Morphogenetic Pattern Formation

Speculative If a periodic anatomical pattern — five digits, a row of palatal ridges, the spacing of hair follicles — is generated by a small network of interacting, diffusing molecules rather than read off a map the embryo already holds, then the pattern’s spacing is a parameter rather than a specification, and a parameter can be set. The condition comes first because it is not established in any vertebrate tissue: no laboratory has yet rebuilt a named vertebrate patterning network from its parts and recovered the pattern at a wavelength predicted from independently measured rate and diffusion constants. Section 7 says what that experiment would have to look like.

Established The magnitude is best stated as an exchange rate in information per cell, because that is the one quantity both sides of this argument have actually measured. The positional-information route is the best-measured patterning system in biology: in the Drosophila blastoderm the four gap genes specify a cell’s position with a standard deviation of about 1% of embryo length, and a single decoder built from wild-type data predicts the positions of nearly all 70 identifiable pair-rule stripes across seven maternal mutant backgrounds to within embryo-to-embryo variability. Holding a coordinate system to 1% of the field costs the embryo roughly 4 to 5 bits per nucleus, and a 2024 analysis puts the residual shortfall at 1.68 ± 0.07 bits per cell, closed only by spatial correlations in the noise extending about 20% of embryo length.

Speculative A self-organising route would buy the same anatomy without maintaining that coordinate system: three genes, two rate constants and a diffusion ratio, with the spacing following from the chemistry. In cultured mouse limb buds, adding a BMP inhibitor and a WNT inhibitor together made the digit period larger as the concentrations of both drugs increased — an anatomical count moved by turning two dials. In engineered human cells, a Nodal–Lefty circuit with effective ranges of 28.7 and 99.8 µm produced stable domains 196 ± 15 µm wide at a periodicity near 400 µm. Those are the units: micrometres of spacing per unit of inhibitor, in a tissue nobody had to draw.

Frontier Who collects this is narrow, and the honest answer is that the near-term gain accrues to developmental biology and synthetic biology rather than to a patient or a manufacturer. There is no therapy at the end of the digit argument. The transferable asset is a design rule for tissue engineering — specify a spacing instead of a geometry — and the briefs that would use it, Lab-Grown Organs and Whole Organ Regeneration, are today blocked by vascular supply and cell sourcing rather than by patterning theory.

Handwave The condition, stated plainly: the upside is real only if the wavelength in a living tissue is set by a mechanism whose parts are known and measurable, and that is exactly what section 5 says nobody can currently measure. A model that fits a pattern is not a mechanism that makes it, and the distance between those two sentences is the whole subject.

Established Turing’s 1952 result is a theorem about linear stability, not a claim about animals. Two or more species that react and diffuse can be stable when well mixed and unstable when allowed to spread, and the instability that grows fastest selects a length scale. Local self-activation with long-range inhibition is the standard cartoon; the mathematics is indifferent to whether the species are proteins, cells, or mechanical stresses.

Frontier Three claims travel under one name in this field, they have wildly different evidential support, and almost every public argument about “Turing patterns” is an argument in which the three have been quietly swapped. They are worth separating before anything else.

Established Claim one: reaction–diffusion is a mathematics that fits periodic biological patterns. This is not in dispute and is close to trivially true. Any pattern with a repeat has a dominant wavelength; a reaction–diffusion model with a handful of free parameters has a tunable wavelength; fitting one to the other costs nothing. A 2018 computational survey put roughly 3 × 10¹¹ parameter combinations across 7,757 distinct networks of up to three species through 8 CPU-years of analysis and found that more than 60% of those networks can exhibit a Turing instability somewhere in parameter space. Pattern-forming networks are common, which is precisely why finding one that fits your data is weak evidence.

Frontier Claim two: a Turing instability is the actual mechanism generating a particular structure in a particular organism. This is strong in a handful of systems and absent everywhere else. The standard list is short — vertebrate skin and fish pigment stripes, mouse digits, mouse hair follicles, mouse palatal ridges, chick feather buds — and Kondo’s 2022 review of the field states the difficulty without softening it: even when a spatial pattern is successfully reproduced by a reaction–diffusion model, it may not be clear whether a diffusing factor is responsible at all.

Frontier Claim three: self-organisation rather than positional information is the right general account of how embryos become patterned. This is a live dispute and it is not close to resolved. Wolpert’s 1969 alternative has cells reading their coordinates from a gradient with a fixed source; Driever and Nüsslein-Volhard’s 1988 demonstration that Bicoid forms an exponential gradient along the fly’s anterior–posterior axis remains the cleanest instructed-patterning result anyone has. A 2026 Nature Reviews Genetics piece frames the modern position as neither model winning outright, with both operating in different contexts.

Speculative The reason claim two is hard is that the observable — a wavelength — is the one thing every candidate model can produce. A gradient with thresholds, a reaction–diffusion instability, a mechanical buckling instability, a chemotactic aggregation and a cell-lineage clock will all deposit stripes at some spacing. Distinguishing them requires a perturbation whose predicted effect differs between models by more than the measurement error, and section 7 is about how rarely that is done.

Handwave A working definition, because the literature does not supply a shared one: a system deserves the label “Turing” when the pattern’s wavelength is set by the interaction of named molecular species whose production, decay and transport rates have been measured, and when changing one of those rates moves the wavelength by the predicted amount. By that definition the count of demonstrated vertebrate Turing systems is zero, and stating it that way is not a rhetorical trick — it is what makes the decisive experiment in section 7 well posed.

Speculative The fourth idea in the room is that the two frameworks are not rivals but layers. Green and Sharpe’s 2015 synthesis names three combinations: a reaction–diffusion pattern generating the gradients that cells then interpret; the two running in parallel, with self-organisation setting periodicity and a gradient setting identity; and a gradient modulating the wavelength of a self-organising pattern. The mouse digit model is an instance of the third, and it is the reason the digit case is cited as a victory by both camps.

