Preclinical vs. Clinical Research: Key Stages of Drug Development

The development of a new medical therapy typically progresses through a pipeline of stages from conceptual discovery to clinical use. Broadly, this pipeline is divided into preclinical (sometimes called “non-clinical”) and clinical phases. Preclinical research includes all work carried out before first-in-human trials. Its primary goal is to evaluate the safety and potential activity of a drug candidate using laboratory and animal models. In contrast, clinical research involves systematic testing in human subjects to confirm safety, determine dosing, and ultimately demonstrate efficacy for regulatory approval. Together, preclinical and clinical testing are “crucial stages in the development of new drugs” that generate the evidence needed for regulators to grant market authorization (^[12]) (^[10]).

Historically, the importance of rigorous preclinical testing was driven home by tragic failures when it was lacking. The infamous 1937 Elixir Sulfanilamide disaster – where over 100 patients died after taking an untested drug formulation (^[6]) – prompted the 1938 U.S. Food, Drug & Cosmetic (FD&C) Act, formally requiring safety data before marketing. Likewise, the 1960s thalidomide tragedy (causing over 10,000 birth defects globally) occurred because the drug had not been tested for teratogenic effects in pregnant animals (^[7]). This led to the 1962 Kefauver-Harris amendments in the U.S., vastly strengthening drug regulations to require proof of both safety and efficacy in well-controlled trials, including specific preclinical studies (such as reproductive toxicity studies) to prevent similar events. These regulatory milestones underscore the role of product-appropriate nonclinical evidence in protecting human subjects and informing clinical trial design.

Over time, standards for nonclinical work have been codified. The studies needed to support an IND depend on the nature and duration of the proposed clinical investigation; GLP is generally expected for pivotal in vitro and in vivo studies submitted in support of an IND. Good Laboratory Practice (GLP) regulations (e.g. 21 CFR Part 58 in the U.S.) help ensure that applicable nonclinical laboratory safety studies are rigorous and reliable (^[13]) (^[14]). GLP not only dictates standard operating procedures and record-keeping, but also mandates the use of validated methods and well-trained personnel. This is a strict contrast to the flexible, exploratory environment often used in basic discovery science. The rationale is simple: data used to support an IND (Investigational New Drug) must be reproducible and auditable. Examples of GLP studies include formal acute and chronic toxicity tests in two animal species, genotoxicity assays, safety pharmacology (e.g. cardiovascular or CNS safety screens), and ADME profiling (^[15]) (^[14]).

In parallel, ethical oversight differs drastically. Animal studies are reviewed by Institutional Animal Care and Use Committees (IACUCs), which enforce the three Rs – Replacement, Reduction, Refinement – to minimize animal use and suffering (^[16]). By contrast, human studies require IRB/ethics committee approval and informed consent, under Good Clinical Practice (GCP) rules. Thus the terms preclinical and clinical also connote fundamentally different ethical frameworks.

Given this context, the central question “What happens before humans are involved?” can be answered by detailing the arsenal of methods and regulatory steps in preclinical development. In brief, preclinical research encompasses target validation, compound optimization, and extensive safety testing. A candidate may advance to first-in-human trials when the submitted nonclinical information supports the sponsor’s conclusion that the proposed investigation is reasonably safe, subject to FDA review and applicable human-subject protections. Nevertheless, as modern data reveal, the average attrition from this stage onward is high. Industry studies estimate that only a few percent of initial discovery leads ever succeed as approved drugs (^[10]) (^[11]). The current state of drug development, still lengthy and costly, drives a continual search for better predictive models and smarter trial designs (e.g. microdosing or seamless phase transitions).

This report explores multiple perspectives on the preclinical versus clinical divide. We first elaborate on the scientific and regulatory activities in preclinical development (including discovery and toxicology), then contrast them with the objectives and practices in clinical trials. We review statistical data on costs, timelines, and success rates, and analyze how these stages contribute to overall R&D productivity. Real-world examples (case studies) illustrate both successes and failures in the handoff. Finally, we discuss emerging technologies and strategies aimed at improving translation from bench to bedside, along with their ethical and practical implications.

Discovery and Lead Optimization. Drug development typically begins with identifying a molecular target (e.g. an enzyme, receptor, or biomarker linked to a disease). High-throughput in vitro screens (using biochemical or cell-based assays) are used to find “hits” – compounds that modulate the target. Medicinal chemistry then optimizes these hits for better potency, selectivity, and drug-like properties. At each step, in vitro ADME screening (e.g. metabolic stability in liver microsomes, inhibition of cytochrome P450 enzymes) informs which molecules are likely to have favorable properties in vivo. Computational (in silico) methods and structure-based design increasingly assist in predicting binding and toxicity. This discovery phase can be iterative and data-driven, but it does not yet involve live subjects. Once a promising “lead” compound is isolated, it undergoes further refinement (so-called “lead optimization”) to balance efficacy with manufacturability and preliminary safety flags.

In Vitro and Ex Vivo Testing. Extensive laboratory testing (all falling under preclinical) aims to predict how the drug will behave in an organism. These include receptor binding assays, enzyme inhibition assays, and in vitro toxicity screens (such as cytotoxicity on cultured cell lines, Ames mutagenicity test, or assays for off-target effects). Advanced models like organ-on-a-chip or 3D cultured microtissues are also used to predict organ-specific toxicity (e.g. cardiotoxicity, hepatotoxicity). As one drug development guide notes, preclinical studies “are designed to assess the feasibility, safety, and biological activity of a new drug… primarily using in vitro (cell-based) and in vivo (animal-based) testing” (^[2]). In this way, in vitro work serves as a filter: roughly it is estimated that only about 1 in 5,000 compounds screened in such assays eventually become marketable drugs (^[10]).

