Turnaround, Batching, and Data Deliverables in a Multi-Site CNS Biomarker Study
A Phase II neurodegeneration trial enrolls across fourteen sites. Blood draws are scheduled at baseline, week 12, and week 24. The biomarker plan reads well on paper. Eighteen months later, the statistician finds that week-24 samples from two sites run 15% higher than everything else, and that half the cohort had baseline and follow-up samples measured on different plates. The assay worked. The study design around it did not.
Key takeaways
- In longitudinal studies of blood-based CNS biomarkers, pre-analytical drift between sites and plate-to-plate variation can look exactly like disease progression or drug effect.
- Vesicle-based readouts add their own storage and handling sensitivities on top of those known for soluble plasma analytes.
- The single most useful batching rule: keep every timepoint from a given participant on the same plate, and balance arms and sites across plates.
- Fast turnaround and tight comparability pull in opposite directions. Decide which interim reads matter before the first sample ships.
- A usable data deliverable includes QC flags, procedural controls, plate maps and bridging data, not just a column of concentrations.
Why logistics decide whether a longitudinal CNS signal survives
What determines data quality in a multi-site CNS biomarker study? Three operational choices, made before enrollment: a single pre-analytical protocol enforced at every site, a batch design that keeps each participant’s timepoints on one plate with bridging controls between runs, and a deliverable specification that ships QC flags and plate metadata alongside results.
Most discussion of clinical trial biomarkers focuses on the assay: sensitivity, specificity, dynamic range. Those properties matter, but they are measured under controlled conditions in one laboratory. A multi-site trial is a different environment. Samples are drawn by dozens of staff, sit on benches for variable times, travel through couriers and freezers, and reach the testing lab months apart.
The EV field has been slow to confront this. A blood EV working group convened by the International Society for Extracellular Vesicles noted that most if not all EV studies to that point had been single-centre, which limits how far their findings translate to clinical use [1]. The same report stressed that blood is a hostile matrix for vesicle work: high concentrations of soluble proteins, lipoproteins and residual cells all interfere with concentration, isolation and detection [1].
For a sponsor using neuron-derived extracellular vesicles (NDEVs) as a lumbar puncture alternative, the stakes are specific. The whole case for blood-based sampling over CSF is that it allows more timepoints, more participants and less burden. That case collapses if the extra timepoints carry more technical noise than biological signal. CSF biomarker limitations are well known; a blood readout has to beat them in practice, not just in principle.
The sections below follow a sample from the draw to the data file and flag the decisions that protect, or erode, the signal along the way.
Figure 1. The sample journey in a multi-site CNS biomarker study, with the main sources of technical variation at each stage.
| Stage | What happens | Main variation risk | Control |
|---|---|---|---|
| 1. Draw | Venipuncture at site | Tube type, fasting state, time of day | Single tube type and draw window in protocol |
| 2. Processing | Centrifugation, aliquoting | Delay to spin, spin settings, residual platelets | Fixed time-to-spin limit; documented spin protocol |
| 3. Storage at site | Freezing, interim storage | Freezer temperature, delay to freezing | -80°C only; time-to-freeze logged |
| 4. Shipping | Courier to central lab | Thaw events, temperature excursions | Dry ice with temperature loggers |
| 5. Batching | Assignment to plates and runs | Plate, reagent lot and operator effects | Participant-on-one-plate rule; bridging samples |
| 6. Analysis | Vesicle isolation and measurement | Run-to-run drift | Procedural and reference controls on every plate |
| 7. Delivery | Data transfer to sponsor | Missing metadata, ambiguous flags | Agreed data dictionary and transfer spec |
Pre-analytical harmonization across sites
Pre-analytical variation covers everything that happens to a sample before it reaches the instrument. For soluble plasma biomarkers in Alzheimer’s disease, it has been studied in unusual depth, and the results are instructive for any blood-based CNS program.
The Standardization of Alzheimer’s Blood Biomarkers working group tested common variations in collection and handling, including tube type, delayed centrifugation of EDTA plasma, centrifugation temperature, aliquot volume, delayed storage and freeze-thaw cycles, and used the results to build a standardized plasma handling procedure [3]. A 2025 follow-up from the Global Biomarker Standardization Consortium extended this work. Collection tube type alone shifted some measurements by more than 10%. NfL and GFAP rose by more than 10% when samples were held at room temperature or -20°C, and amyloid-beta peptides were the most sensitive analytes overall [4]. Phosphorylated tau isoforms, by contrast, were stable across most of the tested conditions [4].