Established Start with the side of this argument that has real numbers, because it is not the self-organisation side. Petkova, Tkačik, Bialek, Wieschaus and Gregor’s 2019 measurement of the Drosophila gap-gene system is the most quantitatively complete account of patterning in any organism. Gap-gene expression levels carry enough information to place a nucleus along the anterior–posterior axis with a standard deviation of about 0.01 of embryo length across almost the whole axis, and the decoder is parameter-free: built once from wild-type data, it predicted pair-rule stripe positions in embryos lacking maternal Bicoid, Nanos or Torso-like, and for nearly all 70 identifiable stripes the prediction agreed with the measurement to within embryo-to-embryo variability.

Established That result was then pushed until it broke, and the break is informative. The 2024 follow-up in PRX Life asked whether the local gap-gene signal alone specifies a unique cell identity and found it does not: a deficit of 1.68 ± 0.07 bits per cell between what is needed and what a purely local readout provides. The gap closes only when spatial correlations in the positional errors are included — correlation lengths around 20% of embryo length, reconciling the accounts to about 2%. Positional information, in other words, is measured, quantified, and already known to be incomplete as a purely local story.

Frontier Nothing on the self-organisation side is measured to that standard, and the four systems usually offered as the strong cases are strong in four different and non-comparable ways. They are worth taking one at a time, because the phrase “Turing patterns have been demonstrated in vertebrates” is doing a great deal of concealed work.

Frontier Zebrafish stripes are the best dynamical case and the worst molecular one. Kondo’s programme established the dynamic signature that matters: ablate a stripe and the pattern regenerates with the same spacing rather than healing back to a stored template, which is what a live wave does and what a pre-pattern does not. The molecular players identified are not a pair of diffusing morphogens at all but pigment-cell interactions, the leopard gene product connexin41.8 and the potassium channel Kir7.1. Speculative The system therefore supports the abstract Turing scheme and refutes the textbook version of it in the same breath: local activation and long-range inhibition are implemented by cells reaching across a distance, not by two molecules with different diffusion constants.

Frontier Mouse digits are the case the field cites, and the primary paper says less than the citation does. Raspopovic and colleagues’ 2014 Science paper screened expression by microarray on sorted cells, identified BMP, SOX9 and WNT as a candidate three-node network, and tested it pharmacologically in cultured limb buds. The reported results are qualitative and they are real: the BMP inhibitor LDN caused loss of digital Sox9-EGFP; the WNT inhibitor IWP2 caused the opposite, with each digit expanding until complete fusion; and applying both drugs together made the digit period larger as the concentrations of both drugs increased, giving fewer digits. Established The model’s Turing instability was restricted by hand to the digital region defined by Hoxd13 expression rather than emerging from the model, which the paper states plainly.

Established A finding about the commission for this brief, since it bears on what the digit case actually shows: the anchor supplied with this commission is not what the cited paper contains. The commission names “a Bmp-Sox9-Wnt Turing network reproduces digit patterning and predicts the observed polydactyly series in Hox-deficient limbs”. The 2014 paper contains no occurrence of the word polydactyly and does not test the model against Hox-mutant digit numbers; it uses the Hoxd13 domain only as a spatial mask on the simulation. The polydactyly-series result is a separate 2012 Science paper by Sheth and colleagues, which reported that progressive loss of Hox function shortens the wavelength of a Turing-type mechanism and yields more, thinner digits. Handwave That paper could not be retrieved this session — the publisher, PubMed Central and the abstract mirrors all refused — so this brief cites it by name, does not quote its digit counts, and does not lean on it. The correction stands on its own: two papers, two models, one anchor sentence that merged them.

Frontier Mouse hair follicles are the case where the molecular identification is cleanest and the mechanism is now actively contested. Sick and colleagues’ 2006 result that WNT and DKK determine follicle spacing through a reaction–diffusion mechanism is the canonical citation, repeated as an established Turing system as recently as a January 2026 Nature Reviews Genetics piece. Speculative Against it, a 2026 PNAS paper argues that chemotactic self-organisation, not reaction–diffusion, captures the dynamics of mammalian follicle patterning; that paper was inaccessible this session and is cited by name only, so this brief records the existence of the challenge rather than its strength.

Speculative Mouse palatal rugae are the case with an identified activator–inhibitor pair and the least independent scrutiny. Green and Sharpe’s review classes the rugae as having strong molecular evidence, with FGF and SHH as the pair identified by Economou and colleagues in 2012, and treats the system as a gradient modulating the wavelength of a self-organising pattern. That primary paper could not be opened this session either; the classification here is taken from the review that cites it, and is flagged accordingly.

Established The strongest recent work is not about any of these systems — it is about what the mathematics permits, and it has dismantled the standard objection to Turing models. The textbook requirement that the inhibitor diffuse substantially faster than the activator was widely treated as biologically implausible. Marcon, Diego, Sharpe and Müller’s 2016 eLife screen of 5,376 three-node and 1,464,320 four-node topologies found 21 and 64 pattern-forming networks respectively, sorted into three types: Type I needs unequal diffusion, Type II tolerates equal diffusion among the mobile signals, and Type III is independent of the diffusion rates altogether. Roughly 70% of the identified networks containing a non-diffusible node fall in Types II or III. Immobile components — receptors, transcription factors — act as capacitors that retain and amplify perturbations regardless of how fast anything diffuses.

Established Diego, Marcon, Müller and Sharpe’s 2018 Physical Review X paper generalised that into a topological statement: which diffusion restrictions apply, how robust the system is, and the phase relations between species are all determined by network topology rather than by the kinetic details. Frontier The practical consequence is a real one for experimentalists — the phase relations are observable in an expression pattern, so topology makes a prediction that does not require measuring a single rate constant.

Frontier The robustness objection survived in a different form and has just been contested again. Scholes and colleagues’ analysis concluded that Turing mechanisms are common but not robust to parametric or structural change, which for a decade was the standard reason to doubt that any real tissue could rely on one. Established A 2026 Development paper by Muzatko, Daga and Hiscock reaches the opposite conclusion from a different direction: symmetry breaking is strongly constrained by the extracellular interactions and is relatively insensitive to the assumed intracellular dynamics, so a necessary condition for patterning can be computed from steady-state input–output response functions alone, agreeing with the full parameter-space calculation about 90% of the time. They identify eight classes of activator–inhibitor system and report that a WNT–DKK architecture still requires a differential diffusive range while a WNT–sFRP architecture does not.