Classic uptake studies, also part of preclinical, measure absorption in gut-like cell layers or through skin/membrane models. Metabolism studies determine whether the compound is stable or quickly broken down by liver enzymes. Importantly, metabolite identification is performed so as not to miss toxic byproducts. If a major metabolite is found, it is often synthesized and tested in parallel (sometimes even in clinical trials) to assess safety. In vitro tests can also reveal receptor cross-reactivity (as happened historically with TGN1412: in monkeys the immunological response was not felt, but in humans it was explosive).

In Vivo Animal Studies. A key hallmark of preclinical research is animal testing. These studies are done only after in vitro data suggest tolerable profiles. Typically two species are used: one rodent (mouse or rat) and one non-rodent. The non-rodent is chosen for physiological similarity to humans. For small-molecule drugs, common non-rodents are dogs or pigs; for large biologic drugs, non-human primates are often chosen because of closer immune/pharmacologic homology (^[17]). As the translational literature explains, “Nonrodent species should be chosen that are most pharmacologically…relevant to humans…Typically, canine and nonhuman primates are used for small and large molecule studies, respectively” (^[17]).

When animal studies are used, they may include pharmacology studies to investigate mechanism or activity and toxicology studies to characterize potential adverse effects. Pharmacology studies might involve, for example, testing an anti-inflammatory compound in rodent models of arthritis, to confirm it modulates the intended pathway. These studies (often not GLP-compliant but still rigorously done) help verify that the drug hits its target in vivo.

Toxicology and Safety Pharmacology. The nonclinical safety program supporting a first-in-human trial is tailored to the product and proposed investigation. GLP is generally expected for pivotal in vitro and in vivo studies submitted in support of an IND. Depending on the program, it may include:

• Single-Dose (Acute) Toxicity: Animals (rodent and non-rodent) receive one escalating dose up to the maximum tolerated dose (MTD) or maximum feasible dose. Observations over ~1-2 weeks reveal organs or systems at risk.
• Repeat-Dose (Subchronic) Toxicity: Animals receive daily doses (often via intended human route, e.g. oral or IV) for a period (e.g. 2–4 weeks for an initial study). This identifies target organ toxicity and helps establish a “No Observed Adverse Effect Level” (NOAEL), the highest dose at which no significant harm is seen (^[15]) (^[18]). The NOAEL (and often, the maximum tolerated dose) are pivotal in calculating a safe starting human dose.
• Safety Pharmacology: Specific studies focus on vital systems. For example, a “CNS safety” study monitors behavioral/neurological effects in animals; a cardiovascular study examines any drug effect on heart rhythm; a respiratory study assesses breathing effects. ICH guidelines recommend core batteries for safety pharmacology, and many of these are conducted as parts of GLP toxicology studies (^[15]).
• Genotoxicity (Mutagenicity) Testing: Short-term tests (such as the Ames bacterial reverse mutation test and mammalian cell assays) determine if the compound can damage DNA. Certain highly sensitive assays may be required if chronic dosing is planned.
• Carcinogenicity & Reproductive Toxicity (as needed): Carcinogenicity assessment is product-specific. Under ICH S1B(R1), the traditional approach generally includes a long-term rodent study—usually a two-year rat study—plus a second rodent carcinogenicity approach; an integrated weight-of-evidence assessment can determine whether a two-year rat study adds value in some cases. Reproductive and developmental toxicity studies are planned according to the product and the populations and clinical trials to be supported.

The nonclinical package for a first-in-human IND is evaluated in the context of the proposed investigation and the product. FDA states that the kind, duration, and scope of animal and other studies depend on the duration and nature of the proposed clinical investigation (^[1]). Under ICH M3(R2), repeated-dose toxicity studies in two species, including one non-rodent, of at least two weeks generally support clinical trials of up to two weeks; longer trials generally require studies of at least equivalent duration, subject to the guideline’s exceptions. Therefore, a one-month, two-species package is not a universal IND-entry requirement (^[19]).

Together, the in vivo studies inform key preclinical decisions. They yield dose–response information (e.g. what dose causes liver enzyme elevations or weight loss in 10% of animals), identify toxicities, and provide PK data (blood levels of drug over time) that can be correlated with effects. Pharmacokinetic and toxicokinetic information is generated as appropriate to characterize exposure in the test species and support the proposed clinical program. Under ICH M3(R2), in vitro metabolism and plasma-protein-binding data for animals and humans, plus systemic-exposure data in species used for repeated-dose studies, generally should be evaluated before initiating human trials; broader ADME information is generally expected before prolonged exposure or large clinical studies. The FDA guidance on pre-IND safety explicitly states that preclinical studies must establish a basic safety profile before human dosing (^[3]).

Chemistry, Manufacturing and Control (CMC). A critical but sometimes overlooked part of development is producing material under controls appropriate to its intended use and stage. For an IND, FDA expects sufficient CMC information to assure the investigational drug’s identity, quality, purity, and strength; the amount of information varies with the phase, formulation, and duration of the proposed investigation (^[20]). Early clinical programs may use phase-appropriate CMC controls rather than a fully mature commercial GMP package. The IND CMC section describes the drug substance and product, manufacturing and controls, stability, and the proposed clinical formulation, including any relevant differences from material used in animal toxicology studies.

