Submitted:
24 October 2026
Posted:
26 October 2026
You are already at the latest version
Abstract
High-dimensional bone marrow immunomics increasingly informs the biological characterisation of multiple myeloma, yet the robustness of aspirate-derived immune phenotypes to peripheral-blood admixture, differential cellular recovery and loss of tissue architecture remains incompletely established. Existing haemodilution indices assess specimen quality but have not been comprehensively evaluated against multidimensional immune-state classification. We propose MARROW-ANCHOR, a spatially anchored validation protocol integrating patient-matched early- and later-pull marrow aspirates, peripheral blood, controlled autologous admixture, multiparameter flow cytometry, selected single-cell transcriptomic and protein profiling, and trephine-based spatial immunophenotyping. The framework distinguishes experimentally introduced blood from unmeasured pre-existing admixture and separates compositional changes from recovery-associated and within-population variation. Prespecified sham controls, participant-level replication, established marrow-quality indices and computational mixture modelling support assessment of classification stability, while independent spatial and functional measurements provide complementary biological corroboration. The primary endpoint quantifies experimentally induced classification reversal; biological misclassification requires additional independent evidence. Explicit validation criteria address technical reproducibility, analytical identifiability, model calibration, overcorrection and external generalisability. Stable classifications, contradictory measurements and failure of adjustment are retained as informative outcomes. MARROW-ANCHOR provides a falsifiable methodological framework for determining when specimen-associated uncertainty warrants qualification of marrow-derived immune interpretations before their translation into precision-medicine applications.
Keywords:
multiple myeloma
; bone marrow haemodilution
; immunomics
; spatial immunophenotyping
; immune-state classification
; single-cell multi-omics
Introduction
The Specimen Precedes the Phenotype
The biological interpretation of an immune signature rests upon an assumption that is rarely interrogated with the same precision as the molecular assay itself: the analysed cellular population must adequately represent the compartment to which its biological meaning is assigned. In multiple myeloma, this distinction is consequential. Bone marrow is not merely the principal anatomical reservoir of malignant plasma cells. It is a spatially organised, immunologically heterogeneous tissue in which malignant, lymphoid, myeloid, endothelial and stromal populations interact within locally differentiated microenvironments. Although contemporary immunomics can resolve these populations at considerable molecular depth, increasing analytical resolution does not independently establish the representativeness of the specimen entering the assay.
Bone marrow aspiration transforms an anatomically organised tissue into a recoverable cellular suspension. This process introduces several potential sources of measurement variation: peripheral-blood admixture, aspiration-order effects, differential recovery of marrow-associated populations, regional differences in tumour infiltration and subsequent processing-associated cellular loss. These mechanisms are neither synonymous nor necessarily independent. Their combined influence becomes particularly important when an aspirate-derived immune signature is interpreted as evidence of marrow-specific dysfunction, cellular exhaustion or a therapeutically relevant biological state.
Pilcher et al. (2026) generated a single-cell immune atlas comprising 1,397,272 marrow cells from 337 participants with newly diagnosed multiple myeloma. Their study demonstrated relationships between immune composition, cytogenetic characteristics and clinical outcomes, establishing the substantial translational potential of high-dimensional marrow immunomics. Nevertheless, the generation of a biologically informative immune atlas does not establish that every aspirate-derived signature is invariant to sampling conditions. The unresolved question is whether a particular immune interpretation survives controlled changes in specimen composition, rather than whether marrow aspirates are generally suitable for immunomics.
Established Haemodilution Assessment and the Remaining Inferential Boundary
Haemodilution is a recognised limitation of marrow-based assessment in plasma-cell disorders, particularly when specimen inadequacy compromises sensitive flow-cytometric measurable residual disease detection. Flores-Montero et al. (2017) established a standardised next-generation flow approach for myeloma MRD assessment, providing an important methodological foundation for the identification of relevant cellular populations.
Óskarsson et al. (2023) subsequently investigated 351 marrow aspirates from 219 participants with plasma-cell disorders and developed a bone marrow quality index based on myeloid precursors and nucleated erythroid cells. These populations distinguished first- from second-pull aspirates, and the combined BMQI achieved an area under the receiver-operating-characteristic curve of 0.88 for this distinction. Their work also incorporated experimental and computational assessments of haemodilution. Consequently, neither controlled peripheral-blood admixture nor objective marrow-quality assessment can legitimately be claimed as originating from the present proposal.
A subsequent review by Badelita et al. (2026) synthesised the methodological difficulties surrounding haemodilution in next-generation flow-cytometric MRD assessment, including limitations of existing markers, specimen-quality indices and standardisation strategies. This literature demonstrates that marrow-quality assessment is an established research field. It does not, however, provide a comprehensive validation of how haemodilution influences multidimensional immune-state classification across transcriptomic, protein, spatial and functional measurements.
The distinction is important. A quality index may identify a relatively haemodiluted aspirate without establishing whether an individual immune signature remains stable, becomes compositionally distorted or crosses a predefined analytical boundary. Conversely, a relatively haemodiluted specimen may retain adequate information for certain immune endpoints. Specimen adequacy and interpretative validity are related properties, but neither can be assumed to determine the other universally.
Blood–Marrow Concordance as Competing Evidence
The proposition that blood admixture necessarily compromises every marrow immune interpretation is contradicted by evidence of cross-compartmental immune concordance. Zabaleta et al. (2026) evaluated immune biomarkers of infection risk in 1,786 patients with multiple myeloma across disease stages and treatment settings. Their study identified immune-population associations with infection and reported significant correlations between corresponding populations detectable in marrow and peripheral blood.
These findings do not establish that blood and marrow are biologically interchangeable. They demonstrate that some clinically relevant immune information may be shared across compartments. A rigorous investigation must therefore permit two competing outcomes: measurable admixture may destabilise marrow-specific immune interpretations, or selected immune features may remain sufficiently concordant across compartments to withstand substantial admixture.
The present protocol does not presume that every circulating-cell contribution represents an error. Instead, it asks which immune measurements are compartment-sensitive, under what conditions that sensitivity becomes analytically consequential, and whether specimen-quality information improves their interpretation.
Composition, Phenotype and Function Are Not Interchangeable
High-dimensional immune profiling generates measurements at several distinct levels: cellular abundance, population-specific molecular expression, aggregate specimen-level signatures and inferred immune function. A change at one level cannot automatically establish a change at another.
The addition of circulating lymphocytes may alter an aggregate cytotoxicity-associated transcriptomic score without changing the expression programme of an individual marrow-associated T cell. Similarly, selective cellular recovery may shift apparent immune-subset frequencies without corresponding biological depletion or expansion. Because immune-cell proportions share a constrained denominator, an apparent increase in one population may arise through the relative reduction of another.
The distinction is especially consequential for exhaustion-associated phenotypes. Potdar et al. (2025) reported enhanced inflammatory cytokine production and proliferative capacity in selected T-cell populations from patients with plasma-cell disorders, including populations displaying markers previously associated with exhaustion. Shasha et al. (2025), examining longitudinal single-cell, immunophenotypic and functional measurements, found no convincing evidence of terminally exhausted, tumour-experienced T cells in their investigated cohort of patients with newly diagnosed myeloma.
These findings do not exclude immune dysfunction across all stages or treatment settings. They establish that checkpoint expression and transcriptional resemblance to previously characterised exhaustion programmes are insufficient, in isolation, to demonstrate functional impairment.
MARROW-ANCHOR therefore distinguishes an altered molecular signature from a change in cellular function and treats functional claims as requiring independent population-appropriate corroboration.