Speculative Their conclusion is the sharpest live statement in the field: simple activator–inhibitor mechanisms alone are capable of robust patterning in vivo, and hair follicles, fingerprints and digit joints may self-organise without additional cellular or mechanical interactions. That is a claim about what is possible, not a measurement of what happens, and the authors present it as such.

Frontier Meanwhile the mechanical alternative has grown from a curiosity into a standing rival. Rombouts, Elliott and Erzberger’s 2023 EMBO Reports review makes the timescale argument concretely: a small molecule takes seconds to diffuse 10 to 100 µm, while mechanical stress propagates the same distance through the cytoskeleton in microseconds. Mechanochemical patterning therefore operates on scales chemical diffusion cannot reach, and the review lists cortical patterns in C. elegans, MDCK monolayers, avian skin and bacterial biofilms as cases where mechanics rather than diffusion sets the scale. Speculative For avian feather buds specifically, the mechanical account — emergent cellular self-organisation and mechanosensation initiating the follicle pattern, reported by Shyer and colleagues in 2017 — competes directly with the chemical one for a system long used as a Turing exemplar.

Frontier The evidence class that could settle any of this is synthetic reconstitution, and it has produced four results worth taking seriously and none that closes the argument. Sekine and colleagues’ 2018 Nature Communications work is the most quantitative: HEK293 cells carrying a Nodal–Lefty activator–inhibitor circuit produced stable alternating domains, with characteristic ranges of 28.7 µm for Nodal and 99.8 µm for Lefty2 — a factor of 3.5 — positive domains 196 ± 15 µm wide and a peak-to-peak periodicity near 400 µm, stable from 36 to 60 hours after induction. Established The authors are explicit about two limits: the diffusion coefficients were inferred rather than measured directly, and their own modelling could not distinguish a classical Turing pattern from a solitary pattern dependent on initial conditions.

Established In bacteria the most careful attempt is the three-node circuit published as a preprint in 2023 and in Cell Systems in 2024. It uses MK01 Escherichia coli, two diffusible quorum-sensing signals (pC-HSL and 3OHC₁₄-HSL) for the self-activation and lateral-inhibition pair and a non-diffusible transcription-factor variant as the third node added for robustness — which is the Marcon Type II/III design principle used as engineering guidance. Established The controls are the part that matters: patterns vanished when diffuser degradation was induced with DAPG and failed at 30 °C, and plasmid sequencing found no mutations, excluding genetic drift. Handwave The preprint reports a constant wavelength but does not state that wavelength in micrometres, and concedes that the patterns and colony dynamics in each experiment were unique — which is the honest description of a stochastic pattern and also exactly what a sceptic would predict from growth noise alone.

Speculative Two further bacterial results sit lower on the evidential ladder and are often cited as though they sat higher. A 2020 engineered-consortium preprint combined MinC-driven cell elongation, a JunA adhesion protein and LuxI/LuxR quorum sensing in E. coli UT5600, and reported branched colonies with a characteristic wavelength near 0.2 cm by Fourier analysis and about five branches per quadrant after five days, with only the full three-gene construct patterning at all. That is a pattern produced by a designed system; it is not a molecular Turing instability, since the slow variable is cell dispersal rather than a morphogen. Frontier The MINPAC circuit, a reaction–diffusion clock-and-wavefront implementation in E. coli, generated sequential rings over 48 to 96 hours and its authors note a mismatch between predicted signal concentrations and measured mCherry intensity attributable to slow reporter degradation.

Established The newest synthetic result, posted in June 2026, is the first to characterise every module separately before assembling them. Landge, Marcon, Soh, Lohrmann, Volkwein and Müller built an E. coli circuit with positive feedback, delayed negative feedback and two orthogonal quorum-sensing signals, measured the signal diffusion rates, and obtained an sfGFP activation front followed by successive mCherry travelling waves in quasi-two-dimensional lawns. Speculative Model-guided perturbations changed timing, amplitude, wavelength and directionality by varying lawn size, lawn position relative to the boundary, and signal degradation. This is the closest anyone has come to the workflow section 7 asks for, and it is in bacteria, in a system with no anatomy to get right.

Speculative One theoretical result deserves a place in the measured record because it removes a constraint everyone assumed: on a discrete cellular lattice a single morphogen suffices. A 2022 analysis showed that an intracellular self-activator paired with a diffusible repressor under one promoter implements local activation and long-range inhibition with a minimum wavelength of two cell diameters, wavelength scaling as the square root of the diffusion coefficient, and growth-coupled patterning improving noise robustness by orders of magnitude. Handwave If one morphogen is enough, then finding two candidate morphogens in a tissue is even weaker evidence for a Turing mechanism than it already was.

Speculative A 2024 preprint adds a more uncomfortable result: linear stability analysis, the tool that defines what a Turing system is, can fail on multistable systems. A state with no Turing instability formed a Turing pattern by attraction to neighbouring Turing steady states; absorbing boundary conditions turned spatially homogeneous solutions into periodic ones and temporal oscillators into stationary patterns; and linear domain growth, widely assumed to improve robustness, did not. Handwave Taken together with the single-morphogen result, the class of models that can produce a stripe is now larger than the class anyone can exclude, which is the structural reason this field has a demonstration problem rather than an imagination problem.

Frontier The summary position, stated as a ledger: the mathematics is established, the general framework question is open, and the mechanism claim is unproven in every vertebrate system where it is asserted. Grodstein, McMillen and Levin’s 2023 paper in Frontiers in Cell and Developmental Biology puts the structural problem cleanly from the modelling side: a field of a given length carrying a pattern of a given wavelength gets about their ratio in repeats, so a larger animal should get extra fingers unless something counts them, and their proposal is an explicit closed-loop controller that measures the repeat number and adjusts the wavelength. Speculative That nobody knows which mechanism does the counting in any real limb is the cleanest statement of how much of this subject remains unbuilt.

Frontier Question one, and the one that organises the rest: what would count as demonstrating a Turing mechanism, and does the field agree? It does not. Kondo’s 2022 review says reproducing a pattern with a complex model may leave the mechanism unknown and that it may not even be clear whether a diffusing factor is responsible; Muzatko, Daga and Hiscock’s 2026 analysis offers a necessary condition computable from steady-state response functions. Speculative A necessary condition is not a demonstration, but it is the first published criterion that an experimentalist could fail, which is more than the field had.

