Nonclinical written summaries, explained.

From a shelf of study reports to a few hundred pages: nonclinical summaries in plain words

Five steps from a finished toxicology study to FDA's 30-day review

1. Laboratories finish their reportsA study director signs each final report, and quality assurance adds its statement. The fictional NWB-4417 has 23 reports.
2. Reports are mapped to headingsEach study is assigned to the M4S sections it feeds. Report TX-311 feeds three sections and the overview.
3. The writer drafts 2.6, then 2.4, and checks every numberWritten summaries, tables and the overview. This is where 22,400 and 24,200 are caught, or carried forward.
4. Named scientists signThe author first, then an approver for each section, and the head of nonclinical development for the overview.
5. FDA has 30 daysThe trial may start after that unless FDA places it on hold, or sooner if FDA says so.

Step 3 is the subject of this guide. The compound, reports and people in this example are fictional.

Retyped tables, blank headings, short page counts, draft reports: what surprises new writers

NOAEL, AUC, exposure margin and tabulated summary: four terms defined

Fourteen terms are defined in the one-page summary.

ICH M4S section by section: 2.4, 2.6.1 to 2.6.7 and the 38 table templates

In shortM4S sets the order and the shape, and it says it does not decide which studies you run. Facts go in 2.6, as text and in tables. Interpretation goes in 2.4. The written summaries should generally stay within 100 to 150 pages, the overview within about 30, and ICH supplies 38 table templates.

M4S states its limits plainly: it "is not intended to indicate what studies are required. It merely indicates an appropriate format for the nonclinical data that have been acquired."4 It also lets authors bend the format: "common sense and a clear focus on the needs of the regulatory authority assessor are the best guides".4

Facts in 2.6, judgment in 2.4

ICH draws the line in its first paragraph. The summaries "should be to provide a comprehensive factual synopsis of the nonclinical data". The interpretation, "the clinical relevance of the findings" and the implications for safe use "should be addressed in the Overview".4 Each summary does end in a Discussion and Conclusions section, where "information should be integrated across studies and across species".4

Species order, route order and the citation format M4S expects

A small rule with large effects sits in the same passage: "Consistent use of units throughout the Summaries will facilitate their review."4 Units are one of the places where correct numbers end up looking wrong, as section 06 shows.

M4S, 21 CFR 312.23, SEND and the 2027 wording change: status in November 2026

In shortThe outline dates from 2002 and has not changed. The rules around it have. Commercial INDs must be electronic, study datasets must arrive with the reports, and FDA has issued a rule that replaces "animal" with "nonclinical" in the IND regulation from February 4, 2027, unless it is withdrawn. That rule adds no requirement. FDA's guidance on non-animal methods is still a draft.

The ICH studies guidelines decide which reports exist; M4S decides where they go

A separate family of ICH guidelines sets what evidence a program should have. A writer rarely needs their detail, and does need to know which one a heading leans on, because M4S defines pivotal-study tables through ICH M3 and the Q&A tells you to keep the heading and explain when a study was not done.4,5

What FDA requires in a first-in-human IND, against what M4S describes

In shortM4S was written for applications to market a drug. An IND has its own rule, 21 CFR 312.23(a)(8), and a 1995 guidance that expects far fewer pages. Commercial INDs still go in the eCTD format, so the Module 2 headings are where the content lands. Most of what the rule names, the summary, the tabulations, the statement of good laboratory practice and the named evaluators, has to be there whichever layout you choose.

