Request allocation Contact 简体中文

ScienceTechnical reports

JM-TR-2025-008 Technical report Library v3.1

Library v3.1: a sauropodomorph build profile for the Brachiosaurus program, constrained by extant reference taxa and bracketing alone

Dr. Rahul Venkataraman 1,* Dr. Anders Kjeldsen 1 Dr. Yuting Zhou 1

1 Jurassic Meat Shenzhen Research Center, Shenzhen, China
* Corresponding author: Dr. Rahul Venkataraman · regulatory@jurassicmeat.com

Contents
  1. Abstract
  2. 1. Summary
  3. 2. Introduction
  4. 3. Materials and Methods
  5. 4. Results
  6. 5. Discussion
  7. 6. Limitations and known issues
  8. 7. Changes in Library v3.1
  9. 8. Data availability
  10. 9. Acknowledgements
  11. 10. References
  12. Supplementary Table S1

Abstract

Library v3.1 introduces a build profile targeting Sauropodomorpha, to support reconstruction of muscle and flavor gene targets for the Brachiosaurus program. The library contains no peptide fragments assigned to Sauropodomorpha. Four sauropodomorph-attributed specimens were examined in eleven extraction attempts; 23 candidate peptide-spectrum matches were obtained, none of which met authentication criteria. The profile therefore constrains the Brachiosaurus target using extant reference taxa (birds and crocodilians, expanded from 63 to 71 taxa in this release) and ancestral-node constraints only. Theropod-assigned peptides are excluded as tip anchors and enter the reconstruction solely through the Saurischia node. We describe the profile, report the resulting target (22 gene targets, 13,273 residues; median per-position posterior 0.81, against 0.93 for the Deinonychus target), and present eight sensitivity analyses. Seven perturbations changed fewer than 1.0% of target positions; removal of crocodilian references changed 3.6%. The reconstructed Brachiosaurus target is 99.7% identical to the reconstructed Saurischia ancestral node and 98.4% identical to the Deinonychus target over shared gene targets, and the positions at which the two species' targets differ are concentrated among those with the lowest posterior support. We set out why we consider a reference-only profile acceptable for the program's stated purpose and record its limitations.

Keywordsextant phylogenetic bracketancestral sequence reconstructionSauropodomorphasensitivity analysisbuild profilesreference taxa

JM-TR-2025-008 · Library v3.1 · 14 October 2025 Dr. Rahul Venkataraman, Dr. Anders Kjeldsen, Dr. Yuting Zhou Computational Genomics and Research, Jurassic Meat Shenzhen Research Center

Public summary of internal report (29 pp.). Library contents are confidential.

1. Summary

  • Library v3.1 adds a sauropodomorph-targeted build profile for the Brachiosaurus program. The library contains 0 sauropodomorph-assigned peptides, and the profile uses extant reference taxa and bracketing constraints only.
  • Eight extant reference taxa were added (four crocodilians, four birds), bringing the total to 71. Theropod fragment count (2,903) and coverage (15,960 unique residues) are unchanged from v3.0.
  • The Brachiosaurus target (22 gene targets, 13,273 residues) has a median per-position posterior of 0.81; 6.2% of positions fall below 0.5.
  • Seven of eight sensitivity analyses changed fewer than 1.0% of target positions. Removing crocodilian references changed 3.6%.
  • The profile is conservative by construction: it makes no claim at the Brachiosaurus tip that is not a consequence of the model and the bracket.

2. Introduction

Every Jurassic Meat reconstruction is an inference across a phylogenetic bracket: the extinct target is placed on a tree whose living members, birds on one side and crocodilians on the other, constrain the states of the ancestral nodes from which the target descends (Witmer, 1995). Fossil-derived peptides, where they exist, add a second kind of constraint. They are direct, if fragmentary, observations of sequence from the extinct lineage itself, and they calibrate the substitution model and branch lengths in the region of the tree where the target sits (JM-TR-2019-004, JM-TR-2023-006).

The Deinonychus program has had both kinds of constraint since its inception. The theropod set has grown from 214 fragments at v1.0 to 2,903 at v3.0. These fragments are assigned no lower than Theropoda. One library specimen (JM-SP-0019, a Hell Creek tooth) is attributed to Dromaeosauridae on morphology (JM-TR-2016-003), but no fragment is assigned below Theropoda on sequence evidence, and the theropod build profile places the Deinonychus tip using theropod-assigned anchors together with the avian bracket, which approaches the Deinonychus lineage closely because birds are themselves theropods.

The Brachiosaurus program has only the first kind of constraint. Sauropodomorpha diverged from the lineage leading to theropods and birds at the base of Saurischia. The avian side of the bracket reaches the sauropodomorph branch only through that node. Above it, along the sauropodomorph branch, there is no living descendant and, in our library, no fossil-derived sequence. This is the sense in which the Brachiosaurus program is the company's first species bracketed at a substantially greater phylogenetic distance from living birds.

Before this release, exploratory Brachiosaurus reconstructions were run under the theropod build profile with the target clade set to Sauropodomorpha and the peptide filter set to Theropoda. From October 2024, when fail-closed filter validation began to reject that mismatch (JM-TR-2025-007), such builds required a recorded exception (EX-2025-004). Under that arrangement theropod-assigned peptides entered Brachiosaurus builds as tip anchors. We do not consider this defensible as a basis for construction, and the v3.1 profile replaces it. Three exploratory constructions made under the earlier arrangement are addressed in Section 4.6.

This report describes the sauropodomorph profile (Section 3), the reconstructed target and its sensitivity to model choices (Section 4), and the reasoning by which we regard a peptide-free profile as acceptable for the program's purpose (Section 5).

