elevated depth and the species count at that station more than doubling (9 vs 4).

Candidatus Actinomarina minuta is among the smallest known free-living bacteria and among the most abundant photoheterotrophs in ocean surface waters, yet not a single complete genome has been published for any member of the order Actinomarinales. Here we present 84 complete Actinomarina genomes assembled from Oxford Nanopore metagenomes of the San Francisco Estuary (SFE), together with 29 additional high-quality (>=95% complete, <5% contamination, single contig) non-circular assemblies. These genomes define 9 species at 95% ANI, three of which are novel, and represent, to our knowledge, the first complete genomes for the entire Actinomarinales order. An expanded phylogeny incorporating 23 high-quality NCBI genomes and a Casp-actino5 outgroup confirms species boundaries and places NCBI genomes among SFE clades. The pangenome of 9,278 clusters is approaching closure (decay ratio 0.55), with a core genome of 227 single-copy genes (2.4%). Singletons (3,858 clusters, 42%) are concentrated in a tRNA-bounded hypervariable region (HVR) at 85-90% of the chromosome from dnaA -- a similar replicative distance to the HVR recently described in Pelagibacter (7-15% from dnaA; Lui & Nielsen, in preparation) but on the opposite replichore. The HVR is flanked by tRNA genes at both boundaries, and in 32 of 84 genomes, the selenocysteine tRNA (selC) is physically located inside the HVR. All 84 genomes encode selenocysteine tRNA (selC) and the selenocysteine biosynthesis genes selA and selD; the elongation factor selB is present but divergent. This has not been previously reported for Actinomarina. Deep learning selenoprotein prediction (deep-Sep) identifies ~5 selenoproteins per genome in 69 families, including a selenoprotein form of selD itself. Retention of dedicated Sec biosynthetic machinery serving multiple targets in a genome of ~1.1 Mbp implies that the catalytic advantage of selenocysteine over cysteine is large enough to justify the cost. KofamScan pathway reconstruction reveals the most extensive set of auxotrophies yet documented from complete genomes in a free-living marine bacterium: arginine, histidine, tryptophan, and thiamine biosynthesis are universally absent, biotin biosynthesis is non-functional (final step absent), and only 5 of 8 TCA cycle steps are retained at the standard detection threshold. Gene order is extensively rearranged between species: no gene adjacency is universal across all 84 genomes. NCBI currently lists 396 Actinomarina genomes, none complete; of the 39 that pass quality thresholds, 41% are misclassified by NCBI, belonging to other genera by GTDB-Tk reclassification.

Long-read metagenomics routinely produces millions of assembled contigs, creating a need for methods that organize sequences into biologically meaningful groups across samples and environments. We present a general-purpose compositional embedding for metagenomic sequences based on a 𝛽-variational autoencoder (𝛽-VAE) trained on multi-scale k-mer frequencies (1-mers through 6-mers; 2,772 features with centered log- ratio transformation). The embedding compresses each contig into a 384-dimensional vector that preserves local compositional similarity, enabling similarity search and graph-based clustering from sequence composi- tion alone. Through systematic comparison of fifteen models trained on up to 17.4 million contigs (525.5 Gbp) from brackish, terrestrial, and reference genome sources, we find that a small set of curated prokaryotic refer- ence genomes (656,000 contigs) outperforms ten-fold larger domain-specific training sets, and that neither reconstruction loss nor Spearman correlation reliably predicts downstream clustering quality. On nearest- neighbor graphs, flow-based clustering (MCL) markedly outperforms modularity-based methods (Leiden), yielding 12,123 clusters from 154,041 contigs (≥ 100 kbp) with 99.2% phylum-level purity confirmed by inde- pendent marker gene phylogenetics. Multi-method taxonomic annotation achieves 87% coverage and reveals that 16.4% of contigs are eukaryotic - the single largest component invisible to standard prokaryotic anno- tation tools. The embedding provides a sample-independent coordinate system for organizing metagenomic sequence space at scale.

