Pangenome core-versus-auxiliary classifications depend not only on gene presence, but also on which genomes are sampled, how many genomes represent a clade, and how completely genes can be linked across datasets. The Conservation vs Fitness — Linking FB Genes to Pangenome Clusters shows that these design choices can make the core boundary more or less discriminative when Fitness Browser fitness data are integrated with KBase pangenome clusters. 1 The Fitness Effects vs Conservation — Quantitative Analysis refines this result by showing that the conservation gradient persists across approximately 194,000 genes from 43 bacteria, but is statistically robust and only weakly predictive of fitness importance. 2
Sampling depth changes the meaning of “core”
In the integrated analysis, 44 of 48 Fitness Browser organisms were mapped to pangenome species clades, yielding 177,863 gene-to-cluster links with 100.0% median protein identity and 94.2% median gene coverage. 1 Thirty-four organisms had at least 90% coverage, while 33 organisms were retained for downstream analysis after Dyella79 was excluded because of a locus-tag mismatch. 1
The classification assigned 145,821 linked genes, or 82.0%, to core clusters and 32,042, or 18.0%, to auxiliary clusters; the 7,574 singleton genes were a subset of the auxiliary category. 1 These proportions describe conservation within the sampled clades, not an unconditional boundary between biologically “core” and “accessory” functions. 1 This supports treating core status as sampling-conditioned: in the separate 43-bacterium analysis, core fractions ranged from 66% among always-neutral genes to 82% among essential genes, rather than defining a sharp fitness boundary. 2
The module-level analysis supports this sampling-conditioned interpretation while showing that conservation can also be assessed for co-regulated fitness units: module genes were 86.0% core versus 81.5% for all genes, an enrichment of 4.5 percentage points (OR=1.46, p=1.6e-87). 3 The modest absolute difference despite strong statistical significance reinforces that a high baseline core rate limits the discriminative value of the boundary. 3 Among 974 modules with at least 3 mapped genes, 577 (59%) were >90% core, 349 (36%) were mixed, and 48 (5%) were <50% core, extending the conservation question from individual genes to functional units without establishing that accessory modules are horizontally transferred or niche-specific. 3
The core_gene_tradeoffs analysis supports this qualification by identifying 25,271 condition-dependent trade-off genes, or 17.8% of the genes examined, and finding that they were enriched among core genes (odds ratio 1.29, p=1.2e-44). 4 Because trade-off status depends on whether fitness is below -1 in some conditions and above 1 in others, this result indicates that a core boundary can capture genes with variable laboratory effects rather than uniformly indispensable genes. 4 The detailed evidence is summarized in Core Gene Paradox — Why Are Core Genes More Burdensome?.
The openness analysis refines this interpretation by finding no significant relationship between a species’ open/closed pangenome summary and either environment effect (Spearman rho = -0.05, p-value = 0.54) or phylogeny effect (Spearman rho = 0.03, p-value = 0.73). 5 Thus, core/accessory boundaries and whole-pangenome openness should not be treated as interchangeable measures of ecological or phylogenetic structuring; the result is a null association among species with both pangenome and ecotype results, not evidence that individual core or accessory genes lack ecological relevance. 5
Small clades can inflate core fractions
Clades containing only 2 genomes can have trivially high core fractions because a gene present in both genomes equals 100% core, reducing the discriminative power of the core-versus-auxiliary classification. 1 Thus, a high core fraction in a small clade may reflect limited sampling rather than broad conservation across the species. 1
The module analysis supports retaining this caveat: its pangenome-linked comparison covered a 29/32 organism subset because Cola, Kang, and SB2B lacked pangenome links, with too few genomes in GTDB for pangenome construction. 3 Within the analyzed modules, family breadth did not predict conservation (Spearman rho=-0.01, p=0.914), redirecting interpretation toward gene-level conservation rather than assuming that module families spanning more organisms are more conserved. 3