3. Materials and Methods

3.1 Sauropodomorph specimen examination

Between 2017 and 2025 the palaeoproteomics group examined four sauropodomorph-attributed specimens: two isolated teeth from the Kem Kem Group attributed at acquisition to Rebbachisauridae (JM-SP-0037 and JM-SP-0038), one vertebral fragment of Middle Jurassic age from Morocco (JM-SP-0049), and one tooth of undocumented provenance (JM-SP-0062) that was withdrawn at intake under the chain-of-custody schema (JM-TR-2021-005) and not extracted. Extraction, demineralisation and LC-MS/MS followed the procedures applied to theropod material (JM-TR-2013-001, JM-TR-2016-003), with procedural blanks in every batch. Spectra were searched against the library alignment and against the contamination exclusion database (human keratins, porcine trypsin, bovine collagen and laboratory-adjacent avian collagens), with peptide-spectrum matches (PSMs) controlled at 1% FDR (Cox and Mann, 2008). Candidate PSMs were required to show deamidation of N and Q residues consistent with the specimen's age, absence from procedural blanks, and no closer match to an exclusion entry.

No Brachiosaurus material has been examined. None is expected to become available.

3.2 Profile definition

The sauropodomorph profile (sauropodomorpha-ref-v1) differs from the theropod profile in four respects (Table 1). Most importantly, its tip-anchor set is empty, and theropod-assigned fragments are admitted only as constraints on internal nodes at or below Saurischia, where they inform the ancestral state that both lineages inherit.

Table 1. Comparison of the theropod and sauropodomorph build profiles in Library v3.1.
Parameter Theropod profile (theropoda-v3) Sauropodomorph profile (sauropodomorpha-ref-v1)
Target clade Theropoda Sauropodomorpha
Tip-anchor peptides Theropod-assigned (2,903) None (0)
Internal-node peptide constraints All nodes on the theropod path Saurischia node and below only
Extant reference taxa 63 (v3.0 set; Section 4.5) 71
Target gene set 19 targets, 11,842 residues 22 targets, 13,273 residues
Unconstrained terminal branch Paraves node to tip Saurischia node to tip
Branch-length source Peptide-calibrated (JM-TR-2023-006) Divergence-time prior, scaled by theropod-path rates
Acceptance threshold for sensitivity analyses 1.0% of target positions 1.0% of target positions

The three additional gene targets in the Brachiosaurus set are involved in intramuscular lipid uptake and storage. They were requested by the muscle biology group in support of the program's marbling objectives and are reconstructed by the same procedure as the shared targets.

3.3 Reference-taxon expansion

Eight extant taxa were added to improve coverage at the two nodes that most constrain the sauropodomorph branch: the crown-crocodilian node, which bounds the archosaur root from the pseudosuchian side, and the base of crown birds, which bounds Saurischia from the avian side (Table 2). Reference sequences for the target gene set were assembled from public sequence resources and verified in-house. Opisthocomus hoazin was included for its position near the base of Neoaves; that it is the only extant bird with a ruminant-like foregut was not a consideration, because diet does not enter the model.

Table 2. Extant reference taxa by group, Library v3.0 and v3.1.
Group v3.0 v3.1 Taxa added in v3.1
Crocodylia 9 13 Mecistops cataphractus, Tomistoma schlegelii, Osteolaemus tetraspis, Paleosuchus palpebrosus
Palaeognathae 5 5 none
Galloanserae 12 15 Anseranas semipalmata, Chauna torquata, Alectura lathami
Neoaves 33 34 Opisthocomus hoazin
Non-archosaur outgroups (turtles, squamate) 4 4 none
Total 63 71 8

3.4 Reconstruction

Ancestral states were inferred by marginal maximum-likelihood reconstruction under an empirical amino- acid substitution model with gamma-distributed rate heterogeneity, on a fixed archosaur topology (Fig. 1) whose extinct tips carry branch lengths derived from divergence-time priors. The Brachiosaurus tip state at each position was taken as the maximum a posteriori state of the marginal distribution obtained by propagating the Saurischia node posterior along the terminal branch. Positions were classed as directly anchored (overlapped by a fossil-derived peptide on the path to the tip), node-anchored (constrained by peptides at an internal node only) or reference-only.

Figure 1. Simplified bracketing topology for the sauropodomorph profileReduced cladogram of the topology used by both build profiles (extant reference taxa collapsed to representatives). Brachiosaurus is highlighted. Theropod-assigned peptides constrain the Deinonychus tip and the theropod path; in the sauropodomorph profile they reach the Brachiosaurus branch only through the Saurischia node. No peptide is placed on the sauropodomorph branch.GallusGallusAnasAnasOpisthocomusOpisthocomusStruthioStruthioDromaiusDromaiusDeinonychusDeinonychusBrachiosaurusBrachiosaurusAlligatorAlligatorPaleosuchusPaleosuchusCrocodylusCrocodylusGavialisGavialisTomistomaTomistomaFigure 1. Simplified bracketing topology for the sauropodomorph profileReduced cladogram of the topology used by both build profiles (extant reference taxa collapsed to representatives). Brachiosaurus is highlighted. Theropod-assigned peptides constrain the Deinonychus tip and the theropod path; in the sauropodomorph profile they reach the Brachiosaurus branch only through the Saurischia node. No peptide is placed on the sauropodomorph branch.GallusGallusAnasAnasOpisthocomusOpisthocomusStruthioStruthioDromaiusDromaiusDeinonychusDeinonychusBrachiosaurusBrachiosaurusAlligatorAlligatorPaleosuchusPaleosuchusCrocodylusCrocodylusGavialisGavialisTomistomaTomistoma
Figure 1.Simplified bracketing topology for the sauropodomorph profile. Reduced cladogram of the topology used by both build profiles (extant reference taxa collapsed to representatives). Brachiosaurus is highlighted. Theropod-assigned peptides constrain the Deinonychus tip and the theropod path; in the sauropodomorph profile they reach the Brachiosaurus branch only through the Saurischia node. No peptide is placed on the sauropodomorph branch.Source: Library v3.1, build profile sauropodomorpha-ref-v1.
Source data
TaxonHighlightedPlacement posterior
Gallus
Anas
Opisthocomus
Struthio
Dromaius
Deinonychus
Brachiosaurusyes
Alligator
Paleosuchus
Crocodylus
Gavialis
Tomistoma

