← all systems · see it in The System
Fitness landscape (adaptive landscape)
Formally, fitness landscapes combine numeric fitnesses and a mutation-graph into a combinatorially structured space where each vertex is a possible genotype (or phenotype). A genotype is a local fitness peak (or local fitness optimum) if no adjacent genotype in the mutation-graph has higher fitness.
Properties published scienceour proposal
Sourced fields, and our own readings.
| aliases | Fitness landscape (biology) |
| born at | with the first replicators (not dated in the sources read); the idea 1932 |
| born precision | bracket only for the structure; the idea dated to the year |
| boundary | Inside: the space of genotypes (or gene frequencies, or phenotypes) joined by single mutational steps, the fitness each point has in a given environment, the shape that results (peaks, valleys, ridges, holes, basins), and the movement of populations across it (adaptive walks, valley crossing, contingency). Outside: the molecular machinery that copies and reads genomes, the environment's own dynamics, the organisms' physiology, and the people who map and use it. |
| nested level | a structure carried by every population of replicators with heritable variation in fitness; the terrain natural selection and evolution move through; also a model family held by evolutionary biology and borrowed by optimisation, management and technology studies |
| scale | organism |
Measures (24) published science
| characteristic clock: generations of the E. coli long-term evolution experiment without reaching a peak | 50000 generations |
| characteristic clock: adaptive walk length to a high DHFR peak (simulated, strong selection weak mutation) | 5.6 ± 2.1 mutational steps |
| dimensions Wright needed to draw a species' field of gene combinations | 5 for 32 combinations of 5 loci (plus one for adaptive value); 9000 for 1000 loci with 10 alleles each dimensions |
| expected number of peaks on a House-of-Cards landscape | 2^L/(L+1) peaks (L loci, two alleles) |
| measurements needed to know a 10-locus landscape | 10 (additive) against 1024 (House-of-Cards) fitness measurements |
| resistance gain from five TEM beta-lactamase mutations | about 100,000 fold |
| accessible mutational trajectories, TEM beta-lactamase | 18 of 120 (102 inaccessible) trajectories |
| accessible mutational trajectories, first five LTEE substitutions | 86 of 120 trajectories |
| genotypes mapped, DHFR landscape | >260,000 (261,382 of 262,144 measured) genotypes |
| fitness peaks, DHFR landscape | 514 (73 high, 33 intermediate, 408 low) peaks |
| share of accessible shortest paths by distance to a high DHFR peak | 86, 62, 39, 21 at 2-5 steps; 1 at 9 steps % |
| share of simulated adaptive walks reaching a high DHFR peak | 76.5 % |
| share of mutant pairs with reciprocal sign epistasis, DHFR | 12.5 (8-22 in other complete landscapes) % |
| variants measured, GB1 four-site landscape | 160,000 variants |
| share of beneficial variants, GB1 | 2.4 % |
| share of background-dependent substitutions, yeast his3 | 85 (over >4 million genotypes) % |
| share of interacting mutation pairs, Hsp90 9-residue region vs yeast double knockouts | almost 50 vs about 3 % |
| fitness peaks, Darwin's ground finches on Santa Cruz | 5 fitness peaks, 4 adaptive-landscape peaks (17-year dataset) |
| share of evolving populations reaching a strong-regulation peak, E. coli binding-site landscapes | about 10 % |
| management simulation studies using the NK model, 1997-2018 | 71 studies |
| best mix of search in an NK service model | 90% local search, 10% long jump share of moves |
| somatic hypermutation steps used to tune K in Kauffman's antibody NK model | about 6-8 steps |
| per-position point-mutation rate in E. coli used by Papkou et al. (×9 for their nine-nucleotide target) | 2.2e-10 per position per generation (× 9 positions) substitutions |
| detection limit of the DHFR fitness assay | differences below 2.5% (present between 16% of mutational neighbours) relative fitness |
Open contentions (5) open contention
Where sources or readings disagree. Both stay on the map.
