A pangenome is a representation of the combined genetic content of a group of organisms, usually members of a species or a defined population. Unlike a single genome, it describes both sequences shared across the group and sequences that differ among its members. Depending on the research question, a pangenome may take the form of a catalogue of gene families, a collection of assembled genomes, a sequence graph, or a combination of these resources. It is therefore both a biological concept—the genomic repertoire of a population—and a framework for organizing observations of that repertoire.
Pangenomes address a fundamental limitation of the reference genome: no individual sequence can contain all the variation found within a species. A reference can provide a useful coordinate system while omitting genes, insertions, rearrangements, and alternative sequence arrangements present in other individuals. Originally developed through comparisons of bacterial genomes, pangenomics now informs research on microbial evolution, crop improvement, human genetic variation, and the interpretation of sequencing data. Its central methodological challenges are distinguishing biological differences from technical errors, sampling diversity adequately, and representing alternative genomes without treating one as the universal norm.
Conceptual foundations and historical development
The modern pangenome concept emerged from comparative genomics, particularly the realization that closely related bacterial isolates can differ substantially in gene content. A landmark study by Hervé Tettelin and colleagues, published in 2005, compared eight genomes of Streptococcus agalactiae, or group B streptococcus. It distinguished a conserved complement of genes from genes found in only some strains and used the term pan-genome for their combined repertoire. The study also argued that sequencing additional strains would continue to reveal previously unobserved genes. This challenged the assumption that one complete bacterial genome could adequately represent the genetic capabilities of an entire species.
In its original, gene-centred formulation, the pangenome is the union of gene families identified across the sampled genomes. If is the set of families present in genome , the observed pangenome of genomes is
The corresponding strict core is their intersection:
These definitions concern families rather than necessarily identical genes. Different versions of a gene can belong to the same family, and a genome may contain several copies. Consequently, a binary gene-presence table does not by itself describe sequence differences or copy number.
The core genome comprises families shared by all sampled members, while the accessory genome contains families present in only a subset. Practical analyses often define a soft core using a high prevalence threshold, such as presence in at least 95% or 99% of genomes, to accommodate incomplete assemblies and occasional gene losses. Some studies subdivide accessory families into a relatively common “shell” and a rare “cloud.” These labels are descriptive conventions rather than universal biological categories. Their boundaries depend on the sampling scheme and analytical thresholds.
The concept subsequently expanded beyond gene inventories. In animals and plants, important variation often lies in noncoding regions, repeated sequences, and chromosome structure. Sequence-based pangenomes therefore represent alternative DNA segments, including different alleles, rather than only lists of genes. A gene can be part of the core while containing extensive sequence variation, and a regulatory insertion can be biologically important without adding a new gene family. Gene-content and sequence-based pangenomes answer related but distinct questions.
A pangenome is also not a fixed, final inventory. An observed resource is a sample-based approximation to a population whose composition changes through reproduction, migration, mutation, and extinction. Its scope must be specified: a clinical lineage, a named species, a crop breeding pool, and a collection of wild relatives yield different pangenomes. Comparisons are meaningful only when those biological boundaries and the underlying representation are explicit.
Evolutionary processes shaping genomic repertoires
Pangenome diversity arises through processes that add, modify, redistribute, or remove DNA. Mutation generates new sequence variants, while gene duplication creates additional copies that may retain the original function or diverge. Deletions remove genes and other segments. Homologous recombination can replace one version of a genomic region with another, and larger rearrangements alter sequence order or orientation. These processes operate alongside natural selection and genetic drift, which influence whether variants persist.
In bacteria and archaea, horizontal gene transfer is especially important. DNA can move through transformation, conjugation, or virus-mediated transfer. A plasmid may introduce several linked functions at once, such as resistance determinants or a metabolic pathway. A bacteriophage can contribute genes that change interactions with hosts or competitors. Integrated mobile elements may subsequently be inherited through cell division, lost, rearranged, or transferred again. Thus, the evolutionary history of an accessory gene need not match the history inferred from the chromosome carrying it.
These mechanisms generate an important distinction between gene presence and gene function. A recently acquired gene may be poorly expressed or incompatible with its new genomic background. Conversely, a conserved gene may acquire a new activity through sequence change without altering its presence–absence classification. Regulatory context, interactions among proteins, and the availability of substrates can determine whether an acquired pathway affects metabolism. Pangenome composition therefore describes potential capabilities, not a complete inventory of expressed traits.