The Spatial Reference Problem
The dissociation of marrow tissue also removes information about cellular localisation and anatomical relationships. Yip et al. (2025) used spatial transcriptomics in human marrow trephine specimens to demonstrate distinct myeloma-associated microenvironments with differing transcriptional profiles and cellular composition. Muiños-Lopez et al. (2025) further characterised the architecture of the myeloma marrow environment, while Koops et al. (2026) described a topographically organised tumour-associated microenvironment emerging during myeloma evolution.
Such findings challenge an assumption that a cellular suspension necessarily preserves the relationships present within intact tissue. Nevertheless, a trephine is not an infallible biological reference. It represents a restricted anatomical region and is subject to sampling heterogeneity, fixation and decalcification effects, section-selection bias and image-analysis limitations.
Accordingly, spatial immunophenotyping is incorporated as an independent anatomical comparator rather than an absolute standard against which every aspirate discrepancy is classified as error. Concordance is informative; discordance requires investigation of sampling geography, measurement scales and assay-specific limitations.
Scientific Objective and Proposed Methodological Contribution
MARROW-ANCHOR is designed to quantify the susceptibility of prespecified marrow-derived immune interpretations to known experimental blood admixture and associated variation in cellular recovery. It further evaluates whether existing specimen-quality measures or independently validated analytical adjustments improve the stability and biological agreement of those interpretations.
Three hypotheses are tested. First, controlled autologous blood admixture may alter immune composition and aggregate signatures in directions predictable from the source populations. Secondly, some sample-level immune classifications may change despite comparatively stable within-population signatures. Thirdly, the relationship between admixture and interpretative stability may differ according to the immune endpoint and its independent anatomical or functional corroboration.
The hypotheses are deliberately falsifiable. Classifications may remain stable across the tested admixture range; existing specimen-quality assessment may perform as well as more complex adjustment; or discordance may primarily reflect spatial heterogeneity or differential recovery rather than peripheral-blood admixture. None of these outcomes should be excluded through retrospective alteration of the study's analytical definitions.
The complete five-stage validation architecture is presented in Table 1. Its proposed contribution lies not in an individual assay, index or computational technique, but in the explicit integration of known specimen perturbation, analytical decomposition, classification-stability testing, orthogonal corroboration and independent falsification.
Materials and Methods
Study Design
MARROW-ANCHOR is proposed as a prospective, paired-specimen, experimentally perturbed methodological validation study supported by an initial computational feasibility phase. Its sequential components comprise computational assessment, experimental admixture, multimodal immune profiling, independent biological corroboration and external validation.
The framework is not presented as a clinically established assay. Experimental performance, reproducibility, correction accuracy and clinical usefulness remain to be determined through implementation.
The core experimental question concerns the effect of additional autologous PB cells upon measurements from an existing marrow aspirate. This is distinct from reconstructing an unobservable, completely blood-free marrow compartment.
Study Population and Eligibility
The principal cohort will comprise adults with newly diagnosed, previously untreated multiple myeloma undergoing clinically indicated marrow examination. Diagnosis and disease classification should follow recognised contemporary criteria, and the relevant clinical characteristics should be recorded before research analysis.
Prior treatment is excluded from the initial cohort to reduce confounding associated with treatment-dependent changes in marrow precursors, immune activation and cellular recovery. This restriction does not eliminate variation attributable to marrow plasma-cell burden, fibrosis, age, inflammation or other baseline biological factors. Treatment-exposed patients should be evaluated subsequently as a separate validation population.
Recorded variables should include demographic characteristics, available disease stage, marrow plasma-cell burden, relevant cytogenetic information, blood counts, active infection or inflammatory conditions, corticosteroid exposure and other clinically important determinants of immune status. These factors should be prespecified as possible modifiers or confounders rather than incorporated selectively after inspection of the results.
All prospective specimen collection requires appropriate institutional approval, informed consent and biological-material governance. Research sampling must not compromise diagnostic morphology, flow cytometry, cytogenetics or other clinically required assessments. An additional invasive sampling procedure should not be performed solely for research completeness without independent clinical and ethical justification.
Cohort Structure and Sample-Size Determination
The study will incorporate a feasibility cohort, an analytical-development cohort and an independent validation cohort.
Approximately 24 participants are proposed for feasibility assessment of specimen yield, technical variability, assay completion and initial endpoint distribution. A provisional analytical-development target of approximately 80 independent participants and an external validation target of approximately 40 participants may be considered. These figures are planning assumptions and must not be represented as established power calculations.
For illustration, if the observed primary classification-reversal proportion were 20% in 80 independent participants, the corresponding descriptive 95% Wilson interval would be approximately 13%–30%. This degree of precision may support initial method characterisation but does not establish adequate power for small between-model differences, subgroup interactions or clinical outcome assessment.
The definitive recruitment target must be based upon feasibility-stage estimates, within-participant variability, anticipated specimen attrition and a prespecified minimum effect considered analytically meaningful. If recruitment or assay yield is insufficient, the study should be reported as feasibility or exploratory validation rather than as definitive establishment of the method.
A subset of participants may undergo advanced single-cell, spatial and functional analyses. Selection must preserve variation in disease burden and specimen adequacy rather than favour specimens displaying particularly pronounced admixture-associated changes.
Specimen Acquisition and Allocation
The principal specimen categories comprise early-pull marrow, later-pull marrow where clinically appropriate, concurrently collected PB and anatomically corresponding trephine tissue when available.
Early-pull marrow serves as the operational baseline. It must not be described as uncontaminated marrow. Later-pull marrow provides a natural acquisition-related contrast, although its composition may reflect selective tissue recovery, aspiration geography and blood admixture simultaneously.
Every specimen requires a stable pseudonymised identifier linked to collection site, sequence, volume, anticoagulant, transport interval, processing conditions, viable-cell concentration and relevant clinical information. Anatomical correspondence between aspirate and trephine sites must be recorded explicitly.
Material necessary for conventional marrow-quality assessment should be retained before erythroid depletion, selective leukocyte enrichment or other processing that alters the populations used in established quality indices.
The required analytical allocation and its associated inferential limitations are defined in Table 2.
Marrow-Quality Assessment
The BMQI described by Óskarsson et al. (2023) will serve as an established comparator:
where (M) represents the percentage of myeloid precursors and (E) the percentage of nucleated erythroid cells, calculated according to the published definition of normal nucleated cellularity after exclusion of tumour plasma cells.
[\mathrm{BMQI}=0.37M+0.14E]
The coefficients, population definitions and denominator must be retained when applying the published index. Measurements from erythroid-depleted, selectively enriched or otherwise incompatible material must not be assigned the interpretation of the original BMQI.
The BMQI estimates relative marrow sample adequacy through associated precursor populations. It is not a direct measurement of the absolute circulating-blood fraction within the aspirate. Its performance may also vary in settings where treatment or disease-associated factors alter the populations upon which it depends.
Additional quality descriptors should include marrow spicules where assessable, morphological cellularity, plasma-cell representation, viable nucleated-cell yield and processing history. These descriptors should be treated as complementary measurements rather than assumed to provide equivalent estimates of haemodilution.
Controlled Autologous Admixture and Experimental Exposure
Matched early-pull marrow and PB suspensions will be prepared using a common validated handling procedure. Viable nucleated-cell concentrations must be determined independently before mixture preparation.
The proposed initial experimental series comprises nominal PB contributions of 0%, 10%, 25%, 50% and 75% of the total viable nucleated-cell input. These levels define a controlled perturbation series, not the expected prevalence or magnitude of haemodilution in routine diagnostic aspirates.
The 25% admixture level is provisionally designated as the primary analytical contrast. Its selection is methodological rather than evidence-based as a clinical threshold; it must be prespecified before the definitive validation analysis and must not be altered on the basis of which level produces the strongest result.
