Tecniplast

Half of laboratory mice are not what scientists think they are. A new genetic survey published in Science analyzed 341 mouse strains from the Mutant Mouse Research and Resource Centers. 47% of strains were genetically inconsistent with how they were described. 7% belonged to an entirely different strain. 26% belonged to a different substrain. Nearly 10% carried genetic changes, including reporter genes, that were not in the strain name at all. The reproducibility debate keeps circling the same target. Animal models. NAMs. Translation. Predictive value. This study reframes the problem. If you cannot identify the model you used, every downstream comparison is compromised. Cross-study replication, regulatory submission, mechanistic interpretation. All of it depends on knowing what the animal actually is. One commentator quoted in the article argues that replication still holds if two labs use the same misnamed strain. True for narrow replication. False for everything else. You cannot generalize to humans, compare across cohorts, or interpret mechanisms when the genetic background is undocumented. This is not only an animal research problem. It is a metadata problem. The same logic applies to organoids, organs-on-chip, and computational models. A NAM with undocumented donor variability or unspecified culture parameters fails for the same reason a mislabelled C57BL/6 fails. Different layer of the stack. Same root cause. Data infrastructure is the precondition for everything else. ARRIVE 2.0. SEND. FAIR. Minimal metadata sets. These are not bureaucratic overhead. They are the only way any model, biological or computational, becomes auditable. The Pistoia Alliance Minimal Metadata Set Working Group exists for this reason. You cannot fix translation without first fixing identification, provenance, and traceability. The next decade of preclinical research will not be decided by the animal vs NAM question. It will be decided by which organizations build the data infrastructure to know what they are actually working with. Half the mice were wrong. The metadata was missing. Reference: Pardo-Manuel de Villena, F. et al. Science 392, 698–700 (2026).

Yesterday in DC. Morning with regulators and NIH. Evening keynote at AAALAC International on AI in research and data infrastructure. Different rooms, same fault line. The fault line is context of use. In the morning, every hard question about animal models and NAMs resolved the moment the discussion moved from "which method is better" to "which method fits this biology, at this stage, for this question." What looks like scientific disagreement in the current regulatory moment is usually categorical disagreement about the question being asked. In the afternoon, the same pattern in a different room. AI does not rescue preclinical research. It exposes whether the data infrastructure underneath it is good enough to be analyzed. Without context of use written into the metadata, machine learning becomes pattern matching on noise. Context of use is the connective tissue between regulators, scientists, and methods. It is also the layer most institutions still treat as an afterthought, then wonder why their AI initiatives stall and their translational signals weaken. Familiar faces. Honest questions. Thanks to Gary for the invitation. #AAALAC #ContextOfUse #AnimalResearch #NAMs #AIinResearch #PreclinicalScience #DataInfrastructure #FAIRdata

The most repeated sentence in preclinical science today is also the most misleading. "NAMs will replace animal models." That sentence collapses three problems into one slogan. It conflates chemical hazard assessment with drug development. It treats validation as a binary instead of a Context of Use performance standard. It assumes human-derived material delivers human relevance without further demonstration. None of these survives contact with the regulatory documents now being published. I spent the last eighteen months writing on this topic. Articles. Manuscripts. Posts. Calls with FDA reviewers, NIH ORIVA leadership, EMA 3Rs Working Party stakeholders. I compiled the synthesis. A few things I no longer believe. NAMs are not the universal solution to drug attrition. Roughly 90% of drugs fail in clinical trials. Animal models account for about 20% of that failure. The rest is commercial, regulatory, and operational. Blaming animals is convenient. It is not accurate. FDORA did not remove the requirement for animal testing. It replaced the phrase "preclinical tests" with "nonclinical tests" and gave examples. FDA already had the authority. The change was language, not regulatory practice. Thalidomide is not an argument against animal testing. The drug was never tested on pregnant animals before being given to pregnant women. When rabbits were tested in 1962, the teratogenic signal was unmistakable. The animal model worked. It was never used. The Emulate Liver-Chip achieves 87% sensitivity and 100% specificity for DILI prediction (Ewart et al., 2022, Communications Medicine). That is a measurable performance advantage over animal models for a specific endpoint. It is also one endpoint. The biggest 3Rs success in safety pharmacology came from refinement, not replacement. Cross-over telemetry design saved more than 4,000 non-rodents across 360 FDA-approved molecules. No platform. No organoid. Better experimental design (Derakhchan et al., 2026). The CHMP qualification opinion of 31 March 2026 is the first EMA NAM qualification for toxicological assessment. Virtual control groups. The bottleneck was never the algorithm. It was the metadata. The Pistoia Alliance MNMS framework (Moresis, Gaburro et al., Lab Animal 2024) addresses exactly that gap. Without harmonized metadata, historical data cannot be pooled. Without pooling, there are no virtual controls. Context of Use governs. Not which tool. Which question. And whether the data, from any source, answer it. Animal models earned their place through insulin, polio, organ transplantation. Defending the status quo without acknowledging documented failures is unsustainable. NAMs are not enemies of animal research. They are part of the same toolkit, applied to different problems. Full synthesis below. #NAMs #AnimalResearch #3Rs #RegulatoryScience #ContextOfUse #VirtualControlGroups #PistoiaAlliance