Frontier Question two: is the digit system a Turing system, or a Turing-shaped model fitted to a gradient-shaped phenomenon? The 2014 result restricted the instability to the Hoxd13 domain by hand. Speculative A model that requires a pre-existing spatial mask to put its pattern in the right place has already conceded that positional information sets the field; what remains at stake is only whether the periodicity inside that field is self-generated. Handwave That is a smaller claim than the one usually made on the paper’s behalf, and it is the claim a serious test should target.

Frontier Question three: do the robustness results of Scholes and of Muzatko actually disagree, or are they answering different questions? The first measures the fraction of parameter space giving patterns for a given topology; the second asks which architectures can break symmetry at all and finds the answer insensitive to intracellular detail. Speculative They are formally compatible — a mechanism can be easy to find and hard to hold — but the field cites them as rival verdicts, and no paper has yet put both calculations on the same network set.

Frontier Question four: how much of the apparent chemical patterning is mechanical? The microsecond-versus-second timescale gap in the 2023 EMBO Reports review is not a detail; it means a mechanical signal can cross a 100 µm field before a morphogen has moved appreciably. Speculative In avian skin the mechanical account is already the competing primary hypothesis, and in mammalian follicles a 2026 chemotactic account has entered the same contest. Handwave There is no system in which the chemical and mechanical hypotheses have been tested against each other with a shared, pre-registered discriminating prediction.

Frontier Question five: scale invariance. Embryos of different sizes make the same number of parts, which a naive reaction–diffusion mechanism cannot do. Speculative The proposed answers — expansion–repression feedback, feed-forward coupling of embryo size to wavelength as in the FGF account of the digit model, and explicit closed-loop repeat counting — are all mechanisms nobody has identified molecularly in a vertebrate. Handwave Until one is, scaling is the strongest single argument that something other than a bare Turing instability is running.

Speculative Question six: does the immobile-node result change what to look for in vivo? If Type II and Type III networks pattern with equal or arbitrary diffusion rates, then the decades of effort spent measuring whether an inhibitor diffuses faster than an activator was aimed at a requirement that applies to only one of three classes. Established The replacement observable — phase relations between species, fixed by topology — is cheap to measure by in situ hybridisation and is almost never reported as a model test.

Speculative Question seven: is positional information ever primary in a vertebrate, or only in insects? The 1% precision result is a fly result, in a syncytium, with no cell membranes to cross. Handwave Whether any vertebrate tissue achieves comparable gradient precision is unmeasured, and the assumption that it does not is one of the least examined premises supporting self-organisation accounts of vertebrate development.

Handwave Question eight, the uncomfortable one: is “Turing mechanism” still a falsifiable category? Single morphogens pattern on lattices, immobile nodes remove the diffusion requirement, multistable systems pattern without a linear instability, absorbing boundaries manufacture periodicity from homogeneity. Speculative Every one of those results is good mathematics and each widens the target; a hypothesis that no observation can embarrass is a vocabulary rather than a theory, and the field has not yet said where it would draw the line.

Frontier Bottleneck one, and the binding one: nobody can measure a morphogen’s diffusion coefficient in a living vertebrate tissue to the accuracy the models need. The Nodal–Lefty synthetic system is the clearest admission — its authors inferred effective ranges of 28.7 and 99.8 µm from pattern geometry and stated that direct measurement of the diffusion coefficients would be necessary to verify those numbers. Speculative In a real embryo the same quantity must be extracted from photobleaching or correlation spectroscopy in tissue that is moving, growing and binding the molecule being tracked, and published values for the same morphogen differ between methods by amounts large enough to change a model’s qualitative behaviour.

Frontier Bottleneck two: there is no way to change one rate constant and nothing else. Every perturbation used in the digit work — LDN, IWP2 — is a pathway inhibitor that alters production, decay and downstream transcription simultaneously. Speculative A quantitative wavelength prediction requires moving one parameter by a known factor; a pathway inhibitor moves an unknown number of parameters by unknown factors, which is why the digit results are dose-monotonic rather than quantitative.

Speculative Bottleneck three: no standard way to report a wavelength with an error bar. Fish stripes, digit periods, follicle spacings and bacterial stripes are reported as images, counts, or Fourier peaks depending on the laboratory, and the 2020 consortium preprint’s 0.2 cm figure is one of the few explicit numbers in the synthetic literature. Handwave Model selection between a reaction–diffusion account and a positional-information account is a statistical comparison, and a field that does not routinely publish an uncertainty on its central observable cannot run one.

Speculative Bottleneck four: the tissues where the claims are made are the tissues hardest to image. Digit patterning happens inside a limb bud over roughly a day; follicle spacing happens in a curved epithelium; palatal rugae form in a closing cavity. Established The synthetic results are in bacterial lawns and cultured human cells precisely because those can be imaged flat, continuously and at single-cell resolution, and that selection effect is why the best-controlled reaction–diffusion data in existence concern no anatomy at all.

Speculative Bottleneck five: no reconstitution platform exists for vertebrate signalling networks. Building the BMP–SOX9–WNT network from defined parts in naive cells, as the bacterial groups build quorum-sensing circuits, would require orthogonal control of three pathways that are pleiotropic, cross-regulating and essential. Handwave No group has claimed this is close, and the honest engineering estimate is that the vertebrate reconstitution is a decade of tool-building behind the bacterial one.

Frontier Bottleneck six: mechanics and chemistry cannot currently be separated in the same specimen. Blocking actomyosin contractility changes tissue geometry, which changes diffusion distances, which changes chemical patterning. Handwave A clean test needs a perturbation that changes tissue mechanics while leaving transport unchanged, and no such perturbation has been demonstrated in any of the four contested systems.

Speculative Bottleneck seven, which is methodological rather than technical: the models are fitted after the data exist. Nothing in the workflow prevents a group from fitting a reaction–diffusion model to a pattern, reporting the fit, and never fitting the competing positional-information model to the same data. Handwave The absence of published head-to-head model comparisons in this field is not a data problem; it is a convention, and conventions are cheaper to change than instruments.