For a mixture containing (N_{\mathrm{PB}}) PB-derived and (N_{\mathrm{BM}}) BM-derived viable nucleated cells:
where (f) denotes the experimentally introduced PB contribution.
[f=\frac{N_{\mathrm{PB}}}
{N_{\mathrm{PB}}+N_{\mathrm{BM}}}]
{N_{\mathrm{PB}}+N_{\mathrm{BM}}}]
Preparation must use measured cell concentrations rather than assuming that equal fluid volumes contain equal numbers of viable nucleated cells. The 0% condition represents the unadmixed operational baseline, which may already contain circulating cells.
A sham-handled aliquot must undergo equivalent manipulation and processing without added PB cells. Technical replication should be incorporated at prespecified conditions, with mixture preparation and assay acquisition balanced or randomised to prevent systematic association between exposure level and handling order.
Cell concentrations, viability and relevant population recovery should be measured both before and after processing. Input proportions may differ from final assayed proportions if particular populations undergo preferential loss.
Artificial admixture isolates the effect of adding circulating cells under specified ex vivo conditions. It does not fully recreate aspiration-induced tissue disruption, preferential recovery or regional sampling.
Identifiability of Baseline Contamination
Absolute pre-existing PB contamination of an aspirate cannot generally be inferred uniquely from its observed composition. Distinct combinations of native marrow cellularity, initial contamination and differential recovery may yield observationally similar distributions.
Accordingly, MARROW-ANCHOR does not designate (f=0) as a blood-free biological reference or claim that successful prediction of experimentally added cells reconstructs authentic pre-aspiration marrow composition.
The principal identifiable exposure is the additional PB fraction introduced under controlled experimental conditions. Any attempt to estimate absolute baseline contamination requires separate identifying assumptions, independent supporting measurements and explicit uncertainty reporting.
Multiparameter Immune Profiling
A validated flow-cytometric panel should distinguish major T-cell, B-cell, NK-cell and myeloid populations, together with selected differentiation or activation-associated phenotypes. Gao et al. (2023) provide a directly relevant Phenomics Protocol describing standardised immunophenotyping, antibody optimisation, instrument characterisation and acquisition quality control.
A provisional marker architecture includes CD45, CD3, CD4, CD8, CD19, CD14, CD16, CD56, CD27, CD45RA, CCR7 and HLA-DR, with appropriate plasma-cell identification or exclusion markers, including CD38 and CD138 where indicated. This is a proposed marker architecture and not a validated ready-to-use antibody panel.
In particular, CD56 must not independently determine NK-cell identity because malignant plasma cells may also express CD56. Population assignment requires suitable lineage exclusion and corroborative phenotypic information.
Antibody clones, fluorochromes, titration, staining procedures, compensation or spectral unmixing, detector settings, viability exclusion and doublet discrimination must be finalised through platform-specific validation.
Both relative cellular frequencies and appropriate recovery measurements should be retained. Frequencies must be reported using explicit denominators, such as viable nucleated cells, CD45-positive cells or defined parent populations.
The denominator itself may influence the apparent admixture response. Prespecified denominator sensitivity analyses are therefore necessary.
Single-Cell Transcriptomic and Protein Profiling
A selected participant subset may undergo scRNA-seq with matched surface-protein characterisation using an established approach such as CITE-seq (Stoeckius et al. 2017). Sample multiplexing may employ validated cell-hashing methods (Stoeckius et al. 2018), provided sample assignment and demultiplexing performance are documented.
The molecular subset should include appropriately paired baseline marrow, PB, sham-handled material and selected admixture conditions. Its composition must be determined prospectively rather than through preferential selection of samples supporting the proposed hypotheses.
QC should assess viable input, captured-cell counts, sequencing complexity, mitochondrial transcript fraction, ambient RNA contamination, potential doublets and representation of predefined immune populations.
Established approaches such as SoupX (Young and Behjati 2020) and DoubletFinder (McGinnis et al. 2019) may be applied when appropriate, with parameters and sensitivity analyses documented.
Computational integration can reduce platform or batch effects but may also attenuate the experimental differences under investigation. Therefore, corrected representations must not replace examination of the original data without explicit justification.
Differential within-population expression should be evaluated using participant-aware methods. The multi-sample framework implemented in muscat (Crowell et al. 2020) provides a relevant precedent for distinguishing cellular observations from biological replication. Treating thousands of cells from one participant as independent study participants would yield misleading precision.
Public-Data Feasibility Assessment
Public datasets should be selected according to documented pairing, specimen type, treatment status, cellular composition, assay platform and availability of participant-level metadata.
GSE288020, associated with the immune-ageing investigation by Potdar et al. (2025), offers material relevant to immune-state interpretation. GSE290061 contains paired marrow and PB CAR-T-cell observations from a small cohort of three patients with treated, relapsed or refractory myeloma. The latter may support limited computational exploration of compartment-dependent immune features but is unsuitable as an independent validation cohort for untreated myeloma, general marrow composition or first- versus later-pull haemodilution.
Where access and metadata permit, the larger myeloma immune atlas of Pilcher et al. (2026) may provide an additional reference for immune annotation and population heterogeneity. Its controlled-access components and any restricted metadata must not be assumed freely available.
Public-data simulations can establish computational feasibility and identify sensitivity to known synthetic mixtures. They cannot substitute for physical admixture experiments or establish natural aspiration-induced haemodilution when specimen provenance is undocumented.
Spatial Immunophenotyping
Matched trephine sections should undergo histomorphological assessment and a validated multiplex immunophenotyping procedure capable of distinguishing plasma-cell-enriched regions and relevant immune populations.
The spatial approach should incorporate suitable plasma-cell and immune-lineage markers, with additional features selected according to the prespecified immune endpoints. Marker combinations must undergo FFPE-specific validation and appropriate pathological interpretation.
Fixation, decalcification and tissue-processing conditions require documentation. Campos-Dopazo et al. (2026) provide a methodological precedent for EDTA-based preparation of marrow specimens for spatial transcriptomics; compatibility with an individual platform must nonetheless be demonstrated.
Regions of interest should be defined prospectively where feasible, including plasma-cell-rich and relatively plasma-cell-sparse regions. Quantification may include cell densities, regional distributions, proximity measurements and within-core heterogeneity.
Independent review of representative segmentation and tissue annotations should be undertaken without knowledge of the corresponding aspirate classification wherever practicable.
Spatial density and aspirate cellular frequency are different measurement constructs. Comparison should therefore employ biologically interpretable features and uncertainty-aware analyses rather than direct numerical substitution between incompatible denominators.
Functional Immune Assessment
Selected T-cell populations may undergo controlled functional measurements, including stimulated cytokine production and proliferation, where adequate viable material and laboratory facilities are available.
Appropriate unstimulated controls, viability assessment and population-specific normalisation are required. An increase or decrease in total cytokine-producing events may arise through altered abundance of contributing populations without a corresponding change in their intrinsic responses.
Checkpoint-marker expression or a transcriptional exhaustion-associated score must not be treated as sufficient evidence of functional exhaustion. Functional measurements themselves remain conditional upon stimulation conditions, cell recovery and assay design.
Procedure
Stage I. Computational qualification
Step 1. Identify public immune-profiling datasets with sufficient participant, compartment and processing metadata for the intended analysis. Document disease stage, treatment exposure and limitations in specimen comparability.
Step 2. Define mutually exclusive principal immune populations, the residual eligible-cell category and the denominators required for compositional analysis.
Step 3. Select a limited set of biologically interpretable immune signatures, recording their source, intended construct, required features and normalisation method.
Step 4. Specify the experimental classification rule before examining prospective admixture responses. Where no clinically validated category exists, identify the result explicitly as an analytical classification.