((((((Gallus,Anas),Opisthocomus),(Struthio,Dromaius)),Deinonychus),Brachiosaurus),((Alligator,Paleosuchus),(Crocodylus,(Gavialis,Tomistoma))));

3.5 Sensitivity analyses

Eight perturbations were applied to the baseline reconstruction, each altering one modelling choice: the terminal branch length (×0.5 and ×2), the substitution model (an alternative empirical matrix), the topology (an alternative arrangement of the dinosaur root in which Ornithischia is sister to Theropoda, leaving the extant bracket unchanged), removal of the palaeognath references, removal of the eight taxa added in v3.1, removal of theropod-peptide constraints at the Saurischia node, and removal of all crocodilian references. For each we counted the target positions whose MAP state differed from baseline and the median change in per-position posterior. Branch-length scaling was additionally run as a series (×0.25 to ×3) with 200 nonparametric bootstrap replicates of the reference alignment.

The acceptance threshold, fixed before the analyses were run, was that no perturbation the program intends to treat as a plausible alternative should change more than 1.0% of target positions (133 of 13,273).

4. Results

4.1 Sauropodomorph peptide recovery

Eleven extraction attempts across three specimens yielded 23 candidate PSMs (Table 3). Fourteen matched an entry in the contamination exclusion database more closely than any library entry. Six showed no deamidation and were treated as modern. Three were also detected in a procedural blank from the same batch. None was authenticated, and none was admitted to the library.

Table 3. Sauropodomorph-attributed specimens examined, 2017–2025.
Specimen Attribution at acquisition Extraction attempts Candidate PSMs Excluded: contamination match Excluded: no deamidation Excluded: in blank Authenticated
JM-SP-0037 Rebbachisauridae, tooth 4 9 6 2 1 0
JM-SP-0038 Rebbachisauridae, tooth 4 11 6 3 2 0
JM-SP-0049 Sauropoda indet., vertebra 3 3 2 1 0 0
JM-SP-0062 Sauropoda indet., tooth 0 (withdrawn at intake) — — — — —
Total 11 23 14 6 3 0

The two Kem Kem teeth were the most promising candidates on preservation grounds. Of the six contamination matches from JM-SP-0038, four were closest to bovine collagen and two to human keratin. This is a typical profile for handled specimens of commercial origin and is consistent with the concerns raised in the wider literature about recent organic material in dinosaur bone (Saitta et al., 2019).

4.2 The reconstructed target

The baseline Brachiosaurus target comprises 13,273 residues across 22 gene targets. Its median per-position posterior is 0.81, and 823 positions (6.2%) fall below 0.5 (Fig. 2). The Deinonychus target under the theropod profile has a median per-position posterior of 0.93 with 1.9% of positions below 0.5. Of Brachiosaurus positions, 4.1% are node-anchored and the remainder reference-only. None is directly anchored.

Figure 2. Per-position posterior probability, Brachiosaurus targetMarginal posterior probability of the MAP state at each of the 13,273 reconstructed positions in the Brachiosaurus target under profile sauropodomorpha-ref-v1. Median 0.81; 823 positions (6.2%) below 0.5.Positions01,0002,0003,0004,0005,0000.20.30.40.50.60.70.80.91.00.2–0.3: 660.3–0.4: 1990.4–0.5: 5580.5–0.6: 7300.6–0.7: 1,3640.7–0.8: 3,4440.8–0.9: 2,4560.9–1: 4,456Posterior probability of MAP stateMedian 0.81Figure 2. Per-position posterior probability, Brachiosaurus targetMarginal posterior probability of the MAP state at each of the 13,273 reconstructed positions in the Brachiosaurus target under profile sauropodomorpha-ref-v1. Median 0.81; 823 positions (6.2%) below 0.5.Positions01,0002,0003,0004,0005,0000.20.40.60.81.00.2–0.3: 660.3–0.4: 1990.4–0.5: 5580.5–0.6: 7300.6–0.7: 1,3640.7–0.8: 3,4440.8–0.9: 2,4560.9–1: 4,456Posterior probability of MAP stateMedian 0.81
Figure 2.Per-position posterior probability, Brachiosaurus target. Marginal posterior probability of the MAP state at each of the 13,273 reconstructed positions in the Brachiosaurus target under profile sauropodomorpha-ref-v1. Median 0.81; 823 positions (6.2%) below 0.5.Source: Library v3.1, baseline reconstruction BR-REF-2025-09.
Source data
Posterior probability of MAP statePositions
0.2–0.366
0.3–0.4199
0.4–0.5558
0.5–0.6730
0.6–0.71,364
0.7–0.83,444
0.8–0.92,456
0.9–14,456

Gene-level support is lower for Brachiosaurus than for Deinonychus in every one of the 19 shared targets, and the shortfall grows as Deinonychus support falls (Fig. 3; slope 1.26, R² = 0.82). Genes that are well constrained in theropods are reasonably well constrained in sauropodomorphs; genes that are poorly constrained in theropods are worse.