“Weeding Out” and Go/No-Go Decisions. Throughout preclinical, decision points abound. At several stages, the team assesses whether there are show-stopper issues (e.g. lethal toxicity, inability to formulate, lack of any efficacy in disease models). As one regulatory science paper notes, “During the early pre-clinical development process, also known as Go/No-Go decision, a drug candidate needs to pass through several steps… determination of drug availability (studies on pharmacokinetics)” (^[14]). In other words, the candidate must “check all the boxes” in early preclinical before advancing. Given the high failure rate, it is common that only a few out of hundreds of synthesized leads ever survive the preclinical gauntlet.

Ethics and Alternatives. Modern preclinical research increasingly emphasizes alternatives to animal use. The “3Rs” principle – Replacement (using cell/tissue/simulations instead of animals when possible), Reduction (using the minimum number of animals by statistical design), and Refinement (minimizing pain and distress) – guides this effort (^[16]). For example, rodent embryonic stem cell assays can sometimes screen for developmental toxicity, potentially reducing the need for full litter studies. Organs-on-chips and multi-organ microfluidic systems aim to replicate human organ-level responses to drugs, which may improve human predictivity and reduce animal use in the future. Regulatory agencies (FDA, EMA, etc.) actively encourage the use of validated alternative methods. The legislative landscape has shifted significantly: the FDA Modernization Act 2.0 (2022) removed the blanket requirement for animal testing, permitting sponsors to use cell-based assays, organoids, organs-on-chips, and computational models as alternatives. The FDA Modernization Act 3.0, which passed the Senate unanimously in December 2025, would, if enacted, direct FDA to replace specified animal-testing terminology with nonclinical terminology in enumerated drug and biologics regulations. In April 2025, FDA Commissioner Makary announced a 5-year roadmap to reduce animal testing, starting with monoclonal antibodies, aiming to make animal testing "the exception rather than the norm." In March 2026, the FDA released draft guidance on New Approach Methodologies (NAMs), describing four general validation considerations: context of use, human biological relevance, technical characterization, and fit for purpose. FDA recommends consultation with the appropriate review division for indication-, disease-, organ-, and endpoint-specific applications. Despite these advances, ICH M3(R2) guidance still recommends certain animal data for safety evaluation before clinical trials, though the regulatory consensus is evolving rapidly.

Regulatory Interaction. Finally, before moving to clinical trials, a preclinical dossier is compiled into the IND (or Clinical Trial Application, CTA, outside the U.S.). The IND includes all animal study reports, CMC data, proposed human protocols, and investigator information. Regulatory agencies often allow a Pre-IND Meeting where the sponsor can present the preclinical data and proposed human trial design to get feedback. A study may begin when the IND is in effect—generally 30 days after FDA receives the application unless FDA places the study on clinical hold, or upon earlier FDA notification—and after applicable IRB review and informed-consent requirements are met (^[21]) (^[22]). In practice, regulators may require additional preclinical work if gaps are identified. For example, if animal toxicology showed potential liver effects, regulators may ask for specialized liver PK or histology. Thus, regulatory review is an integral component of “what happens before humans are involved” – it is the formal checkpoint ensuring preclinical rigor.

Once preclinical milestones are achieved, a candidate drug enters clinical development. Clinical trials are conducted in sequential phases to test safety and efficacy in humans, always under strict ethical and regulatory oversight. Unlike the controlled laboratory environment of preclinical work, clinical trials involve living patients (or healthy volunteers) and must account for human variability, placebo effects, and behavioral/compliance factors.

• Phase 0 (Exploratory Trials): In some cases, a very small dose (“microdose”) is given to a few humans (usually healthy volunteers) to quickly assess basic pharmacokinetics and target engagement without expecting therapeutic effect (^[23]). This is an optional early step to help calibrate animal-to-human dose extrapolations, but is used only in special programs. Phase 0 still requires IND and ethical approval.

• Phase I (First-in-Human): The first formal phase I trial typically involves 20–100 healthy volunteers (for non-oncology drugs) or sometimes patients (for toxic therapies or cancer drugs). The primary goals are safety and tolerability: researchers give ascending doses to small cohorts to determine the maximum tolerated dose (MTD) and observe any adverse effects. Secondary objectives often include preliminary pharmacokinetics (half-life, clearance in humans) and pharmacodynamics (biomarkers of drug activity). A classic 3+3 design or other dose-escalation rules is used. According to recent data, only about ~47% of drugs entering Phase I progress to Phase II (^[4]). IRB approval and informed consent are mandatory. Applicable clinical trials must be registered and reported on ClinicalTrials.gov as required by U.S. law; registration may also be required by other jurisdictions, funders, or journals. Trials are governed by GCP. Dropouts or unexpected toxicities in Phase I can halt a development program early, reflecting residual uncertainty after preclinical safety.

• Phase II (Proof-of-Concept): Phase II studies (often still randomized but sometimes uncontrolled) enroll several hundred patients who have the target disease. The goals are to obtain preliminary efficacy data and to further assess safety and dose-response. These trials can be subdivided into Phase IIa (pilot proof-of-concept, often open-label) and Phase IIb (dose-ranging, randomized). Phase II is statistically powered to find signals of benefit (or futility) but is not definitive. It is often the biggest “hurdle”: only roughly 25–30% of Phase I candidates succeed in Phase II (^[4]). Failure often occurs due to lack of efficacy or unacceptable side effects in the patient population that were not seen in animals or healthy volunteers. Clinical endpoints (e.g. symptom score, biomarker change) and safety data from Phase II inform dose selection for Phase III.