Step 5. Construct synthetic mixtures from appropriately paired BM and PB profiles. Preserve source identities and known experimental proportions.
Step 6. Perform repeated subsampling to assess computational variability, with random seeds and sampling procedures documented.
Step 7. Evaluate the consequences of idealised proportional mixing and separately specified differential-recovery scenarios.
Step 8. Freeze the primary estimand, analytical procedures, development–validation allocation and planned sensitivity analyses before definitive prospective evaluation.
The computational stage qualifies the analytical workflow; it does not establish physical-assay performance or biological validity.
Stage II. Specimen acquisition and qualification
Step 9. Enrol eligible participants following appropriate approval and consent, recording prespecified clinical and pathological variables.
Step 10. Obtain clinically appropriate early-pull marrow, later-pull marrow where permissible, concurrent PB and corresponding trephine tissue.
Step 11. Record aspiration site, collection sequence, sample volume, needle repositioning, processing interval and relevant deviations.
Step 12. Determine viable nucleated-cell concentrations, morphology and cellular adequacy before selective processing.
Step 13. Apply BMQI-compatible assessment using the published denominator and population definitions. Record circumstances preventing valid calculation.
Step 14. Allocate available material according to Table 2, prioritising core analyses and documenting omissions.
Stage III. Admixture preparation and technical controls
Step 15. Prepare matched autologous BM and PB suspensions using validated handling procedures.
Step 16. Measure viable nucleated-cell concentrations independently and calculate the required cellular contributions for each mixture.
Step 17. Prepare the prespecified graded admixture series while maintaining comparable total viable-cell input where feasible.
Step 18. Prepare the sham-handled marrow control using equivalent handling and processing conditions.
Step 19. Balance or randomise preparation and acquisition order across conditions.
Step 20. Measure post-preparation viability and recovery, retaining both nominal input and observed output measurements.
Step 21. Perform technical replication at prespecified conditions and evaluate the variability of sham controls.
Step 22. Apply assay-specific QC rules determined during qualification, recording exclusions independently of the preferred biological outcome.
Stage IV. High-dimensional immune acquisition
Step 23. Verify instrument performance, staining controls and fixed acquisition settings.
Step 24. Acquire flow-cytometric data using a validated gating hierarchy and appropriate lineage exclusion.
Step 25. Record major immune populations, selected subpopulations, viable-cell recovery and relevant event counts.
Step 26. Evaluate population-specific measurement adequacy and avoid quantitative interpretation of insufficiently represented rare subsets.
Step 27. Preserve raw flow files, gating definitions, instrument QC, processing metadata and analysis versions.
Step 28. Select the prespecified molecular-profiling subset according to documented criteria.
Step 29. Perform single-cell preparation, sequencing and platform-specific QC using a validated laboratory SOP.
Step 30. Generate participant-level immune population and within-population signature estimates without pseudoreplication.
Stage V. Anatomical and functional comparison
Step 31. Prepare matched trephine sections using validated tissue processing.
Step 32. Identify plasma-cell-associated and immune-cell compartments through morphological and multiplex measurements.
Step 33. Assess spatial heterogeneity and segmentation performance, with blinded independent review where feasible.
Step 34. Perform selected immune-functional assays using appropriate population-specific controls.
Step 35. Compare aspirate-derived phenotypes with independently assessed anatomical and functional features, preserving the distinct measurement scales.
Stage VI. Compositional analysis
Step 36. Calculate the expected idealised mixture composition for each defined population:
where (b_k) and (c_k) are the baseline BM and PB proportions, respectively, of population (k), measured using compatible denominators.
[a_k(f)=(1-f)b_k+fc_k]
This represents the expected input composition under proportional mixing and does not account for differential recovery or unknown baseline contamination.
Step 37. Where independently measured recovery data permit, evaluate the relationship:
where (r_k(f)) denotes the relative recovery of population (k) under the relevant experimental condition.
[p_k(f)=\frac{a_k(f)r_k(f)}{\sum_{j=1}^{K}a_j(f)r_j(f)}]
Recovery parameters must not be derived from the target classification and subsequently presented as independent evidence that the classification has been corrected.
Step 38. Quantify distributional changes using a measure such as Hellinger distance:
[H(p,q)=\sqrt{\frac12\sum_{k=1}^{K}(\sqrt{p_k}-\sqrt{q_k})^2}]
This measure summarises differences between consistently defined cellular compositions but does not independently determine whether a biological interpretation is erroneous.
Step 39. For a linear aggregate immune signature, define:
where (p_k) is the proportion of population (k) and (\mu_k) its corresponding mean signature value.
[S=\sum_{k=1}^{K}p_k\mu_k]
For two experimental conditions, the exact symmetric decomposition is:
and
such that:
\sum_k(p_{k1}-p_{k0})\frac{\mu_{k1}+\mu_{k0}}{2}]
\sum_k(\mu_{k1}-\mu_{k0})\frac{p_{k1}+p_{k0}}{2}]
[\Delta S=\Delta S_{\mathrm{comp}}+\Delta S_{\mathrm{within}}]
The decomposition distinguishes the mathematical contributions of population weighting and changes in measured population-specific values. It does not independently establish the causal origin of either component.
The identity should not be transferred uncritically to nonlinear classifiers, latent embeddings or data-dependent scores whose definitions change between conditions.
Stage VII. Classification stability and analytical benchmarking
Step 40. Apply the frozen analytical classifier to sham-handled baseline marrow and the prespecified primary admixture condition.
Step 41. Estimate the participant-level frequency of classification reversal, retaining continuous scores, classification-boundary distances and measurement uncertainty.
Step 42. Evaluate secondary admixture conditions and dose–response behaviour using methods that respect within-participant dependence.
Step 43. Compare observed changes with predictions from source composition, with additional assessment of measured recovery and processing-associated variation.
Step 44. Benchmark unadjusted interpretation against BMQI-informed assessment and any independently developed mixture-aware adjustment.
Stage VIII. Independent validation and reporting
Step 45. Apply the frozen analytical pipeline to participants excluded from model development.
Step 46. Evaluate agreement with relevant spatial and functional observations without using validation outcomes to refit the model.
Step 47. Assess calibration, reproducibility, classification stability and potential overcorrection.
Step 48. Determine whether complex adjustment improves independent measurement beyond simpler comparators.
Step 49. Report prespecified endpoints, uncertainty, technical failures, missing observations and protocol deviations.
Step 50. Classify the final evidence according to analytical feasibility, analytical validity, biological corroboration and independent reproducibility. No clinical benefit should be inferred without appropriate outcome-based validation.
Primary estimand
The primary endpoint is the participant-level probability that a frozen analytical immune classification changes after 25% experimentally added autologous PB compared with sham-handled baseline marrow:
where (C_i) denotes the prespecified classification for participant (i).
[\theta=P{C_i(0.25)\neq C_i(\mathrm{sham})}]
This endpoint measures operational classification instability. It does not estimate the probability of true biological misclassification or clinical diagnostic error.
The primary analysis should report a point estimate and confidence interval. Participant identity is the independent biological unit, and multiple aliquots or mixture conditions must not be counted as independent participants.
Continuous signature changes, compositional distances, within-population differences, spatial concordance and functional corroboration constitute secondary endpoints.
Compositional Statistical Analysis
Because immune-population frequencies are constrained by a shared total, conventional interpretation of isolated percentage changes may be misleading. The Bayesian compositional approach scCODA, developed by Büttner et al. (2021), offers an established method for evaluating differential cellular abundance while accounting for compositional constraints and uncertainty.