Ecology helps explain why particular repertoires are maintained. Organisms occupying different habitats may benefit from different nutrient transporters, surface structures, or stress-response systems. In Escherichia coli, for example, the species-wide repertoire includes genes associated with diverse ecological and host-associated lifestyles. A gene that is accessory at the species level can be nearly universal within a specialized lineage. Conversely, apparent associations with a habitat can arise because closely related organisms share ancestry, rather than because the habitat independently selected each gene.
Genome reduction creates a contrasting pattern. Host-restricted bacteria may lose functions supplied by the host environment and have limited opportunities to acquire DNA from unrelated organisms. Such lineages can display relatively restricted gene repertoires. This is not a universal property of host association, because mobile elements and host transitions can also generate substantial variation.
In eukaryotes, sexual reproduction reshuffles variation among chromosomes, while gene duplication, transposable element activity, and polyploidy can profoundly alter genome content. Plant genomes frequently contain large repeat-rich regions and duplicated gene families. Human genomes share most genes but differ in sequence, copy number, and structural organization. These contrasts explain why a bacterial gene-family matrix, a crop structural-variation catalogue, and a human haplotype graph may all be called pangenomes despite emphasizing different evolutionary phenomena.
Sampling, core size, and open or closed pangenomes
Every empirical pangenome depends on which genomes were included. Sampling many nearly identical outbreak isolates provides a different view from sampling across geography, ecological niches, and deep evolutionary lineages. The first collection may characterize recent transmission extremely well while missing much of the species-wide repertoire. Accordingly, the number of genomes alone is not a sufficient measure of coverage. Their relatedness, provenance, assembly quality, and distribution across the population are equally important.
Researchers often examine accumulation curves by adding genomes sequentially and counting the gene families discovered. Because the order of addition changes the curve, analyses commonly average over many random orders or use combinations of subsets. The observed pangenome generally grows with sample size, while the strict core can only remain constant or shrink. Neither trend alone demonstrates a biological mechanism: adding a divergent lineage or introducing systematic annotation errors can strongly alter both curves.
An open pangenome is one for which the inferred repertoire continues to grow substantially as additional genomes are sampled. A closed pangenome is one whose accumulation curve appears to approach a finite repertoire under the chosen model and sampling conditions. These terms describe an extrapolation, not proof that a species has an infinite or permanently exhausted gene pool. Evolution can introduce new genes even after a particular dataset appears saturated.
One commonly used family of models expresses the expected number of newly detected families in the th genome as
where and . Under an indefinitely extrapolated version of this model, the cumulative sum diverges for and converges for . Other studies model total pangenome size directly, often with a power law plus an intercept. Parameter values cannot be compared without checking the formula: an exponent for total repertoire growth does not have the same interpretation as an exponent for newly discovered families.
Rare genes create particular statistical difficulties. A family observed once may be genuinely restricted, part of a mobile element, a contaminant, or an annotation artifact. Because rare families strongly influence estimates of unseen diversity, errors can inflate apparent openness. Conversely, highly stringent family clustering can merge distinct genes and conceal diversity. Models also become unreliable when they assume equivalent sampling opportunities but the dataset heavily overrepresents a few lineages.
Core size is similarly conditional. If a gene has a small probability of being missed in each assembly, demanding its detection in every genome becomes increasingly restrictive as sample size grows. Soft-core thresholds reduce this problem but introduce a convention that must be reported. Robust studies examine sensitivity to thresholds, genome completeness, and population structure rather than treating a single core count or openness estimate as an intrinsic species constant.
Sequencing, assembly, and gene-family construction
Pangenome construction begins with biological sampling and DNA sequencing, but the critical computational decisions depend on the intended representation. A bacterial gene-content analysis usually requires assembled genomes and consistently predicted genes. A sequence-resolved human or plant pangenome requires accurate assemblies, identification of corresponding regions, and explicit treatment of homologous chromosomes. Combining datasets produced with different methods is possible, but differences in laboratory and computational processing can masquerade as biological variation.
Genome assembly reconstructs longer sequences from sequencing reads. Short reads are economical and can provide accurate base-level information, but repeated regions often exceed their length, leaving assemblies fragmented. Long reads can span repeats and large insertions, while high-accuracy long-read technologies reduce the trade-off between read length and sequence accuracy. Additional information, including parental data or chromosome-conformation measurements, can help assign sequences to chromosome-scale structures. Remaining errors include collapsed duplications, incorrectly joined regions, and missing sequence.
