Monitoring clonal and mutational drift across serial mouse passages
Every genomic drift monitoring program begins with a reference point: the earliest available passage of a PDX line, sequenced and characterized as thoroughly as possible before any mouse-to-mouse passaging has had the chance to reshape its clonal composition.
The scientific value of a PDX model rests on the assumption that it faithfully represents the patient tumor it was derived from. Establishing this fidelity requires deep molecular characterization of the earliest achievable passage — whole-exome or targeted panel sequencing, copy-number profiling, and often RNA-seq — performed on the founder (F1) tumor and, whenever tissue allows, directly compared back to the original patient specimen.
This baseline characterization becomes the fixed reference point that every later passage will be measured against. Without it, "drift" cannot be quantified at all — a program monitoring genomic stability without a captured baseline is only able to compare passages to each other, not to the ground truth of the original patient tumor.
Best practice is to deep-sequence and then cryopreserve substantial stock of the earliest possible passage (F1-F2), so that if a working colony later drifts beyond acceptable limits, the line can be re-derived from a passage still close to the original patient genome rather than being lost entirely.
Comparative studies sequencing matched patient tumor and F1 PDX tissue typically report high — but not perfect — concordance: the large majority of somatic single-nucleotide variants and structural rearrangements present in the patient tumor are detectable in the F1 xenograft, with concordance rates commonly exceeding 90-95% for major driver mutations. Some discordance is expected even at this earliest passage, reflecting intratumoral heterogeneity in the original specimen (different tumor regions can carry subtly different mutational profiles) and stochastic sampling of which subclones happen to be present in the specific fragment implanted.
This means "drift" is not purely an artifact introduced by mouse passaging — some baseline clonal heterogeneity is already present at F1, which is why longitudinal monitoring compares each subsequent passage back to this captured F1 baseline rather than to an idealized, unmeasured "true" patient genome.
A rigorous PDX genomic surveillance program defines its monitoring protocol before serial passaging even begins: which passages will be re-sequenced (commonly every 2-3 passages, or at defined intervals such as F1, F3, F5, F8), which assay panel will be used consistently across time points (to avoid confounding technical batch effects with true biological drift), and what quantitative drift thresholds will trigger a "watch" or "fail" QC classification, as covered in Stage 5.
Establishing this plan upfront, and archiving sufficient baseline material for future reference, is what allows a biobank to eventually publish reliable passage-limit recommendations for each individual PDX line rather than relying on generic, one-size-fits-all cutoffs.
To keep a PDX line alive and expand it into a usable research resource, tumors must be repeatedly harvested from one generation of mice and re-implanted into the next — a mechanically simple but biologically consequential process repeated many dozens of times over a model's working life.
When an F(n) tumor reaches a defined harvest size (commonly 1000-1500 mm³), it is surgically excised from the host mouse under sterile conditions. The tumor mass is trimmed of any obviously necrotic core tissue, then cut into multiple small fragments (typically 3×3×3 mm, identical to the original implantation protocol), each of which is implanted subcutaneously into a new, naive recipient mouse — generating the F(n+1) generation.
Because a single donor tumor can typically yield 10-30 viable fragments, each passage represents an expansion step: one F(n) tumor becomes many F(n+1) tumors, which is how a single successful patient engraftment is eventually built into a colony large enough to support drug efficacy studies across dozens of treatment arms.
At every passage, a portion of the harvested fragments is cryopreserved rather than immediately re-implanted, building a frozen archive spanning the model's full passage history. This serves two purposes directly relevant to genomic drift management: first, it provides a practical backup against colony loss (infection, husbandry failure, or simple attrition); second — and more importantly for drift monitoring — it means that if later-passage tumors are found to have drifted unacceptably far from the original patient genome, researchers can thaw and re-establish the line from an earlier, less-drifted passage rather than losing access to that patient's tumor model entirely.
Well-run biobanks maintain freezer inventories indexed by passage number specifically so that "resetting" a drifted line to an earlier passage is a routine, low-friction operation.
Because each passage compounds on the last, a line allowed to passage unchecked to F15 or F20 cannot simply be "corrected" — the only way back to a lower-drift state is retrieving an earlier cryopreserved passage, which is why archiving at every single passage (not just periodically) is standard practice.
