Species, Route and Sample Size: Why Every Figure Carries Them

A mouse dose is not a quantity of exposure

Billon and colleagues gave six male C57BL/6J mice per group, twelve weeks old, 50 mg/kg of SLU-PP-332 by intraperitoneal injection for six days, then ran them on a treadmill to exhaustion. The treated animals ran roughly 70 percent longer than vehicle-treated animals. What survives repetition of that experiment is the phrase 50 mg/kg. Species, route and group size drop away with each retelling, until the figure reads as a property of the molecule rather than a record of one protocol in a single strain and sex.

Each discarded field changes what the number means. Species and body mass determine whether a milligram per kilogram in a 25 gram animal represents comparable exposure in a 60 kilogram one. Route determines whether the compound reached circulation at all, by which path, and whether it met the liver before the rest of the body. Group size determines how much of the reported magnitude is signal rather than noise.

The body surface area conversion and what it was built for

The conversion applied to rodent figures traces to one regulatory document: the FDA guidance for industry of July 2005, Estimating the Maximum Safe Starting Dose in Initial Clinical Trials for Therapeutics in Adult Healthy Volunteers, from the Center for Drug Evaluation and Research. Its Table 1 assigns each species a km factor and directs that an animal dose in milligrams per kilogram be divided by a species-specific figure to obtain a human equivalent dose for a 60 kilogram adult; the divisors are stated in the guidance itself and are not reproduced here. Reagan-Shaw, Nihal and Ahmad restated that table in FASEB Journal in 2008, after resveratrol mouse studies were reported with the animal dose converted on body weight alone.

What the document is for is narrower than its circulation implies. Its algorithm begins at a no observed adverse effect level from animal toxicity studies and works toward a maximum recommended starting dose for a first-in-human trial, with a safety factor applied afterwards. The guidance states that it does not address dose escalation or maximum allowable doses, and that it is not pertinent to endogenous hormones and proteins used at physiologic concentrations or to prophylactic vaccines. The empirical basis it cites is work from 1966 and 1970 showing that lethal and maximum tolerated doses of antineoplastic drugs correlated across species when expressed per square metre. That is a finding about toxicity thresholds.

No source was found establishing that the same table, applied to a dose that produced an effect in a rodent, predicts a dose that would produce that effect in a person. PubMed was searched on 17 August 2026 for validations of human equivalent dose estimates against efficacy endpoints, for body surface area dose translation, and for interspecies scaling of pharmacodynamic rather than toxicological endpoints; the retrieved work scales pharmacokinetics. Running an efficacy figure through the safety table is recorded here as untraced.

The exponent is contested inside the guidance itself

Later analyses disagreed with the exponent the table uses. The guidance records that work published in 1988 and 1992 found maximum tolerated doses for that drug set were fitted more closely by body weight raised to 0.75 than by the 0.67 exponent implied by body surface area, and that surface area normalisation was retained regardless because it returned a more conservative starting-dose estimate. Part of the convention is an empirical claim and part is a safety preference, and the document separates the two. Quoting the output without that separation makes the arithmetic look more settled than its source.

Method choice moves the result by more than rounding. The same guidance states that scaling on a milligram per kilogram basis returns human equivalent doses several-fold higher than the surface area approach, by a margin that widens as the species gets smaller, and it names three categories in which surface area scaling is not recommended at all: agents limited by local toxicity at the site of administration, agents placed into anatomical compartments with little onward distribution, and proteins above 100,000 daltons given intravascularly. Chae and colleagues showed the spread empirically in 2025, back-calculating a first-in-human dose three ways from the same three-species data; the predictions differed by roughly fourfold.

Intraperitoneal injection is a laboratory route

Al Shoyaib, Archie and Karamyan reviewed the intraperitoneal route in Pharmaceutical Research in 2020. Veins draining the visceral peritoneum empty into the portal vein, so material absorbed there meets the liver before systemic circulation, and the authors note that the metabolic fate of intraperitoneally administered small molecules resembles that of orally administered drugs. Scale differs as well: peritoneal fluid volume is 50 to 75 millilitres in humans against 0.02 to 0.1 millilitres in mice, while volumes up to 10 millilitres per kilogram can be given to rodents by this route. The review states that the route is minimally used in the clinic, mostly for treatment of peritoneal cancers.