This sampling effect is important when interpreting associations between conservation and phenotype. Essential genes were 86.1% core compared with 81.2% for non-essential genes, with a median odds ratio of 1.56, and 18 of 33 organisms showed statistically significant enrichment by Fisher’s exact test after Benjamini-Hochberg false discovery rate correction at BH-FDR q < 0.05. 1 The enrichment supports a relationship between essentiality and within-clade conservation, but the clade-size limitation means that the magnitude and discriminative value of the core category can vary with genome sampling. 1 The broader analysis refines that interpretation: fitness breadth was positively but weakly associated with core status (Spearman rho=0.086, p=8.1e-230), so large sample sizes do not make conservation a strong standalone predictor of fitness importance. 2
Coverage failure creates a second boundary problem
Coverage determines whether a gene can be assigned to a pangenome cluster at all, so incomplete joins can move genes out of the core-versus-auxiliary comparison or make apparent absence ambiguous. 1 Ten organisms were excluded from Phase 2 because they had less than 90% DIAMOND coverage, reducing taxonomic breadth. 1
Four organisms—Cola, Kang, Magneto, and SB2B—were unmatched because their species had too few genomes in GTDB for pangenome construction. 1 The main Escherichia coli clade was absent because it contained too many genomes, while Keio, an Escherichia coli BW25113 strain, mapped to the small s__Escherichia_coli_E clade at only 26.1% coverage. 1 These exclusions and uneven mappings make the observed core fraction partly a property of which clade was available and successfully joined, not solely a property of gene biology. 1
Dyella79 illustrates a distinct failure mode: it was excluded from Phase 2 because the Fitness Browser gene table used the locus-tag format N515DRAFT_* whereas the protein sequences used ABZR86_RS*, producing a 0% join rate. 1 This case shows that identifier incompatibility can alter the analyzed population before biological conservation is assessed. 1 The newer analysis supports retaining this caveat because singleton genes showed near-zero mean fitness under tested laboratory conditions, while explicitly noting that poor transposon coverage could produce the apparent neutrality. 2
The module analysis further refines the coverage boundary: its module memberships came from ICA, or independent component analysis, using a |Pearson r| >= 0.3 threshold and a maximum of 50 genes per module, and essential genes were absent because essential loci lack usable transposon-insertion fitness data. 3 Consequently, its 86.0% module-gene core fraction describes the measurable, non-essential portion of the genome and cannot be used to characterize essential-gene modules. 3
Implications for integrated fitness analyses
The conservation-versus-fitness result should therefore be interpreted as a clade- and coverage-conditioned association. 1 Essential genes were generally more likely to be classified as core, but the analysis also found that most genes in well-characterized bacteria were core regardless of essentiality, making the enrichment modest. 1 The broader comparison supports this qualification: essential genes were 82% core versus 66% for always-neutral genes across approximately 194,000 genes, a 16-percentage-point gradient that remained weak as a predictor. 2
The report’s stratification by clade size and lifestyle indicated that essential-core enrichment was robust across the genomic contexts examined, but the report also cautioned that small clades reduce the discriminative power of core-versus-auxiliary status. 1 This combination supports using core status as evidence of conservation while treating it as a sampling-dependent measurement rather than a fixed universal label. 1 Condition-specific fitness provides a further refinement: strong condition-specific effects were 77.3% core versus 70.3% for genes without specific phenotypes, indicating that conditionally important genes are not necessarily accessory, although the laboratory condition space remains limited. 2
The core_gene_tradeoffs result refines this condition-specific interpretation: core-gene burden was function-specific, with core genes more burdensome in Protein Metabolism, Motility, and RNA Metabolism but with the pattern reversed for Cell Wall genes. 4 Its selection-signature matrix included 28,017 Costly + Conserved genes, defined as genes conserved despite laboratory-measured cost, but this is evidence from laboratory fitness and conservation patterns rather than direct measurement of natural selection. 4 Thus, core status can mark conditionally costly functions without establishing universal essentiality or a uniform burden boundary. 4