Every fragment implanted is logged against its full lineage: which F(n) donor tumor it came from, the harvest date, recipient mouse ID and strain, and implantation site. This passage-tracking metadata is what allows a later genomic QC finding — say, a concerning copy-number event detected at F6 — to be traced back through the line's history to determine whether the same event is present in earlier cryopreserved stock (suggesting it was present from early on and is not new drift) or absent (suggesting it arose specifically during more recent passaging).
This lineage record is foundational to everything downstream: without knowing precisely which passage a given tumor sample represents and where it sits in the line's history, quantitative drift monitoring is not possible.
Each passage is, from a population-genetics perspective, a bottleneck event: only a small fragment of the total tumor cell population is carried forward into the next generation, and that fragment must then re-establish itself in a new, foreign host — conditions that reward some clones and eliminate others.
A patient tumor is rarely a single genetically uniform cell population — it is typically a heterogeneous mixture of related but distinct subclones, each carrying a somewhat different complement of mutations, having diverged from a common ancestor over the course of the tumor's development in the patient. When a 3×3×3 mm fragment is cut for implantation, it captures only a small, essentially random sample of this total clonal diversity — and that sampling is repeated at every subsequent passage.
Compounding this sampling bottleneck is genuine selection: clones that happen to be better adapted to growing in a mouse host microenvironment (different growth factor availability, different stromal signaling, different immune context even in immunodeficient mice) will outcompete less-adapted clones within each passage, gradually shifting the dominant clonal composition of the line away from whatever was dominant in the original patient tumor.
Longitudinal sequencing studies across passaged PDX cohorts generally find that early, truncal driver mutations (the founding oncogenic events present in essentially all tumor cells, such as a KRAS or TP53 mutation) tend to remain stably detectable across many passages, since they are usually present in every subclone and therefore cannot be "selected away." What tends to drift more substantially is the relative abundance of later, subclone-specific passenger mutations and copy-number variants — these can rise, fall, or disappear entirely as the underlying subclone that carries them expands or is outcompeted.
This distinction matters enormously for interpreting a drug efficacy study: a study evaluating a therapy targeted against a stable truncal driver mutation is much less vulnerable to passage-related drift than one relying on a marker or vulnerability only present in a specific subclone that may not persist across passages.
A 2017 large-scale PDX genomic stability study (Ben-David et al.) found that copy-number profiles could diverge measurably from the founder tumor within just a handful of passages in a meaningful fraction of models — a finding that reshaped how the field thinks about passage-limit recommendations.
Drift is quantified by comparing genome-wide copy-number profiles and variant allele frequencies between the baseline passage and each subsequent passage under surveillance. A simple and widely used summary metric is a genome-wide copy-number "drift score" — essentially a distance metric between the CNV profile of the current passage and the F1 baseline, aggregated across the genome. Rising drift scores over successive passages indicate progressive divergence, while a stable, flat drift score across many passages indicates a genomically robust, low-drift line.
Variant allele frequency tracking of specific known driver and passenger mutations, layered on top of the genome-wide CNV metric, gives a more granular picture of exactly which clonal populations are expanding or contracting passage to passage.
Genomic drift is not the only thing changing across passages — the tumor's entire cellular ecosystem is transforming in parallel, as the human stromal and immune cells present in the original patient fragment are steadily replaced by mouse-derived counterparts.
When a tumor fragment is first excised from a patient, it is not pure tumor parenchyma — it carries along a substantial complement of the tumor's native stroma: cancer-associated fibroblasts, tumor-infiltrating immune cells, and fragments of the original human vasculature. This is part of what makes early-passage PDX models such valuable microenvironment research tools — for a brief window, researchers have access to a tumor growing alongside at least some of its authentic human stromal partners.
But none of these human stromal or immune cell populations can self-renew indefinitely inside a mouse host — they have no ongoing source of replenishment (no human bone marrow, no human stem cell niche), so with each passage that dilutes and eventually eliminates them, the tumor's human stromal compartment inexorably shrinks.