• Phase III (Pivotal Trials): Successful phase II drugs move to large-scale Phase III trials, which typically involve hundreds to thousands of patients across multiple centers. These are randomized, controlled trials (often placebo- or standard-care–controlled) designed to definitively demonstrate efficacy on pre-specified primary endpoints, as well as to gather extensive safety data. Phase III trials can take 2–4+ years. According to industry data, about 50–60% of drugs that enter Phase III will meet their primary efficacy endpoints (^[4]). A positive Phase III is used to support a regulatory application.

After Phase III, a sponsor submits a New Drug Application (NDA) or Biologics License Application (BLA) with all data. The regulatory review process involves additional inspections of GMP and GLP practices and may require Phase IV (post-marketing) studies for long-term safety monitoring.

Throughout clinical development, the regulatory focus shifts from establishing basic safety (preclinical) to demonstrating a favorable benefit-risk in humans. Ethical oversight intensifies (multiple IRB reviews, data safety monitoring boards, patient consent). The standards for evidence are also higher: randomized controlled trials with statistical rigor are expected.

The key differences between preclinical and clinical testing are summarized in Table 1. Preclinical research is exploratory, model-driven, and focused on risk avoidance in future human trials (^[3]) (^[24]). Clinical trials are confirmatory, hypothesis-driven in patients, and focused on measuring human effects under GCP. For example, preclinical studies define a safe starting dose (often by computing the human equivalent dose from the animal NOAEL with safety factors); clinical phase I actually tests that dose and escalates it in humans. Preclinical endpoints are typically biomarker changes or pathological findings in animals, whereas clinical endpoints are patient outcomes (e.g. survival, relief of symptoms, functional improvement).

Aspect	Preclinical Research	Clinical Trials
Subjects/Models	In vitro systems (cells, tissues, computer models) and animal models (rodents, non-rodents) ([24]) ([17])	Human volunteers/patients with IRB approval
Primary Goals	Assess toxicity, pharmacokinetics (ADME), biologic mechanism, and formulation ([24]) ([2])	Assess safety (Phase I) and efficacy (Phases II/III) in target population; refine dosing
Regulations/Ethics	Good Laboratory Practice (GLP) compliance ([13]); Animal ethics (3Rs) ([16])	Good Clinical Practice (GCP); IRB/consent; registration
Outcome Metrics	NOAEL, MTD in animals; organ toxicities; biochemical endpoints (enzyme levels, histopathology) ([15])	Adverse event rates; pharmacokinetics in humans; clinical endpoints (symptom scores, biomarkers, survival)
Scale (Subjects)	Dozens to hundreds of animals (typically <50 per study) and unlimited cell assays ([25])	Phase I: ~20–100; Phase II: ~100–300; Phase III: ~300–3000+; >3,000+ in Phase IV surveillance
Timeline	1–5 years (discovery through IND submission)	~1–2 yrs (Phase I) + 1–3 yrs (Phase II) + 2–4 yrs (Phase III) ([26])
Success Probability	No universal screening-to-IND or screening-to-approval ratio; estimates depend on modality, program definitions, and where counting begins	Phase-transition and approval estimates vary by dataset, time window, and definition. A 2025 analysis found recent overall success of about 5% for all clinical development programs, not a universal phase benchmark ([5])
Cost Distribution	Lower absolute cost per study (though discovery costs significant); contributes to ~20–30% of R&D spend	Major cost driver (especially Phase III), roughly ~70–80% of late-stage R&D budget
Control vs. Variability	Highly controlled lab conditions; homogeneous animal strains	Inherent human variability (genetic, environmental, comorbidities)

Table 1: Comparison of Preclinical versus Clinical Research. Key features, goals, and scales differ markedly between the two stages (^[12]) (^[3]).

As Table 1 indicates, preclinical work is limited to non-human systems. The “costs” of failure at this stage manifest as lost experiments and wasted chemistry efforts, whereas clinical failure incurs the far greater cost of patient trials. For example, Tufts CSDD estimated that developing one successful new drug (including failures) averaged about $1.24 billion (in 2005 USD) (^[27]). More recent analyses paint an even starker picture: a January 2025 RAND study published in JAMA Network Open found a median cost of $708 million per approved drug (mean $1.3 billion, skewed by outliers), while Deloitte's 15th Annual Report (2024 data) placed the average cost for large pharma at $2.23 billion per asset—with $7.7 billion spent on terminated candidates alone in a single year. Much of this cost is absorbed in the clinical phases, but it depends on thorough preclinical weeding. In practice, a lead chemical often fails in preclinical development due to toxicity, solubility problems, or pharmacokinetics, so that only a handful of candidates ever reach IND. When one does, the clinical phases carry it forward or not.

Drug development attrition is dramatic and data-driven analysis offers insight into each phase. Industry studies of pipeline statistics show that failure in clinical trials, especially in Phase II, is much more common than success. For drugs entering Phase I trials in the 2014–2023 period, the overall odds of approval was only ~6.7% (^[11]). This is consistent with earlier Tufts and Nature reviews that cited single-digit likelihoods. By therapeutic area, oncology drugs have especially steep attrition (Phase III failure rates ~50–60% (^[28]) (^[29])).