The five things 312.23(a)(8) and FDA's guidance expect of the toxicology summary

Draft reports, and the 120-day update they bring

Writers often assume every report must be final and audited before the IND goes in. FDA's guidance names that assumption ("Most sponsors have concluded") and then allows the alternative: "an integrated summary report of toxicologic findings based on the unaudited draft toxicologic reports of the completed animal studies may be submitted".6 The price is a second document. SEND datasets are still due when a draft report is submitted, and they are replaced if the final report changes them.12

The investigator's brochure has to tell the same story

An IND also carries an investigator's brochure, which summarizes "the pharmacological and toxicological effects of the drug in animals" for the doctors running the trial.14 FDA may place a Phase 1 trial on hold if the brochure is "misleading, erroneous, or materially incomplete".27 The same NOAELs and margins therefore appear in 2.4, 2.6, the brochure and often the protocol, usually from different authors on different dates.

Which pharmacology, pharmacokinetic and toxicology reports feed which heading

In shortToxicology findings travel furthest. One toxicology report and its separate toxicokinetic report can feed four or five documents. One preprint's two examples run to 58 and 61 reports for one IND, and the fictional case has 23. Map each report to its headings before writing a sentence, and any report left without a heading becomes visible.

Where toxicology, toxicokinetic and pharmacology reports catch a writer out

One AUC can appear in six sections, and each copy has to be found

ICH foresaw this: "One tabular format can contain results from several studies. Alternatively, it may be appropriate to cite the data resulting from one study in several tabular formats."4 In an eCTD, the report file itself is filed once and referenced from each place that needs it.5 So one AUC can be printed in 2.6.4, 2.6.5, 2.6.6, 2.6.7.3, 2.6.7.7 and 2.4, and each copy has to be found when the report changes.

Exposure margin, human equivalent dose and starting dose, worked through on one fictional report

In shortThese sections carry two separate sums. A starting dose comes from doses, using FDA's conversion factors. An exposure margin comes from measured exposure. A writer does not choose either, and does have to confirm that each figure in the text can be recomputed from what the reports print. These are document checks on a fictional compound; the scientific judgment belongs to named people.

The dose sum: from NOAEL to maximum recommended starting dose

FDA's 2005 guidance covers first doses in adult healthy volunteers. It converts each species' NOAEL to a human equivalent dose (HED) with a set divisor, 6.2 for a rat and 1.8 for a dog, and applies "a safety factor of at least 10" to reach the maximum recommended starting dose (MRSD).7 In the fictional case, the projection report PK-230 prints the rat sum, and the 60 kg body weight is the one it uses.

The exposure sum: where 37 and 40 come from

The margin uses a different kind of number. FDA's guidance is firm that measured exposure "cannot be employed for setting a safe starting dose" in a first IND.7 M4S then asks for the comparison anyway, in 2.4 and in each Discussion section, because it tells a reviewer how far the animal data reach.4

The human number in the margin: first dose or highest planned dose

M4S asks that exposure in animals "be related to exposure in humans given the maximum intended doses".4 A margin quoted only at the first dose answers a narrower question. The fictional overview adds the other half in words: dose increases stop at the exposure seen at the dog NOAEL. FDA's starting dose guidance "does not address dose escalation or maximum allowable doses",7 so a writer checks that the text says which human exposure each margin uses.

Six small checks every exposure sentence should pass

Swapped digits, the wrong sex, a NOAEL nobody stated: how nonclinical summaries go wrong

In shortMost of these errors are small and read well. The fictional case holds four: swapped digits, a male value printed for females, a margin built on a wrong input and a mistyped protein binding figure. Public data on real filings are missing, so a team's own quality records are the best evidence it can get. No single check finds them all, as the grid below shows.

What a weak nonclinical section can cost in FDA's 30-day IND review

In shortMost initial INDs proceed. In the one fiscal year that FDA staff studied in full, 8.9% went on clinical hold, and manufacturing was the most cited reason, then clinical, then toxicology. Those data say nothing about summary errors. They do show the window: a problem found in those 30 days stops a planned trial.