The openness null result refines the integration boundary: a single whole-pangenome openness metric did not predict whether environment or phylogeny dominated gene-content variation, and the report identifies its limited representation of pangenome structure as a hypothesis-generating caveat rather than a demonstrated mechanism. 5 The module-family null result points to a related limitation: module-family breadth was not associated with core fraction, so cross-organism module scope should not be substituted for direct gene-level conservation. 3 This concept therefore refines pangenome-integration by identifying clade composition, join coverage, metric choice, and observability of fitness data as interpretive constraints on cross-dataset integration. 1,5,3 It also qualifies the essentiality relationship discussed in gene-essentiality, because the observed 1.56 median odds ratio depends on the sampled clades and the resulting conservation categories. 1 The proposed extension to condition-specific fitness connects this issue to condition-specific-fitness, since stress-dependent fitness effects may be compared against conservation only after sampling and coverage are controlled. 1
Prevalence thresholds, flag accounting and denominators
The core boundary is a prevalence threshold, not a biological category. In the pangenome gene_cluster table, is_core = 1 marks clusters present in ≥95% of genomes, is_auxiliary = 1 marks clusters present in <95% and >1 genome, and is_singleton = 1 marks clusters present in exactly 1 genome. The three flags are mutually exclusive within a row. 6
Counts follow a different convention from the row flags. In the pangenome table, no_core + no_aux_genome = no_gene_clusters, and no_singleton_gene_clusters is a subset of no_aux_genome. Core plus auxiliary therefore equals the total, and singletons must not be added a second time. 6 This supports the accounting used in the integrated fitness analysis, where the 7,574 singleton genes were reported as a subset of the 32,042 auxiliary genes rather than as a third category. 1
A reported core percentage also depends on its denominator. Counting all genes, including those not mapped to pangenome clusters, gave essential genes as 82% core. Counting mapped genes only gave 86% core. 6 The same source records a critical review issue in which figures showed 86%/78% while the prose said 82%/66%, and attributes the discrepancy to inconsistent denominators. No figure number is given. The source advises documenting which denominator is used and keeping it consistent within any synthesis or comparison. 6 This refines the gradient comparison recorded under Tensions below. It does not by itself establish which denominator each analysis on this page used. 6
Other projects define and use core differently. The structural-annotation analysis treated a gene as conserved when is_core = true, which it describes as presence in the majority of genomes within each of 14,768 species clades. It paired this with structural novelty, defined as msa_depth < 10: fewer than 10 detectable homologs in the AlphaFold training set, where msa_depth is the depth of the multiple sequence alignment. 7 Core status here means prevalence within a species clade, not universality across bacteria. The report reads conservation as implying strong purifying selection and likely biological function, and reads low alignment depth as implying a genuinely unprecedented fold. These are interpretations, not direct measurements. 7
The level of pangenome calling and aggregation changes the label again. In the prophage–AMR analysis, core/accessory labels depend on species-level pangenome calling with motupan, so the same gene can have different conservation status in different species. 8 In the plant-microbiome analysis, the 46.8% genome-wide core baseline is a gene-cluster-level fraction across all species. It is the share of all gene clusters classified as core by their respective species pangenomes, not a species-level average. 9 A gene-cluster baseline pooled across all species is therefore not directly comparable with gene-level core fractions among Fitness Browser-linked genes unless the unit of aggregation is matched. 9
Sparse genome sampling across projects
The clade-size caveat recurs beyond the integrated fitness analysis. Most Fitness Browser species have few genomes in GTDB, the Genome Taxonomy Database (median 9, range 2–399). This makes the ≥95% prevalence threshold for core designation imprecise. With only 2–9 genomes, a gene present in all 9 is labeled core but could be accessory at larger sampling depth. 10 Only a few species are well sampled, for example K. michiganensis with 399, B. thetaiotaomicron with 287 and S. meliloti with 241 genomes. 10 The project’s core-versus-accessory null result should therefore be interpreted cautiously for species with <20 genomes. 10