A comprehensive analysis by Keith Kaitin (Tufts CSDD) examined long-term metrics. Kaitin et al. reported an average of 7.2 years from the start of clinical trials to market approval, and an overall clinical success rate of only ~16% for products that entered human trials (^[26]) (^[30]). These figures reaffirm that even with good preclinical safety, most candidates fail due to insufficient efficacy or unforeseen issues. Kaitin’s data also noted variability by area; for instance, neuropharmacologic agents had especially low success (~8.2% (^[30])). More recent analyses from Citeline suggest success rates may even be declining: Phase I-to-approval odds have fallen to 6.7% (2014–2023) compared to previous decades (^[11]), largely because Phase II success has dropped. A 2025 Nature Communications study ("Dynamic clinical trial success rates for drugs in the 21st century") confirmed this trend, showing that Phase I success rates dropped from over 75% (2006–2008) to below 40% in recent years, though overall rates have recently plateaued and begun to recover slightly. Notably, the study found that repurposed drugs now have lower success rates than novel drugs—a counterintuitive finding. At the company level, leading pharmaceutical firms achieve an average likelihood of approval of 14.3% from Phase I—roughly double the industry-wide average—reflecting the value of institutional expertise in preclinical candidate selection and clinical trial design.

By comparison, there is no single “publication” of preclinical success rates because the entire discipline of preclinical is more diffused. However, one metric is the fraction of high-throughput screened compounds that yield an IND candidate. CSIRO (Australia’s research agency) estimates roughly 1 in 5,000 active compounds ultimately becomes a marketable drug (^[10]). In practical terms, for every thousand molecules tested in vitro, only a few move to animal studies, and far fewer reach IND. Among biopharmaceutical companies, reports suggest roughly 1–5% of lead candidates entering preclinical end up submitting an IND (^[10]). Attrition in preclinical often occurs due to “no efficacy in any animal model,” unacceptable toxicity (e.g. genotoxicity or organ-specific damage), or poor ADME (e.g. too rapidly metabolized or insoluble).

Another quantitative lens is time and cost. While exact preclinical durations vary (often 1–4 years from lead optimization to IND), the clinical phases typically span significantly longer (Phase I–III combined ~6–8 years (^[26])). A landmark Tufts analysis (2001) estimated the total time from discovery to launch at ~10–15 years. More granular analysis indicates about 7.2 years are spent after IND (clinical), implying 3–5 years in pre-IND development (^[26]). Published development-cost estimates vary substantially because they use different samples and account differently for failed programs and the cost of capital. For example, a RAND analysis of 38 drugs approved by FDA in 2019 estimated a median adjusted R&D cost of $708 million and a mean of $1.3 billion; these results should not be generalized as a fixed cost split between preclinical and clinical development (^[31]).

It is worthwhile noting power and sample size differences. Preclinical toxicology studies might involve only small cohorts (e.g. 3–10 animals per sex per dose group) and focus on detecting gross toxicities. Clinical trials must be statistically powered to detect much smaller treatment effects, requiring tens to thousands of patients. The regulatory standards for data analysis differ: preclinical results are often assessed qualitatively (detailed pathology reports, NOAEL determination) or with simple statistics, while clinical trial results demand rigorous statistical analysis (confidence intervals, p-values) and often have independent data monitoring boards.

Examining real-world examples highlights the strengths and limitations of preclinical work. Table 2 summarizes some notable cases (both failures and successes) illustrating what can happen before humans are involved.

Case/Drug	Preclinical Findings	Outcome in Humans	Lessons Learned
Thalidomide (1960s)	Passed non-GLP tests; no routine teratology studies done ([7])	Caused severe birth defects (~10,000 cases worldwide) ([7])	Prompted regulations requiring formal reproductive toxicity testing; historic cautionary tale.
Fialuridine (FIAU, 1990s)	Rodent and primate studies did not predict toxicity; mitochondria not tested explicitly	In a hepatitis B trial, 5 of 15 patients suffered fatal liver failure ([9])	Uncovered that some human-specific toxicities (mitochondrial) can be missed; led to new preclinical screening for mitochondrial toxicity.
TGN1412 (2006)	Extensive monkey studies showed no severe toxicity; high NOAEL; lacked tests for cytokine release in human immune cells ([8])	First-in-human trial: 6 volunteers given 0.1 mg/kg (500× below animal NOAEL) all developed life-threatening “cytokine storm” ([8])	Taught importance of cautious dose escalation, in vitro human immune tests (e.g. on human blood) for immunotherapies, microdosing steps.
Rofecoxib (Vioxx, 1999)	Standard toxicity tests in rats/dogs showed no obvious cardiovascular issues; initial 3-month dog studies	Merck voluntarily withdrew Vioxx in September 2004 after a long-term controlled clinical trial found an increased risk of serious cardiovascular events versus placebo ([32]).	Highlighted that some long-term or idiosyncratic toxicities (e.g. prothrombotic effects) may not appear in short animal studies.
Brillinta (ticagrelor)	Safe in animal models with adequate antithrombotic effect; metabolized to active form	In large Phase III PLATO trial, showed superiority to clopidogrel with manageable bleeding risk; approved	Example of concordance: animal models of platelet aggregation reasonably predicted human efficacy.
PCSK9 inhibitors (e.g. evolocumab)	Monoclonal antibody showed robust LDL-lowering in mice and primates with no toxicity	In humans, large trials confirmed potent cholesterol lowering and cardiovascular benefit ([33])	Success story reflecting that a well-chosen animal model (primates with lipid metabolism similar to humans) can translate to human disease.

