Whether animal models of type 2 diabetes mellitus predict the human efficacy of glucose-lowering drugs can only be established retrospectively, by comparing treatment effects in animals with those in clinical trials. The first such assessment (Varga et al. 2015, using rosiglitazone as a case study) found rats closer than mice to human outcomes and streptozotocin-induced models best approximating the human dose-effect relationship, but left the central questions open: HbA1c, the primary human endpoint, was too rarely reported in animal studies for a powered comparison; the contributions of cross-species dose scaling and species pharmacokinetics versus model biology were not separated; the assembled evidence is publication-biased and based on small samples; strain-level selection remained statistically inconclusive; and extrapolation beyond thiazolidinediones was untested. The only direct successor study extended the method to DPP4 inhibitors on glucose alone, confirming that HbA1c under-reporting persists and that model performance is drug-class dependent. The open problem is a bias-aware, HbA1c-inclusive, multi-drug-class predictive-validity assessment of diabetic animal models with strain-level resolution and pharmacokinetic attribution of animal-human deviations.

Background

Type 2 diabetes mellitus (T2DM) is a disorder of chronically elevated blood glucose driven by insulin resistance and progressive failure of the insulin-secreting pancreatic beta cells. Before a glucose-lowering drug reaches patients, its efficacy is tested in animal models of T2DM: chemically induced models in which the beta-cell toxin streptozotocin (STZ), with or without a high-fat diet, destroys part of the insulin-producing cells; genetic models such as the db/db mouse (a leptin-receptor mutant that becomes obese and diabetic) and the ob/ob mouse (leptin-deficient); and diet-induced obese models. Which of these models to trust for predicting human drug response directly shapes which drug candidates enter clinical trials.

Three dimensions are used to judge an animal model. Face validity asks whether the model resembles the human disease phenotypically; construct validity asks whether it arises from the same cause; predictive validity asks whether the model responds to drugs the way humans do. The first two can be argued from biology, but predictive validity can only be measured a posteriori, by comparing treatment effects observed in animal studies with those observed in human trials. At the level of biomedicine as a whole such comparisons have been sobering: systematic comparisons of animal experiments with the corresponding clinical trials found the treatment effects frequently discordant (Hackam and Redelmeier 2006; Perel et al. 2007; van der Worp et al. 2010), and in analgesia, cross-species comparison of pharmacokinetic-pharmacodynamic relationships (how drug exposure translates into effect) showed rat and human sensitivity can diverge substantially (Whiteside et al. 2008). For T2DM specifically, reviews of animal models discussed almost exclusively face and construct validity until Varga et al. (2015) performed the first retrospective predictive-validity assessment.

Varga, Zsíros and Olsson (2015) took rosiglitazone, a thiazolidinedione (an agonist of PPAR-gamma, a nuclear receptor that improves insulin sensitivity), as a case study. They fitted regressions of treatment effect (change in fasting glucose, or in glycated hemoglobin HbA1c, the modified hemoglobin fraction that reflects average glycemia over preceding weeks and is the primary efficacy endpoint of human diabetes trials) on dose and treatment duration in human data, including studies that GlaxoSmithKline was compelled to disclose during litigation; used the fitted coefficients to predict the corresponding animal effects; and scored each species, strain, and diabetes-induction method by the deviation between observed and expected effects. Rats deviated significantly less from the human-expected effects than mice; STZ-induced models best approximated the human exposure-effect relationship, especially when the effect was ascertained through HbA1c; the best strain could not be settled statistically; the HbA1c analysis was in places 'not practicable' because animal studies rarely reported it; and the authors cautioned that extrapolation outside the thiazolidinedione class 'should be avoided'. Animal doses (6-20 mg/kg) were higher than human-equivalent doses, and although the analysis converted doses across species by the FDA-guideline scaling factors (dividing rat doses by 2.25 and other murine doses by 11.25) and also checked a categorical low/high-dose cut-off at 6.5 mg, the authors noted that 'in an ideal case, free plasma concentration of rosiglitazone could be compared between human and animal subjects' before attributing deviations to model biology.

The case study exposed the general methodological problem: predictive-validity estimates for diabetic animal models rest on a heterogeneous, small-sample, publication-biased literature, and the retrospective procedure had been demonstrated on a single drug class and essentially a single outcome. The natural completion of the program, a multi-drug-class, HbA1c-inclusive, bias-aware predictive-validity assessment that resolves strain-level model selection and separates pharmacokinetic from biological sources of animal-human deviation, has not been delivered.

