This week I tuned into ARDD 2025 from home. Alex Zhavoronkov’s choice to livestream the whole thing was such a win. It felt like I could be part of the conversations without getting on a plane.
One talk really stuck with me. I don’t remember the speaker’s name or even the title, but the idea hit hard. It wasn’t just about microbes and mice - it was about how the plain, unchanging lab environment itself shapes what we see in animal studies.
For decades, laboratory mice have been the quiet workhorses of biomedical research. Everything about them is designed for consistency. But here’s the problem: life in the real world is anything but consistent.. And yet we treat those studies as the ultimate gold standards.
It made me think deeply - How many assumptions like that do we carry around in science without questioning? Or in many other cases, when we question - we receive a defensive answer.
The speaker showed how lab bred mice suffer from depauperate microbiomes and feeble immune systems that skew the conclusions of preclinical results. Slides from the talk showed how generations of SPF maintenance lead to mice with distorted microbiomes, making them poor proxies for human responses. For instance, exposure to pet store mice or natural environments "normalizes" their microbiota and boosts differentiated memory CD8+ T-cells in blood, female reproductive tract, and other tissues, mimicking adult human or neonatal patterns far better than sterile lab conditions. This is backed by studies showing that "wildling" mice—those with natural microbiomes—better capture human immune dynamics in inflammation and infection models, where conventional lab mice fail. And there was a case shared of the CD28 superagonist antibody TGN1412 tested well in lab mice (ex vivo and in vivo) but caused no proinflammatory issues; however, it led to cytokine storms in humans during Phase I trials.
One talk really stuck with me. I don’t remember the speaker’s name or even the title, but the idea hit hard. It wasn’t just about microbes and mice - it was about how the plain, unchanging lab environment itself shapes what we see in animal studies.
For decades, laboratory mice have been the quiet workhorses of biomedical research. Everything about them is designed for consistency. But here’s the problem: life in the real world is anything but consistent.. And yet we treat those studies as the ultimate gold standards.
It made me think deeply - How many assumptions like that do we carry around in science without questioning? Or in many other cases, when we question - we receive a defensive answer.
The speaker showed how lab bred mice suffer from depauperate microbiomes and feeble immune systems that skew the conclusions of preclinical results. Slides from the talk showed how generations of SPF maintenance lead to mice with distorted microbiomes, making them poor proxies for human responses. For instance, exposure to pet store mice or natural environments "normalizes" their microbiota and boosts differentiated memory CD8+ T-cells in blood, female reproductive tract, and other tissues, mimicking adult human or neonatal patterns far better than sterile lab conditions. This is backed by studies showing that "wildling" mice—those with natural microbiomes—better capture human immune dynamics in inflammation and infection models, where conventional lab mice fail. And there was a case shared of the CD28 superagonist antibody TGN1412 tested well in lab mice (ex vivo and in vivo) but caused no proinflammatory issues; however, it led to cytokine storms in humans during Phase I trials.
Reasons why lab bred mice(or other species?) are not the gold standards to First in Human scaling
- Metabolic and Transporter Mismatches: beneath the environmental layer sits a more intractable difference: metabolism and transport. Inbred mouse strains are artificially homogeneous, lacking the genetic and metabolic diversity of human populations. Their responses to high-fat diets, their nutrient processing, and even circadian regulation vary not just by strain but by sex and should be questioned on their role in predicting human metabolic disease outcomes.
- Molecular transporters are equally species-specific: glucose transporters in brain metabolism function differently in mice, leading to poorly aligned models of Alzheimer’s and other neurological disease. A huge evidence is there on the expression and function of transporters, for example, organic anion transporters (OAT1/OAT3) differ, undermining predictions about drug uptake, metabolite clearance, and toxicity in humans. Many such differences are actually predicted or hypothized by PBPK models. Transcriptomic atlases further depict differences in energy metabolism across organs, while differences in circadian biology (mice are nocturnal) introduce confounders in metabolic translation to diurnal humans.
- Most salient: over 90% of drugs that succeed in animal models fail in clinical trials for humans, with metabolic and transporter issues as frequent culprits. Moreover, poor experimental design- small sample sizes, lack of blinding, and underreported variables are usually the largest gaps of translation and reproducible work.
Why NAMS?
Faced with these gaps, NAMs increasingly make sense for anchoring contemporary biomedicine. Organoids, Organs-on-a-chip, High-throughput approaches and computational omics offer human relevant data without ethical and practical drawbacks of animal testing. However, they require the open mind of the industry to establish rigorous practices and investigation to validate their predictive power, specifically for complex processes like metabolism and transport or immunogenicity, that directly impacts the drug drug interactions scenarios.
High throughput toxicogenomic screening are now becoming more common in drug discovery pipeline which enables some scalable alternatives to animal testing. Organ on chip have shown promise in modelling human transport and reabsorption in many cases. A lot of these techniques still outline strong need for regulatory validation. On the other hand regulatory agencies (FDA, EMA) now actively encourage NAM development and validation, recognizing that rodent extrapolation is too unreliable for high-stakes toxicology, metabolic syndrome modeling, and personalized medicine. Strategies include integrating patient data, AI-driven modeling etc.
NAMs, coupled with PBPK modeling and advanced bioengineering, can and will enable human-centric science and have the potential to partially accelerate the discovery pipeline and we must try to do more to keep those models alive even at the edges of translation to clinics. The field now requires not just technical skill, but systems-level reasoning and methodological humility: recognizing the boundaries and blind spots of our models, and actively seeking alternatives that move discovery toward human health, not just convenience of animal studies.
High throughput toxicogenomic screening are now becoming more common in drug discovery pipeline which enables some scalable alternatives to animal testing. Organ on chip have shown promise in modelling human transport and reabsorption in many cases. A lot of these techniques still outline strong need for regulatory validation. On the other hand regulatory agencies (FDA, EMA) now actively encourage NAM development and validation, recognizing that rodent extrapolation is too unreliable for high-stakes toxicology, metabolic syndrome modeling, and personalized medicine. Strategies include integrating patient data, AI-driven modeling etc.
NAMs, coupled with PBPK modeling and advanced bioengineering, can and will enable human-centric science and have the potential to partially accelerate the discovery pipeline and we must try to do more to keep those models alive even at the edges of translation to clinics. The field now requires not just technical skill, but systems-level reasoning and methodological humility: recognizing the boundaries and blind spots of our models, and actively seeking alternatives that move discovery toward human health, not just convenience of animal studies.
The sterility we proud for control is also what makes the results less transferable to humans. The cost of ignoring this isn’t abstract.