We read the abstracts of the two papers in this section, not the full text. FDA staff who analyzed that year reported that "more than half" of the held applications became active within a year, and that holds did not differ much between rare and common diseases or between commercial and research sponsors.28

A later FDA analysis of new commercial INDs at its oncology office put holds and withdrawals in the 30-day window at under 10%, again with nonclinical reasons behind clinical and quality ones, and found held INDs were mostly for first-in-human trials or from sponsors with limited regulatory experience.29 Neither abstract gives a share for toxicology, and both papers are years old. We found no newer full-year study of initial INDs at FDA's drug center.

Three grounds for a Phase 1 hold that a nonclinical writer can touch, and one reassurance

The first three grounds are quoted from the regulation.27 FDA can also clear a trial sooner than 30 days.2

GLP inspections reach the study report, and the summary built on it answers to the sponsor

In shortGood laboratory practice (GLP) rules govern how a study is run and reported. Quality assurance reviews the final report against the raw data and signs a statement. Those rules stop at the report. Accuracy of the copies in Module 2 is the sponsor's own job, with no outside audit step built in.

FDA lists its most common GLP observations, among them missing procedures, deviations without authorization and documentation, and a study director who "failed to assure that all experimental data" were accurately recorded and verified, in FDA's summarized wording.30 Those concern the laboratory. A sponsor writing summaries from an outside laboratory's reports inherits the report as signed, and a study's GLP status is itself a fact the summary has to state.4

FDA proposed revising the GLP rule in 2016. We found no final rule, and FDA's notice of April 30, 2026 still collects information under the current requirements.18

EU trial dossiers want critical analysis where Module 2.6 wants a factual synopsis

In shortA sponsor filing a first trial in Europe writes an investigational medicinal product dossier and a brochure, under different rules from an FDA IND. The EU text asks for summaries and a critical analysis in one place, says to discuss margins by exposure and asks for a statement of GLP status. It also suggests the structure of Module 4 for the data. The M4S outline comes back in force at the marketing application.

The European Medicines Agency's (EMA) guideline on first-in-human trials, in effect since February 1, 2018, adds its own expectations for how a sponsor reasons about early dose risk.32 We read the EU and FDA texts for this guide. Teams filing elsewhere should check each agency's own format, because re-using a US summary unchanged can leave gaps in the critical analysis the EU text asks for.

FDA's commercial IND counts, 2014 to 2025: the volume behind these sections

In shortFDA's drug center received 1,210 commercial INDs in 2025, up from 782 in 2014. Each needs this content in some form. The one public time estimate we found is about 100 hours for three written summaries, and that estimate has no measurement behind it. No credible public figure exists for the market or for typical length.

AI drafting of Module 2.6: the one public test, and how far the FDA and EMA texts reach

In shortThe one public test we found, a preprint, cut first-draft time to under four hours. The same paper found toxicology sections weak on completeness and emphasis, invented NOAELs for in vitro studies and dose formulation analysis left out of every draft. Its authors are tied to the tool's maker and the drugmaker. FDA's draft guidance does not address drafting that leaves safety, quality and study reliability untouched, and the sponsor still answers for every number.

The preprint's test: fast first drafts that still need a writer

Eser and colleagues posted this paper on arXiv on September 10, 2025. We found no peer-reviewed version. They ran one AI tool on the study reports of two historical INDs from one drugmaker, timed the drafts, compared them with writers' estimates, and had one experienced writer score them.1 The drafts were never filed.

The paper's error list says more than its scores. It lists numeric slips ("11 h vs 7 h onset"), values assigned to in vitro studies and a "100% omission rate for GLP-required dose formulation analysis", with text three to five times longer than expert-written documents.1

It defined a critical error to include omitting that same analysis, and it also reports that no critical errors were detected, so read that line with care.

Its authors state the limits: two INDs, one company, a single assessor, first drafts only. All authors work for or hold shares in the tool's maker or the drugmaker, one is a consultant, and both companies funded it.1 We found no independent test of AI on Module 2.6.