Figure 3. Gene-level posterior support, Brachiosaurus vs Deinonychus targetsMean per-position posterior for each of the 19 reconstruction targets shared by both profiles. Every point lies below the line of equality. Linear fit: slope 1.26, R² = 0.82.Brachiosaurus target, mean posterior per gene0.650.700.750.800.850.900.951.000.840.860.880.900.920.940.960.981.00Deinonychus target, mean posterior per geneShared reconstruction targets linear fit: y = 1.26x − 0.361, R² = 0.82Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.881 Brachiosaurus target, mean posterior per gene: 0.756Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.945 Brachiosaurus target, mean posterior per gene: 0.827Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.862 Brachiosaurus target, mean posterior per gene: 0.72Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.878 Brachiosaurus target, mean posterior per gene: 0.834Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.937 Brachiosaurus target, mean posterior per gene: 0.793Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.917 Brachiosaurus target, mean posterior per gene: 0.799Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.876 Brachiosaurus target, mean posterior per gene: 0.742Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.882 Brachiosaurus target, mean posterior per gene: 0.736Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.853 Brachiosaurus target, mean posterior per gene: 0.703Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.972 Brachiosaurus target, mean posterior per gene: 0.892Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.901 Brachiosaurus target, mean posterior per gene: 0.785Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.984 Brachiosaurus target, mean posterior per gene: 0.896Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.859 Brachiosaurus target, mean posterior per gene: 0.696Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.939 Brachiosaurus target, mean posterior per gene: 0.805Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.896 Brachiosaurus target, mean posterior per gene: 0.754Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.938 Brachiosaurus target, mean posterior per gene: 0.793Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.972 Brachiosaurus target, mean posterior per gene: 0.855Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.859 Brachiosaurus target, mean posterior per gene: 0.707Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.978 Brachiosaurus target, mean posterior per gene: 0.858Figure 3. Gene-level posterior support, Brachiosaurus vs Deinonychus targetsMean per-position posterior for each of the 19 reconstruction targets shared by both profiles. Every point lies below the line of equality. Linear fit: slope 1.26, R² = 0.82.Brachiosaurus target, mean posterior per gene0.70.80.91.00.850.900.951.00Deinonychus target, mean posterior per geneShared reconstruction targets linear fit: y = 1.26x − 0.361, R² = 0.82Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.881 Brachiosaurus target, mean posterior per gene: 0.756Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.945 Brachiosaurus target, mean posterior per gene: 0.827Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.862 Brachiosaurus target, mean posterior per gene: 0.72Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.878 Brachiosaurus target, mean posterior per gene: 0.834Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.937 Brachiosaurus target, mean posterior per gene: 0.793Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.917 Brachiosaurus target, mean posterior per gene: 0.799Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.876 Brachiosaurus target, mean posterior per gene: 0.742Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.882 Brachiosaurus target, mean posterior per gene: 0.736Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.853 Brachiosaurus target, mean posterior per gene: 0.703Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.972 Brachiosaurus target, mean posterior per gene: 0.892Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.901 Brachiosaurus target, mean posterior per gene: 0.785Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.984 Brachiosaurus target, mean posterior per gene: 0.896Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.859 Brachiosaurus target, mean posterior per gene: 0.696Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.939 Brachiosaurus target, mean posterior per gene: 0.805Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.896 Brachiosaurus target, mean posterior per gene: 0.754Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.938 Brachiosaurus target, mean posterior per gene: 0.793Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.972 Brachiosaurus target, mean posterior per gene: 0.855Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.859 Brachiosaurus target, mean posterior per gene: 0.707Shared reconstruction targets Deinonychus target, mean posterior per gene: 0.978 Brachiosaurus target, mean posterior per gene: 0.858
Figure 3.Gene-level posterior support, Brachiosaurus vs Deinonychus targets. Mean per-position posterior for each of the 19 reconstruction targets shared by both profiles. Every point lies below the line of equality. Linear fit: slope 1.26, R² = 0.82.Source: Library v3.1, baseline reconstructions BR-REF-2025-09 and DN-v3.1-REF.
Source data
Deinonychus target, mean posterior per geneBrachiosaurus target, mean posterior per gene
0.8810.756
0.9450.827
0.8620.72
0.8780.834
0.9370.793
0.9170.799
0.8760.742
0.8820.736
0.8530.703
0.9720.892
0.9010.785
0.9840.896
0.8590.696
0.9390.805
0.8960.754
0.9380.793
0.9720.855
0.8590.707
0.9780.858

4.3 Relationship to the ancestral node and to the Deinonychus target

Over the 19 shared reconstruction targets (11,842 positions), the Brachiosaurus target is 99.7% identical to the reconstructed Saurischia ancestral node and 98.4% identical to the Deinonychus target (Fig. 4). The Deinonychus target is 98.6% identical to the Saurischia node. In other words, the Brachiosaurus target departs from the common ancestor of the two lineages at about a fifth as many positions as the Deinonychus target does.