For diploid organisms, separating the two parental chromosome copies is particularly important. A haplotype-resolved assembly preserves variants that occur together on the same chromosome. If the two copies are collapsed into a single consensus, one allele may disappear, alternative segments may be merged, and copy-number estimates can become misleading. Polyploid organisms add the challenge of distinguishing closely related chromosome sets as well as ordinary allelic variation.
Gene-centred workflows next predict coding sequences and group related genes into families, often using similarities among their translated protein sequences. This step must distinguish orthology, the relationship between genes separated by speciation, from paralogy, the relationship created by duplication. Similarity thresholds alone do not always resolve that distinction. Closely related paralogues may be merged, whereas rapidly evolving versions of the same ancestral gene may be split. Genomic neighbourhood, gene trees, and conserved domain structure can improve clustering.
Consistent annotation is essential. A gene interrupted at a contig boundary may be predicted as two fragments, creating spurious families. A genuine frameshift can produce a pseudogene, but a sequencing error can produce the same apparent signal. Some pipelines revisit the original DNA near expected gene locations to distinguish absence from failed prediction. Widely used microbial tools, including Roary, Panaroo, and PPanGGOLiN, make different choices about clustering, neighbourhood information, and partitioning the repertoire.
The resulting resource often includes a presence–absence matrix, representative family sequences, annotations, and quality-control information. Sequence-based projects additionally retain assemblies and alignments so that alternative structures can be examined directly. Reproducibility requires recording software versions, parameters, sample identities, and filtering decisions. A high family count is not inherently evidence of a richer biological pangenome; it may instead indicate inconsistent annotation, contamination, or excessively fragmented clustering.
Graph representations and the problem of reference bias
A conventional reference genome presents a linear sequence for each chromosome, sometimes supplemented with alternative loci. This arrangement is convenient for naming positions and comparing datasets, but it privileges the sequence included in the reference. Reads from an individual carrying a divergent allele or a large insertion may align poorly, align to an incorrect location, or fail to align. The resulting reference bias can distort variant discovery, estimates of allele frequency, and measurements of gene expression when sequence differences affect read mapping.
A variation graph represents alternative sequences within a connected structure. In a common formulation, nodes carry sequence, edges specify permitted adjacencies, and paths trace individual chromosomes or haplotypes through the graph. A region with two alleles can branch into alternative paths and rejoin downstream. Insertions, deletions, duplications, and some rearrangements can therefore be represented alongside single-base substitutions rather than being described only as differences from one privileged sequence.
Not all pangenome graphs have the same structure or purpose. Gene-order graphs represent relationships among gene families and their neighbourhoods; sequence graphs operate at nucleotide resolution. Some graphs are constructed by adding variants to an existing reference. Others derive from alignments among independently assembled genomes. De Bruijn graph approaches organize sequence through overlapping words of fixed length. A graph built mainly around large structural differences will not necessarily represent every small variant in its constituent genomes.
Graph-based mapping seeks locations where sequencing reads match one or more paths. Representing alternative alleles can improve alignment in regions poorly captured by a linear reference, but the benefit depends on graph quality and indexing methods. Repeated sequences may create many similar routes. Excessive graph complexity increases computational demands, and a graph can permit combinations of edges that no sampled chromosome actually carries. Retaining haplotype paths helps distinguish observed chromosome arrangements from merely possible traversals.
Coordinates are another major difficulty. A linear reference gives a position a simple chromosome-and-base address. In a graph, an equivalent location may require a node, an offset, an orientation, and a path context. Rebuilding the graph can change node identifiers even when much of the biological sequence is unchanged. Stable naming, translation back to established references, and versioned path systems are therefore central infrastructure rather than minor formatting details.
Graph representation does not automatically eliminate bias. A graph constructed from an unrepresentative collection still omits unsampled diversity, while difficult repeats and duplicated regions can remain unresolved. Nor does every research question require a graph: a curated gene-family catalogue can be more interpretable for some microbial analyses. The appropriate representation depends on whether the task concerns gene repertoire, structural organization, read alignment, or population-level variant inference.
Microbial ecology, pathogens, and gene-content diversity
Microbial pangenomics connects genomic variation with ecological differentiation. The accessory repertoire can encode nutrient acquisition, environmental sensing, surface modification, and interactions with other organisms. In marine bacteria such as Prochlorococcus, comparisons among isolates and environmental sequences have revealed extensive diversity associated with different environmental conditions. Genes involved in nutrient utilization can vary among populations even when basic cellular functions remain highly conserved. Such patterns help explain how related organisms occupy distinct ecological niches.