Frontier Dependency one: quantitative transport measurement in tissue. Every mechanistic claim in this subject reduces to a diffusion coefficient, a decay rate and a production rate for named species in a named tissue. Speculative Progress here does not require new physics; it requires a method whose systematic errors are characterised well enough that two laboratories measuring the same morphogen agree, and that comparison has not been published for any vertebrate morphogen at the precision the models need.

Speculative Dependency two: orthogonal, graded, cell-autonomous control of individual pathway parameters. Optogenetic and degron tools give timing control; what the models need is amplitude control over one reaction rate with a known dose–response. Established Synthetic-biology practice already provides this in bacteria — the three-node circuit used DAPG-induced degradation of a diffuser as a control, which is exactly the right shape of experiment — and it is the missing capability in vertebrates.

Frontier Dependency three: model-selection statistics adapted to spatial pattern data. Comparing a reaction–diffusion model to a positional-information model on the same images is a nested-model problem with strongly correlated residuals, and the field has no agreed criterion. Speculative The information-theoretic approach used on the fly gap genes — asking how many bits a candidate mechanism supplies and comparing that to the bits the anatomy requires — is the one framework that has already produced a falsifiable shortfall, namely the 1.68 bit deficit, and nobody has applied it to a self-organising system.

Speculative Dependency four: a benchmark dataset. There is no shared corpus of patterns with measured wavelengths, uncertainties, perturbation responses and raw images against which competing models can be scored. Handwave Fields that acquired such a benchmark — protein structure being the obvious case — converged fast afterwards, and the reason this one has not is that no funder owns the problem.

Speculative Dependency five: results from adjacent briefs this page does not own. The bioelectric route to pattern control, where voltage rather than a morphogen carries the spatial information, is treated in Bioelectric Medicine and is a genuinely different candidate mechanism rather than a variant of this one. Frontier The self-organising capacity of stem-cell aggregates is treated in Embryo Models, and its efficiency figures bound how well any reconstitution experiment in mammalian tissue can currently be expected to work.

Handwave Dependency six, stated as an absence: nothing in this subject waits on a theoretical result that does not exist. The mathematics is ahead of the measurements by a wide margin and has been since 2016. That is unusual for a Category XI topic and it changes what the bottleneck is: this is an experimental and conventional problem wearing the costume of a theoretical one.

Frontier The tests below are ranked by what a result would settle, not by how hard they are, and the first is the one this brief treats as decisive.

Speculative Tier one, the decisive experiment: an in-vivo perturbation that changes the predicted Turing wavelength by a quantitatively predicted amount, scored against a positional-information model fitted to the same data, with the prediction registered before the measurement. Every element of that sentence is load-bearing. Frontier A quantitative prediction, not a direction: the digit work predicted and observed that the period increases with combined BMP and WNT inhibition, but a gradient model with a threshold also predicts fewer digits under the same treatment, so a monotonic result discriminates nothing. Speculative A rival model fitted to the same data, not merely mentioned: the comparison is the experiment. Handwave And registration first, because a wavelength model has enough free parameters that a post-hoc fit is nearly guaranteed, which is the specific failure this design exists to exclude.

Speculative The cheapest version that would still be decisive: take the mouse limb-bud culture system already in use, measure the digit period and its uncertainty under a dose series of a single-pathway perturbation, publish the predicted period at each dose before running the assay, and publish alongside it the period predicted by a positional-information model fitted to the untreated controls. Frontier No new instrument is required; the apparatus, the drugs and the imaging all exist and were used in 2014. Handwave What does not exist is a group willing to publish a number in advance, and nobody has funded the version of this experiment whose result could be negative.

Speculative Tier two: full reconstitution of a named vertebrate network in naive cells, with the wavelength predicted from independently measured constants. This is the decisive evidence class for the general question, as distinct from the decisive single experiment for this brief. Frontier The template is already written in bacteria: the June 2026 travelling-wave circuit characterised positive feedback, delayed negative feedback and two orthogonal quorum-sensing signals separately, measured the diffusion rates, then predicted how lawn size, boundary position and signal degradation would change timing, amplitude, wavelength and directionality. Handwave Doing that with BMP, SOX9 and WNT in a naive mammalian cell line is the experiment that would end the digit argument, and no group has announced an attempt.

Frontier Tier three: direct measurement of the diffusion coefficients that every model assumes. The Nodal–Lefty ratio of 3.5 is the single most cited number in synthetic patterning and it was inferred from the pattern it is supposed to explain. Speculative An independent measurement that confirmed it would convert a circular argument into a prediction; one that contradicted it would invalidate the best-quantified synthetic Turing result in mammalian cells. Established Either outcome is publishable, the method exists, and this is the lowest-cost high-information experiment on the list.

Speculative Tier four: the mechanical control that nobody runs. In any system claimed as chemically self-organising, block contractility by a defined amount, measure the pattern wavelength, and report whether it moved. Handwave The confound is real — changing mechanics changes geometry changes transport — but a null result under a graded mechanical perturbation would be the strongest available evidence that a pattern is chemical, and the field has published essentially none of them.

Speculative Tier five: the phase-relation test, which is nearly free. Diego and colleagues’ 2018 result fixes the phase relations between species from network topology alone. Frontier For any tissue with a proposed three-node network, in situ hybridisation for all three species on the same section either shows the predicted phase relations or does not, at the cost of a week and no new equipment. Handwave That this test is not standard practice in papers proposing Turing networks is the clearest evidence available that the field is not yet trying to falsify itself.

Speculative Tier six: the scaling experiment. Grow the same tissue at two sizes differing by a known factor and measure whether the repeat number or the wavelength is conserved. Frontier A conserved repeat number at changed size requires a counting mechanism of the kind Grodstein, McMillen and Levin model explicitly; a conserved wavelength with changed repeat number is what a bare Turing instability predicts and what no vertebrate does. Speculative This is the one test on the list that an existing comparative dataset across related species might already be able to answer.