An FDA Commissioner resigns. The NAMs roadmap does not. Marty Makary stepped down yesterday after roughly thirteen months at the FDA. Kyle Diamantas, the agency's top food regulator and a lawyer by training, takes over as acting commissioner. Many will read this as a setback for the new approach methodologies agenda. That reading is incomplete. The April 2025 Roadmap to Reducing Animal Testing in Preclinical Safety Studies was not a personal initiative. It was institutionalized inside CDER. It moved through cross-center scientific reviews, a permanent tool-qualification pathway, an NIH partnership, and a sequence of draft guidances. Monoclonal antibody non-human primate testing in December 2025. Weight-of-evidence NAMs in March 2026. A year-one progress report in April 2026. That infrastructure does not disappear when a Commissioner leaves. What can change is pace and political amplification. Commissioners set tone. They unlock attention. They protect controversial science from internal pushback. With an acting commissioner from food regulation, and a confirmation calendar not yet defined, the NAMs file will compete with mifepristone, vapes, rare-disease approvals, and whatever the next Commissioner inherits. The risk is not reversal. The risk is deprioritization. Two reminders worth keeping in front of every discussion this week. First, FDORA (2022) replaced "preclinical tests including tests on animals" with "nonclinical tests." It did not remove the requirement for animal testing. The Roadmap is an FDA strategy, not a statutory mandate. Strategies survive leadership transitions. They also slow down without sponsors. Second, NAMs are not a replacement-by-default. They are surrogates of human disease. So are animal models. Context of use determines the method. The substantive work in 2026 is not "animals versus organoids." It is qualification, lifecycle management, and digital biomarkers that bridge in vivo data with human-relevant readouts. What I will be watching: - Whether the March 2026 draft guidance moves to final within the original timeline. - Whether the NIH partnership continues to receive funding signals. - Whether CDER review divisions keep accepting NAMs data in IND submissions at the rate observed in 2025–2026. - Who the next Commissioner is, and whether the nominee has a position on nonclinical strategy. The NAMs trajectory was never about one person. It rests on regulatory science inside CDER, scientific consensus across academia and industry, and the precompetitive infrastructure that makes data interoperable in the first place. Leadership changes the speed. It does not change the direction. Watch the guidances. Watch the funding. Watch who replaces Makary.