A useful distinction is between a species pangenome and a community metagenome. A pangenome organizes variation among members of a defined group, whereas metagenomics samples DNA from a community containing potentially many species. The approaches intersect when metagenomic data are used to reconstruct genomes or measure the distribution of known gene families. However, community gene abundance cannot be interpreted directly as the number of strains carrying a gene: abundance also depends on organism frequency, gene copy number, and the ability of reads to distinguish similar sequences.
Pathogen studies use pangenomes to investigate variation in host interaction, surface antigens, and antimicrobial resistance. Resistance determinants may occur on plasmids, transposons, or chromosomal regions, and their mobility can produce similar resistance phenotypes in otherwise distinct lineages. Conversely, closely related isolates may differ because one has acquired or lost a mobile element. A gene-content comparison therefore complements, rather than replaces, sequence-based analysis of conserved genomic regions.
Pangenome thinking also influenced reverse vaccinology, in which genomic information is used to identify candidate antigens. Comparisons across strains can show whether a candidate is widely distributed and how variable its sequence is. In group B streptococcus, early comparative work emphasized that a single strain could omit antigens present elsewhere in the species. Presence alone is insufficient for establishing a useful target: surface exposure, expression during infection, sequence variation, and experimental evidence of immune recognition also matter.
In outbreak investigation, core-genome variation can provide a relatively stable basis for estimating relatedness, while accessory elements offer additional information about recent acquisition and loss. Mobile elements can evolve too rapidly or move too freely to serve as straightforward markers of ancestry. Their absence in a draft assembly is also less reliable than a confidently observed sequence difference in a well-covered conserved locus.
Microbial pangenomes further complicate the relationship between taxonomy and function. Members of a named species may share a recognizable genomic backbone yet differ substantially in ecological capabilities. At the same time, broad functional categories can conceal important distinctions: two strains may both encode a transporter but recognize different substrates. Interpreting microbial diversity therefore requires integrating gene content with sequence, genomic context, experimental physiology, and environmental distribution.
Plant pangenomes and crop improvement
Plant pangenomes developed partly in response to the limited diversity represented by individual crop reference genomes. Domestication and breeding preserve some alleles while reducing others, and a reference assembled from one cultivar may omit sequences found in landraces, wild relatives, or geographically distinct breeding material. Crop pangenomes aim to describe this broader repertoire, including presence–absence variation, copy-number differences, and structural variation such as large insertions, deletions, inversions, and translocations.
Plant genome architecture makes the task demanding. Many crops have repeat-rich genomes, large gene families, and histories of whole-genome duplication. Wheat combines a large genome with closely related subgenomes; maize contains extensive transposable-element variation; rice varieties can differ in substantial stretches of sequence absent from a single reference. Distinguishing allelic differences from duplicated copies is essential because an apparent gene absence may instead reflect an unresolved assembly or incorrect assignment to a related chromosome.
Several major datasets demonstrate the scale of the field. A 2020 tomato study used 100 diverse accessions to construct a pangenome-scale catalogue of structural variation and investigated relationships with gene expression and agricultural traits. In 2021, chromosome-scale assemblies of the 26 founders of the maize nested association mapping population provided a resource for comparing genome structure and gene content across diverse lines. These projects went beyond adding short sequences to a reference: they enabled direct examination of alternative genomic arrangements.
Plant disease-resistance genes are especially relevant because they frequently occur in variable, duplicated clusters. Some encode receptors involved in detecting pathogens, and their number and sequence can differ among accessions. A linear reference lacking a particular resistance-associated region may prevent conventional analyses from observing it directly. Pangenome resources can expose that missing variation, although demonstrating a causal resistance function still requires genetic and experimental evidence.
Pangenomes also support trait association and breeding-marker development. An insertion that contains a gene, or a duplication that changes dosage, may explain variation not captured well by nearby single-base markers. Integrating presence–absence variants with a genome-wide association study can therefore broaden the set of testable genetic hypotheses. The analysis must still account for relatedness, breeding history, and environmental effects; a newly visible variant is not automatically causal.
A super-pangenome extends sampling across several related species, often including wild relatives of a crop. This can reveal sequences lost during domestication or absent from cultivated breeding pools. Its interpretation is more difficult because evolutionary divergence blurs the boundary between allelic alternatives and different genes. Practical use also depends on whether a desirable sequence can function in the crop’s genomic background and whether linked traits accompany its introduction. A sequence catalogue identifies possibilities, not guaranteed agronomic outcomes.
The human pangenome and population representation
The human pangenome extends reference genomics from a largely linear framework toward a representation of multiple human chromosome sequences. The established human reference has been indispensable, but it is not the genome of an average person, nor does it contain every sequence present in human populations. Much of its original sequence came from a limited donor set, and some regions were incomplete or difficult to resolve. Human pangenomics addresses both missing sequence and the asymmetric treatment of variants as departures from one reference arrangement.