The environmental resistome analysis independently supports the same caveat. Its 95% prevalence threshold for core depends on species genome count, and because many species have few genomes, its core/accessory AMR labels are imprecise. 11
The central digest refines the clade-size picture for essentiality. Across 33 bacteria, putative essential genes (those with no transposon insertions in RB-TnSeq, random barcode transposon sequencing) were 86.1% core versus 81.2% for non-essential genes (median OR=1.56, 18/33 significant after BH-FDR, Benjamini-Hochberg false discovery rate correction). The digest describes this enrichment as real but modest, and reports that the signal is strongest in organisms with larger clades, where more genomes give more reliable core classification. 12 At the gene-family level, essentiality penetrance had a weak positive association with core fraction (rho=0.123, p=1.6e-17; 97.1% core for >80% penetrance vs 92.8% for <20%). Clade size did not predict essentiality rate (rho=-0.13, p=0.36). 13
Small pangenomes also compress effect sizes. The Nitratidesulfovibrio vulgaris pangenome has relatively few genomes, which yields a coarse core/auxiliary classification with a high baseline core fraction (76.3%) that compresses effect sizes. 14 Both that project and the costly-dispensable gene analysis note that the binary core/auxiliary classification obscures the fraction of genomes carrying each gene. Both propose a quantitative conservation metric to gain resolution and statistical power. No such reanalysis has yet been reported. 14,15
Some core-versus-non-core contrasts are partly structural. A singleton contrast is partly built into the definitions, because core genes cannot be singletons. 15 Similarly, the ≥95% prevalence definition gives core genes near-zero variance in Jaccard distance, a presence/absence dissimilarity between genomes. This partly suppresses their distance-based phylogenetic signal. 12
Species-level pangenomes may be too coarse for strain-level questions. The A. baylyi analysis relied on a species-level pangenome, which may have limited resolution compared with a population-level analysis. 16
The choice of clade can change conservation calls within a single named species. In the cystic fibrosis formulation design, only 7 G. sanguinis genomes were evaluated. Of these pathways, 7/18 (39%) amino-acid pathways and 37/39 (95%) carbon pathways met the conservation criterion. 17 The pangenome contains two N. mucosa clades. The primary analysis used s__Neisseria_mucosa_A (15 genomes), but the PROTECT isolate reference genome (GCA_003028315.1) maps to s__Neisseria_mucosa (8 genomes). The 8-genome clade shows stronger conservation: 18/18 amino-acid pathways at >95% versus 16/18, and 37/62 carbon pathways versus 27/62. It also contains 1 respiratory genome. 17 The report concludes that the primary analysis, if anything, understates the conservation claims for the anchor species. 17 Read against the small-clade caveat above, however, stronger conservation in the smaller clade cannot be cleanly separated from sampling depth. Whether the primary analysis understates conservation therefore remains a hypothesis. 17,1
Sampling is also taxonomically uneven. The 293K genomes underlying the structural-annotation analysis are not phylogenetically balanced: common taxa such as Pseudomonas and E. coli are over-represented, which influences core gene counts and MSA-depth distributions. 7 At the other extreme, E. coli was excluded from GTDB pangenome construction because it had too many genomes for species-level analysis. This supports the earlier observation that the main Escherichia coli clade was unavailable for fitness integration. 18
Fitness coverage adds a further sampling asymmetry. Fitness experiments cover only 48 organisms, so many members of the same species clade may never have been fitness-tested. The metabolic-dependency analysis notes that this may inflate the apparent conservation of pathways that are complete but untested. 19
The functional dark-matter analysis shows the sampling distribution directly. Its figure fig38_pangenome_species_distribution.png compares the full-pangenome species-count distribution with the older Fitness Browser-only distribution and shows the phylum distribution. 20
High baselines and prevalence ceilings
A high baseline core rate limits what any gradient can show. Module families were nearly all core regardless of how many organisms they spanned. The module analysis proposes that the high core-genome baseline (~82%) leaves little room for a gradient. 3 The conservation–fitness synthesis attributes its null breadth–conservation trend to the same cause: the baseline is already so high that there is no room for a gradient. 21 Both are interpretive explanations of null or modest results, not tests of them. 3,21