Where appropriate to the study design, scCODA may be used as a prespecified secondary analytical comparator. Its estimates must remain distinct from the primary classification-reversal endpoint. Compositional associations do not directly establish absolute cellular expansion, depletion or functional change.
Population-specific transcriptomic comparisons should use participant-aware approaches, such as pseudobulk or suitable hierarchical models, consistent with the principles examined by Crowell et al. (2020).
A prespecified hierarchical testing approach or appropriate FDR adjustment should address multiplicity among secondary immune signatures. Exploratory analyses must be identified separately.
Missing observations should be classified by cause, including insufficient specimen material, viability failure, incomplete tissue availability, assay QC failure or missing clinical metadata. Because these events may be associated with disease burden or sampling difficulty, complete-case estimates should undergo sensitivity analysis where attrition is substantial.
Independent Model Evaluation
Model training, tuning and final validation must remain separate at the participant level. All samples, mixtures, sequencing libraries and derived observations belonging to one individual must remain in the same analytical partition.
A correction cannot be credited with improved spatial or functional validity if those same outcomes were used to select or optimise its parameters.
Prespecified comparisons should include unadjusted measurements, established specimen-quality-informed interpretation and the proposed experimental adjustment, where developed.
The standards of evidence and criteria capable of weakening or falsifying each claim are presented in Table 3.
Troubleshooting
The principal troubleshooting challenge is distinguishing technical inadequacy from scientifically informative disagreement. A result that contradicts the hypothesis must not be removed solely because it is inconvenient. Exclusion should follow prespecified QC criteria, and corrective actions should be documented before interpretation of the relevant primary endpoint wherever feasible.
The principal threats, corrective measures and remaining limitations are detailed in Table 4.
Reproducibility and Operational Qualification
The proposed framework requires a documented specimen metadata dictionary, validated laboratory SOPs, fixed gating hierarchies, appropriate technical controls, raw-data retention and version-controlled computational procedures.
Before experimental implementation, the laboratory must finalise antibody clones, fluorochrome allocations, reagent concentrations, processing intervals, instrument configurations, molecular-library specifications, assay-specific thresholds and acceptable technical-replicate variation.
Such details must be established empirically, not invented for manuscript completeness. Their absence from the present proposal limits claims of immediate cross-laboratory reproducibility.
Likewise, no model, assay or correction procedure should be described as validated before the corresponding qualification and independent evidence have been obtained.
Anticipated Results
The anticipated outputs concern analytical behaviour under defined experimental perturbations. No experimental results or performance estimates are claimed.
Controlled Admixture and Cellular Recovery
The first anticipated output is a participant-specific relationship between known PB inputs and measured immune-cell composition.
Under idealised proportional mixing, the distribution should reflect the measured source populations. Deviations may arise from differential cellular recovery, processing effects or limitations of the assumed population definitions.
Accurate documentation of input and recovered composition is essential because a technically imprecise admixture series cannot support reliable biological interpretation. A failure to recover the expected mixture should trigger investigation of the experimental process rather than automatic fitting of a more complex model.
Aggregate Immune-Signature Responses
Aggregate signatures may change through altered cellular composition even when within-population measurements remain comparatively stable.
Where observed changes are substantially accounted for by source-population weighting, the findings would favour a compositional explanation. Additional variation would require consideration of selective recovery, annotation instability, technical processing and intrinsic biological differences.
The analytical decomposition should not be interpreted as proving a particular cellular mechanism without independent corroboration.
Classification Stability
The primary analysis will determine whether experimental admixture causes a prespecified immune classification to change relative to the sham-handled baseline.
Possible outcomes include robust classification stability, reproducible composition-associated reversals or heterogeneous threshold sensitivity.
Classification reversal should be accompanied by continuous scores and uncertainty intervals. A minor perturbation near a boundary may produce a category change, whereas a larger continuous effect may occur without reversal.
Accordingly, the categorical endpoint cannot replace interpretation of the underlying measurements.
Anatomical and Functional Corroboration
Spatial immunophenotyping may demonstrate agreement with selected aspirate-derived immune features, regionally restricted discordance or insufficient correspondence for a particular biological construct.
A trephine-derived observation should not automatically override a cellular measurement; the relative strengths and limitations of the two modalities must be considered.
Functional assays may corroborate or contradict inferred immune states. Molecular resemblance to exhaustion, for example, may coexist with preserved functional responses under the tested conditions.
Such discrepancies are informative because they identify the limits of phenotype-based functional inference.
Correction Performance
A specimen-aware adjustment may improve interpretation, provide no additional value or introduce biological overcorrection.
A positive analytical result would require reproducible improvement in a prespecified endpoint, acceptable calibration and independent confirmation. Improved fit within the model-development dataset is insufficient.Conversely, stable immune classifications or failure of correction to outperform established quality assessment would constitute meaningful negative results.
No prediction of clinical treatment benefit is warranted from the proposed analytical study alone.
Discussion
The central problem is biological representativenessThe methodological challenge addressed by MARROW-ANCHOR is not a deficiency of contemporary immunomics technology. It is the potential disparity between the resolution of an assay and the representativeness of the biological material subjected to that assay. In multiple myeloma, aspiration can alter the cellular composition available for analysis before sequencing, immunophenotyping or computational classification begins.
A technically reproducible result may therefore remain conditional upon acquisition and processing. The appropriate scientific response is not to dismiss marrow aspirates but to determine which interpretations are stable, which are sensitive to sampling and whether that sensitivity affects their intended biological meaning.Novelty must be demonstrated against established methods
The foundations of the proposed study already exist. Objective marrow-quality assessment, controlled admixture, high-dimensional flow cytometry, single-cell profiling, spatial transcriptomics and compositional statistical analysis are established methodological domains.
In particular, Óskarsson et al. (2023) provide the essential comparator for haemodilution assessment, while Gao et al. (2023) demonstrate the procedural detail expected of a reproducible immunophenotyping protocol in Phenomics.
MARROW-ANCHOR's proposed distinction is the endpoint-directed integration of these approaches: experimentally quantified admixture is related to immune-classification stability, decomposed into constituent analytical effects and assessed against independent anatomical and functional evidence.
Whether this integration offers a genuine methodological advantage remains an empirical question. A novel framework cannot establish originality merely through a new name or a more elaborate computational model.Classification instability and biological misclassification must remain separate
The term misclassification can obscure an essential evidentiary distinction. An analytical category may change because a measured score crosses a threshold. Such reversal demonstrates instability of the assigned category under the tested perturbation. It does not independently establish that either category represents the patient's true biological condition.
The distinction is particularly important for exploratory immune clusters or transcriptional signatures without a validated biological or clinical reference. A mathematical boundary may be reproducible without corresponding to a discrete functional state.
The proposed hierarchy therefore distinguishes continuous measurement change, operational classification reversal, independently corroborated biological discordance and clinical diagnostic error. Only the first two can be established directly through the controlled admixture design.
Compositional Effects Must Not Be Mistaken for Intrinsic Cellular Dysfunction
An aggregate immune signature is influenced by the relative abundance of its contributing populations. Consequently, changing the proportions of circulating and marrow-derived cells can alter a specimen-level score without inducing a new programme within any constituent population.
The symmetric decomposition proposed here separates population-weighting changes from differences in measured population-specific values for linear aggregate signatures. Nevertheless, even the within-population component is not inherently causal. It may reflect selective recovery of substates, differential capture, annotation differences or measurement error.
This is why sham controls, recovery measurements and participant-aware analysis are integral to the protocol rather than optional refinements.
Marrow–Blood Agreement Is a Legitimate Outcome
Cross-compartmental immune concordance is not merely a theoretical possibility. Zabaleta et al. (2026) demonstrated relationships between marrow and PB immune populations in a large clinically characterised myeloma cohort, supporting the relevance of selected circulating immune biomarkers.