- C-3839 How much do interactions between mutations steer evolution?
booked: The map keeps the competing readings; none is ruled.
the other reading: a transient, second-order effect that matters only at intermediate allele frequencies, treated as noise in quantitative genetics (as Bank reports the other view) - C-3859 Is the 'fitness landscape' picture a help or a hindrance to understanding evolution?
booked: The map keeps the competing readings; none is ruled.
the other reading: a superficial metaphor to be replaced (de Visser and Krug; Kaplan 2008, title only; multiscapes, Catalan et al.); a unifying conceptual tool that should not be abandoned when its models fail (Petkov 2015) - C-3884 Does a rugged fitness landscape trap evolution on low peaks?
booked: The map keeps the competing readings; none is ruled.
the other reading: classical theory and some measured landscapes: ruggedness traps populations and closes paths (Wright's worry; NK models; 102 of 120 TEM paths closed; RNA peaks largely isolated); hard landscapes (Kaznatcheev 2019): worst-case constructions where reaching even a local peak takes exponential time; for all dynamics conditional on FP vs PLS; how common they are is an open empirical question - C-3889 Does evolution cross fitness valleys by Wright's 'shifting balance' among small local populations?
booked: The map keeps the competing readings; none is ruled.
the other reading: Coyne, Barton and Turelli 1997: drift often unnecessary for peak shifts; phase III faces major obstacles; not an important explanation of adaptation; Phase III 'one of the most controversial aspects' (Wade 2013, reporting the critics: migration must be intermediate; interdemic selection seen as unnecessary) - C-3896 Are evolution's fitness peaks separated by valleys or joined by ridges?
booked: The map keeps the competing readings; none is ruled.
the other reading: peaks joined by long neutral ridges or networks across a nearly flat, holey surface (Gavrilets; neutral networks; Catalan et al.); compensation after one or more neutral steps: with neutral mutations pairwise reciprocal sign epistasis may be absent from a multi-peaked landscape (Ivankov and Zorin 2026)
Cross-scale patterns seen here (17) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Fitness landscape (adaptive landscape). Each links to every other system that shows it.
- nested ladder (structural)
Kinds of epistasis, each a stronger form of the last: magnitude epistasis (diminishing returns) without extra peaks; sign epistasis that removes paths; reciprocal sign epistasis, necessary for multiple peaks when no neutral mutations exist (Poelwijk's theorem as Ivankov and Zorin 2026 qualify it; wi - same role, different builds (structural)
The same job, resisting trimethoprim, done by genotypically different peaks: walks 'reach genotypically different high fitness peaks with similar trimethoprim resistance'; parallel populations evolve similar phenotypes via distinct mutations. - same toolkit, independent builds (structural)
Several insect lineages adapting to cardiac glycosides (monarch butterflies among them) used the same three substitutions, in an order constrained by epistasis (Karageorgi et al. 2019, through Bank). - templating (structural)
One formal module, the NK landscape, reused unchanged across systems: from evolutionary biology into antibody maturation (Kauffman), management (Levinthal 1997; 71 studies), technology design (Frenken), customer service (Li et al. 2022) and cancer models; and the landscape itself shared with physics - trade one property for another (structural)
Resistance bought with growth: 'The genotype with the highest IC50 also exhibits the lowest drug-free growth rate, suggesting a trade-off, or cost, to resistance.' Also in theory: more peaks, shorter walks, less of the landscape explored (Bank). - Umwelt: each system carves its own slice of one shared medium (structural)
One landscape, different populations see different slices: in E. coli 'even tiny differences in fitness are visible to natural selection'; in small populations bottlenecks 'smoothen the shape of the fitness landscape perceived by an evolving population'; the epistasis a population meets through its - causal ceiling (partial)
Kaznatcheev's hard landscapes, as worst-case constructions: for strong-selection weak-mutation dynamics some starts need exponentially long paths; for all evolutionary dynamics the claim rests on the open FP vs PLS conjecture ('this conjecture remains open'); how common hard landscapes are is 'an op - coordination-before-mapping (partial)
Populations reach high peaks with no map: selection 'must rely only on local information from the limited genetic variation of nearby mutants', yet 76.5% of DHFR walks reach a high peak. - namespace completeness (the map predicts its gaps) (partial)
Combinatorially complete landscapes enumerate every genotype of a sequence space and find members nature lacks: functional 27Cys DHFR variants absent from 5000 proteobacteria; none of the high-fitness Hsp90 genotypes found in natural sequences. - same shape, opposite sign (partial)