The module classification adds cutoffs that were chosen rather than derived. The 90% and 50% thresholds separating core, mixed and accessory modules were convenient but not biologically motivated. 3
Pathway-level validation shows the same compression. Only 7 of 48 Fitness Browser organisms have data in GapMind, a pathway-completeness predictor, which limits Tier 1 to Desulfovibrio vulgaris Hildenborough, Shewanella oneidensis MR-1, Pseudomonas putida, Pseudomonas stutzeri, Caulobacter crescentus, Sinorhizobium meliloti and Azospirillum brasilense. These are well-studied model organisms with near-complete, well-annotated core genomes. As a result, the gap between categories is small and the conservation-validation signal is compressed. 22
Applying a core-level threshold to pathway scoring can fail outright. Using gapmind_pathways with sequence_scope = 'core' for broad taxonomic comparisons yielded 0% completeness across all compartments. Core-level scoring is too stringent for genus-level or cross-species comparisons, where pathway genes may not be universally present. The recommended alternative is max-aggregation at the genus level. 6
Prevalence ceilings also limit co-inheritance tests. Most Fitness Browser genes map to core clusters (>95% prevalence). There, the phi coefficient, a correlation of gene presence/absence, approaches 0 for both cofit and random pairs because there is almost no variance to correlate. 23 Stronger co-fitness scores weakly predicted lower phi (Spearman rho=-0.109, p<1e-300 across 1.04M pairs). The report reads this as a prevalence ceiling effect, in which the strongest co-fitness pairs are near-universal core genes. That explanation is interpretive rather than tested. 23
Near-clonal species behave differently. Ddia6719 (average nucleotide identity, ANI, 99.47%) showed the strongest pairwise signal (delta=+0.093), and pseudo3_N2E3 (ANI 99.66%) also showed a positive delta (+0.026). The report attributes this to sufficient accessory gene variation despite low overall genome diversity. 23 However, the high baseline phi in these species makes their absolute phi values less interpretable. 23
Strain-resolution check of conservation classes
Where strain-level data exist, atlas conservation classes track within-species prevalence. Among AMR genes, atlas-defined Core genes were 77.3% fixed (>=95%), 22.7% variable (5-95%) and 0.0% rare (<=5%). Auxiliary genes were 0.0% fixed, 57.3% variable and 42.7% rare, and Singletons were 0.0% fixed, 21.3% variable and 78.7% rare. 24 The central digest records the same result as Atlas Core genes being 77% fixed within species and Singletons 79% rare. 12 The report concludes that this supports the core/accessory distinction as capturing real biological variation in AMR gene inheritance. Even so, 22.7% of Core AMR genes were variable rather than fixed at strain resolution, so core status is a strong but imperfect proxy for fixation. 24
Tensions
The analysis supports essential genes being more conserved within sampled clades, with 86.1% of essential genes classified as core versus 81.2% of non-essential genes and a median odds ratio of 1.56. 1 At the same time, clades containing only 2 genomes can produce trivially high core fractions, and the main Escherichia coli clade was unavailable because it contained too many genomes. 1 The tension is therefore between interpreting core status as biological conservation and recognizing that the estimate is shaped by both undersampling and exclusion of overlarge or poorly joinable clades. 1
The 86.1%-versus-81.2% enrichment also differs from the 82%-versus-66% essential-to-always-neutral gradient in the broader analysis. 1,2 These values should not be reconciled silently: they arise from different organism sets, phenotype categories, and integration procedures, so the difference is evidence that the estimated conservation boundary is analysis-dependent. 1,2
The module analysis introduces a related observability tension rather than resolving the essentiality comparison: module genes were 86.0% core versus 81.5% of all genes, but zero essential genes appeared in modules because essential genes lack usable transposon-insertion fitness data. 3 Thus, the module-level enrichment cannot be compared directly with essential-versus-non-essential core fractions as if they represented the same observable gene universe. 3