Sources: Historical and pharmacological case reports (row 1–4) and modern clinical trial publications (row 5–6) (^[7]) (^[9]) (^[8]).

The above cases illustrate diverse facets of “preclinical vs. clinical.” For thalidomide, the lack of rigorous animal teratology testing (and regulatory guidance) led to disaster. This case helped drive modern reproductive and developmental toxicity assessment. The timing and design of those studies are tailored to the product, intended population, and clinical program; ICH permits staged approaches and product-specific study designs (^[34]). In the fialuridine case, standard animal tests failed to reveal the drug’s mitochondrial toxicity; only after patient deaths did investigators find the underlying cause. This has since led to inclusion of in vitro tests for mitochondrial toxicity in certain drug classes (especially nucleoside analog antivirals). For TGN1412, the culprit was a superagonist antibody: animal models, even non-human primates, did not predict the massive cytokine response in humans. Investigators recommended new in vitro tests using human immune cells (e.g. measuring cytokine release) before giving immunomodulatory agents to people. Both TGN1412 and Vioxx underscore that some human-specific or long-term risks are inherently hard to gauge in preclinical stages.

Conversely, some modern successes (not in the table) show alignment. For example, many cancer drugs targeting human tumor pathways are tested in mouse xenograft tumor models. Those that show a strong effect often do modestly well in early human trials, though even here the relevance is limited by differences in tumor microenvironment and immune system. The development of PCSK9 inhibitors (a cholesterol-lowering class) is cited as a case where animal (primate) models correctly predicted a large LDL reduction in humans, facilitating faster development.

The dichotomy “preclinical vs. clinical” has practical implications for researchers, industry, regulators, and patients alike. Given the high stakes, there are strong incentives to improve preclinical predictivity and streamline early development. Several current trends bear highlighting:

• Advanced Preclinical Models. New technologies aim to bridge the gap between animal models and humans. Human induced pluripotent stem cell (iPSC) platforms can produce patient-specific cell types (heart, liver, neurons, etc.) for toxicity screens. Organ-on-a-chip microphysiological systems combine multiple cell types with perfusion to mimic organ function; these have shown promise in detecting cardiotoxicity and liver toxicity that standard animal tests sometimes miss. CRISPR-engineered mice carrying human drug-metabolizing enzymes or immune receptors are being developed to make animal responses closer to human. Another approach is in silico modeling: computer simulations of drug-target interactions and animal-to-human dose scaling can prioritize leads and flag potential risks before any lab testing. While none of these fully replace GLP animal tests today, they can reduce reliance on animals and provide earlier insight. Federal initiatives (e.g. NIH’s Tissue Chip program) and regulatory interest (FDA’s Predictive Toxicology Roadmap) are accelerating these tools. In July 2025, NIH announced that it would no longer develop new funding opportunities focused exclusively on animal models of human disease. Applicants may still propose animal-only research when it is scientifically appropriate and justified. The FDA’s March 2026 NAMs draft guidance provides general validation considerations when nonclinical NAM data are submitted in support of a drug application. It does not address specific methodologies, and sponsors are encouraged to consult the relevant FDA review division about a proposed use.

• Microdosing and Accelerated Trials. Phase 0 microdosing (as mentioned) exposes humans to sub-pharmacologic doses (typically <1/100th of therapeutic exposure) to gather PK data and early PD signals. This can validate some preclinical assumptions. Likewise, some first-in-human trials now use adaptive designs or accelerated titration to more cautiously escalate doses, especially for biologics or oncology drugs. Regulatory agencies (FDA, EMA) have provided guidance for “safe harbor” of exploratory IND microdosing, reflecting confidence that microdosing poses negligible risk but high informational value. These strategies aim to expose fewer patients to ineffective doses and detect failure earlier.

• Regulatory Science and Biomarkers. The preclinical/clinical transition is aided by stronger biomarkers of drug effect. For example, if a preclinical model identifies a blood biomarker that increases when the drug hits its target, clinical trials can measure the same biomarker to see if the drug engages the target in humans at expected doses. Consistency of PD markers between animals and humans increases confidence. Agencies now encourage use of translational biomarkers when available. Similarly, computational toxicology (quantitative structure–activity relationships) is used to predict off-target binding (e.g. hERG channel assays for cardiac risk) before clinical testing.

• Data Integration and Open Science. Large databases of preclinical and clinical results (such as PubChem for compounds or repositories of failed clinical trials) help researchers learn what failed and why. Collaborative data sharing initiatives (e.g. TransCelerate) aim to share preclinical tools and standards across industry. Meta-analyses of attrition data (like the Norstella/Citeline report) keep the community informed of risk profiles by disease area.

• Ethical and Patient Considerations. From the patient perspective, better preclinical testing ideally means safer trials. However, excessive animal requirements can delay life-saving therapies. A central dilemma is how much risk is acceptable for patients in first-in-human trials. The well-publicized TGN1412 tragedy has made regulators extremely cautious with novel biologics. Conversely, in areas of high unmet need, FDA programs such as Fast Track and Breakthrough Therapy designation are intended to facilitate development and expedite review for qualifying products. They provide additional interaction and, for eligible products, features such as rolling review; they do not categorically waive product-appropriate nonclinical requirements or guarantee smaller clinical trials. International clinical research remains subject to the requirements of the jurisdictions involved, and harmonization efforts such as ICH guidelines seek to promote common technical standards.