Selection on one trait run in reverse: selecting first for low cefotaxime resistance (negative selection) in TEM-15 beta-lactamase opened a gateway that selection for high resistance could not reach. Also the formal one: physicists minimise a potential, biologists maximise fitness ('taking the inver - seed (partial)
Biparental reproduction as the unpacker: 'a limited number of mutations ... furnish an almost infinite field of possible variations'; the landscape is the combinatorial field (sequence space larger than the molecules of the known universe). The Seed here is sexual recombination; the landscape is wha - symmetry breaking (name the cut) (partial)
One start, many ends: from the same variant 1000 walks reached 31 different high peaks; the cut is the chance first steps inside overlapping basins (47.5% of variants can reach all 74 high peaks). - The carry (partial)
The wait face: 'long periods of stasis in which the population waits for new adaptive paths to open up'; Wright: after a wholly new favourable mutation 'the time required for attainment of sufficient frequency to be subject to selection to an appreciable extent is enormous'; no peak after 50,000 LTE - The Payment to Entropy (partial)
Genetic form: mutation is the landscape's move and its toll. More mutation means 'an average lowering of the adaptive level' (Wright); adaptation needs 'a surplus of beneficial over deleterious changes', a theorem linked to stochastic nonequilibrium thermodynamics. - the whole sets the local reference (partial)
Fitness is read against the population: 'a landscape is in general not an absolute but a relative function'; DHFR fitness relative to wild type; frequency dependence makes one genotype's fitness depend on the others present. - who opens the door (partial)
Doors opened ahead of time by an earlier step: in influenza NP individually deleterious substitutions 'enabled the virus to subsequently accumulate beneficial immune-system evading substitutions'; in TEM-15 an initially deleterious mutation is 'an initial gateway' to otherwise inaccessible sequence - red-queen (surface)
Red Queen effects named as one cause of seascapes: 'an interaction between two species which dynamically affects each one's fitness landscape'. No stalemate data in the sources read.
Valence: what it trades, and with whom (14) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Protein via sys:protein · Fitness landscape (adaptive landscape) gain
- Human (Homo sapiens) via sys:human · Fitness landscape (adaptive landscape) mixed
- Natural selection via sys:gene-pool · Fitness landscape (adaptive landscape) gain
- Statistical mechanics via sys:mental-representation · Fitness landscape (adaptive landscape) gain
- Mutation via sys:mutation · Fitness landscape (adaptive landscape) gain
- Habitat and ecosystem via sys:ecosystem · Fitness landscape (adaptive landscape) mixed
- Genetic recombination via sys:genetic-recombination · Fitness landscape (adaptive landscape) mixed
- dihydrofolate reductase via sys:protein · Fitness landscape (adaptive landscape) gain
- Evolution (biological evolution across time) via sys:gene-pool · Fitness landscape (adaptive landscape) gain
- Genetic drift via sys:gene-pool · Fitness landscape (adaptive landscape) mixed
- Technology via sys:mental-representation · Fitness landscape (adaptive landscape) gain
- Organization via sys:mental-representation · Fitness landscape (adaptive landscape) gain
- Beta lactamase via sys:protein · Fitness landscape (adaptive landscape) gain
- Natural environment via sys:natural-environment · Fitness landscape (adaptive landscape) mixed
Partners and parts published science
Built from / builds (4)
Sits inside / holds (2)
expressed by (6)
- Recombination/Composition
- Niche Construction (alter environment)
- Quality Improvement/Refinement
- Search/Foraging
- Uncertainty Reduction/Learning
- Quality Control & Error Correction
floor filled by (19)
- DNA and genes
- Protein
- Human (Homo sapiens)
- Natural selection
- String-theory vacua (the landscape; projected model)
- Mutation
- Evolution (biological evolution across time)
- Genetic drift
- Speciation
- Natural environment
- Genotype space
- Epistasis
- Genotype-phenotype map
- Holey fitness landscape
- NK model
- Rough Mount Fuji model
- Fisher's geometric model
- Neutral network (genotype network)
- Adaptive walk
represents (1)
Kinds (12)
- String-theory vacua (the landscape; projected model)
- Potential energy surface
- Genotype-to-fitness landscape
- Allele-frequency landscape (mean fitness over gene frequencies)
- Phenotype-to-fitness landscape
- Fitness seascape
- Holey fitness landscape
- Adaptive multiscape
- Dual landscape (adaptation with niche construction)
- Adaptive landscape (population mean fitness; Simpson-Lande)
- Performance landscape (management and technology)
- Waddington's epigenetic landscape
Notes published science
- Wright saw the seascape in 1932, about sixty years before it got its name: the environment is in continual change, he wrote, so 'certain of the high places are gradually being depressed and certain of the low places are becoming higher' (src:wright-1932).