Two lessons follow. First, plasma NfL variability is partly a handling problem, not only a biology problem. Second, analytes differ in fragility, so a protocol written for the most stable marker in a panel will under-protect the rest. The Alzheimer’s Association summary of the consensus protocol caps samples at two freeze-thaw cycles [4]. That is a sensible floor for any multi-analyte study.
Why vesicle readouts need extra care
Extracellular vesicles add sensitivities that soluble analytes do not have. In a systematic study of storage conditions, -80°C storage reduced EV concentration and sample purity in a time-dependent manner and increased particle size and size variability. Flow cytometry data also suggested that vesicles fuse during storage, producing artefactual particles [5]. None of the eight storage strategies tested prevented these effects. The authors recommend storing the biological matrix rather than isolated vesicles when storage is unavoidable [5].
For a multi-site study, this has a practical implication. Sites should freeze plasma, not attempt any vesicle processing locally, and the central lab should isolate vesicles from all samples under one protocol. It also argues for recording storage duration per sample, because a baseline sample stored for 18 months and a follow-up stored for three are not equivalent even when both stayed at -80°C.
Residual platelets are the other EV-specific concern. Platelets left in plasma after a weak or delayed spin can release vesicles ex vivo, diluting the minor neuronal fraction with non-neuronal material [1]. A documented, double-spin protocol applied identically at every site is cheap insurance. MISEV2023, the current ISEV reporting framework, sets out the pre-analytical and characterization details that studies of body-fluid EVs should report, and it is a reasonable template for the site manual [2].
Batch design: keep each participant on one plate
Batch effects arise when measurements depend on laboratory conditions, reagent lots or personnel rather than biology. They become a serious problem when they correlate with the outcome of interest, because then they produce wrong conclusions rather than just noise [6]. In a longitudinal neurodegeneration trial, the outcome of interest is change over time. If timepoints are measured in different batches, change over time and batch are confounded by design.
The fix is simple to state. Every sample from a given participant goes on the same plate, measured in the same run. Within each plate, treatment arms, sites and disease-severity strata are balanced, and sample positions are randomized so that edge effects do not line up with any group. If a participant’s timepoints cannot fit on one plate, they at least share a run and reagent lot.
Figure 2. Two ways to batch the same 40-participant, three-timepoint study.
| Design choice | Batch by timepoint (avoid) | Batch by participant (prefer) |
|---|---|---|
| Plate 1 contents | All baseline samples | All three timepoints for participants 1–13, arms balanced |
| Plate 2 contents | All week-12 samples | All three timepoints for participants 14–27, arms balanced |
| Plate 3 contents | All week-24 samples | All three timepoints for participants 28–40, arms balanced |
| Is plate effect confounded with time? | Yes, completely | No |
| Is plate effect confounded with arm? | No | No |
| Can plate effect be estimated? | Only from bridging samples | From bridging samples and within-plate structure |
Bridging and reference samples
Even a well-balanced design needs a way to measure drift between runs. Two kinds of control do this. A pooled reference plasma, aliquoted at study start and included on every plate, tracks run-to-run shifts in the whole workflow, isolation included. Bridging samples, a small set of study samples re-measured across consecutive runs, show whether the shift is uniform across the concentration range or worse at the low end.
Statistical correction after the fact is possible, and tutorials for proteomic data set out a full sequence of assessment, normalization, correction and quality checks on the adjustment itself [7]. Our earlier piece on normalization and batch effects in clinical RNA-seq covers that analytical side. But correction relies on the design. When plate and time are fully confounded, no model can separate them.
Turnaround as a design trade-off
Sponsors want results quickly. Interim analyses, dose-escalation decisions and go/no-go reviews all depend on biomarker data arriving on schedule. Comparability, on the other hand, favors waiting: the most comparable dataset is the one measured in the fewest runs, with the fewest reagent lots, as close together in time as possible. Every study sits somewhere between these poles, and the right point depends on how the biomarker will be used.
Three broad models cover most programs.