The present framework must therefore distinguish compartment-sensitive biological signals from features that are stable or meaningfully shared across compartments.
Where immune classifications remain robust despite substantial experimental admixture, this should be interpreted as evidence of measurement stability within the tested conditions, not as a failure to support the preferred hypothesis.
Anatomical Information Improves Context but Does Not Establish Absolute Truth
Spatial profiling reveals relationships that are lost when tissue is converted into a cell suspension. Myeloma-associated niches differ in local cellular composition and transcriptional characteristics, as demonstrated by Yip et al. (2025), Muiños-Lopez et al. (2025) and Koops et al. (2026).
However, a trephine is itself a limited and processed biological sample. Anatomical mismatch, focal infiltration, section selection and tissue-preservation effects can generate differences unrelated to aspiration haemodilution.
Spatial measurements should therefore corroborate defined anatomical properties without being promoted into an infallible whole-marrow reference. Where different modalities interrogate non-equivalent constructs, discordance may be irreducible without additional sampling or measurement.
Functional Inference Requires Independent Evidence
The interpretation of exhaustion-associated immune states is particularly sensitive to extrapolation. Potdar et al. (2025) and Shasha et al. (2025) illustrate the limitations of assigning functional failure from molecular or phenotypic resemblance alone.
MARROW-ANCHOR consequently requires a defined distinction between markers of cellular differentiation or activation and experimentally measured function.
Functional assays can provide orthogonal evidence but remain restricted by their stimulation conditions, population definitions and inability to reproduce the complete marrow microenvironment.
The appropriate conclusion is therefore bounded by the specific biological process measured, rather than extended to global immune competence.
Identifiability and the Danger of Overcorrection
The controlled experimental design identifies the effect of additional PB cells on an existing aspirate. It does not uniquely reconstruct the composition of the intact tissue before aspiration.
An apparently successful correction may recover known added-cell proportions while remaining incapable of determining the unknown baseline contamination. If its assumptions about source populations and differential recovery are invalid, the corrected output may be biologically less representative than the unadjusted measurement.
Overcorrection is especially concerning when circulating immune populations carry genuine disease-associated information. Removing their contribution indiscriminately could suppress clinically relevant biological variation rather than eliminate artefact.
Accordingly, any adjustment must demonstrate an advantage over simpler approaches and retain performance when applied to independent measurements and participants.
Reproducibility and Implementation Limitations
The proposed study is technically demanding. Complete implementation requires clinically appropriate paired sampling, sufficient viable cellular material, reliable specimen-quality assessment, validated immunophenotyping, selected molecular and spatial assays, and an analytical workflow capable of preserving independence between development and validation data.
Specimens from patients with extensive marrow disease, fibrosis or low cellularity may be particularly difficult to process. Their exclusion could produce a systematically selected study population. Sample attrition must therefore be reported and evaluated as a potential source of bias.
A staged approach is scientifically preferable. Initial flow-cytometric admixture experiments can determine whether meaningful classification instability exists and justify subsequent molecular and spatial investigation. Advanced assays should resolve defined uncertainties rather than be included solely to increase technological breadth.
The most substantial limitation of the present manuscript is its implementation status. It specifies an integrated experimental and analytical design, but final laboratory settings, empirical precision, assay performance and independent results have not been established. These must be supplied before the framework can legitimately be presented as a fully validated operational protocol.
Implications for Precision Immunomics
The translational importance of MARROW-ANCHOR lies in establishing a defensible relationship between specimen measurement and biological interpretation.
If an immune signature remains stable under relevant perturbation, the evidence supporting its analytical robustness is strengthened. If it changes predictably with altered cellular composition, its interpretation may require specimen-quality qualification. If discordance primarily reflects selective recovery or anatomical heterogeneity, blood-admixture correction alone may be inadequate. If an adjustment fails independent validation, its additional complexity should not be adopted.
These are distinct scientific outcomes, each capable of refining the interpretation of marrow immunomics without requiring an artificial presumption of methodological success.
Precision immunomics must ultimately distinguish the ability to measure a signal from the justification for assigning that signal biological or clinical authority.
MARROW-ANCHOR operationalises this distinction through controlled perturbation, explicit analytical assumptions, independently assessed corroboration and falsifiable endpoints.
Conclusions
MARROW-ANCHOR proposes an integrated experimental framework for determining whether peripheral-blood admixture and acquisition-associated variation alter multidimensional immune interpretations in multiple myeloma. By separating compositional perturbation from within-population variation, distinguishing classification instability from independently substantiated biological misclassification, and benchmarking specimen-aware adjustment against established quality assessment, the protocol addresses an unresolved methodological boundary between aspirate measurement and precision immunomics.
Its scientific validity remains contingent upon experimental implementation, reproducibility and independent validation. The central question is not simply whether marrow and blood differ, but whether those differences materially change the immune conclusion attributed to the patient.
Author Contributions
Birupaksha Biswas (BB): Conceptualisation; clinical and pathological formulation of the research question; development of the biological rationale; study design; experimental protocol architecture; interpretation of the relevant haematopathological and immunological literature; preparation of the original manuscript; and critical scientific revision. Twisha Titirsha (TT): Mathematical formulation; development and derivation of the analytical equations; formal mathematical analysis; validation of the proposed mathematical framework; computational and methodological evaluation; and critical review of the quantitative foundations of the protocol. Both authors are responsible for reviewing, approving and accepting accountability for the final submitted manuscript.
Funding
Funding disclosures are to be confirmed by both authors before submission.
Conflicts of Interest
Financial and non-financial competing-interest declarations are to be confirmed by both authors before submission.
Ethics Approval
The manuscript describes a proposed experimental protocol and mathematical validation framework. No completed prospective human-participant investigation is reported.
Implementation
involving human biological specimens will require appropriate institutional ethics approval before commencement.
Consent to Participate
Not applicable to preparation of the present methodological manuscript. Appropriate informed consent will be required for prospective human-participant investigations.
Consent for Publication
Not applicable. No identifiable participant information is included.
Data and Materials Availability
No new patient-derived experimental dataset is reported. The mathematical framework has been developed and assessed by the second author. Supporting analytical derivations and validation materials should be made available where appropriate to permit independent verification.
Code Availability
Any computational code used to validate the mathematical framework will be documented and made available in accordance with the journal's reproducibility requirements and the authors' agreed arrangements.
Abbreviations
BM, bone marrow; BMQI, bone marrow quality index; CITE-seq, cellular indexing of transcriptomes and epitopes by sequencing; EDTA, ethylenediaminetetraacetic acid; FDR, false discovery rate; FFPE, formalin-fixed, paraffin-embedded; GEO, Gene Expression Omnibus; MARROW-ANCHOR, proposed spatially anchored marrow immunomics validation framework; MM, multiple myeloma; MRD, measurable residual disease; NK, natural killer; PB, peripheral blood; QC, quality control; scRNA-seq, single-cell RNA sequencing; SOP, standard operating procedure.