Figure 4. Pairwise sequence identity across shared gene targetsPercent amino-acid identity over the 19 reconstruction targets shared by both profiles (11,842 positions), between reconstructed targets, reconstructed ancestral nodes and two extant reference taxa.…Deinonychus targetSaurischia nodeArchosauria nodeGallus gallus…Brachiosaurus targetBrachiosaurus target × Brachiosaurus target: 100100Brachiosaurus target × Deinonychus target: 98.498.4Brachiosaurus target × Saurischia node: 99.799.7Brachiosaurus target × Archosauria node: 96.996.9Brachiosaurus target × Gallus gallus: 93.393.3Brachiosaurus target × Alligator mississippiensis: 90.190.1Deinonychus targetDeinonychus target × Brachiosaurus target: 98.498.4Deinonychus target × Deinonychus target: 100100Deinonychus target × Saurischia node: 98.698.6Deinonychus target × Archosauria node: 96.296.2Deinonychus target × Gallus gallus: 94.494.4Deinonychus target × Alligator mississippiensis: 89.289.2Saurischia nodeSaurischia node × Brachiosaurus target: 99.799.7Saurischia node × Deinonychus target: 98.698.6Saurischia node × Saurischia node: 100100Saurischia node × Archosauria node: 97.197.1Saurischia node × Gallus gallus: 93.693.6Saurischia node × Alligator mississippiensis: 90.490.4Archosauria nodeArchosauria node × Brachiosaurus target: 96.996.9Archosauria node × Deinonychus target: 96.296.2Archosauria node × Saurischia node: 97.197.1Archosauria node × Archosauria node: 100100Archosauria node × Gallus gallus: 92.192.1Archosauria node × Alligator mississippiensis: 9393Gallus gallusGallus gallus × Brachiosaurus target: 93.393.3Gallus gallus × Deinonychus target: 94.494.4Gallus gallus × Saurischia node: 93.693.6Gallus gallus × Archosauria node: 92.192.1Gallus gallus × Gallus gallus: 100100Gallus gallus × Alligator mississippiensis: 85.185.1Alligator mississippiensisAlligator mississippiensis × Brachiosaurus target: 90.190.1Alligator mississippiensis × Deinonychus target: 89.289.2Alligator mississippiensis × Saurischia node: 90.490.4Alligator mississippiensis × Archosauria node: 9393Alligator mississippiensis × Gallus gallus: 85.185.1Alligator mississippiensis × Alligator mississippiensis: 100100Figure 4. Pairwise sequence identity across shared gene targetsPercent amino-acid identity over the 19 reconstruction targets shared by both profiles (11,842 positions), between reconstructed targets, reconstructed ancestral nodes and two extant reference taxa.…Deinonychus targetSaurischia nodeArchosauria nodeGallus gallus…BrachiosaurustargetBrachiosaurus target × Brachiosaurus target: 100100Brachiosaurus target × Deinonychus target: 98.498.4Brachiosaurus target × Saurischia node: 99.799.7Brachiosaurus target × Archosauria node: 96.996.9Brachiosaurus target × Gallus gallus: 93.393.3Brachiosaurus target × Alligator mississippiensis: 90.190.1Deinonychus targetDeinonychus target × Brachiosaurus target: 98.498.4Deinonychus target × Deinonychus target: 100100Deinonychus target × Saurischia node: 98.698.6Deinonychus target × Archosauria node: 96.296.2Deinonychus target × Gallus gallus: 94.494.4Deinonychus target × Alligator mississippiensis: 89.289.2Saurischia nodeSaurischia node × Brachiosaurus target: 99.799.7Saurischia node × Deinonychus target: 98.698.6Saurischia node × Saurischia node: 100100Saurischia node × Archosauria node: 97.197.1Saurischia node × Gallus gallus: 93.693.6Saurischia node × Alligator mississippiensis: 90.490.4Archosauria nodeArchosauria node × Brachiosaurus target: 96.996.9Archosauria node × Deinonychus target: 96.296.2Archosauria node × Saurischia node: 97.197.1Archosauria node × Archosauria node: 100100Archosauria node × Gallus gallus: 92.192.1Archosauria node × Alligator mississippiensis: 9393Gallus gallusGallus gallus × Brachiosaurus target: 93.393.3Gallus gallus × Deinonychus target: 94.494.4Gallus gallus × Saurischia node: 93.693.6Gallus gallus × Archosauria node: 92.192.1Gallus gallus × Gallus gallus: 100100Gallus gallus × Alligator mississippiensis: 85.185.1AlligatormississippiensisAlligator mississippiensis × Brachiosaurus target: 90.190.1Alligator mississippiensis × Deinonychus target: 89.289.2Alligator mississippiensis × Saurischia node: 90.490.4Alligator mississippiensis × Archosauria node: 9393Alligator mississippiensis × Gallus gallus: 85.185.1Alligator mississippiensis × Alligator mississippiensis: 100100
Figure 4.Pairwise sequence identity across shared gene targets. Percent amino-acid identity over the 19 reconstruction targets shared by both profiles (11,842 positions), between reconstructed targets, reconstructed ancestral nodes and two extant reference taxa.Source: Library v3.1, baseline reconstructions.
Source data
Brachiosaurus targetDeinonychus targetSaurischia nodeArchosauria nodeGallus gallusAlligator mississippiensis
Brachiosaurus target10098.499.796.993.390.1
Deinonychus target98.410098.696.294.489.2
Saurischia node99.798.610097.193.690.4
Archosauria node96.996.297.110092.193
Gallus gallus93.394.493.692.110085.1
Alligator mississippiensis90.189.290.49385.1100

This is the expected consequence of a terminal branch without data. With no observation at or near the tip, the marginal distribution at the Brachiosaurus tip is the Saurischia node posterior, broadened by substitution along the branch. Where the node posterior is strong, the tip MAP state equals the node state. Where it is weak, broadening can shift the MAP state to a different residue, and those are the positions at which the target departs from the node. The Deinonychus target departs from the node for a different reason: theropod-assigned anchors and the nearby avian bracket supply information that moves the theropod tip away from the shared ancestor.

The two species' targets differ at 189 of 11,842 shared positions (1.6%). At 147 of these (77.8%), the Brachiosaurus state carries a posterior below 0.6. The positions at which the profiles differ are, by construction, the positions at which the reconstruction is least certain.