End-of-study batch. All samples are held at -80°C and measured together once the last participant completes. This gives the cleanest comparability but no interim visibility. Because the vesicle storage effects described above are time-dependent [5], a long study also leaves early samples stored far longer than late ones.
Completed-participant batches. Samples are measured once a set of participants has finished all timepoints, with each set on its own plates. This preserves the participant-on-one-plate rule, gives rolling visibility, and suits most pharmacodynamic monitoring in neurodegeneration trials. It needs strong bridging controls, because the dataset now spans many runs.
Rolling measurement by timepoint. Samples are measured as they arrive. This is the fastest model and the riskiest. It suits safety signals or dose decisions that need a quick read on a single timepoint, and should be paired with a plan to re-measure the samples that feed the primary longitudinal analysis in participant-balanced batches later. That plan only works if enough plasma volume was banked to allow it, which is a protocol-stage decision.
Whichever model is chosen, write it into the lab manual alongside rerun criteria: which QC failures trigger a repeat, how much volume a repeat consumes, and what happens when a sample has none left. Reruns in vesicle work are not free, because isolation consumes plasma that cannot be recovered.
What a usable data deliverable contains
A data file with one concentration per sample is easy to produce and hard to use. A biostatistician analyzing a longitudinal neurodegenerative biomarker endpoint needs to know which values to trust, which to down-weight and which batch each came from. The deliverable should be specified in the lab contract, not negotiated at database lock.
Method validation work in neurology biomarkers makes a related point: the intended use of a method has to be defined before validation begins, because it determines which parameters matter [8]. The same logic applies to deliverables. A file built for exploratory discovery looks different from one feeding a prespecified pharmacodynamic endpoint.
Vesicle-specific quality evidence
For NDEV readouts, the deliverable should also show that the isolation worked on each run. Procedural controls are central here. In the published validation of the ExoSORT™ neuron-derived EV isolation workflow, neuronal markers in the isolated fraction were 35 times higher than in material pulled down with a non-specific IgG procedural control [9]. The same study reported that variability between two trained operators was lower than variability between donors, and that repeated analysis of clinical samples stayed below 20% CV [9]. Reporting the procedural-control result for each plate lets the sponsor confirm that enrichment held across the whole study, not just in the validation set.
This matters because no immunoaffinity method captures a perfectly pure population. A 2026 multi-institution study validating immunoaffinity enrichment of nervous system, alveolar and hepatocyte vesicles from human plasma states plainly that the targeted surface markers do not by themselves prove tissue of origin, and that capture likely enriches a subset of each tissue’s vesicles [10]. Per-run enrichment evidence is what turns that caveat into a measured quantity.
Figure 3. Minimum contents of a data deliverable for a multi-site NDEV biomarker study.
| Component | Contents | Why the sponsor needs it |
|---|---|---|
| Results file | Sample ID, participant, visit, analyte, value, units, LLOQ/ULOQ | Primary analysis input |
| Sample QC flags | Hemolysis, below-LLOQ, failed replicate, protocol deviation | Decide inclusion before unblinding |
| Run metadata | Plate ID, run date, reagent lots, operator code | Model or check batch effects |
| Plate maps | Well position for every sample and control | Detect edge or position effects |
| Procedural controls | Per-plate IgG control and enrichment readout | Confirm vesicle isolation held on each run |
| Reference and bridging data | Pooled reference values; bridging sample values across runs | Quantify and, if needed, correct drift |
| Pre-analytical record | Time to spin, time to freeze, storage duration, thaw count | Covariates for sensitivity analyses |
| Data dictionary | Variable definitions, formats, controlled terms | Map cleanly to the sponsor’s data standard |
Agree the transfer format early. Most sponsors map laboratory data to CDISC standards for submission, so variable names and controlled terminology in the lab file should be chosen with that mapping in mind. Agree also on the audit trail: which file versions exist, what changed between them, and who approved each release.
What to settle before first patient in
Most of the problems described above cannot be fixed after enrollment. They can be prevented cheaply at protocol stage. Before the first sample is drawn, the sponsor and the testing lab should have agreed on the following:
- One site manual. Tube type, draw window, time-to-spin limit, spin protocol, aliquot volume, freezer temperature and a two-cycle freeze-thaw cap, with a log for each.