Problem Statement

Determine the predictive validity of animal models of type 2 diabetes mellitus for the human clinical efficacy of glucose-lowering drugs, at species, strain, and diabetes-induction-method resolution. As posed by the rosiglitazone case study and left open by it, the question comprises: (i) whether the observed concordance between animal and human treatment effects generalizes from thiazolidinediones to the other major antidiabetic drug classes, measured on both fasting glucose and HbA1c, the primary human endpoint that animal studies rarely report, including how broadly the STZ-induction advantage holds across strains and species; (ii) how much of the observed animal-human deviation is attributable to cross-species dose scaling and species-specific pharmacokinetics rather than to the biology of the model; (iii) how the predictive-validity estimates change when the publication and small-sample biases of the assembled literature are quantified and corrected; and (iv) which specific strains within species consistently predict human outcomes across multiple drugs and outcomes, so that model selection can be prescriptive rather than tradition-based. These components jointly serve one objective: an evidence base stating which diabetic animal models predict human antidiabetic-drug efficacy, with quantified uncertainty. A complete answer addresses all four components, each a distinct required part of the objective; no subset of them resolves the task. Where a component cannot be answered affirmatively because the assembled literature lacks the requisite data, a documented data-insufficiency finding resolves that component only when it is accompanied by the estimate the available data do support and by an explicit statement of what remains unassessable: for component (ii), the share of the animal-human deviation attributable to cross-species dose scaling, with the pharmacokinetic contribution separated as far as exposure data permit; and for component (iv), a strain-level ranking, or a stated statistical null, over the strains the data support. Because the objective is HbA1c-inclusive, a glucose-only assessment does not by itself resolve components (i) or (iii): each is resolved only when the HbA1c comparison is addressed in its own right, affirmatively where the literature permits a powered comparison and otherwise by a data-insufficiency finding specific to HbA1c, and, where HbA1c cannot be compared, the corresponding fasting-glucose-based estimate is supplied as well.

References

  1. Varga OE, Zsíros N, Olsson IAS (2015) Estimating the predictive validity of diabetic animal models in rosiglitazone studies. Obesity Reviews 16(6):498-507. doi:10.1111/obr.12278, https://doi.org/10.1111/obr.12278
  2. Franco NH, Miranda SB, Kovács N, Nagy A, Thiện BQ, Reis F, Varga O (2021) Assessing Scientific Soundness and Translational Value of Animal Studies on DPP4 Inhibitors for Treating Type 2 Diabetes Mellitus. Biology 10(2):155. doi:10.3390/biology10020155, https://doi.org/10.3390/biology10020155
  3. Franco NH, Miranda SB, Kovács N, Nagy A, Akinsolu FT, Olsson IAS, Varga O (2020) Trends in animal model preference for preclinical drug testing for type-2 diabetes and future directions. bioRxiv preprint. doi:10.1101/2020.03.03.973230, https://doi.org/10.1101/2020.03.03.973230
  4. Ferreira GS, Veening-Griffioen D, Boon WPC, Hooijmans CR, Moors EHM, Schellekens H, van Meer PJK (2020) Comparison of drug efficacy in two animal models of type 2 diabetes: A systematic review and meta-analysis. European Journal of Pharmacology 879:173153. doi:10.1016/j.ejphar.2020.173153, https://doi.org/10.1016/j.ejphar.2020.173153
  5. Veening-Griffioen DH, Ferreira GS, Boon WPC, Gispen-de Wied CC, Schellekens H, Moors EHM, van Meer PJK (2021) Tradition, not science, is the basis of animal model selection in translational and applied research. ALTEX 38(1):49-62. doi:10.14573/altex.2003301, https://doi.org/10.14573/altex.2003301
  6. Hackam DG, Redelmeier DA (2006) Translation of Research Evidence From Animals to Humans. JAMA 296(14):1731-1732. doi:10.1001/jama.296.14.1731, https://doi.org/10.1001/jama.296.14.1731
  7. Perel P, Roberts I, Sena E, Wheble P, Briscoe C, Sandercock P, Macleod M, Mignini LE, Jayaram P, Khan KS (2007) Comparison of treatment effects between animal experiments and clinical trials: systematic review. BMJ 334(7586):197. doi:10.1136/bmj.39048.407928.be, https://doi.org/10.1136/bmj.39048.407928.be
  8. van der Worp HB, Howells DW, Sena ES, Porritt MJ, Rewell S, O'Collins V, Macleod MR (2010) Can Animal Models of Disease Reliably Inform Human Studies? PLoS Medicine 7(3):e1000245. doi:10.1371/journal.pmed.1000245, https://doi.org/10.1371/journal.pmed.1000245
  9. Whiteside GT, Adedoyin A, Leventhal L (2008) Predictive validity of animal pain models? A comparison of the pharmacokinetic-pharmacodynamic relationship for pain drugs in rats and humans. Neuropharmacology 54(5):767-775. doi:10.1016/j.neuropharm.2008.01.001, https://doi.org/10.1016/j.neuropharm.2008.01.001