4.4 Sensitivity analyses

Seven of the eight perturbations changed fewer than 133 positions (1.0%), the pre-specified acceptance threshold (Table 4, Fig. 5). Removing the crocodilian references changed 482 positions (3.6%). Without the crocodilian side, the bracket no longer brackets: the archosaur root is constrained from the avian side only, and its posterior collapses toward the avian state. This is not a perturbation the program intends to make, and it is reported to show that the crocodilian references are doing the work they are there to do.

Table 4. Sensitivity of the Brachiosaurus target to single modelling perturbations.
Perturbation Positions changed % of target Median change in posterior Within threshold
Terminal branch length ×0.5 16 0.12 +0.02 Yes
Terminal branch length ×2 47 0.35 −0.04 Yes
Alternative empirical substitution matrix 36 0.27 −0.01 Yes
Alternative dinosaur-root topology (Ornithischia sister to Theropoda) 26 0.20 0.00 Yes
Palaeognath references removed 15 0.11 −0.01 Yes
Eight taxa added in v3.1 removed 61 0.46 −0.03 Yes
Theropod-peptide constraints at Saurischia removed 85 0.64 −0.05 Yes
Crocodilian references removed 482 3.63 −0.11 No

Figure 5. Positions changed by each sensitivity perturbationNumber of Brachiosaurus target positions (of 13,273) whose MAP state differs from baseline under each single perturbation. Pre-specified acceptance threshold: 133 positions (1.0%).0100200300400500Branch length ×0.5Branch length ×0.5 Positions changed: 1616Branch length ×2Branch length ×2 Positions changed: 4747Alternative substitution matrixAlternative substitution matrix Positions changed: 3636Alternative dinosaur-root topologyAlternative dinosaur-root topology Positions changed: 2626No palaeognathsNo palaeognaths Positions changed: 1515No v3.1 additionsNo v3.1 additions Positions changed: 6161No theropod constraints atSaurischiaNo theropod constraints at Saurischia Positions changed: 8585No crocodiliansNo crocodilians Positions changed: 482482Positions changedFigure 5. Positions changed by each sensitivity perturbationNumber of Brachiosaurus target positions (of 13,273) whose MAP state differs from baseline under each single perturbation. Pre-specified acceptance threshold: 133 positions (1.0%).0200400Branch length ×0.5Branch length ×0.5 Positions changed: 1616Branch length ×2Branch length ×2 Positions changed: 4747Alternativesubstitution matrixAlternative substitution matrix Positions changed: 3636Alternativedinosaur-roottopologyAlternative dinosaur-root topology Positions changed: 2626No palaeognathsNo palaeognaths Positions changed: 1515No v3.1 additionsNo v3.1 additions Positions changed: 6161No theropodconstraints atSaurischiaNo theropod constraints at Saurischia Positions changed: 8585No crocodiliansNo crocodilians Positions changed: 482482Positions changed
Figure 5.Positions changed by each sensitivity perturbation. Number of Brachiosaurus target positions (of 13,273) whose MAP state differs from baseline under each single perturbation. Pre-specified acceptance threshold: 133 positions (1.0%).Source: Library v3.1, sensitivity set BR-SENS-2025-09.
Source data
PerturbationPositions changed
Branch length ×0.516
Branch length ×247
Alternative substitution matrix36
Alternative dinosaur-root topology26
No palaeognaths15
No v3.1 additions61
No theropod constraints at Saurischia85
No crocodilians482

The branch-length series shows a minimum at the baseline length and a slow rise on either side (Fig. 6). Shortening the branch pulls the tip toward the node, and lengthening it pulls the tip toward the stationary distribution of the substitution model. Neither is informed by any observation from the sauropodomorph lineage. The baseline length is a divergence-time prior of 79 million years from the Saurischia node to a Late Jurassic tip, scaled by substitution rates estimated along the theropod path.

Figure 6. Sensitivity of the Brachiosaurus target to terminal branch lengthPositions differing from the baseline target as the sauropodomorph terminal branch is scaled from 0.25× to 3× its baseline length. Line: point estimate; band: 95% interval over 200 bootstrap replicates of the reference alignment.Positions changed from baseline0204060801001201401600.51.01.52.02.53.0Terminal branch-length scaling factorAcceptance threshold (1.0%)Baseline (79 Myr)Positions changedTerminal branch-length scaling factor: 0.25 Positions changed: 33 (24–43)Terminal branch-length scaling factor: 0.5 Positions changed: 16 (11–23)Terminal branch-length scaling factor: 0.75 Positions changed: 7 (3–12)Terminal branch-length scaling factor: 1 Positions changed: 0 (0–5)Terminal branch-length scaling factor: 1.5 Positions changed: 26 (18–34)Terminal branch-length scaling factor: 2 Positions changed: 47 (37–60)Terminal branch-length scaling factor: 3 Positions changed: 92 (75–112)Figure 6. Sensitivity of the Brachiosaurus target to terminal branch lengthPositions differing from the baseline target as the sauropodomorph terminal branch is scaled from 0.25× to 3× its baseline length. Line: point estimate; band: 95% interval over 200 bootstrap replicates of the reference alignment.Positions changed from baseline0501001500.51.01.52.02.53.0Terminal branch-length scaling factorAcceptance threshold (1.0%)Baseline (79 Myr)Terminal branch-length scaling factor: 0.25 Positions changed: 33 (24–43)Terminal branch-length scaling factor: 0.5 Positions changed: 16 (11–23)Terminal branch-length scaling factor: 0.75 Positions changed: 7 (3–12)Terminal branch-length scaling factor: 1 Positions changed: 0 (0–5)Terminal branch-length scaling factor: 1.5 Positions changed: 26 (18–34)Terminal branch-length scaling factor: 2 Positions changed: 47 (37–60)Terminal branch-length scaling factor: 3 Positions changed: 92 (75–112)
Figure 6.Sensitivity of the Brachiosaurus target to terminal branch length. Positions differing from the baseline target as the sauropodomorph terminal branch is scaled from 0.25× to 3× its baseline length. Line: point estimate; band: 95% interval over 200 bootstrap replicates of the reference alignment.Source: Library v3.1, sensitivity set BR-SENS-2025-09.
Source data
Terminal branch-length scaling factorPositions changedPositions changed interval
0.253324–43
0.51611–23
0.7573–12
100–5
1.52618–34
24737–60
39275–112

