The first time I helped a biotech put a monoclonal antibody into the clinic, the CSO walked into my office holding a printed copy of ICH M3(R2) and asked which sections we could skip. Fair question. The answer is "most of them," and the second question, the one that actually matters, is what replaces them.
If you're running a small molecule program, ICH M3(R2) is your bible. Two species, rodent carc, genotox battery, the whole machinery. Then you switch to a biologic and someone hands you ICH S6(R1) and you realize the rules don't translate. They were never supposed to.
Here's what actually applies, what doesn't, and where FDA has been quietly moving the goalposts.
Why ICH M3 doesn't fit a mAb
Small molecule tox is designed around metabolism and off-target chemistry. You give a rat a 1000 mg/kg dose of a kinase inhibitor and you find out which organs the parent compound and its metabolites torch. You run Ames, micronucleus, and chromosome aberration because small molecules can bind DNA. You run two-year carc studies because they're around long enough for tumors to grow.
A monoclonal antibody is a 150 kDa protein. It doesn't get metabolized into reactive species. It doesn't fit in a DNA groove. It's cleared by FcRn recycling and target-mediated drug disposition, not CYP3A4. Most of the standard small molecule battery tells you nothing useful.
ICH S6(R1) — the harmonized guideline for biotech-derived pharmaceuticals, finalized in 2011 as an addendum to the original 1997 S6 — exists because regulators figured this out about 25 years ago. The core idea is simple. The toxicity of a biologic is mostly exaggerated pharmacology, not chemistry. So you don't run a generic tox battery. You run studies in a species where the drug actually works, at doses that meaningfully engage the target, for long enough to see what happens when you turn the pathway up or down.
That's the whole frame. Everything else flows from it.
The two principles that drive every decision
Read S6(R1) carefully and two phrases come up over and over.
Pharmacologically relevant species. A species where your antibody binds the target, with comparable affinity, and where the downstream biology is similar enough to humans to mean something. If the target epitope isn't conserved, the species is useless for safety prediction, no matter how many animals you dose.
Homologous proteins (surrogates). When no relevant species exists, S6(R1) opens the door to a surrogate molecule, a rodent-targeted antibody against the rodent ortholog of your target. It's not the clinical candidate. It's a proxy that tells you what happens when you hit that pathway in vivo.
Both principles get abused. Teams pick NHP because "everyone uses cyno" without checking cross-reactivity. Other teams refuse surrogates because "FDA won't accept it." That's wrong, FDA accepts them when justified. I've seen both mistakes cost a year.
Species selection: NHP is the default, not the rule
The cynomolgus macaque is the default for one reason: most human mAbs cross-react with cyno targets because the proteins are usually >95% sequence identical. That's the empirical truth, not a regulatory preference. S6(R1) doesn't say "use NHP." It says use a relevant species.
Here's the decision tree I actually walk teams through:
- Test cross-reactivity in vitro first. SPR or ELISA against the target ortholog from rat, dog, mini-pig, cyno, marmoset. Not just binding. Affinity comparable to human (within ~10-fold is the rough bar). Then a functional assay: does the antibody trigger the same downstream effect in the species' cells?
- If one rodent species qualifies, use it. S6(R1) explicitly allows a single relevant species when justified, and rodents are cheaper, faster, and ethically lighter. Don't run NHP for a cytokine antibody just because it's traditional.
- If no rodent qualifies but NHP does, that's your species. Often a single cyno study is enough; S6(R1) section 3.3 says "the use of one species may be sufficient" when the second species adds no information.
- If nothing qualifies, build a surrogate or use a transgenic model. This is where programs lose six months. Plan it before you finish lead optimization.
Transgenic mice expressing the human target are the right tool for cases where the antibody only binds human, the biology is well understood, and a surrogate would be a different molecule entirely. CD20, PD-1, and TNF-targeting programs have all used this approach. The catch: the model has to be validated, the human transgene has to be expressed where it's expressed in humans, and FDA will ask about copy number and tissue distribution.
For more on what nonclinical packages should contain, our nonclinical overview section walks through the CTD Module 2.4 structure FDA expects.