Handwave What would change this brief’s assessment most, in one sentence: a registered, quantitative, model-comparative wavelength prediction that survives in a living vertebrate tissue. Speculative Absent that, the field can continue producing better fits indefinitely, and a better fit is not a mechanism.

Frontier The engineering requirements here are unusual because the subject has no product, so the specification is for instruments and for design rules rather than for a device.

Speculative Requirement one: a light-sheet or lattice imaging setup that holds a mouse limb bud, a skin explant or a palatal shelf alive for 24 to 48 hours at single-cell resolution while reporting three fluorescent channels. Frontier Every published vertebrate pattern claim rests on fixed tissue at a handful of timepoints; the dynamical signature that distinguishes a live wave from a stored template — the signature Kondo used in fish by ablation and regeneration — requires continuous observation, and nobody has that for digits.

Speculative Requirement two: graded, orthogonal, quantitatively calibrated control of one pathway parameter at a time. The bacterial precedent is a diffuser whose degradation rate is set by an inducer; the mammalian equivalent would be a degron or optogenetic module on a single node with a published dose–response. Handwave Until a vertebrate perturbation has a calibration curve, no wavelength prediction in a vertebrate can be quantitative, and every result in that literature will remain directional.

Frontier Requirement three: a reconstitution chassis. The bacterial work shows what the finished article looks like — modules characterised separately, diffusion measured, then assembled — and the mammalian version needs a naive cell line with the three target pathways knocked out and re-introducible under inducible promoters. Speculative The HEK293 Nodal–Lefty system is the existence proof that a mammalian cell line can be made to pattern at all; extending it from an engineered pair to a native network is the step.

Speculative Requirement five, which is software rather than hardware: an open model-comparison pipeline. Given images, wavelengths with uncertainties and a perturbation series, it should fit both a reaction–diffusion and a positional-information model and report a scored comparison. Frontier Grodstein, McMillen and Levin released their simulation code openly, which is the right convention; what is missing is not code but an agreed scoring rule.

Speculative Where the engineering actually pays off first is not in anatomy but in materials and manufacture. A three-gene circuit that lays down stripes at a set spacing in a bacterial lawn is a patterning primitive for living materials, and the briefs that would consume it are Synthetic Biology and Biological Computing rather than any clinical page. Handwave That reordering — bacteria and materials before tissue and medicine — is the opposite of how the field presents itself, and it is what the evidence supports.

Established The nearest existing page on this map is Whole Organ Regeneration (FR-III-03), and the relationship is one of near-total non-overlap, which is worth stating because the commission named it as this brief’s nearest neighbour. That page is about producing tissue: it reports that mammals have one documented case of whole-organ regeneration, that a human liver sits at 81% of its starting volume a year after donation, and that the binding constraint is funding direction rather than biology. Speculative This page is about which formal theory of patterning is correct, and would be the same brief if no organ were ever regenerated. The honest cross-reference is that regeneration is the most demanding consumer of a patterning theory and the least dependent on having one, since existing regeneration work proceeds by scaffold and cell source rather than by wavelength.

Established Embryo Models is the closer neighbour in substance. Stem-cell aggregates that organise themselves into embryo-like structures are the most striking demonstration of self-organisation in mammalian tissue and the most dangerous evidence to cite here: that page records about one per cent efficiency, no demonstrated developmental potential and no agreed fidelity test. Speculative A structure that forms itself is evidence that self-organisation happens; it is not evidence about which self-organising mechanism, and the one-per-cent figure is the reason a reconstitution experiment in mammalian tissue is currently a statistics problem as well as a biology problem.

Frontier Bioelectric Medicine carries the genuinely rival mechanism. If anatomical pattern is stored in membrane voltage distributions rather than in morphogen concentrations, the entire reaction–diffusion framework is the wrong level of description for the phenomena it claims. Speculative That page reports the voltage claim as unreplicated outside the laboratory that found it, which is the correct weight; the point for this brief is that a third candidate exists and is rarely included in model comparisons alongside reaction–diffusion and positional information.

Speculative Basal Cognition shares this brief’s central pathology in a different domain: a mathematics or a vocabulary that fits a behaviour is repeatedly mistaken for a mechanism that produces it. Frontier The remedy is also the same — specify in advance what observation would be incompatible with the label — and both pages end at the same place.

Established Synthetic Biology supplies this brief’s decisive evidence class and also its cautionary statistic. That page reports that fabrication became an engineering discipline while design did not, with 19 of 88 designed genome segments lethal in the recoded Escherichia coli project. Speculative Reconstituting a patterning network is a design problem of exactly that kind, which is a reason to expect the vertebrate reconstitution to be slow even after the tools arrive.

Speculative Extended Evolutionary Synthesis is adjacent through developmental bias. If pattern is generated by a self-organising network with a small number of accessible wavelengths, then the variation evolution sees is structured before selection acts. Handwave That page reports developmental bias as having mechanism and no prospective prediction, which is the same verdict this page reaches about Turing mechanisms in vertebrates, arrived at independently.

Frontier This subject has no institution. There is no morphogenesis programme with a budget line, no consortium, no facility, and no funder who owns the question of which patterning theory is correct. Speculative The work is done inside developmental-biology laboratories funded to study a tissue and inside physics and synthetic-biology groups funded to build circuits, and the model-comparison experiment section 7 calls decisive falls between both mandates.

Speculative The publication incentive runs against the decisive experiment in a specific, checkable way. A paper reporting that a reaction–diffusion model reproduces a pattern is publishable in a high-impact journal; a paper reporting that both a reaction–diffusion model and a positional-information model reproduce it equally well is a null result about model selection and has no natural venue. Handwave The near-absence of published head-to-head comparisons in a field forty years old is consistent with that incentive and hard to explain otherwise.

Frontier Preregistration is the cheapest available institutional fix and it has no constituency here. The decisive test requires only that a predicted wavelength be posted before the assay runs, which costs nothing and requires no new infrastructure. Speculative Clinical trials adopted registration under regulatory compulsion and psychology adopted it after a replication crisis; developmental biology has had neither trigger, and the mechanism-attribution error this brief documents is not the kind that produces a public scandal.

