A friend of mine works in a preclinical lab. His company runs the kind of studies that are supposed to tell the FDA whether a drug is safe enough to test in humans. Over the past couple of years each round has been taking longer because the number of parameters keeps going up, which is a polite way of saying regulatory requirements have been accumulating faster than anyone's ability to keep track of them.
They hired a consultant. The consultant was supposed to make sure the IND package (the application you file with FDA before you can dose your first patient) included everything it needed. Some of the required studies still didn't get done. They had to go back, repeat work, re-validate. In an industry where reproducibility is its own KPI, "we had to redo the study" is not a scheduling inconvenience. It is an existential problem.
I'm a software person. I got into this the way I get into most things: a friend with a regulatory background pointed me at the problem, and I started reading. ICH M3 first (the master guideline for what preclinical work you need before human trials), then S7A and S7B for safety pharmacology, then S2 for genotoxicity, then the actual Code of Federal Regulations, Title 21 Part 312, which is the legal text governing IND applications and which reads like it was written to be deliberately incomprehensible.
They're all like that, by the way. Long, dense, cross-referencing each other. I doubt anyone reads any of them start to finish voluntarily. The people who know these documents best are the regulatory lawyers who bill by the hour to interpret them for you.
At some point I found a 90-page NIH manual from 2019 about IND submissions that I'm fairly sure has been downloaded fewer times than it has pages.
After all that reading I went looking for a checklist. A simple list of what you need, with links to the rules that require it. It does not exist. FDA guidance runs hundreds of pages. Consulting firm blogs give you four paragraphs of advice so vague it could apply to any regulated industry and then a contact form. So I made one.
What FDA reviewers actually care about
When a reviewer picks up your IND, the questions are not complicated. Can this drug be safely given to a human, based on the animal data you've submitted. Can you manufacture it the same way twice. Is your clinical trial designed to produce useful data without injuring the participants.
Preclinical, CMC, protocol.
If the package doesn't clearly answer all three you get a clinical hold. FDA tells you to stop, fix the gaps, and resubmit. For a company spending half a million dollars a month on runway is not an abstraction. It is the scenario that comes up in every conversation I have with regulatory affairs people, and I've had about fifteen of those conversations since January.
The IT architecture analogy. The rules don't change that often. The checks aren't even that complicated if you know where to look. This is an IT architecture problem, fundamentally. If you think about your system architecture at the end of development, after you've already built everything, it's going to be bad. Same dynamic here. Plan your study architecture before you start running studies, or discover the gaps when it's expensive to fix them.
The problem isn't that the requirements are secret or impossible. The problem is that there are a lot of documents, each one is long, and there are nuances you only learn about by going through the submission process. At which point you've already spent the money.
Paperwork
Form 1571 is the main IND application. Filling it out for the first time is roughly the experience of doing your taxes if your taxes were longer and the penalty for a mistake was the FDA telling you to go away.
You also need the 1572, the 3674, a cover letter, a table of contents, an introductory statement, and a general investigational plan. None of this is intellectually difficult. All of it is tedious in the way that only government forms can be, and nobody tells you about any of it until you're trying to file and googling "FDA Form 3674 fillable PDF" at midnight. The full Module 1 breakdown is at regfo.com/library/m1.
Preclinical
Most clinical holds originate here.
Toxicology
You need single-dose and repeat-dose toxicity studies, both conducted in two species: one rodent, one non-rodent. This is the gap I hear about most often, and it's not close. Teams run their rat studies, get good results, start assembling the IND, and somewhere around month three somebody actually reads ICH M3 Section 7 and discovers they also need dog or primate data. Beagle studies do not schedule themselves. CRO slots for large-animal tox are booked months out.
This is essentially what happened at my friend's lab. The consultant was supposed to catch it. Consultants do help, I should say. I'm not dismissing the entire profession. But as in any industry, a consultant is partly an insurance policy, and how much that policy is worth depends entirely on how much you trust the specific person. In this case, the studies that should have been flagged weren't, and the work had to be repeated. In preclinical research, repeating a study means re-establishing reproducibility, re-validating batches, re-spending money you already spent. It compounds.
The batch comparability trap
There's a related problem that isn't obvious until you see it happen. Your toxicology batch and your clinical batch need to be comparable. The material you tested in animals and the material you plan to give humans. That's technically a manufacturing question, but it surfaces as a preclinical problem when your non-rodent study is delayed and you've changed your manufacturing process in the interim. Now you're trying to argue that different batches are equivalent while simultaneously waiting for toxicology data you don't have. I've seen this more than once. The conversation always involves someone saying "the process change was minor" and FDA asking for data to support that claim, which they don't have.