Whether co-occurring microbial populations share similar evolutionary dynamics remains poorly understood, because strain-level resolution across many lineages simultaneously has been unachievable from metagenomic assembly. Here we apply 720 Gbp of Nanopore sequencing and the assembler myloasm to a South San Francisco Bay microbiome, recovering 488 high-quality single-contig genomes including 328 circular chromosomes — without manual curation. Strain-level resolution revealed striking contrasts in population structure: Pelagibacter yielded 78 high-quality single contig genomes spanning 18 species with no two sharing >=99% average nucleotide identity (ANI) — every genome a distinct strain, consistent with phage-driven frequency-dependent selection — while HIMB114 was nearly monotypic, with 9 of 11 high-quality genomes in a single species, suggesting periodic selection or a recent sweep. Using unsupervised Markov clustering of variational autoencoder embeddings, we assigned 99.3% of contigs >=5 kbp to a taxonomic domain and resolved nearly 50% of the community to species level. The assembly also captured nearly 95,000 viral contigs (>=5 kbp) including 502 giant viruses, novel eukaryotic lineages, and mercury resistance genes reflecting the Bay's contamination legacy. These contrasting patterns — coexisting in a single 10-liter sample — reveal that fundamentally different evolutionary strategies operate simultaneously within a complex microbiome, a level of resolution not previously demonstrated from de novo assembly. We'd like to note that the results of this study mark a milestone in a 10 year hunt for Torben to get as many complete Pelagibacter genomes from a metagenome as possible.

The seed is a habitat with limited resources and space. Although it is widely accepted that microbial competition is a key driver of the assembly of seed-associated microbial communities, the underlying mechanisms of this competition are not well understood. The initial objective of this work was to assess the importance of contact-independent microbial competition between the phytopathogenic bacterium Xanthomonas campestris pv. campestris 8004 (Xcc8004) and 30 strains representative of the bacterial populations most commonly associated with radish (Raphanus sativus) seeds. We identified Pseudomonas lactucae CFBP 13502 as a potent inhibitor of Xcc8004, mediated by exometabolites, specifically induced by certain seed-borne strains. Transcriptomic analysis linked this inducible activity to the upregulation of a gene cluster encoding a lipopeptide siderophore. Targeted gene deletion in P. lactucae CFBP 13502 confirmed that this cluster is essential for antagonism against Xcc8004. Furthermore, iron supplementation abolished this inhibitory effect, strongly supporting iron chelation as the underlying mechanism. Through comparative metabolomics, we elucidated the structure of a family of lipopeptide siderophores, produced by P. lactucae CFBP 13502, which we named lactuchelins. Our findings provide molecular evidence of competitive exclusion mechanisms at the seed microbiome interface, highlighting lactuchelins as a promising avenue for the development of seed-based biocontrol strategies against seed-borne phytopathogens.

This protocol provides a methodology for extracting high-molecular weight (HMW) DNA from biomass from water samples collected on PES filters and is designed to support microbiome studies requiring strain-level resolution. Developed for applications such as long-read metagenomics, we describe gentle handling techniques to prevent DNA shearing. We outline a detailed procedure spanning sample storage, enzymatic lysis, and purification using a carboxyl bead mixture. This specific protocol was featured in research published in mSystems in August 2024 and a preprint posted on bioRxiv. Please cite our papers and this protocol if you use this method or felt inspired by it in your experimental methodology.

Functional annotation of microbial genomes is often biased toward protein-coding genes, leaving a vast, unexplored landscape of non-coding RNAs (ncRNAs) that are critical for regulating bacterial and archaeal physiology, stress response and metabolism. Identifying ncRNAs directly from genomic sequence is a paramount challenge in bioinformatics and biology, essential for understanding the complete regulatory potential of an organism. This paper presents RNAMunin, a machine learning (ML) model that is capable of finding ncRNAs using genomic sequence alone. It is also computationally viable for large sequence datasets such as long read metagenomic assemblies with contigs totaling multiple Gbp. RNAMunin is trained on Rfam sequences extracted from approximately 60 Gbp of long read metagenomes from 16 San Francisco Estuary samples. We know of no other model that can detect ncRNAs based solely on genomic sequence at this scale. Since RNAMunin only requires genomic sequence as input, we do not need for an ncRNA to be transcribed to find it, i.e., we do not need transcriptomics data. We wrote this manuscript in a narrative style in order to best convey how RNAMunin was developed and how it works in detail. Unlike almost all current ML models, at approximately 1M parameters, RNAMunin is very small and very fast.