Speculative The citation pathology is institutional rather than individual, and the anchor error in this brief’s own commission is an instance of it. Two 2012 and 2014 papers with different models and different evidence became, in general circulation, one result: a Turing network that predicts a mutant digit series. Handwave Nobody did anything wrong at any step; the compression happened in review articles, abstracts and teaching slides, and there is no mechanism anywhere in the literature for decompressing it.

Speculative What an institution could usefully own: a benchmark. A curated set of patterns with measured wavelengths, uncertainties, perturbation series and raw images, hosted openly, with a scoring protocol for candidate models. Frontier The cost is that of a small data facility and a standards committee; the obstacle is that the benefit accrues to the field rather than to any laboratory, which is the standard signature of something that only gets built when a funder decides to build it.

Handwave One structural observation with no fix attached: the synthetic-biology groups now producing the best-controlled reaction–diffusion data are not developmental biologists and do not publish in developmental-biology venues. Speculative The 2026 travelling-wave preprint is a European Research Council and Deutsche Forschungsgemeinschaft product in a bacterial system; whether its methods migrate into vertebrate laboratories is a question about careers and reviewers rather than about science.

Frontier The direct ethical stakes of this subject are low and should be stated as low rather than inflated to match the framework. Nobody is harmed by an incorrect theory of digit spacing, no patient is treated on the basis of one, and the experiments in section 7 use cultured limb buds, engineered bacteria and cell lines.

Speculative The indirect stake is real and is about claim inflation. “We understand how the embryo builds a hand” is a sentence that funds tissue-engineering programmes, recruits students and reassures regulators, and this brief’s finding is that the sentence is not supported by the paper usually cited for it. Handwave The harm is diffuse, deferred and impossible to attribute to any individual decision, which is exactly the profile of harm that institutions systematically fail to price.

Frontier Where the work does touch a live ethical boundary, it does so through embryo models rather than through pattern theory. A reconstitution experiment in mammalian tissue that produced a genuine self-organised anatomical pattern would sit close to the structures Embryo Models covers, and that page records the oversight problem as jurisdictional rather than ethical. Speculative The patterning experiment most likely to raise the question is also the one most likely to be done in a cell line nobody classifies as an embryo model, which is how jurisdictional gaps are usually first exercised.

Speculative Animal use is the one concrete ethical cost and it is modest but not zero. The limb-bud culture experiments require mouse embryos; the scaling test requires comparative material across related species. Handwave A registered, model-comparative design would reduce the total number of animals used relative to the current practice of serial exploratory perturbation, which is a genuine and rarely made argument for methodological reform in this field.

Speculative Finally, an epistemic obligation rather than an ethical one: a field whose central hypothesis has become unfalsifiable owes its students a statement to that effect. Handwave Teaching Turing patterns as the explanation of digits, stripes and follicles, without teaching that the strongest vertebrate evidence is a dose-monotonic drug response in cultured tissue, transmits a confidence the literature does not contain.

Speculative If self-organisation turns out to be the general account, the practical consequence is a change in how biological manufacture is specified: by parameters rather than by geometry. A scaffold-based tissue programme specifies where every feature goes; a parameter-based one specifies a spacing and lets the tissue place the features. Frontier The difference is the difference between a drawing and a recipe, and it matters most where the drawing is large, as in Lab-Grown Organs, where four manufacturing routes currently fail at four different walls and none of them is a patterning wall.

Handwave If positional information turns out to be the general account, the consequence is the opposite and arguably larger: biological form is written down somewhere, and can in principle be read and rewritten. Speculative The 1% precision and roughly 4 to 5 bits per nucleus measured in the fly is the only quantitative estimate anyone has of how much information an embryo actually uses to place a cell, and if a comparable figure held in vertebrates it would put a number on how much specification a synthetic organism needs.

Speculative The most likely outcome is that neither wins and the civilizational consequence is an accounting change rather than a capability change. The 2026 Nature Reviews Genetics framing — both mechanisms operating in different contexts — implies that engineering biological form will require knowing, tissue by tissue, which layer is doing which job. Handwave That is a slower and less quotable future than either pure story, and it is the one the evidence currently supports.

Frontier There is a durable intellectual consequence independent of which side wins: this subject is the clearest available case study in how a scientific community mistakes a successful model for a discovered mechanism. Speculative The pattern is not confined to biology — it recurs wherever a flexible formalism meets an under-determined observable — and the corrective this brief arrives at, registered quantitative predictions with a rival model fitted to the same data, is general rather than local.

Handwave What this subject will not do, stated so that no reader infers it: it will not produce a therapy, a material, or a capability on any horizon this brief can defend. Speculative The nearest thing to a deliverable is a patterning primitive for engineered living materials, and that arrives through bacteria, not through anatomy.

These horizons track what would have to be measured, not what anyone has scheduled, because no programme in this subject has a timetable:

• 10 yr: Frontier Expect the synthetic route to keep outrunning the developmental one, with bacterial and cultured-cell circuits reporting measured diffusion rates, predicted wavelengths and model-guided perturbations, as the June 2026 travelling-wave work already does. Speculative Expect at least one of the four canonical vertebrate systems — on current form, hair follicles — to be reassigned from reaction–diffusion to a chemotactic or mechanochemical account, and expect that reassignment to be contested rather than settled. Speculative Expect the phase-relation test to become standard practice, because it is cheap and because the 2018 topology result makes it a real prediction.
• 25 yr: Speculative This is the window in which a vertebrate reconstitution is plausible: a named network rebuilt in naive mammalian cells with a wavelength predicted from independently measured constants. Speculative A success would make claim two established for the first time in any vertebrate system and would put the design rule in reach of the tissue programmes in Lab-Grown Organs. Handwave A failure — the network rebuilt, the pattern absent — would be the most informative negative result this field could produce and would be extremely hard to publish.
• 50 yr: Handwave On this horizon the plausible development is consolidation rather than victory: a tissue-by-tissue map of which layer sets periodicity and which sets identity, with the framework dispute dissolved rather than decided. Speculative If no vertebrate reconstitution has succeeded by then, the sober reading is that the vertebrate systems are not implemented the way the models assume, and the label will quietly narrow to the fish and bacterial cases where the dynamics were actually observed.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The defensible statement concerns shape rather than date: a theory that survives because every observation can be accommodated does not get refuted, it gets replaced when a competing formalism makes a cheaper prediction. Handwave Any claim here about designing organisms to a specified morphology is an assertion, and this page does not make one.