What regulators say, and which rules do not reach this work

Margin sums and table cells suit software; integrated conclusions stay with people

The writer, the toxicologist and the head of nonclinical: who holds 2.4 and 2.6 today, and where first-draft hours go

In shortOne writer usually holds the pen, while each scientific section has its own approver. The writer chases reports, retypes tables and checks the numbers, and the head of the function signs the overview. In one fictional estimate, the toxicology written summary alone takes 52 of 130 first-draft hours.

Writer to head of nonclinical: who touches the summaries

Where a dog AUC travels: seven steps from signed report to eCTD build

Where about 130 hours go on a first draft of the summaries

Who holds the pen: your own writers, a contract writer or a software tool

The swapped digits on the NWB-4417 screens: 23 study reports, one binder, a signed 2.6.6

In shortThis is the example from the top of the page, inside the product. Lena Ortiz, the regulatory writer, opens her binder. The table builder fills the pivotal-study table from five reports and finds that it disagrees with her text. She opens the sentence beside the source page, takes the source value, the number checker recomputes the margin, and she signs as author.

1

The M4S outline as a binder, with each heading's reports

Sign-off is due October 30, and the IND goes to FDA on November 20. Lena's home screen is the Module 2 nonclinical binder: a tab for 2.4 and for each of 2.6.1 to 2.6.7, with each section's approver, due date and sign-off track.

27 headings need her, 4 numbers are to resolve and 2 coverage gaps are open. She selects 2.6.6.3, repeat-dose toxicity. It draws on five reports and cites 46 of the 48 findings extracted for it by the report reader, the agent that reads each report, and two findings wait for her decision.

2

2.6.7.7 builds itself, and it disagrees with the text

Lena opens 2.6.7.7, the pivotal-study table. The table builder fills it from TX-311, TX-312, BA-511, BA-512 and HP-311: 18 rows and 55 cells, each cell linked to its page, and 55 of 55 match their source.

Then the number checker compares the finished table with her written summary. A banner says so: 2.6.6.3 gives the dog AUC at the NOAEL as 22,400 ng·h/mL, and this table took 24,200 from BA-512, page 11.

3

The typed 22,400 beside BA-512's printed 24,200

One click opens 2.6.6 at the sentence. A line runs from the red 22,400 to the highlighted 24,200 in BA-512, Table 3, page 11. The panel gives the reason: digits transposed.

At the top, 425 of 428 numbers match. The other flagged value in view reads 1.8-fold for ALT in females, the male value; TX-311 Table 11 gives 1.5.

4

The source value goes in, and the margin follows

Lena chooses the source value. The margin in 2.6.6.9 was built on the typed figure: 22,400 ÷ 610 gave 37-fold. The number checker recomputes it from BA-512 and the projected human AUC of 610 ng·h/mL in PK-230: 24,200 ÷ 610, so 40-fold.

The number checker cannot edit the draft on its own. It proposes the source value, and the section stays unsigned while any number disagrees. Each change goes into the section's history.

5

Lena signs 2.6.6 as author, and it goes to Dr. Haddad

With every number matching, Lena signs 2.6.6. The dialog states the meaning of her signature: "I drafted this section with the agents and checked every number against its source." The AI use is recorded with the section: drafted by the summary writer, 428 numbers checked, 3 corrected from the sources.

The section goes to Dr. Omar Haddad, the study toxicologist, for approval. Module 2 is not finished: one number still disagrees, in 2.6.4, and the overview is still to come.

Coverage gaps, the full number check, the 2.4 overview and your rules: thirteen more screens

In shortOne section is the story. These screens show the rest: what the report reader extracted and nobody cited, the table as it builds, any cell back to its page, the full number check, the section history, the overview and its recomputed margins, all 23 reports, the dashboard and the settings.

The writer and four named approvers sign what six agents read, draft and check

In shortOne agent reads the reports, one plans the sections, two draft, and two check. None of them may sign, make an adverse or non-adverse call the report does not make, or choose between conflicting sources. The writer resolves every flag and signs as author, and each section has its own approver.