A key distinction separates a more complete single reference from a population pangenome. In 2022, the Telomere-to-Telomere Consortium published the T2T-CHM13 assembly, resolving many regions missing from earlier human assemblies. This resource greatly improved access to centromeric, duplicated, and other difficult sequences, but one highly complete assembly cannot represent population-wide diversity. Completeness along a chromosome and diversity across people are complementary goals.
In 2023, the Human Pangenome Reference Consortium published a draft reference based on assemblies from 47 genetically diverse individuals, representing 94 haplotypes. The associated analyses reported approximately 119 million bases of euchromatic polymorphic sequence added relative to GRCh38. That figure describes sequence represented across the resource, not an increase of that size in every individual genome. The project also demonstrated improved representation and detection of variation when compared with conventional reference-based approaches.
Human pangenomes are particularly relevant to regions containing segmental duplications, variable gene copy numbers, or highly divergent haplotypes. The major histocompatibility complex is a prominent example of extensive allelic diversity and complex genomic organization. Other difficult regions contain gene families with closely related copies, where short reads may map ambiguously. Accurate assemblies and haplotype-aware comparisons help distinguish genuine copy-specific differences from alignment artifacts.
The biological interpretation of human variation requires care. A pangenome can improve the observation of a variant without establishing its relationship to disease. A sequence absent from an older reference is not necessarily rare, harmful, or newly evolved. Likewise, a structurally complex allele may be benign. Clinical interpretation depends on independent evidence, including population frequency, functional effects, inheritance patterns, and the strength of associations with a phenotype.
Representation cannot be reduced to a small set of continental labels. Human variation reflects overlapping histories of migration, admixture, relatedness, and local population structure, with substantial diversity within named populations. Increasing donor diversity reduces some forms of reference bias but does not make a resource exhaustive. Rare variants, underrepresented communities, difficult genomic regions, and the distinction between inherited and somatic variation remain important boundaries of current reference resources.
Association, functional interpretation, and responsible infrastructure
A major analytical use of pangenomes is relating genomic variation to phenotypes. In microbial studies, the tested feature may be a gene family, a short DNA word, or a compacted sequence segment. In plants and humans, analyses may include structural alleles, copy-number states, or graph-defined variants. These approaches broaden the observable variant space beyond sites already present in a linear reference. They also increase the number and complexity of statistical tests.
Population structure is a central source of confounding. If a bacterial lineage both carries a gene and commonly exhibits a phenotype, a simple association may implicate the gene even when another inherited feature is responsible. Similar problems occur in crop panels and human cohorts. Methods based on kinship, lineage structure, or phylogenetic relationships can reduce this problem, but closely linked variants may remain difficult to separate. Independent populations, recombination events, and experimental perturbations can provide additional evidence.
Functional interpretation must distinguish several levels of claim. Detecting a sequence establishes its observed presence; annotation proposes a function; association links it statistically to a trait; experimental work may establish a mechanism. A family labelled with a broad metabolic function may contain members with different substrate specificities. A gene may also be present but disrupted, silenced, or dependent on another locus. Integrating pangenomes with transcriptomic measurements, proteomics, and targeted experiments can narrow these gaps without making all genomic differences functionally equivalent.
As resources expand, computational maintenance becomes a scientific issue. Adding genomes can change family boundaries, graph topology, coordinate mappings, and variant identifiers. Analyses require versioned releases, traceable sample metadata, documented quality criteria, and clear distinctions between absent sequence and missing information. Benchmarking is complicated because no universally complete truth set exists for every genomic region or variant class. Results measured in relatively accessible regions may not describe performance in repeat-rich or highly duplicated regions.
Human resources additionally involve data privacy, consent, and governance. Genome sequences can reveal information about relatives as well as donors, and restricting direct identifiers does not remove every identification risk. Public release, controlled access, and downstream reuse must follow the conditions under which samples were obtained. Engagement with participating communities is particularly important when historical underrepresentation is addressed through new sampling; inclusion in a dataset does not by itself settle questions of control, benefit, or appropriate use.
The interpretation of absence remains a practical limit across all pangenome systems. A sequence not observed may be biologically absent, unassembled, unannotated, filtered out, or simply outside the sampled population. Reliable resources preserve enough evidence to distinguish these possibilities, including read support, assembly quality, annotation provenance, and the precise population to which an inference applies.