- A landscape can be wildly rugged and still easy to climb: the fitness of more than 260,000 versions of one bacterial enzyme under an antibiotic gave 514 peaks, yet three in four simulated populations still reached one of the high ones, and walks from the very same start ended on 31 different peaks (src:papkou-2023-dhfr).
- Going downhill first can be the way up: in a lab directed-evolution experiment, people selected a resistance gene first for LOWER antibiotic resistance; that let an initially harmful mutation open a gateway to a part of sequence space the uphill-only route could not reach, and produced better alleles than three other strategies (src:steinberg-2016-valleys).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
fitness landscape: from genotype space to adaptation, as evolutionary geneticists describe it (reader's view)
- L1: Genotype space: every possible genotype (Wright's 'field of possible gene combinations'), much larger than the number of molecules in the known universe
- L2: The genotype-phenotype map: many genotypes to one phenotype; ridges in two dimensions are neutral networks in many
- L3: The phenotype meets its environment: the phenotype-fitness map (Fisher's single optimum; a dose-response curve per genotype)
- L4: Fitness of each genotype and its epistasis: magnitude, sign and reciprocal sign, higher-order and global epistasis
- L5: Topography: peaks, valleys, ridges, plateaus, holes and basins; smooth, rugged or holey
- L6: Dynamics on the landscape: adaptive walks, accessible paths, trapping, valley crossing, contingency
- L7: Evolutionary outcomes: adaptation, predictability or contingency, divergence and speciation, and people's steering of evolution
the boardroom (shared floors)
- L5: The landscape's shape: fitness over genotype space with peaks (genotypes fitter than all their one-mutant neighbours), valleys, ridges and basins; smooth (additive, one peak) through rugged (many peaks; on average 2^L/(L+1) at maximum epistasis) to holey (a nearly flat surface with holes of incompatible genotypes). Peaks are defined against the move set (single mutations; recombination changes the connections). Reciprocal sign epistasis is necessary for multiple peaks only when no neutral mutations exist
- L6: Movement on it: adaptive walks up accessible (monotonically fitness-increasing) paths, trapping on local peaks, valley crossing by drift, subdivision, indirect paths or environmental change, overlapping basins of attraction and contingency. Two surfaces: adaptive walks of single genotypes (strong selection, weak mutation) move on the genotype graph; gradient climbing of mean phenotypes and Wright's drift of gene frequencies move on the mean-fitness adaptive landscape
- L7: What it yields: adaptation to (often local) peaks, repeatability or contingency of outcomes, divergence and reproductive isolation, open-ended evolution on hard landscapes, and predictions people use to steer evolution
way in: Environment: what sets each genotype's fitness, and moves it
- L1: The environment, living and non-living, in continual change: high places depressed, low places raised (Wright); fluctuating optima turn the landscape into a seascape
- L2: The phenotypic optimum: in Fisher's geometric model fitness falls (Gaussian) with distance from one optimum; a new environment moves the optimum
- L3: Dose-response: genotypes mapped to drug dose-response curves (IC50, drug-free growth); the most resistant genotype grows slowest without drug
- L4: Environment-dependent epistasis: the same 32 E. coli genotypes give vastly different landscapes across eight environments
way in: Molecular structure: interactions inside the gene product (the inner channel)
- L1: The physical organisation of a protein: locally interacting residues and fine balances in its sequence composition
- L2: Biophysical pleiotropy: a missense mutation changes several properties at once; held responsible for TEM beta-lactamase's closed paths
- L3: Frustration between stability and aggregation: extensive sign epistasis and local peaks galore (lattice proteins)
- L4: Epistasis in the genotype-phenotype map: about half of Hsp90 mutation pairs interact, against about 3% of double gene knockouts
way in: Mutation: the steps that join genotypes
- L1: Mutation supply: new genotypes arise one copying error at a time; Papkou et al. multiply a per-position point-mutation rate for E. coli of 2.2e-10 per generation by the nine positions of their target ('the extremely low mutation rate for the nine-nucleotide mutational target'); more mutation lowers the species' average adaptive level (Wright)