• Emerging Therapies: New modalities—gene therapies, cell therapies, nanomedicines—bring fresh preclinical/clinical challenges. For example, viral vectors for gene therapy require specialized biodistribution studies, and immune response assays, that go beyond standard toxicology. Stem cell transplants or CAR-T cell therapies are essentially “living drugs” with unique considerations. Preclinical models for such therapies are inherently limited, so clinical translation often hinges on very conservative phase I protocols.

• AI-Driven Drug Discovery and Regulatory Frameworks. Artificial intelligence may support candidate selection, data analysis, and regulatory submissions. In January 2025, FDA issued draft guidance proposing a risk-based framework for establishing the credibility of AI models used to support regulatory decisions about a drug or biologic’s safety, effectiveness, or quality. The appropriate credibility activities depend on the model’s context of use.

Future Directions: The ultimate goal is to reduce failure rates and development times. If preclinical models could more reliably predict clinical success, many costly trials might be avoided. Integration of AI/machine learning to mine chemical and biological data could suggest better targets and flag risky compounds. In the next decade, it is plausible we will see increased use of human-based models (e.g. patient-derived organoid libraries) to complement or, in some cases, supplant animal tests—a trajectory supported by the enacted FDA Modernization Act 2.0, the Senate-passed FDA Modernization Act 3.0, and FDA's 2026 draft guidance on validating new approach methodologies (NAMs). On the clinical side, adaptive trial designs, real-time data monitoring, and even digital biomarkers (wearable sensors) may speed the evaluation of early safety and efficacy, reducing patient exposure to ineffective approaches. Recent high-profile clinical failures in 2025—including semaglutide in Alzheimer's (biomarker improvement but no clinical effect in 3,800+ patients) and latozinemab in frontotemporal dementia (target engagement without clinical benefit)—continue to underscore the gap between biological plausibility demonstrated in preclinical models and therapeutic efficacy in humans.

However, these advances come with challenges. Regulatory frameworks will need to adapt to accept non-traditional data (Can an organ-on-chip result count as equivalent to a rat study in an IND?). Ethical debates will continue (should microdosing be more widely allowed? How to protect trial participants while expediting new treatments?). Cost pressures also remain: the industry must invest in more preclinical studies that could double early costs, in exchange for hopefully reduced failure later. Some critics argue that truly first-in-human drug exposures always carry uncertainty, and no amount of preclinical data can foretell rare idiosyncratic effects. Others point to successes like improved study design or patient selection (e.g. only enroll patients with the target biomarker) as equally important to bridging the gap.

From a broader perspective, science aims to make the leap from “bench to bedside” more efficient. The concept of translational medicine has gained traction: it formalizes feedback loops in which clinical observations inform new preclinical hypotheses, and preclinical findings quickly guide clinical strategy. For example, if a Phase I trial uncovers a mild liver enzyme elevation, researchers might immediately run additional animal or cell tests to understand it. Conversely, if animal studies suggest a possible new indication, a “phase IIa” trial may test that hypothesis in a small patient cohort. This tight coupling reflects a systems-thinking approach to drug development, rather than a rigid handoff at IND.

Practically, improvements in preclinical prediction could have large implications for public health. Lower failure rates mean lower costs for successful therapies, and potentially faster access for patients. They might also reduce reliance on animal testing, which is an ethical and societal benefit. New regulatory pathways (e.g. “animal rule” for drugs where human trials are impossible, or “adaptive approvals”) provide models for how less conventional data might suffice when urgent need is present.

In the context of global pandemics (e.g. vaccine development for COVID-19), the usual preclinical timelines were compressed by overlapping steps and massive parallelization. Traditional distinctions between preclinical and clinical were blurred: some vaccine platforms began human trials based on small animal data plus mechanistic reasoning, because the alternate was many more deaths. This raises the question of how rigid the “preclinical/clinical barrier” should be in crises—certainly, science can accelerate, but it still largely follows the principle that some safety data must precede broad human use.

The gap between preclinical promise and clinical reality has wide-ranging implications:

• For Researchers: It underscores the need to choose the right models. A model is only as good as how well it mimics human disease. For example, many Alzheimer’s treatments worked in mouse plaques but failed in humans, likely because rodent models do not recapitulate human neurodegeneration fully. Thus, a preclinical researcher must critically evaluate whether an animal model has face, predictive, and construct validity for the human condition. There is also an increasing role for biomarkers and systems biology, to better translate mechanistic understanding into dosing and patient selection strategies.

• For Industry: High attrition means enormous sunk costs on failures. Thus, pharmaceutical and biotech companies are investing in de-risking strategies. Many firms now require “human-proof-of-concept” even in preclinical decisions (for instance, requiring proof that a target is linked to disease by genetic evidence or genome-wide association studies) to avoid costly blind alleys. In addition, strategic portfolio management becomes crucial: firms dynamically allocate resources away from projects that stumble in preclinical or early clinical. The concept of “go/no-go” gates (with objective criteria including target engagement, animal efficacy data, etc.) is formalized at upper levels of development teams to make disciplined kill decisions early.

• For Regulators: Agencies continue to balance the dual aims of protecting human subjects and enabling innovation. They issue guidances clarifying what preclinical package is needed for various first-in-human scenarios. Regulators also hold “pre-IND” or scientific advice meetings to align expectations. Recent shifts include more acceptance of model-informed drug development (MIDD): using pharmacometric models built on animal and early human data to predict dosing and responses, thereby potentially reducing trial sizes. The FDA and EMA have published documents on using MIDD and on new toxicology paradigms (e.g. use of genomic biomarkers in animal toxicity studies). In March 2026, FDA released its general NAMs draft guidance. It provides a general validation framework and recommendations for NAM data used in drug development; it is not intended to address specific NAMs. Sponsors should consult the appropriate FDA review division about a proposed use. Regulatory science – the study of how to best evaluate data – is itself an evolving field, with both the FDA Modernization Acts and new guidance documents formally acknowledging that the rigid “animal → then human” model is being augmented for novel therapies.