As the fragment re-establishes itself in a new host at each passage (recall the angiogenic race described in the companion PDX Tumor Engraftment simulation), the new blood vessels, fibroblasts, and any immune infiltrate that grow into the tumor are, by definition, mouse-derived — the host's own cells responding to the same hypoxic and paracrine signals that originally recruited the (now largely absent) human stroma. Each successive passage repeats this process, so that by roughly the fourth or fifth passage, the tumor's entire non-malignant compartment is essentially of murine origin, while only the malignant epithelial/tumor cells themselves remain of human origin.
This creates an increasingly chimeric model: a genomically human tumor growing within an entirely mouse-derived vascular and stromal scaffold — a state that persists stably for many further passages once fully established.
This stromal transition has a direct practical consequence: any study of stromal-tumor crosstalk, tumor-associated fibroblast biology, or human immune-stroma interactions must use only very early passage PDX material (F1-F2), since by later passages the relevant human stromal cell populations are simply no longer present to study.
Because later-passage PDX tumors contain a mixture of human tumor cells and mouse stromal cells, molecular analyses (particularly RNA-seq and any bulk sequencing approach) must computationally deconvolve the two species' contributions to avoid misattributing mouse stromal gene expression to the human tumor. Species-specific read alignment (mapping sequencing reads separately against human and mouse reference genomes and discarding ambiguous reads) is standard practice, as is confirming that a given passage's tumor purity (fraction of human-origin cells) meets a minimum threshold before that sample is used for expression-based analyses.
Histological confirmation — staining with human-specific versus mouse-specific vimentin or CD31 (endothelial) antibodies — provides an orthogonal, visual check on the same stromal replacement process being tracked genomically.
The practical output of a genomic drift monitoring program is a concrete, defensible answer to a simple question every user of a PDX model needs answered: up to which passage number can this specific line still be trusted to represent the original patient tumor?
Left unchecked, the combined effects of clonal selection (Stage 3) and stromal replacement (Stage 4) mean that a PDX line passaged indefinitely will eventually diverge substantially from the original patient tumor it was meant to represent — potentially losing or dramatically altering the abundance of the very driver mutations or biomarkers that made it a scientifically valuable model in the first place. A drug efficacy result generated in a heavily drifted, high-passage tumor risks being a study of an evolved laboratory artifact rather than a study relevant to the original patient's cancer.
Setting and enforcing a validated passage limit is the practical safeguard against this failure mode — a way of stating explicitly, with genomic evidence behind it, "this model remains a trustworthy representation of the founder patient tumor up to passage N, and should not be used beyond that point without re-validation."
A defensible passage-limit determination combines several lines of evidence gathered across the surveillance program:
• Genome-wide CNV drift score trajectory: passages are flagged once the drift score crosses a pre-defined threshold relative to the F1 baseline, distinguishing a "PASS" (low drift, safe to use) from a "WATCH" (rising drift, use with caution and increased monitoring) from a "FAIL" (drift exceeds acceptable limits, line should not be used without re-derivation from an earlier archived passage) • Key driver mutation and biomarker persistence: confirming that the specific mutations or expression markers a given research program depends on remain detectable at the passage being used • Human stromal purity: for studies depending on any human stromal component, confirming sufficient purity remains at the passage in question • Histological and growth-kinetic consistency: confirming the tumor still resembles its original histopathological grade and growth rate, as a phenotypic cross-check against the genomic data
Because drift rate varies substantially between individual PDX lines — some remain remarkably stable through 10+ passages while others drift measurably by passage 3-4 — passage limits should be set on a per-line basis from that line's own monitoring data, not applied as a single blanket rule across an entire biobank.
For a biobank distributing PDX models to external researchers (see the companion PDX Biobank Molecular Annotation simulation), the validated passage limit for each line is essential metadata that must travel with every distributed vial — alongside passage number itself, so that a receiving lab can immediately determine whether the material they have received falls within the line's validated window, and plan their own further passaging accordingly.
Many programs also recommend that end users performing their own extended passaging periodically re-verify key genomic features locally, since drift can in principle continue to accumulate even within a lab's own passaging after receipt of validated early-passage stock — quality control is an ongoing commitment, not a one-time certification.