Safety pharmacology: usually integrated, sometimes standalone
ICH S7A and S7B were written with small molecules in mind. S6(R1) explicitly says safety pharmacology endpoints should be incorporated into general toxicology studies for biologics, not run as standalone studies.
In practice that means:
- Cardiovascular telemetry (blood pressure, heart rate, ECG) gets recorded during the repeat-dose tox study in NHP, not in a separate dog cardio study.
- Respiratory measurements (rate, tidal volume) added to the same animals where feasible.
- CNS observations from the FOB or modified Irwin screen, layered onto the daily clinical observations.
The hERG assay? Skip it. Antibodies don't engage cardiac ion channels because they don't get into the cytoplasm. The S7B paradigm isn't relevant. I still get asked this every few months.
The exception is when your target has known cardiac, CNS, or respiratory expression. Then you owe FDA a tighter look. An antibody against a GPCR expressed on cardiomyocytes? You're going to want dedicated telemetry, possibly an isolated heart assay using species-relevant tissue, and a clear exposure-response. Document why you did or didn't run a standalone study. Both ICH S6(R1) section 5 and FDA reviewers want the reasoning, not just the result.
For deeper context on what safety pharmacology endpoints belong where, see our library section on safety pharmacology.
Tox study design: doses, frequency, duration
This is where small molecule habits do the most damage. Teams default to "high, mid, low + control, daily dosing, 13 weeks" because that's what they did in their last CRO contract. Wrong frame.
Dose levels. For a mAb you're not chasing an MTD. You're characterizing the exposure-response across the pharmacologic range. Pick doses that bracket saturation of the target. Typical layout: a low dose near the projected human therapeutic, a mid dose 3-10x above that, a high dose at or above expected target saturation. The high dose justification under S6(R1) is exposure margin and pharmacology, not toxicity at all costs.
Frequency. Most therapeutic mAbs have half-lives of 1-3 weeks in NHP. Daily dosing is biologically meaningless and creates unnecessary immunogenicity. Weekly or every-other-week dosing matches the clinical regimen and maintains target coverage. S6(R1) section 4.2 explicitly says dosing frequency should reflect intended clinical use and pharmacokinetics.
Duration. For a chronic indication, a 26-week (6-month) repeat-dose study in the relevant species is the typical pivotal package. For acute or short-course indications, 4 to 13 weeks may be enough. ICH S6(R1) doesn't require a 9-month study. That's a small molecule reflex from M3.
If you're heading into IND with a single NHP species at 4-week duration to support a Phase 1 single-ascending-dose study in oncology patients, that's a defensible package. For a chronic autoimmune indication with multi-year dosing, you'll need the 26-week study before you can run beyond a few months of clinical exposure.
For the structural conventions FDA expects in your tox reports, our repeat-dose toxicity and single-dose toxicity library pages cover the data fields reviewers actually look for.
Immunogenicity: monitor it or explain why you didn't
Here's the part I see programs underestimate. Anti-drug antibodies (ADAs) in animal tox studies don't predict human immunogenicity. Humans don't recognize a humanized mAb the way a cyno does. FDA knows this. So why does S6(R1) still require ADA monitoring?
Because ADAs in your tox study can confound the exposure data. If 6 of 8 animals in the high-dose group develop neutralizing ADAs by week 8, your week 12 exposures are not what you think they are. Your NOAEL based on dose might be wrong. Your safety margins are wrong.
Practical approach:
- Validated ADA assay before study start, not after.
- Sampling timepoints aligned with PK draws so you can correlate.
- Pre-specified rules for how you handle ADA-positive animals in the analysis (include, exclude, sensitivity analysis: pick before you have the data).
- Report ADA incidence, titer, and impact on exposure in the study report.
If an animal becomes ADA-positive and exposure crashes, that animal stops contributing safety information at that point. FDA reviewers will read the exposure curves and the ADA tables together. If the story doesn't hold up, you'll get a question.
What FDA's 2023-2024 draft on mAb nonclinical changes
In late 2023 FDA published a draft guidance addressing nonclinical considerations for therapeutic protein products, with mAb-specific commentary that built on S6(R1). The themes from that draft and the agency's public statements at BIO and DIA meetings:
- One species is increasingly acceptable when justified, particularly for oncology indications and for targets where cyno is the only relevant species. Don't reflexively design two-species programs for biologics.