What each of the six agents does on 2.4 and 2.6, and what it may never do

The writer signs first; toxicology, DMPK and pharmacology approve their own sections

Agent steps and human actions are kept apart in each section's history, and the AI use stays with the section for your own policy file. These are controls that support your compliance with 21 CFR Part 11; your own validation and procedures complete it. Botminds is SOC 2 Type II audited, with information security managed to ISO 27001.

Lena Ortiz and the four approvers on the NWB-4417 screens

What 23 study reports become: cited sections, page-linked tables and a number-check report

In shortThe solution reads what an IND already runs on: pharmacology, pharmacokinetic and toxicology reports, the dose projection and your template. It returns sections ready for your eCTD build, with each table cell linked to its page. Its outcome figures here are design targets, to be measured in the live solution on your own reports.

eCTD publishing, document management and your Word template keep their jobs

Your writers stay too. In the one public test, the lowest stated scores were on emphasis,1 and the agents here leave emphasis to the author.

Three controls that hold a number in Module 2 to its report page

A wrong exposure figure or a late IND summary: what is published on the cost, and what only you can measure

In shortPublished figures are few, and none of them prices an error in these sections. The largest cost in your own program is likely the one nobody records: senior scientists re-reading numbers in the last weeks. Measure your own hours and catches. The table below shows where to look.

Hours, catches and amendments: pull your own figures from the last IND

The first-dose rationale, the brochure and your toxicologist's week: costs no spreadsheet counts

The return on nonclinical summary software: writing hours, checking hours and INDs a year

In shortPut in your own numbers. The calculator values writing hours and checking hours released on each IND, then takes off what the tool and your testing cost. The first-draft hours alone rarely pay for a tool at low volume. Checking hours and any real change in date are what change the answer, and the date stays on its own line.

Reading, drafting, tables, checking, amendments: where IND summary work saves

INDs a year, hours by hand and how many numbers you check

Score your IND writing team: INDs ahead, report versions, named approvers, house template and AI on the record

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

Scores of 0 to 6 start with counting numbers; 12 to 16 plan the next live IND

NOAELs, table cells and amended reports: the questions that test a nonclinical writing tool

In shortAsk every vendor the same questions, us included. The answers show how a value is captured, how a table is built, whether a stated NOAEL stays the study director's, what happens when a report is amended, and how you can measure it on your own reports.

We found no independent test of any nonclinical writing tool's accuracy, as of November 2026. A planted-fault test on your own closed IND is the fairest comparison, and it measures your current quality review too.

An eight-week nonclinical writing pilot: plant faults in a closed IND, grade against the page, then decide

In shortMeasure first, then test on a closed IND whose answers are settled, with faults you planted and one report amendment. Then run alongside one live section. None of these steps commits you to a vendor.

1. Trace ten values through a filed IND: two NOAELs, two exposures, a margin, a first dose sum and four findings. That is your baseline for errors that got through.
2. Choose a closed IND and two sections: 2.6.6 with 2.6.7, for numbers, and 2.4, for argument. Keep the reports in the environment your laboratory contracts allow.
3. Plant 40 faults from your own quality history: swapped digits, wrong sex, wrong day, a stale margin, a NOAEL nobody stated. Lock the key, and keep planted text away from any live file.
4. Run and grade: two graders who did not choose the tool, against the page, with a third to settle disputes. Sample 60 unflagged numbers too, to see what was missed.
5. Decide with four numbers: numbers matched to a page, catch rate by fault type, false flags on clean text, and approver minutes to a section they will sign.

Numbers matched, faults caught and approver minutes on a closed IND

The four decision numbers are the ones in step 5. Amendment catch and words kept are supporting measures.

What writers and sponsors ask about M4S, page counts, draft reports, margins and AI

In shortThe questions writers and buyers search for most, each answered on its own.