- L2: The mutational neighbourhood: each genotype one remove from its single mutants (5 of 32 in Wright's five-locus figure; 27 one-mutant neighbours for a functional DHFR variant)
- L3: The mutation graph: genotypes joined by single mutational steps, a hypercube of 2^L genotypes for L biallelic loci and a Hamming graph for sequences with more letters per site (20^L for proteins); a peak is defined against these neighbours
- L4: Which neighbour is taken: under strong selection and weak mutation the population sits at one genotype and hops to a neighbour by rare substitution
way in: Other organisms and the population's own effects: niche construction, coevolution, frequency dependence
- L1: Other species and the population's own effects on its world
- L2: Niche construction: organisms alter the environment, so moving across a landscape is like moving on a trampoline or water bed (Lewontin)
- L3: Red Queen effects: an interaction between two species dynamically affects each one's landscape
- L4: Frequency dependence: competition makes one genotype's fitness depend on the frequency of others; a landscape is relative, not absolute
way in: People: drugs, directed evolution, breeding and measurement
- L1: People's acts on evolving populations: drug regimens, directed evolution, gene editing
- L2: Drug exposure varies in time and space through absorption, distribution, metabolism and clearance
- L3: Selection schemes: selecting first for low resistance opened a gateway in TEM-15 beta-lactamase
- L4: Measurement: site-directed mutagenesis, CRISPR and deep mutational scanning measure thousands of genotypes; the fitness proxy chosen (growth, MIC, fluorescence, binding) shapes the measured landscape
way in: Recombination: new combinations of alleles already present
- L1: Biparental reproduction: a limited number of mutations furnish an almost infinite field of possible variations
- L2: Recombination produces new combinations of (multiple) segregating alleles, changing the connectedness of genotypes in the landscape
- L3: On rugged landscapes recombination dilutes the genotypes needed to cross valleys; recombining populations can stall at local optima
- L4: On holey landscapes, in large populations, recombination selects for genetic robustness, slowing the build-up of incompatibilities
way in: The population itself: its size, subdivision and drift
- L1: The population on the landscape: its numbers, subdivision into local races, and the field of genotypes it occupies
- L2: Accidents of sampling: gene frequency differs a little each generation by chance and may wander far from equilibrium (Wright)
- L3: Drift against selection: in E. coli (effective size above 1e8) even tiny fitness differences are visible to selection
- L4: The landscape the population perceives: bottlenecks smoothen it; the epistasis met through segregating variation differs from the complete landscape
way out: A substitution fixes: the population moves one step (the next gene pool)
- L1: The population now sits at a neighbouring genotype, the start of the next step
- L6: Each step taken with Kimura's fixation probability for the mutant's fitness difference
way out: Genomes carry the path taken
- L1: Epistasis creates detectable contingencies between substitutions in distantly related species
- L7: Contingent trajectories and environment-dependent epistasis may leave the population's ecological history in its genomes
way out: Landscape knowledge used by people: drug sequencing, evolutionary control, protein engineering, optimisation and management search
- L1: Treatment changes the selective environment itself (drug choice, dose, timing)
- L7: Evolutionary control: treatment as a dynamic decision problem updated as the population evolves; in optimisation, a fitness function over all candidate solutions
way out: Reproductive isolation: populations on opposite sides of a hole
- L1: Lineages split: incompatibilities accumulate with genetic distance; drift promotes and recombination hinders it
- L7: Populations diverge along clusters of well-fit genotypes and become species on opposite sides of a hole
way out: The population's own effects back on its environment
- L1: The environment altered by the organisms themselves
- L6: As the organism moves up, the plank tips the other way (the seesaw case of the dual landscape model)