- Plasma volume budget. Enough banked volume per timepoint for the planned panel, one rerun and any re-measurement planned under a rolling turnaround model.
- Shipping plan. Dry ice, temperature loggers and a rule for samples that arrive with an excursion.
- Batching model. End-of-study, completed-participant or rolling, chosen against the interim decisions the biomarker must support.
- Plate design rules. Participant-on-one-plate, balanced arms and sites, randomized positions, and fixed positions for reference and bridging controls.
- Acceptance and rerun criteria. Written QC thresholds, rerun triggers and handling of exhausted samples.
- Deliverable specification. File contents, data dictionary, transfer format, delivery schedule and version control.
None of this is exotic. It is the same discipline that made plasma p-tau measurements comparable across cohorts. What changes with neuron-derived vesicle readouts is that the matrix is more sensitive and the analytes include proteins such as TDP-43 and alpha-synuclein that have no established handling literature of their own yet.
Where the field is heading
As blood-based CNS biomarkers move from exploratory endpoints toward patient stratification and pharmacodynamic decision-making in ALS, Parkinson’s, Alzheimer’s and FTD trials, the operational layer will face the same scrutiny the assays already do. Expect pre-analytical studies built specifically for vesicle analytes, shared reference materials that let labs compare runs across studies, and sponsors writing batching and deliverable requirements into requests for proposals as standard. Programs that design for comparability from the first draw will be the ones whose longitudinal data hold up when it matters.
References
- Clayton A, Boilard E, Buzas EI, et al. Considerations towards a roadmap for collection, handling and storage of blood extracellular vesicles. J Extracell Vesicles. 2019;8(1):1647027. https://doi.org/10.1080/20013078.2019.1647027
- Welsh JA, Goberdhan DCI, O’Driscoll L, et al. Minimal information for studies of extracellular vesicles (MISEV2023): From basic to advanced approaches. J Extracell Vesicles. 2024;13(2):e12404. https://doi.org/10.1002/jev2.12404
- Verberk IMW, Misdorp EO, Koelewijn J, et al. Characterization of pre-analytical sample handling effects on a panel of Alzheimer’s disease-related blood-based biomarkers: Results from the Standardization of Alzheimer’s Blood Biomarkers (SABB) working group. Alzheimers Dement. 2022;18(8):1484-1497. https://doi.org/10.1002/alz.12510
- Verberk IMW, Gouda M, Antwi-Berko D, et al. Evidence-based standardized sample handling protocol for accurate blood-based Alzheimer’s disease biomarker measurement: Results and consensus of the Global Biomarker Standardization Consortium. Alzheimers Dement. 2025;21(10):e70752. https://doi.org/10.1002/alz.70752
- Gelibter S, Marostica G, Mandelli A, et al. The impact of storage on extracellular vesicles: A systematic study. J Extracell Vesicles. 2022;11(2):e12162. https://doi.org/10.1002/jev2.12162
- Leek JT, Scharpf RB, Bravo HC, et al. Tackling the widespread and critical impact of batch effects in high-throughput data. Nat Rev Genet. 2010;11(10):733-739. https://doi.org/10.1038/nrg2825
- Čuklina J, Lee CH, Williams EG, et al. Diagnostics and correction of batch effects in large-scale proteomic studies: a tutorial. Mol Syst Biol. 2021;17(8):e10240. https://doi.org/10.15252/msb.202110240
- Andreasson U, Perret-Liaudet A, van Waalwijk van Doorn LJC, et al. A practical guide to immunoassay method validation. Front Neurol. 2015;6:179. https://doi.org/10.3389/fneur.2015.00179
- Eitan E, Thornton-Wells T, Elgart K, et al. Synaptic proteins in neuron-derived extracellular vesicles as biomarkers for Alzheimer’s disease: novel methodology and clinical proof of concept. Extracell Vesicles Circ Nucleic Acids. 2023;4(1):133-150. https://doi.org/10.20517/evcna.2023.13
- Pierri B, Jackson GL, Gololobova O, et al. Immunoaffinity-based protocol to enrich nervous system cell-, lung alveolar cell-, and hepatocyte-derived extracellular vesicles from human plasma. J Extracell Biol. 2026;5(7):e70171. https://doi.org/10.1002/jex2.70171
Leave a Reply