The core_gene_tradeoffs finding adds a related tension rather than resolving it: core genes were more burdensome in several functional categories, whereas non-core Cell Wall genes were more burdensome. 4 This functional reversal means that a core-enrichment or costly-conserved signal cannot be interpreted as a single genome-wide burden rule. 4
The openness analysis does not contradict these within-clade conservation findings, but it refines their scope: its null correlations concern species-level environment and phylogeny effect sizes, whereas the conservation analyses compare gene categories within sampled clades. 5,1,2 Whether openness fails to predict ecological dominance because the metric is too coarse, because relevant adaptation is concentrated in particular functional categories, or because of limited power remains unresolved. 5
The clade-size evidence is not fully aligned. The integrated analysis states that essential-core enrichment was robust across clade-size and lifestyle strata. 1 The central digest reports that the same enrichment was strongest in organisms with larger clades. 12 Both statements could hold if the enrichment persists in every stratum but varies in magnitude. The sources supplied here do not report stratum-specific effect sizes that would show this. 1,12
The sources also describe the core definition differently. The pangenome tables and the AMR fitness-cost analysis use a ≥95% prevalence threshold. 6,10 The structural-annotation analysis describes is_core as presence in the majority of genomes within each species clade. 7 The evidence here does not resolve whether this reflects a different threshold or a loose description of the same flag. Conservation statements from these projects should therefore not be pooled without checking the definition. 7,6
Open Directions
- Recompute core fractions across controlled subsamples of each clade, using repeated genome down-sampling to ask how the essential-versus-non-essential odds ratio changes with clade size. 1
- Reanalyze the 10 organisms excluded for less than 90% DIAMOND coverage with identifier harmonization and orthology-based recovery, asking whether their exclusion changes the 82.0% core and 18.0% auxiliary proportions. 1
- Compare clades with different genome counts using confidence intervals or hierarchical models, asking whether essential-core enrichment remains after explicitly modeling clade size and coverage. 1
- Extend the comparison from binary essentiality to condition-specific fitness, especially genes with fitness < -2 under stress conditions, asking whether conditionally important genes show the same sampling-sensitive conservation pattern. 1
- Reconcile the 86.1%-versus-81.2% and 82%-versus-66% gradients by rerunning both phenotype classifications on the same clades and coverage-filtered gene set, testing whether category definitions or sampling explain the discrepancy. 1,2
- Test whether the 86.0%-versus-81.5% module-gene enrichment persists after matching module and non-module genes by organism, callability, module size, and pangenome clade size, and separately quantify the essential genes that remain unobservable to ICA. 3
- Stratify pangenome openness by functional category and lifestyle, and compare openness, auxiliary fraction, Heap’s law alpha, and pangenome fluidity against environment and phylogeny effects to test whether the null result is specific to the single summary metric. 5
- Reclassify the core-gene trade-off and costly-conserved signals within matched clades and functional categories, testing whether the observed burden reversals persist after clade size, coverage, and condition space are controlled. 4,1
- Replace binary core/auxiliary labels with the fraction of genomes carrying each cluster in the N. vulgaris field-versus-lab and costly-dispensable analyses, testing whether quantitative conservation recovers effect sizes compressed by the high core baseline. 14,15
- Rerun the AMR core-versus-accessory fitness-cost comparison restricted to species with at least 20 genomes, or on down-sampled genome sets from well-sampled species such as K. michiganensis, to test whether the null result reflects threshold imprecision. 10
- Restrict co-inheritance tests to gene pairs in which both clusters are auxiliary (<95% prevalence), as the co-inheritance report proposes, to maximize presence/absence variance and escape the prevalence ceiling. 23
- Down-sample the 15-genome N. mucosa clade to 8 genomes and repeat the pathway-conservation analysis, separating clade identity from clade size in the 18/18 versus 16/18 amino-acid pathway difference. 17
- Recompute essential-gene core fractions under both the all-gene and the mapped-only denominator for the same organism set, testing how much of the gradient discrepancy recorded under Tensions is a denominator effect. 6,1