Their conclusion is the operative sentence. Intraperitoneal administration in rodents is described as justifiable for pharmacological and proof-of-concept studies whose object is target engagement, and not as a basis for translating formulation properties or pharmacokinetics. Turner and colleagues, surveying routes of administration in laboratory animals, make the parallel point that route is a design decision with consequences for the result rather than a neutral delivery detail. Recording the route is therefore recording which class of question an experiment was capable of answering.

Several entries in this record turn on that. Every published in vivo figure for SLU-PP-332, including its tissue and plasma concentrations, came from intraperitoneal injection into male mice. The MOTS-c work of Lee and colleagues in 2015 used daily intraperitoneal injection in outbred male mice on a 60 percent fat diet, ten per group. BAM15 is the contrasting case: Alexopoulos and colleagues used oral gavage and dietary admixture in 2020, which is why the BAM15 record carries an oral maximum plasma concentration and a half-life while the SLU-PP-332 record carries neither.

Swallowing a peptide is a different experiment

For peptides the oral figure usually does not exist at all. Renukuntla and colleagues, reviewing oral delivery of peptides and proteins in 2013, described the gastrointestinal epithelium as a physical and biochemical barrier and reported oral bioavailability across that class as below 1 to 2 percent, without a figure for any individual peptide. Absorption across the stomach is limited by low surface area, by pepsin, and by the acidic environment; material surviving the stomach meets pancreatic proteases downstream. Those constraints are properties of the route rather than of any particular sequence.

A further enzymatic layer sits at the absorptive surface itself. Picariello and colleagues applied peptidases from porcine jejunal brush border membrane to gastro-pancreatic digests and measured a degree of hydrolysis of 70 to 77 percent. That study used milk proteins rather than a therapeutic peptide, so it establishes that the brush-border layer performs substantial additional cleavage without supplying a figure for any given compound. PubMed was searched on 17 August 2026 for brush-border hydrolysis measurements of the peptides recorded on this site and returned no records; that figure is recorded here as untraced.

Formulation work has been undertaken to cross that barrier. Buckley and colleagues reported in 2018 that oral semaglutide is coformulated with the absorption enhancer SNAC and that absorption occurs in the stomach by a transcellular route, a mechanism specific to that formulation rather than general to peptides. Twarog and colleagues, comparing SNAC and sodium caprate across more than thirty years of study, found the mechanisms still in dispute and reported single-digit, highly variable increases in oral bioavailability of macromolecule payloads in human trials. An injected figure carries no information about the same material swallowed.

What an n of six supports

Group sizes behind the rodent work on this site run from six to ten. Billon's treadmill experiment used six animals per group; Lee's diet-induced obesity arm used ten. Numbers in that range can support a reported direction of effect within the protocol, strain and sex tested. What they support far less well is magnitude, and magnitude is the part that travels. Button and colleagues, in Nature Reviews Neuroscience in 2013, reported a median statistical power of 21 percent across 49 meta-analyses covering 730 studies, and identified overestimation of effect size and low reproducibility as direct consequences.

The mechanism is arithmetic rather than misconduct. Ioannidis showed in 2008 that when a discovery is claimed by crossing a significance threshold in an underpowered study, the observed effect is expected to exceed the true one, and that flexible analysis with selective reporting inflates it further. Sena and colleagues measured the consequence in animal work: across 16 systematic reviews covering 525 publications of interventions in animal models of acute ischaemic stroke, only 2 percent reported no significant effect on infarct volume, and trim-and-fill analysis suggested publication bias accounted for around a third of reported efficacy, with the pooled estimate of reduction in infarct volume falling from 31.3 percent to 23.8 percent.

That reading is contested, and the contest belongs on the record. Nord and colleagues reanalysed the same 730 studies in 2017 using Gaussian mixture modelling and found they do not form a single distribution, that power varies substantially across subfields, and that it is lowest in candidate gene association studies. Their conclusion is that low power is a real problem and not a universal one. Both readings are kept here, because averaging them would produce a confidence neither paper supports.

None of this makes a rodent result uninformative. It makes the unlabelled version of one uninformative. Species carries the scaling problem, route the absorption problem and group size the inflation problem, and those three largely determined the size of the number to begin with. A row that carries all three can be checked against the report it came from; a row stripped of them cannot. The fields sit on every row because removing them removes the finding.