Removing the theropod-peptide constraints at the Saurischia node changed 85 positions. This is the largest effect among the perturbations within threshold, and it means the strongest empirical constraint on the Brachiosaurus target, beyond the extant bracket itself, is theropod peptide evidence entering through the shared ancestor.

4.5 Library-level statistics

The theropod fragment set, its coverage and its placement statistics are unchanged from v3.0: 2,903 fragments covering 15,960 unique residues, with a median theropod placement posterior of 0.74. Adding eight reference taxa changed the library-level placement of no theropod fragment by more than 0.02 in posterior. The eight taxa are enabled in the sauropodomorph profile only: the theropod profile (theropoda-v3) retains the 63-taxon v3.0 reference set in this release, so that theropod build outputs are identical to v3.0 for identical configuration (verified by checksum across all production profiles).

4.6 Exploratory constructions

Three exploratory Brachiosaurus constructions (BR-2024-01 to -03) were made under Library v2.8, before fail-closed filter validation, using theropod peptides as tip anchors (Table 5). They were re-derived under the v3.1 profile. BR-2024-01 and BR-2024-02 had been inactive since early 2025 and were retired. BR-2024-03 differed from the v3.1 target at 61 positions, all below a posterior of 0.6, and was re-attested to v3.1 under the procedure described in JM-TR-2025-007. Two new research constructions, BR-2025-01 and BR-2025-02, were built from a v3.1 release candidate whose target set is identical to the release, and reference the v3.1 commit e06a7c4 directly.

Table 5. Brachiosaurus constructions and their v3.1 disposition.
Construction Built under Tip anchors Positions differing from v3.1 target Disposition
BR-2024-01 v2.8 Theropod (2,236) 208 Retired
BR-2024-02 v2.8 Theropod (2,236) 214 Retired
BR-2024-03 v2.8 Theropod (2,236) 61 Re-attested to v3.1
BR-2025-01 v3.1 RC None 0 Active, research
BR-2025-02 v3.1 RC None 0 Active, research

5. Discussion

The question this release has to answer is whether a build profile for an extinct species can be acceptable when it contains no sequence evidence from that species or from any member of its clade. We think it can, for the program's purpose, and we set out the argument in five parts so that each can be assessed separately.

First, the bracket is the same bracket. Birds and crocodilians bracket Sauropodomorpha exactly as they bracket Theropoda: both are saurischian dinosaurs, and both lie inside the bracket defined by crown archosaurs (Witmer, 1995). What changes is not which taxa bracket the target but how far the target lies from the nearest bracketing node. The method is unchanged; the distance is greater.

Second, the targets are conserved. The 22 gene targets are skeletal-muscle structural and regulatory genes and lipid-handling genes, chosen because they are what is grown. Most are deeply conserved across amniotes, and the posterior at the Saurischia node is high for most positions. A long branch with no data degrades a well-constrained ancestral estimate slowly.

Third, the profile does not pretend. A profile that used theropod peptides as tip anchors, as the exploratory arrangement did, would give a Brachiosaurus target with higher reported posteriors and less justification for them. The reference-only profile reports lower support (median 0.81 against 0.93) and states where that support comes from. We regard the lower number as the more accurate one.

Fourth, the result is robust to every perturbation we intend to treat as plausible. No change to branch length within a factor of two, to the substitution model, to the arrangement of the dinosaur root or to the composition of the avian reference set moved more than 0.64% of target positions. The one perturbation that exceeded the threshold removes half of the bracket.

Fifth, the purpose is food. The Brachiosaurus program aims to produce a muscle tissue with the eating qualities the reconstruction predicts. It does not aim to establish sauropodomorph phylogeny or to recover the sequence of any individual animal, and it makes no claim to have done so. The reconstruction is a reconstruction of what we intend to grow.

Taken together, these points support a narrower conclusion than might be hoped: the Brachiosaurus target is a well-characterized estimate of a saurischian muscle gene set, propagated to a sauropodomorph tip under a stated model, differing from the corresponding theropod estimate chiefly where neither is confident. That is what the profile delivers, and it is what the release notes describe as reliance on reference sequences and bracketing constraints only.

The phenotypic differences observed in early Brachiosaurus prototypes, principally in intramuscular lipid accumulation, are the subject of ongoing work by the muscle biology group. The present report does not attribute them to particular reconstructed positions. Culture and maturation conditions for the two programs differ.

The sauropodomorph profile, the reconstruction and the sensitivity analyses were presented to the Ethics & Bioresponsibility Council on 23 September 2025: reviewed; no concerns raised. Dr. Zhou is a member of the Council.