- Tissue cross-reactivity (TCR) studies are deprioritized. FDA has been signaling for years that the IHC-based TCR study has poor predictive value. The draft moves toward "consider whether TCR adds information beyond what target biology already tells you." For many programs, that's no.
- Justification of high dose should be exposure-based, not toxicity-based. If you've achieved a multiple of human exposure that covers target saturation, you don't need to push higher to find an MTD that doesn't exist for that mechanism.
- Flexibility on study duration for accelerated programs. Earlier and shorter pivotal studies, when paired with strong PK/PD modeling and a clear FIH starting dose justification.
Read the draft itself before it changes. Drafts move. The principles, though, are aligned with where S6(R1) was already pointing. Design the program around the biology, not the checklist.
What this means for an IND-stage mAb program
If you're putting together a nonclinical package for an IND, the rough shape:
- Cross-reactivity package across candidate species with justification of your chosen species
- A pivotal repeat-dose tox study in the relevant species at clinically relevant dosing frequency, with integrated safety pharmacology endpoints
- TK sampling and ADA monitoring throughout
- Local tolerance at the planned route of administration
- Tissue cross-reactivity only if you can articulate what question it answers
- A clean Module 2.4 nonclinical overview that explains your design choices in terms of S6(R1) principles, not boilerplate
No genotox. No carc unless you have specific cause for concern. No two-year rodent. No daily dosing. No hERG.
That's the part that feels wrong to people coming from small molecules. It is supposed to.
How Regfo handles biologics nonclinical
When we built the rules engine, we kept the small-molecule and biologic rule sets separate on purpose. The S6(R1) rules don't fire on a small molecule program and the M3 rules don't fire on an mAb program. Today we have 373 active preclinical requirements; a chunk of those are S6(R1)-specific (species justification, dosing frequency rationale, ADA monitoring, integrated safety pharm). When you upload a tox report, the engine reads it and checks which of those requirements are addressed and which aren't.
We don't tell you what your NOAEL should be. We tell you whether your study report contains what FDA wants to see in it. Different problem. Cheaper to fix.
If you're putting together an IND for a mAb in the next quarter and want a second pair of eyes on the nonclinical package before you submit, that's the workflow we built for.
FAQ
Do I need a rodent species for an mAb tox program? No, not by default. ICH S6(R1) requires a pharmacologically relevant species. If your antibody doesn't bind the rodent target with comparable affinity, the rodent study tells you nothing useful. Often NHP alone is sufficient. Sometimes a single rodent species is enough. The question is relevance, not species count.
Is the cynomolgus macaque always required? No. Cyno is the default because most humanized mAbs happen to cross-react with cyno targets. If your antibody works in a rodent, use the rodent. NHP is not a regulatory requirement, it's an empirical convenience.
What about genotoxicity and carcinogenicity studies for mAbs? ICH S6(R1) explicitly says standard genotoxicity studies are not appropriate for biologics. Carcinogenicity studies are not routinely required either. Instead, you do a weight-of-evidence assessment based on the biology of the target and the duration of clinical exposure. If your target is involved in growth regulation or immune surveillance, FDA will want a specific argument.
Can I use a surrogate antibody if no species cross-reacts with my clinical candidate? Yes. ICH S6(R1) section 3.3 supports the use of a homologous protein in a relevant species when no species cross-reacts with the clinical candidate. The surrogate is not the clinical molecule, so the data are supportive rather than pivotal in the strict sense, but FDA accepts this approach when properly justified.
How long does my pivotal repeat-dose tox study need to be for an mAb? For chronic indications, 26 weeks in the relevant species is the typical pivotal package. For acute indications, 4 to 13 weeks may be sufficient. ICH S6(R1) does not require the 9-month study that ICH M3(R2) requires for small molecules. Match the duration to the intended clinical use.
Does FDA still want tissue cross-reactivity (TCR) studies? Less than they used to. FDA has been signaling for several years that IHC-based TCR studies have limited predictive value, and the 2023-2024 draft on mAb nonclinical reinforces a "consider whether it adds information" framing. Many recent INDs have proceeded without a TCR study when the target biology was well characterized.