• For Patients and Society: Ultimately, preclinical research exists to protect patients and produce effective therapies. Patients generally benefit from thorough preclinical testing because it reduces the likelihood of unexpected harm. However, there is a trade-off: stringent requirements can lengthen development time and increase costs, potentially slowing the availability of new drugs. Stakeholder groups (patient advocates, bioethicists) debate how to balance these factors. For example, terminally ill patients sometimes demand expanded access or “right-to-try” even before all preclinical data is in, arguing that potential benefit outweighs unknown risks. Society must also weigh animal welfare concerns: strong preclinical safeguards reduce risks to trial participants but involve animal experiments. The public’s confidence in the drug approval process depends on both dimensions: feeling safe in trials while also trusting that researchers are not recklessly exposing humans.

• For Future Therapies: Personalized medicine and gene therapy blur preclinical/clinical lines. For a one-off gene therapy for a rare disease, the “traditional path” may not be followed due to ethical imperatives. Regulatory frameworks for such cases are under active discussion (e.g. FDA’s approach to emergency INDs or the EU’s Compassionate Use programs). On the technological front, artificial intelligence is increasingly used to predict off-target effects and to optimize chemical structures in silico before any wet-lab test. The FDA's January 2025 draft guidance on AI in drug development proposes a risk-based framework for establishing the credibility of AI models used to support regulatory decisions. AI may assist candidate selection, data analysis, and regulatory submissions, but clinical-development performance should be assessed using clearly defined datasets that distinguish AI-discovered programs from programs that merely use AI-enabled tools.

In addition to Table 1 (above), Table 3 provides a snapshot comparison of typical timelines and attrition rates through the pipeline. This illustrates the drastic narrowing of candidates as one moves from preclinical evaluation to market approval.

Stage	Typical Duration	Attrition / Success	Likelihood of Approval From Entry Into Stage
Discovery/Lead Optimization (Pre-IND)	~1–3 years (often overlapping with early preclinical)	Very high attrition; only a few percent of leads reach IND ([10])	Not reported in this clinical-stage dataset
Preclinical (GLP Toxicology)	~1–2 years (can overlap with discovery)	Many INDs are abandoned here if toxicity is unacceptable; success undefined	Not reported in this clinical-stage dataset
Phase I	~0.5–1 year	~53% of INDs fail in Phase I (Phase I success ~47%) ([4])	~6.7%
Phase II	~1–2 years	~72% of Phase I survivors fail here (success ~28%) ([4])	~14.2%
Phase III	~2–4 years	~45% of Phase II survivors fail (success ~55%) ([4])	~50.6%
NDA/BLA Review	~1–2 years post-Phase III	~92% of Phase III completed programs succeed ([4])	~92%
Post-Approval (Phase IV)	Ongoing post-market study	Shows long-term safety/rare events, not part of initial success metric	–

Table 3: Timeline and Attrition of Drug Development Stages. The chances of moving to the next stage are indicated (data from recent industry analyses (^[35])). Note how only a small fraction of IND-eligible molecules ultimately become approved drugs.

Understanding the interplay between preclinical and clinical phases is crucial for appreciating how new therapies reach (or fail to reach) patients. Preclinical research is an indispensable gatekeeper: by rigorously testing drug candidates in non-human systems, it seeks to prevent human harm and optimize the candidates that move forward. These studies span from early discovery screening to GLP animal toxicology to manufacturing quality control. Clinical research then takes the baton, assessing whether the drug’s promise holds true in humans and quantifying its benefit–risk. The two stages employ different methods, ethical oversight, and success criteria (summarized in Table 1).

Quantitative data underline that attrition is steep at every stage. Industry analyses consistently show that only single-digit percentages of new compounds reach the market (^[11]) (^[30]). This stark reality has driven a dual imperative: to reduce avoidable failures (by better preclinical models and predictive tests) and to make development faster and more patient-centric. The tragic examples of thalidomide, TGN1412, and others serve as stern reminders that preclinical work – however comprehensive – can never guarantee human safety. Each new case of unexpected toxicity prompts scientists and regulators to refine the preclinical toolkit further.

Looking ahead, the frontier lies in closing the preclinical-clinical gap. FDA’s March 2026 draft guidance describes general validation considerations for New Approach Methodologies (NAMs) submitted in support of drug applications and encourages consultation with the relevant review division; it does not establish a universal replacement for product-appropriate nonclinical evidence (^[36]). The practical requirements remain product- and program-specific. Human-relevant models, translational science, and carefully evaluated computational tools may strengthen development programs, but their use must fit the regulatory context of use.

Ultimately, while clinical trials in humans are the definitive test, they rest on the foundation of preclinical research. Ensuring that foundation is solid – through rigorous science, transparency of data, ethical care for animals and people, and innovative tools – will shape the future landscape of medicine. The journey before human involvement is complex, highly regulated, and ever-evolving. Its mastery is essential if the transition from laboratory promise to patient benefit is to be smooth, efficient, and safe (^[3]) (^[30]).