Most preclinical labs are proud of their experiments. Few are proud of their metadata. That gap is the real bottleneck in translational research. Not the animal model. Not the technology stack. The data infrastructure underneath. I worked with Damien Huzard while editing two chapters of the Springer book on home cage monitoring at Pistoia Alliance . He coordinated content across multiple contributors and held a quality bar that did not move. A short profile on someone the field should know better. Background PhD at EPFL. Two postdocs at CNRS/INSERM Montpellier. First-author papers in Science Advances and Translational Psychiatry. EPNA Award. About a thousand citations. That is the academic surface. What he is actually building In 2024 he founded Neuronautix, a scientific consultancy in Montpellier. The premise is unfashionable. Most consultancies sell strategy decks. Damien ships working pipelines, ontology files, and Python libraries that other labs can run. All open source: . HCMO. Home-Cage Monitoring Ontology with SHACL validation. . MBO. Mouse Behavior Ontology with operational definitions. . LWTools. Python package for LiveMouseTracker, in use across European labs. . MetaDatApp. API-first FAIR metadata platform, CNRS-funded. . FAIR-VCG. LLM-assisted Virtual Control Group generator. . Science Agent Squad. Multi-agent framework for HCM literature synthesis. The argument The animal models versus NAMs debate is a distraction from a more boring fact. Most preclinical data is not reusable. Until it is, neither approach can be properly evaluated against the other. Virtual control groups. Historical control reuse. Cross-laboratory replication. All require metadata that nobody is generating at scale yet. Damien has chosen to build that layer. His recent piece in Neuroscience Applied, "Data welfare is animal welfare," makes the point bluntly. Wasted data from well-conducted experiments is a 3Rs problem. That reframing matters. It moves data infrastructure from the IT department to the ethics committee. Why it deserves attention Field-built ontologies survive contact with the lab. Library-built ontologies often do not. A consultant who ships open-source code and validated ontologies is doing something different from selling slides. The deliverable outlives the engagement. Damien is the kind of scientist the field needs more of and rewards less than it should. If you run home cage monitoring studies or want your preclinical data reusable in five years, look at his work. neuronautix.com Sometimes the most important contributions are made by people who decided to fix the boring problem first. #PreclinicalResearch #HomeCageMonitoring #FAIRdata #DigitalBiomarkers #3Rs #OpenScience

"Generative AI will reduce animal use in preclinical research." A new paper in Pharmacological Research tests this claim properly. The result is more interesting than the slogan. The setup. A lipidomics dataset from 26 mice in an experimental autoimmune encephalomyelitis model. Three groups. 62 lipid mediators. The authors artificially shrink the dataset to n=6 per group. Statistical significance collapses. The known top hits disappear. Then they apply genESOM, a generative AI based on emergent self-organizing maps, to augment the reduced data. The original top hits reappear. Lysophosphatidic acids. Sphingolipids. Ceramide species. Treatment groups separate again. Gaussian mixture models and CTGAN fail under the same conditions. The conclusion the press release wants. AI can replace animals. The conclusion the authors actually write. Synthetic data cannot substitute for biological replication. Augmentation does not create new biological information. It stabilizes detection of signals already present in the observed data. Statistical inference must be interpreted conditionally, because synthetic points do not increase true degrees of freedom. This is not an animal reduction tool in the way 3Rs language usually implies it. It is an exploratory analysis stabilizer. The mice were still used. The signal was still generated by biology. The AI recovers what is already there. It cannot recover what was never sampled. For exploratory pharmacology, this is useful. Hypothesis generation. Lipidomics screens where the question is "is there anything here worth following up." The genuine methodological contribution is the embedded error inflation control via dimensionality modulation, with a data-driven stopping criterion that halts augmentation at one synthetic point per original. That is what stops genESOM from doing what GMM and CTGAN did in the same paper. For confirmatory work, the answer flips. You cannot augment your way out of an underpowered pivotal study. The degrees of freedom do not exist. The misuse pathway is obvious. Someone reads "30 to 50 percent animal reduction" in the abstract and runs a confirmatory study at n=6 with a genAI rescue plan. The signal recovered will be the signal already there. The signal that was absent will stay absent. The honest reading. This paper is a clean contribution to small-sample exploratory analysis. It does not reduce the mice in the experiment that generated the data. It might reduce the next experiment, by clarifying which signals are worth pursuing. That is reduction by better study design, not by replacement. Translation is not only a sample size problem. It is also an endpoint problem, a power problem, a metadata problem, and an incentive problem. Generative AI is one tool. It sits inside the analytical stack. It does not sit instead of it. Lötsch et al. 2026, Pharmacological Research. Open access.