6. Limitations and known issues

  • No sauropodomorph peptides. The library contains no sequence from any sauropodomorph. Every claim about the Brachiosaurus tip is a consequence of the model, the extant bracket, and theropod evidence entering through the Saurischia node.
  • The branch length is a prior. The 79-million-year terminal branch is taken from divergence-time estimates and scaled by theropod-path rates. No sauropodomorph observation informs it. Fig. 6 shows how much this matters.
  • Shared ancestry with the Deinonychus target. Both targets inherit the Saurischia node, including its theropod-peptide constraints. Errors at that node propagate to both species. The two targets cannot be used to validate each other.
  • Positions of interest are the least certain. The 189 positions at which the Brachiosaurus and Deinonychus targets differ are concentrated among the lowest-posterior positions in the Brachiosaurus target (77.8% below 0.6).
  • Lipid-handling targets. The three targets unique to the Brachiosaurus set have mean posteriors of 0.71, 0.74 and 0.77, the lowest of the 22. They are the targets most directly associated with the program's marbling objective.
  • Specimen attribution. The two Kem Kem teeth examined were attributed to Rebbachisauridae at acquisition. This attribution was not tested, because no authenticated sequence was obtained from which to test it.
  • Prior exploratory builds. Exploratory Brachiosaurus builds run between 2022 and the v3.1 freeze used theropod tip anchors. Their outputs remain in research notebooks and are not referenced by any active construction.

7. Changes in Library v3.1

As published in the Peptide Library Release Notes:

  • Added a sauropodomorph-targeted build profile to support the Brachiosaurus program. No sauropodomorph-derived peptides are included in this release; the profile currently relies on reference sequences and bracketing constraints only.
  • Documentation updates to build-filter configuration.

Also listed in the release notes for v3.1:

  • Eight extant reference taxa (63 → 71) (Section 3.3).
  • profile: key in build configuration. Builds with no profile: key default to the theropod profile; this implicit default is deprecated (earliest removal 2027-10).
  • Manual override flags older than 24 months, deprecated in v2.8 (JM-TR-2023-006), are now ignored at build time unless renewed. At the v3.1 freeze 866 overrides lapsed on this basis, including 512 of the 1,207 legacy records attributed to jm-curate. Their lapse changed no position in any production target set (verified by checksum), and the median theropod placement posterior is unchanged at 0.74.
  • The validation report states, for each profile, the number of fossil-derived entries contributing to the target sequence set. For the sauropodomorph profile this number is zero.
  • Known issue: the sauropodomorph profile reports "N/A" for median fossil placement posterior (KI-0588).

Library v3.1 contains 2,903 theropod-assigned fossil-derived fragments (+0 sauropodomorph) covering 15,960 unique residues, with a median theropod placement posterior of 0.74, and reference sequences from 71 extant taxa. The build profile is described in Sections 3.2 to 4.4. The documentation updates close CAPA-2024-038 (JM-TR-2025-007) and describe the fail-closed filter behaviour, the target-clade check and the recorded-exception mechanism, which is no longer required for Brachiosaurus builds.

8. Data availability

Library contents are confidential. The build profile, reconstruction outputs (BR-REF-2025-09) and sensitivity set (BR-SENS-2025-09) are retained at the Shenzhen Research Center and are available to authorized research staff. Library access is not available to third parties.

9. Acknowledgements

We thank Dr. Hao Lin and the palaeoproteomics group for eleven extraction attempts on sauropodomorph material, and for reporting their outcome without qualification. We thank Dr. Sofia Marchetti for discussions of intramuscular lipid targets, and Dr. Elena Popescu for review of the sensitivity pipeline.

This work was funded internally. All authors are employees of Jurassic Meat and may hold equity. The profile was reviewed by the Ethics & Bioresponsibility Council (23 September 2025): reviewed; no concerns raised.

10. References

  1. Witmer LM (1995). The extant phylogenetic bracket and the importance of reconstructing soft tissues in fossils. In: Thomason JJ (ed.), Functional Morphology in Vertebrate Paleontology. Cambridge University Press, 19–33.
  2. Schweitzer MH, et al. (2007). Analyses of soft tissue from Tyrannosaurus rex suggest the presence of protein. Science 316:277–280.
  3. Schweitzer MH, et al. (2009). Biomolecular characterization and protein sequences of the Campanian hadrosaur B. canadensis. Science 324:626–631.
  4. Saitta ET, et al. (2019). Cretaceous dinosaur bone contains recent organic material and provides an environment conducive to microbial communities. eLife 8:e46205.
  5. Cox J, Mann M (2008). MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nature Biotechnology 26:1367–1372.
  6. JM-TR-2013-001. Endogenous peptide fragments recovered from a Kem Kem Group theropod tooth. Jurassic Meat internal report.
  7. JM-TR-2016-003. Fragments from three additional specimens with documented provenance. Jurassic Meat internal report.
  8. JM-TR-2019-004. Library v2.0: a versioned multiple-sequence-alignment architecture for fragmentary ancient peptides. Jurassic Meat internal report.
  9. JM-TR-2021-005. Library v2.3: keratin-associated peptides and the chain-of-custody metadata schema. Jurassic Meat internal report.
  10. JM-TR-2023-006. Library v2.8: diagenetic modification in placement scoring. Jurassic Meat internal report.
  11. JM-TR-2025-007. Library v3.0: commit-hash provenance for cell-line constructions, and a retrospective audit of the cross-clade build-filter defect. Jurassic Meat internal report.

Supplementary Table S1

Table S1. Distribution of Brachiosaurus target positions by anchoring class and posterior band.
Anchoring class Positions Posterior < 0.5 0.5–0.8 ≥ 0.8 Median posterior
Directly anchored 0 — — — —
Node-anchored (Saurischia and below) 544 9 116 419 0.88
Reference-only 12,729 814 5,422 6,493 0.81
Total 13,273 823 5,538 6,912 0.81