• Depends on No typed dependency edge on this map blocks this subject, and the absence is informative. Nothing here waits on a result another brief produces: the mathematics has been ahead of the measurements since 2016 and the missing inputs are instruments and conventions. Embryo Models is the nearest page that supplies material rather than context, because a mammalian reconstitution experiment would use the same aggregates and inherit their roughly one per cent efficiency; Bioelectric Medicine supplies a rival mechanism rather than a prerequisite.
• Requires (not on this map) Five conditions, in the order of the tokens above. The transport numbers. Every model reduces to a diffusion coefficient, a decay rate and a production rate for named species in a named tissue, and the flagship synthetic result inferred its 28.7 and 99.8 µm ranges from the pattern they were meant to explain. The registered comparison. A predicted wavelength posted before the assay, with a positional-information model fitted to the same data — the cheapest item on this list and the one whose absence does the most damage. The chassis. A naive mammalian line with the three target pathways removable and re-introducible under calibrated induction, which is the vertebrate equivalent of what bacterial groups already have. The benchmark. A curated corpus of patterns with wavelengths, uncertainties, perturbation series and raw images, plus a scoring protocol, owned by nobody today. The imaging. Continuous observation of an intact limb bud, skin explant or palatal shelf for 24 to 48 hours at single-cell resolution in three channels, without which the live-wave signature that convinced the field in fish cannot be sought in mammals.
• Enables A confirmed mechanism would give tissue engineering a parameter-based specification for periodic structure — set a spacing rather than draw a geometry — and would give engineered living materials a patterning primitive that works in bacterial lawns today. It would also put a number on how much positional specification a synthetic organism requires, which no programme currently has.
• Adjacent Whole Organ Regeneration is the commission’s named neighbour and overlaps least; Embryo Models and Synthetic Biology overlap most, the first for self-organising mammalian material and the second for the reconstitution methods this brief calls decisive; Basal Cognition shares the failure mode of a vocabulary mistaken for a mechanism; Extended Evolutionary Synthesis consumes the result through developmental bias.

Frontier Claim: “Turing explained how the leopard got its spots.” Established Turing proved that a reaction–diffusion system can be stable without diffusion and unstable with it, and that the instability selects a wavelength. Speculative He did not show that this is how any animal is patterned, and seventy-four years later the strongest vertebrate evidence is a dynamical signature in fish skin whose implementing molecules are cell-surface interactions and ion channels rather than the two diffusing morphogens of the theory.

Frontier Claim: “Digit patterning was shown to be a Turing mechanism in 2014.” Established What was shown is that a BMP–SOX9–WNT model, with its instability restricted by hand to the Hoxd13 domain, predicted the direction of three drug perturbations that were then observed: loss of digital Sox9 under BMP inhibition, digit fusion under WNT inhibition, and a larger digit period with both. Speculative Direction is not magnitude, a masked instability is not a self-organised field, and a positional-information model with a threshold predicts the same three directions.

Established Claim: “The same 2014 network predicted the polydactyly series in Hox-deficient limbs.” This is the anchor this brief was commissioned with, and it is wrong. The 2014 paper does not use the word polydactyly and does not test against Hox-mutant digit counts. Handwave The polydactyly result is Sheth and colleagues’ separate 2012 paper, which this brief could not retrieve and therefore does not quote. Two papers merged into one sentence is the most common way a field’s strongest-sounding claim comes to have no single paper behind it.

Frontier Claim: “Turing models are biologically implausible because they need the inhibitor to diffuse much faster than the activator.” Established False since 2016. Of the 21 three-node and 64 four-node pattern-forming topologies identified in the eLife screen, about 70% of those containing an immobile node fall into classes that tolerate equal diffusion rates or are independent of diffusion rates altogether. Speculative The correction cuts both ways: it rescues the models from an old objection and makes them harder to falsify, since a failed diffusion-ratio measurement no longer excludes anything.

Frontier Claim: “Turing patterns are not robust, so biology cannot use them.” Established The robustness analysis behind that claim examined 7,757 networks over roughly 3 × 10¹¹ parameter combinations and found Turing instabilities common but the parameter regions narrow. Frontier The 2026 Development analysis reaches the opposite practical conclusion by asking a different question and reports that simple activator–inhibitor mechanisms alone are capable of robust patterning in vivo. Speculative Both results stand; the field has not yet run them on a common network set, so citing either as settled is premature.

Frontier Claim: “Positional information has been refuted by self-organisation.” Established It is the best-measured mechanism in the field: 1% precision along the fly axis, a parameter-free decoder that predicts nearly all 70 pair-rule stripes across seven mutant backgrounds. Speculative Its known limitation is quantitative and specific — a 1.68 bit per cell shortfall from a purely local readout, closed by correlated noise — which is a refinement of the theory, not a refutation of it.

Speculative Claim: “Synthetic Turing patterns have been built, so the mechanism is proven.” Established Synthetic circuits have produced periodic patterns in bacteria and in HEK293 cells, with real controls — loss of pattern under induced diffuser degradation, loss at 30 °C, no plasmid mutations. Handwave Building a pattern with a designed circuit proves the mechanism is available to biology; it says nothing about whether any embryo uses it, and the HEK293 authors could not distinguish their own result from a solitary pattern set by initial conditions.

Speculative Claim: “Organoids and gastruloids prove development is self-organising.” Frontier They prove that mammalian cells can organise without an external template, which nobody disputes. Handwave They do not identify which self-organising mechanism operates, and at the efficiencies reported in Embryo Models they are a poor assay for any quantitative question.

Handwave Claim, made mostly in passing and worth naming: “It is all just reaction–diffusion at some level.” Speculative With single-morphogen lattice patterning, multistable patterning without a linear instability, absorbing boundaries generating periodicity, and mechanical stress propagating a hundred thousand times faster than diffusion, the phrase has expanded to cover nearly any spatial mechanism. Frontier A category that excludes nothing predicts nothing, and the repair is the one section 7 specifies: a registered number, a rival model, the same data.