Validation and utilization of preclinical models for drug development

(Steve Perrin, ALS Therapy Development Institute)

Amyotrophic lateral sclerosis (ALS) case study: The hSOD1G93A transgenic ALS mouse model emulates human disease pathologies; the model’s disease course also recapitulates the human disease.

  • However, whether or not the model has demonstrated translational validity remains unclear.
    • Riluzole, a drug approved for ALS treatment, has a marginal effect in the mouse model and in human patients.
    • Many other putative therapeutics have demonstrated efficacy in the model yet do not translate to patients.

Drugs that have preclinical efficacy in delaying disease onset and extending survival in ALS mouse models (based on published studies):

  • Celebrex, a COX-2 inhibitor approved as a nonsteroidal anti-inflammatory drug (See REF, Drachman 2002).
  • Minocycline, a tetracycline antibiotic (See REF, Zhu 2002).
  • Ceftriaxone, a cephalosporin antibiotic (See REF, Rothstein 2005)
  • Lithium, approved as a drug for manic episode of manic depression (See REF Fornai 2008).

However, when Celebrex, minocycline, and lithium were tested in ALS clinical trials, there were no beneficial effects on clinical outcomes in patients.

  • Celebrex, although safe and well tolerated, had no impact on clinical outcomes (See REF Cudkowicz 2006).
  • Minocycline actually exacerbated disease progression and mortality in a phase III clinical trial (See REF, Gordon 2007).
  • In all cases, the resource cost was high (e.g. $50+ million cost, time)

In contrast to the published preclinical studies, animal model data from the ALS Therapy Development Institute (TDI) have matched clinical trial data in demonstrating no beneficial clinical impact with the various drug treatments (at the doses stated in the published studies).

  • Celebrex, ceftriaxone, and lithium all had no effect on disease onset or survival.
  • As observed in the clinical trial, minocycline exacerbated the disease.
  • Looking back at the previously published preclinical studies, there were various caveats including limited numbers of animals per experimental group (10 or fewer) and limited information on study design parameters (e.g. gender, littermates).

There are likely unrecognized confounding variables that affect study outcomes. A consistent positive control drug is needed.

  • Among the variables that contribute to noise in animal models (based on TDI database):
    • Censoring – Need to remove non-disease related deaths.
    • Low copy number – Genetic drift to low copy number transgenic animals can lead to false positives. Copy number has a big significant impact on median survival in the ALS animal models.
    • Gender
    • Litter

Example of an optimized study design:

  • n=48 total mice
    • A larger N decreases the noise frequency of apparent effects (see FIG. 4).
  • Experimental group of 12 males and 12 females, control group of 12 males and 12 females
  • Same gender litter matching
  • Observers blind to treatment
  • Single, uniform endpoint criterion
  • Confirmation of transgene copy number prior to enrollment
  • Tracking and censoring from final analysis all non-ALS deaths
  • Statistical analysis: A log rank method is typically necessary for any survival analysis; since we have shown the hSOD1G93A model has multiple variables, the Cox Proportional Hazards model is the most appropriate.
  • No multi-arm studies
  • An important problem is the expense – for this optimized study design for testing one drug at one dose, the cost is about $120,000, which is prohibitive for many academic labs.

Computational approaches can be used to drive unbiased discovery and derive hypotheses for druggable targets.

TDI novel target discovery example: Microarray profiling (Affymetrix) of multiple tissue types from the hSOD1G93A transgenic mouse at various time points, followed by computational analyses of interacting signaling pathways:

  • In the three most relevant tissues (muscle, spinal cord, sciatic nerve) and at the time of disease onset, 5 pathways consisting of 95 total genes were discovered with this “interactome” approach. These pathways are involved in costimulatory and humoral response immune system signaling, which are novel and never previously described as possible ALS targets.
  • These signaling pathways were validated with other methods to examine expression patterns. It was discovered that macrophages progressively infiltrate and accumulate in nerves from disease onset.
  • This was further validated in ALS patients – blood samples and muscle biopsies from 300 patients demonstrated that the costimulatory pathway is highly upregulated in 60% of the group.
  • Preclinical testing of a blocking antibody targeting CD40L (a costimulatory molecule) demonstrated efficacy in hSOD1G93A transgenic mice in delaying disease progression and improving survival and body weight loss. This antibody was the only treatment that was beneficial for these three parameters (out of all the compounds tested by TDI).
  • PK and tolerability studies were done in the hSOD1G93A mice.
    • It is strongly recommended to do these studies in the actual mouse line that you will be working with (instead of using previous data from other lines), since the results can be significantly different depending on the model.
  • Looking at the mechanism of action, it was found that the CD40L antibody treatment significantly decreased costimulatory pathway gene expression by about 40%. The treatment reduced axonal recruitment of macrophages and astrocytosis and improved motor neuron survival.

Performing PD studies in parallel, using multiple methods, can help lower attrition later on.

  • The TDI CD40L antibody had opposite effects depending on the dosing time point (before/at onset); an in-situ hybridization study demonstrated differential inflammatory responses that were dependent on dosing time point and therefore informed the dosing of other experiments.

In determining the suitability of an animal model, it is very important to look at the variance in the intended measure.

  • EXAMPLE: In the prpTDP43A315T mouse model, there is a 150 day variance in survival – this means that 400 animals are needed per group in order to detect a 5% drug effect.
  • To reduce the variance, backcross to create a cogenic line and characterize the noise variables

Decent quality preclinical data can be obtained from animal disease models but care needs to be taken to minimize variance and implement optimized study design guidelines.

Working Effectively with Your TTO: Technology Evaluation Process and IP Protection >