References
- Badelita, S.N.; Sandu, H.M.; Coriu, D.; Popa, D.C. Bone marrow hemodilution assessment in multiple myeloma MRD by next generation flow cytometry - a mini review. Front Oncol. 2026, 16, 1849714. [Google Scholar] [CrossRef]
- Büttner, M.; Ostner, J.; Müller, C.L.; Theis, F.J.; Schubert, B. scCODA is a Bayesian model for compositional single-cell data analysis. Nat. Commun. 2021, 12, 6876. [Google Scholar] [CrossRef]
- Campos-Dopazo, L.; Cócera, M.; Aguirre-Ruiz, P. Preparation of formalin-fixed, paraffin-embedded bone marrow samples for spatial transcriptomics using EDTA-based decalcification. J. Vis. Exp. 2026, 234. [Google Scholar] [CrossRef]
- Crowell, H.L.; Soneson, C.; Germain, P.L.; Calini, D.; Collin, L.; Raposo, C.; Malhotra, D.; Robinson, M.D. muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data. Nat. Commun. 2020, 11, 6077. [Google Scholar] [CrossRef]
- Flores-Montero, J.; Sanoja-Flores, L.; Paiva, B. Next Generation Flow for highly sensitive and standardized detection of minimal residual disease in multiple myeloma. Leukemia 2017, 31, 2094–2103. [Google Scholar] [CrossRef]
- Gao, J.; Luo, Y.; Li, H.; Zhao, Y.; Zhao, J.; Han, X.; Han, J.; Lin, H.; Qian, F. Deep Immunophenotyping of Human Whole Blood by Standardized Multi-parametric Flow Cytometry Analyses. Phenomics 2023, 3, 309–328. [Google Scholar] [CrossRef]
- Koops, M.; Bertamini, L.; Papazian, N. Spatial profiling identifies a distinct and topographically-defined tumor microenvironment that emerges during multiple myeloma evolution. Blood Cancer Discov. 2026. [Google Scholar] [CrossRef]
- McGinnis, C.S.; Murrow, L.M.; Gartner, Z.J. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 2019, 8, 329–337.e4. [Google Scholar] [CrossRef]
- Muiños-Lopez, E.; Lopez-Perez, A.R.; Sudupe, L. Characterization of the bone marrow architecture of multiple myeloma using spatial transcriptomics. Commun. Biol. 2025, 8, 1620. [Google Scholar] [CrossRef]
- Óskarsson, J.Þ.; Rögnvaldsson, S.; Thorsteinsdottir, S. Determining hemodilution in diagnostic bone marrow aspirated samples in plasma cell disorders by next-generation flow cytometry: Proposal for a bone marrow quality index. Blood Cancer J. 2023, 13, 177. [Google Scholar] [CrossRef]
- Pilcher, W.C.; Yao, L.; Gonzalez-Kozlova, E. A single-cell atlas characterizes dysregulation of the bone marrow immune microenvironment associated with outcomes in multiple myeloma. Nat. Cancer 2026, 7, 224–246. [Google Scholar] [CrossRef]
- Potdar, S.V.; Kapadia, S.; Azeem, M.I. Immune-aging is linked to clinical malignancy, racial ancestry, and vaccine responses in myeloma. Blood Adv. 2025, 9, 2438–2442. [Google Scholar] [CrossRef]
- Shasha, C.; Glass, D.R.; Moelhman, E. Hallmarks of T-cell exhaustion and antigen experience are absent in multiple myeloma from diagnosis to maintenance therapy. Blood 2025, 145, 3113–3123. [Google Scholar] [CrossRef]
- Stoeckius, M.; Hafemeister, C.; Stephenson, W. Simultaneous epitope and transcriptome measurement in single cells. Nat. Methods 2017, 14, 865–868. [Google Scholar] [CrossRef]
- Stoeckius, M.; Zheng, S.; Houck-Loomis, B. Cell Hashing with barcoded antibodies enables multiplexing and doublet detection for single cell genomics. Genome Biol. 2018, 19, 224. [Google Scholar] [CrossRef]
- Yip, R.K.H.; Er, J.; Qin, L. Profiling the spatial architecture of multiple myeloma in human bone marrow trephine biopsy specimens with spatial transcriptomics. Blood 2025, 146, 1837–1849. [Google Scholar] [CrossRef]
- Young, M.D.; Behjati, S. SoupX removes ambient RNA contamination from droplet-based single-cell RNA sequencing data. GigaScience 2020, 9. [Google Scholar] [CrossRef]
- Zabaleta, A.; Tamariz-Amador, L.E.; Kostopoulos, I.V. Immune biomarkers of increased infection risk in multiple myeloma. Blood 2026, 147, 2081–2088. [Google Scholar] [CrossRef]
Table 1.
MARROW-ANCHOR: integrated experimental and analytical validation matrix.
| Stage and objective | Specimen inputs and operations | Essential controls | Prespecified outputs | Interpretative limits and falsification |
|---|---|---|---|---|
| A. Matched acquisition and provenance | Collect clinically appropriate early-pull BM, later-pull BM, PB and corresponding trephine tissue; document anatomical site, pull order, volume and clinical context | Diagnostic prioritisation, acquisition metadata, documented site concordance and sample attrition | Specimen-provenance record, baseline cellular characteristics and paired-sample structure | Early-pull BM is not necessarily blood-free; later-pull variation cannot be assigned uniquely to haemodilution |
| B. Quality assessment and controlled admixture | Measure marrow-associated precursor populations, calculate compatible BMQI values, prepare sham-handled controls and graded autologous BM–PB mixtures | Pre-enrichment quality aliquots, independent cell counting, balanced processing, technical replication and recovery measurements | Known added-blood fraction, specimen-quality indicators, viability and post-processing cellular recovery | Experimental addition is identifiable; absolute pre-existing contamination is not; poor mixture fidelity limits inference |
| C. Multimodal immune profiling | Undertake validated multiparameter flow cytometry, selected scRNA-seq/CITE-seq and relevant immune-subset analyses | Fixed gating, lineage exclusion, platform QC, batch balancing, ambient-RNA assessment, doublet detection and participant-level replication | Major population frequencies, within-population phenotypes, aggregate signatures and assay-specific uncertainty | Compositional change is not proof of altered intrinsic function; platform discordance requires independent investigation |
| D. Decomposition and classification testing | Model expected mixture composition, assess differential recovery, decompose aggregate signatures, apply frozen classifiers and compare analytical approaches | Independent development–validation separation, prespecified endpoints, continuous-score retention, uncertainty estimation and leakage prevention | Admixture-response estimates, classification-reversal frequency, compositional contributions, model calibration and benchmark performance | Classification reversal is not automatically biological misclassification; failure to outperform simpler methods argues against additional correction |
| E. Biological corroboration and external validation | Evaluate matched spatial immune features, selected functional measurements and independently recruited participants | Blinded interpretation where feasible, region-level tissue QC, population-appropriate functional controls and frozen analytical procedures | Spatial and functional agreement, reproducibility, generalisability and evidence for or against specimen-aware adjustment | Spatial tissue is not absolute ground truth; functional assays are context-dependent; clinical utility requires separate outcome validation |
a The table describes a proposed experimental framework and contains no measured results. b Admixture refers to an experimentally defined contribution of PB-derived viable nucleated cells; pre-existing contamination of the baseline aspirate remains an unobserved quantity. c Operational classification instability, biologically corroborated misclassification and clinical diagnostic error are distinct endpoints. d All aliquots and experimental conditions belonging to one participant must remain within the same development or validation partition. e Spatial and functional measurements provide complementary reference information rather than a universally valid hierarchy of biological truth. f Negative findings, analytical non-identifiability and failure of correction must be reported without outcome-dependent exclusion.
Table 2.