Pfizer killed danuglipron in April 2025. A Nature paper just published last week explains why the field needed a different mouse. Most people assume that if a drug works in humans, it must work in mice. For the new oral GLP-1 weight-loss drugs (danuglipron, orforglipron, Eli Lilly's just-approved Foundayo), that assumption is wrong. Standard mice have a serine at position 33 of their GLP-1 receptor. Humans have a tryptophan. Small-molecule GLP-1 drugs bind the human version. Not the mouse. So for years, the preclinical pharmacology of an entire class of weight-loss medications was effectively running blind. Godschall and colleagues at UVA fixed that. CRISPR-Cas9. One amino acid swap. A humanized Glp1r mouse that responds to oral small-molecule GLP-1 agonists the way humans do (Godschall et al., 2026, Nature). A second group at Gubra and Terns published a parallel humanized line in EBioMedicine the same window (Sonne et al., 2026). What they found, once the right mouse existed, matters. A new reward circuit. Central amygdala GLP1R neurons project to the ventral tegmental area and dampen dopamine release in the nucleus accumbens. The result: small-molecule GLP-1 drugs selectively suppress palatable food intake without touching standard chow consumption. This is the mechanistic basis for the clinical signals showing GLP-1RAs reduce alcohol and cannabis use disorder incidence (Wang et al., 2024). It is also why long-term effects on motivated behavior in chronic users deserve serious attention. Three things stand out for anyone working in translational science. One. The CRISPR humanized mouse is not a substitute for animal research. It is animal research, done right. Context of use is the governing framework. The wild-type mouse was the wrong model for this molecular class. The humanized mouse is the right one. Two. Conditioned taste avoidance and open field tests could not distinguish nausea from satiety for danuglipron and orforglipron. Continuous home-cage video monitoring with pose tracking (SLEAP) and unsupervised behavioral segmentation (Keypoint-MoSeq) could. Digital phenotyping is not optional anymore. It is the signal-to-noise upgrade that turns weak readouts into discriminative ones. Three. Pfizer discontinued danuglipron after a drug-induced liver injury in a dose-optimization study. The Nature authors state plainly that preclinical profiling in the right model "might have identified" the side-effect risk earlier. Not certain. But plausible. And avoidable. The animal-research-versus-NAMs debate is not the right debate. The right debate is which model for which question, with which endpoints, structured under FAIR data, ready for regulatory scrutiny. This paper is one of the cleanest demonstrations of that argument I have read this year. #TranslationalScience #GLP1 #DigitalBiomarkers #ContextOfUse #AnimalResearch #Pharmacology #DrugDevelopment

Most CCS documents reference local isolates. Few CCS documents actually use them. The difference is what an inspector finds when the CCS is opened next to the environmental monitoring trend report and the GPT release records. Annex 1 paragraph 2.5 lists sixteen CCS elements. The local isolate panel touches at least five directly. Monitoring systems. Cleaning and disinfection. Prevention mechanisms. Process risk management. Continuous improvement. A CCS that uses local isolates as connective thread reads as one document. A CCS that mentions local isolates in three sections without cross-reference reads as a checklist. The structural difference is visible to an inspector within fifteen minutes. The bridge between the data and the CCS is trend analysis. Without trend analysis, EM data is a stack of CFU counts. With trend analysis, EM data is a signal that drives panel decisions. A sustained shift in Gram-negative recovery in a specific Grade C location is a signal. The signal triggers a review of cleaning and disinfection in that location and a review of the panel composition for the qualification studies covering that location. The review may or may not change the panel. The review itself is the audit-relevant action. The CCS is not binary between integrated and fragmented. It is a maturity continuum. The diagnostic is not the number of isolates. The diagnostic is the connectivity. #ContaminationControl #Annex1 #GMP #QCMicrobiology #SterileManufacturing

A local isolate stops being a local isolate the moment it stops resembling the organism on the original environmental monitoring plate. Phenotype drift is silent. Inventory loss is silent. Panel staleness is silent. All three are visible at audit. Three biological mechanisms drive drift in cleanroom isolates. Loss of accessory function. Selection for fast-growing variants. Reversion of the biofilm and stress-adapted phenotype to the planktonic, unstressed phenotype. None of these mechanisms is hypothetical. All three are documented in environmental microbiology literature on cleanroom and water-system organisms. The procedural firewall is the master and working cell bank architecture. Master cell bank. Original aliquoted population. Frozen at earliest possible passage. Never used for routine QC challenge. Working cell bank. Generated from a master aliquot, minimal expansion, used for QC challenge organisms. When the working bank is exhausted, a fresh working bank comes from a new master aliquot. Never from a thawed working aliquot. USP 1117 recommends keeping passages from seed to working culture to five or fewer. Five is the discipline. Drift increases beyond it. Without this architecture, every freeze-thaw cycle and every sub-culture moves the panel a small distance from the original recovery. After three years the panel resembles the laboratory more than the cleanroom. The audit observation is not about the isolate that drifted. It is about the SOP that did not catch the drift. #PharmaceuticalMicrobiology #QualityControl #Annex1 #SterileManufacturing #GMP