Study duration
Your tox studies have to run at least as long as your planned clinical dosing. ICH M3 is explicit about this. If your Phase 1 protocol specifies 28 days of dosing, you need 28-day tox data. Submitting 14-day data with a promise to do the longer study later is a clinical hold. It is arithmetic and people still get it wrong.
Safety pharmacology and hERG
Safety pharmacology covers cardiovascular function (including a cardiac ion channel assay called hERG), central nervous system effects, and respiratory function. All studies must be conducted under Good Laboratory Practice.
The hERG trap. A company runs a non-GLP hERG assay early in development, for internal decision-making. Reasonable. They get a clean result. Then they file the IND with that same non-GLP data because "the result is the same either way, GLP is just more paperwork." GLP is not just more paperwork. GLP is the thing that makes FDA accept your data. Non-GLP hERG = clinical hold, every time. The assay costs ~$30-40K. Companies delay their entire program by months to avoid that. I have stopped trying to understand the logic.
Tox requirements: CTD M4.2.3.1. Safety pharmacology: CTD M4.2.1.
Genotoxicity and PK
Genotoxicity is an Ames test plus one additional in vitro assay. A positive result in either triggers an in vivo follow-up study. Pharmacokinetics is single-dose in two species with plasma protein binding. I have never heard of an IND failing on either of these.
Reproductive toxicology
If your Phase 1 trial enrolls only men, ICH M3 lets you skip reproductive toxicity studies entirely. The moment you include women of childbearing potential, you need embryo-fetal development studies in two species. The trap is the mid-study decision. Enrollment is slow, someone suggests broadening the eligible population, and now you need studies that weren't budgeted, weren't scheduled, and will add months to the timeline. I know of a rare disease program where exactly this happened. Added five months and roughly $400,000 in unplanned costs. The investigator wasn't angry at FDA. He was angry at his own team for not asking "what if we expand enrollment?" during planning and working out what that would cost. Specifics at CTD M4.2.3.5.
CMC
Chemistry, manufacturing, and controls. Not my area. I've been deep in tox and pharmacology, not drug formulation, and I'd rather say that than get the details wrong. FDA requires drug substance specifications, a manufacturing process description, stability data, a drug product formulation, and certificates of analysis. The batch comparability issue I wrote about above usually shows up first as a tox problem, which is why I covered it there. Requirements are at regfo.com/library/m3.
The math
Your protocol includes the expected elements: objectives, endpoints, patient selection criteria, safety monitoring, stopping rules, IRB approval, informed consent. None of that is where the tension is.
The tension is in the dosing rationale. Reviewers want to see a no-observed-adverse-effect level from your animal studies, the drug exposure at that level, your proposed human starting dose, and the safety margin between them. How wide that margin needs to be is — remarkably — not specified in the guidance, which creates exactly the confusion you'd predict. The informal consensus among regulatory affairs professionals, based on every conversation I've had, is that you want at least a 10-fold margin for a standard small molecule.
One team told me they'd submitted with a 3-fold margin on a non-cancer drug. "Our compound is well-tolerated," they said, as if that meant the math didn't apply. FDA sent questions.
Cancer drugs under ICH S9 have more flexibility on starting dose and toxicology requirements, but I tried to summarize oncology IND rules once and got corrected on three points, so I'm leaving that for a separate post.
Where things break
The two most common gaps, by a wide margin, are the missing non-rodent toxicology study and the non-GLP hERG assay. Both covered above.
Recovery groups are a judgment call. That's where you include a group of animals that stops receiving the drug partway through the study, so you can demonstrate that toxic effects are reversible. A toxicologist I've worked with recommends them by default on any repeat-dose study longer than 14 days. She argues that adding recovery groups costs about $50,000, while a six-week delay from FDA asking reversibility questions costs multiples of that. I think she's right when you have organ findings. On a clean study, I've heard of people filing without recovery groups and getting no questions, so the honest answer is that it depends.
I don't love that answer. But it's the accurate one.
What this doesn't cover
Cancer drugs, biological products, gene therapies, cell therapies — each operates under a different regulatory framework and compressing any of them into a paragraph would do more harm than good. Pediatric development plans are not required at the initial IND stage. The CTD Library at regfo.com/library has 218 sections covering all of these for anyone who wants to go deeper.
What to do with this
This is a list of what goes in the box. Whether your studies actually satisfy these requirements is a harder question, and in my experience it's the one that teams confront about three weeks before their planned filing date, which is not a great time to discover that you're missing a study.
If a friend asked me what Regfo does — no marketing, just over a beer — I'd say it tells you upfront which studies you need so you only run the ones that actually move you toward approval. Instead of hiring someone expensive to read guidelines on your behalf, you upload your reports, check them against 24 guidelines and 1,054 rules, and know where you stand in about three minutes. The other approach involves two weeks and a spreadsheet that becomes less useful the bigger it gets.