Specimen-to-assay allocation and hierarchy of biological interpretation.
| Specimen or experimental condition | Core measurement | Additional measurement | Principal purpose | Limitation |
|---|---|---|---|---|
| Early-pull BM | Morphology, specimen adequacy, BMQI-compatible analysis and immune flow cytometry | scRNA-seq/CITE-seq | Operational aspirate reference | May contain pre-existing PB admixture |
| Later-pull BM | Matched specimen-quality and immune flow measurements | Selected single-cell profiling | Acquisition-related comparison | Haemodilution, selective recovery and anatomical effects may coexist |
| Concurrent PB | Differential count, viability, cellular concentration and immune phenotyping | Paired single-cell profiling | Autologous circulating-cell source | Does not define the original intact-marrow composition |
| Sham-handled BM | Repeated viability, cellular recovery and flow cytometry | Selected molecular measurements | Measurement of handling-related variability | Cannot remove pre-existing baseline sampling bias |
| Controlled BM–PB mixtures | Known viable-cell input, recovered composition and immune profiling | Selected single-cell measurements | Experimentally identifiable admixture response | Does not reconstruct every mechanism of aspiration |
| Matched trephine | Histomorphology and multiplex spatial immunophenotyping | Spatial transcriptomics | Independent anatomical comparison | Restricted tissue geography and processing artefacts |
| Functional aliquots | Population-appropriate controlled functional assays | Extended immune-functional assessment | Corroboration of selected functional interpretations | In vitro responses incompletely represent in vivo immune behaviour |
| Independent validation cohort | Frozen primary analytical workflow | Targeted multimodal replication | Reproducibility and transportability | Performance may depend upon population and platform differences |
a Clinical diagnostic requirements take precedence over research allocation. b The early-pull specimen is an operational comparator, not an absolute biological reference. c Assay denominators and processing histories must be preserved; measurements obtained after selective enrichment are not automatically comparable with measurements from unprocessed aspirates. d Missing advanced assays must be documented and do not automatically invalidate the prespecified core analysis.
Table 3.
Prespecified endpoints, evidentiary requirements and falsification criteria.
| Domain | Endpoint | Required evidence | Unsupported inference | Result weakening or falsifying the claim |
|---|---|---|---|---|
| Mixture fidelity | Agreement between intended and recovered mixture composition | Independent cell counts, viability and replicate QC | Assuming fluid-volume proportions equal recovered-cell proportions | Unexplained input–output differences beyond prespecified tolerance |
| Compositional response | Change in major immune-population frequencies | Paired source profiles, compatible denominators and participant-level uncertainty | Calling relative-frequency changes absolute depletion or expansion | Differences indistinguishable from sham variation |
| Aggregate immune signature | Continuous score response to admixture | Fixed signature definition and reproducible analysis | Equating aggregate expression with intrinsic cellular dysfunction | No material perturbation beyond technical variability |
| Within-population phenotype | Population-specific expression or protein changes | Stable annotations, adequate recovery and replicated analysis | Attributing residual expression variation uniquely to biological transformation | Changes explained by annotation, recovery or processing effects |
| Classification stability | Primary classification-reversal frequency | Frozen rule, independent participants and uncertainty intervals | Treating every reversal as clinical misdiagnosis | Stable categories or variation confined to threshold uncertainty |
| Spatial corroboration | Agreement with independently measured tissue features | Matched anatomy, tissue QC and appropriately comparable biological features | Treating a single trephine as whole-marrow ground truth | No reproducible association or dominant anatomical sampling effects |
| Functional corroboration | Agreement between inferred state and measured function | Population-appropriate controlled functional assays | Inferring exhaustion from markers alone | Preserved function despite an unsupported dysfunction claim |
| Adjustment performance | Independent improvement over unadjusted and quality-informed analysis | Frozen model, calibration and independent comparators | Claiming success from development-data fit | No gain, poor calibration or biological overcorrection |
| External reproducibility | Performance in held-out participants | Independent biological replication and fixed analysis | Treating resampled cells as an external cohort | Failure to reproduce performance or substantial transportability loss |
| Clinical usefulness | Improvement in validated clinical prediction or decisions | Separate patient-outcome validation | Inferring clinical benefit from analytical correction | No demonstrated benefit; outside the present protocol's principal claim |
a Classification reversal is an operational endpoint and does not establish biological error without independent justification. b Statistical significance alone is insufficient to establish analytical or clinical importance. c Failure of a correction model does not imply that the underlying specimen is universally invalid. d Tissue and suspension measurements must retain their distinct denominators and uncertainty structures.e Negative findings must be reported without retrospective alteration of endpoint definitions.
Table 4.
MARROW-ANCHOR troubleshooting and quality-assurance matrix.
| Step | Potential problem | Likely explanation | Corrective procedure | Residual limitation |
|---|---|---|---|---|
| 9–14 | Inadequate early-pull material | Hypocellularity, fibrosis, difficult aspiration or diagnostic prioritisation | Preserve clinical material; document insufficiency; retain available core measurements | Attrition may correlate with disease features |
| 12–14 | Invalid BMQI | Selective erythroid loss, incompatible denominator or inadequate precursor identification | Reassess original sample requirements; report unavailable values rather than substitute incompatible measures | BMQI cannot be reconstructed reliably from every processed aliquot |
| 15–20 | Nominal admixture differs from recovered composition | Cell-counting error, unequal viability or differential recovery | Verify source concentrations and post-processing cellular recovery | Analysed-cell composition may differ from intended input |
| 18–22 | Excessive sham variation | Handling stress, processing delay or instrument instability | Evaluate repeat preparations and technical controls | Poor baseline precision limits detectable admixture effects |
| 23–27 | Ambiguous NK-cell identification | Aberrant plasma-cell CD56 expression or inadequate lineage exclusion | Apply validated plasma-cell discrimination and NK-cell gating | Some populations may remain phenotypically ambiguous |
| 24–27 | Unstable rare-cell estimates | Insufficient event counts or gate sensitivity | Apply prespecified minimum-event requirements | Rare-subset inference may remain unavailable |
| 28–30 | Discordant flow and single-cell abundances | Capture bias, doublets, ambient RNA or unequal recovery | Examine platform QC and independent flow measurements | The platforms may interrogate different recovered populations |
| 29–30 | Apparent within-population transcriptomic changes | Annotation instability, substate selection or batch effects | Assess marker identity, population subdivision and processing balance | Residual variation may not reflect intrinsic biological change |
| 31–35 | Weak spatial signal | Fixation, decalcification or segmentation failure | Apply tissue QC, validated staining and independent review | Uninterpretable tissue cannot provide negative biological corroboration |
| 35, 46 | Aspirate–trephine disagreement | Anatomical mismatch, regional heterogeneity or different measurement scales | Examine site provenance and region-specific findings | Discordance cannot automatically be assigned to blood admixture |
| 36–39 | Mixture prediction fails | Differential recovery or invalid proportional-mixing assumptions | Use independently measured recovery and sensitivity analyses | Absolute baseline contamination may remain non-identifiable |
| 40–42 | Frequent boundary-sensitive reversals | Scores clustered near a classification threshold | Retain continuous scores, confidence intervals and threshold distances | Category change may be disproportionate to underlying biological variation |
| 44–48 | Correction fails independently | Overfitting, leakage or population differences | Freeze model and assess calibration in held-out participants | Non-generalisable adjustment should not be recommended |
| 34, 46 | Functional results contradict molecular phenotype | Marker–function discordance or assay-context dependence | Report findings at the appropriate cellular level with controls | No single functional assay establishes global immune competence |
| 45–50 | Missing multimodal observations | Insufficient sample or technical assay failure | Record causes and undertake prespecified sensitivity analyses | Complete-case estimates may be biased |
| 45–50 | No advantage over existing quality assessment | Immune-profile robustness or insufficient added model value | Report the negative comparison without retrospective model replacement | Complex adjustment is not justified solely by conceptual appeal |
a Technical exclusions require documented criteria independent of the preferred outcome. b Repeated technical observations do not replace independent biological replication. c Failure of an optional molecular or spatial assay need not invalidate an otherwise interpretable core analysis. d Stable classifications, failed correction and contradictory measurements may constitute informative negative findings.e Troubleshooting must not permit retrospective selection of favourable analytical procedures.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.