Single-Cell vs Single-Nucleus RNA Sequencing: Choosing the Right Approach

Researchers comparing transcriptomic methods face a practical fork: dissociate fresh tissue into single cells or isolate single nuclei from fresh, frozen, or fixed samples. The choice changes which cells you capture, how you prepare tissue, and what biological questions you can answer. Single-cell RNA sequencing (scRNA-seq) profiles RNA from cells dissociated from fresh tissue, while single-nucleus RNA sequencing (snRNA-seq) profiles RNA from nuclei isolated from fresh, frozen, or hard-to-dissociate samples. This article gives students, researchers, analysts, and life-science professionals a comparison framework based on sample type, tissue characteristics, and research question, with a decision tree grounded in published applications.

What Each Method Measures

scRNA-seq captures the full transcriptome of intact cells after enzymatic or mechanical dissociation of fresh tissue. The method requires viable cells at the point of capture, which limits its use to samples that can be processed quickly after collection. snRNA-seq captures nuclear RNA from isolated nuclei, which means it can work with frozen tissue, fixed tissue, and samples that resist standard dissociation protocols.

The biological difference matters. Cytoplasmic mRNA is abundant in most cells, so scRNA-seq captures a broad view of gene expression. Nuclear RNA includes precursor and nascent transcripts, which can reveal regulatory activity but may not fully represent cytoplasmic mRNA pools. For tissues where dissociation triggers transcriptional stress responses, snRNA-seq avoids the activation artifacts that can appear during cell isolation.

Tissue Type Drives the Decision

The most important variable is your starting material. Fresh tissue with soft extracellular matrix, such as blood, bone marrow, or cultured cells, suits scRNA-seq because dissociation is straightforward and cell viability remains high. Frozen tissue, archived clinical specimens, and tissues with dense or fibrous architecture favor snRNA-seq because nuclei survive freezing and mechanical disruption better than intact cells.

A systematic toolbox developed for fresh and frozen human tumors analyzed 216,490 cells and nuclei from 40 samples across 23 specimens spanning eight tumor types. The authors found that scRNA-seq and snRNA-seq from matched samples recovered the same cell types but at different proportions. This finding has direct implications for study design: if your research question depends on accurate cell-type proportions, you need to validate that your chosen method preserves the populations of interest.

At a Glance

Core Principles of Method Selection

Cell Recovery and Composition

The proportion of recovered cell types differs between methods. In the tumor toolbox study, matched samples profiled by both methods recovered the same cell types but at different proportions. This means your cell-type composition results depend on your choice of method. If you are studying a rare population that is underrepresented in one method, you may need to switch approaches or add enrichment steps.

For tissues with large or fragile cells, such as adipocytes or cardiomyocytes, scRNA-seq often loses these populations during dissociation. snRNA-seq captures nuclei from these cells more reliably because the nucleus is smaller and more resistant to mechanical stress. Studies of heart failure used snRNA-seq to profile myocardial biopsies and recovered cardiomyocytes with sufficient depth to identify thousands of differentially expressed genes. A study of human heart failure with preserved ejection fraction recovered 48,886 nuclei from pooled septal myocardial biopsies and identified 14 cell types, with cardiomyocytes and fibroblasts showing the most differentially expressed genes.

Dissociation Challenges

Enzymatic dissociation required for scRNA-seq can activate stress response genes and alter the transcriptome before capture. This artifact is especially problematic in neural tissue, where the dissociation process itself can induce immediate early genes and other activity-dependent transcripts. snRNA-seq avoids this problem because nuclei are isolated rapidly in cold buffers without prolonged enzymatic digestion.

The kidney provides a clear example. A study of mouse acute kidney injury used snRNA-seq to characterize cell states during repair and identified a distinct proinflammatory and profibrotic proximal tubule cell state that fails to repair. The authors detected this failed-repair state in other models of kidney injury and found it increased during aging in rat kidney and over time in human kidney allografts. Working with frozen kidney tissue would have been difficult with scRNA-seq because the tissue is dense and the proximal tubule cells are vulnerable to dissociation stress.

Frozen and Archived Samples

Biobanked frozen tissue is a major resource for clinical research. snRNA-seq is the standard choice for these samples because nuclei can be isolated from cryopreserved tissue without requiring viable cells. The tumor toolbox study demonstrated that snRNA-seq profiles frozen tumors effectively, enabling analysis of clinical specimens that cannot be processed fresh.

FFPE tissue presents additional challenges because formalin cross-links RNA. A cryogenic enzymatic dissociation method developed for FFPE samples produced ten times more nuclei than the homogenate method, with higher gene detection sensitivity and RNA coverage. The method retained more RNA molecules within nuclei and showed gene expression correlation up to 94 percent with frozen samples. This approach enables profiling of archived clinical specimens that were previously inaccessible to single-cell transcriptomics.

Practical Workflow for Method Selection

Step 1: Assess Your Sample Type

List your sample characteristics before choosing a method. Record whether the tissue is fresh, frozen, or fixed. Note the extracellular matrix density, the presence of large or fragile cells, and the time between collection and processing. If you have any doubt about tissue condition, test a small piece with both methods before committing your full cohort.

Step 2: Define Your Biological Question

Ask whether you need cytoplasmic mRNA or nuclear RNA. If your question concerns mature mRNA abundance, translation, or cytoplasmic processing, scRNA-seq provides more relevant data. If your question concerns transcriptional regulation, nascent RNA, or cell-type identity in archived samples, snRNA-seq is appropriate.

Step 3: Check Published Precedents

Search the literature for your tissue type and disease context. The studies cited here provide precedents for kidney, heart, brain, eye, muscle, liver, lung, and plant tissues. If a published study used snRNA-seq successfully on your tissue type, that method is a reasonable starting point. If no precedent exists, run a pilot comparison.

Step 4: Pilot Test on Representative Samples

Process at least two biological replicates per condition with both methods if feasible. Compare cell-type recovery, gene detection sensitivity, and data quality metrics. The tumor toolbox study provides criteria for testing and selecting methods across tumor types, and the same logic applies to other tissues.

Step 5: Validate With Independent Methods

Reference-free deconvolution methods can estimate cell-type proportions from bulk RNA-seq data, but they require biological validation. A comparative evaluation of two unsupervised deconvolution approaches found that one method achieved moderate concordance with independent snRNA-seq astrocyte data while the other showed substantially lower concordance. This finding underscores the importance of validating computational predictions against direct single-cell measurements.

Options and Tradeoffs

Fresh Tissue With scRNA-seq

The main advantage of scRNA-seq on fresh tissue is access to the full cytoplasmic transcriptome. Gene detection per cell is generally higher because both nuclear and cytoplasmic mRNA contribute to the library. The main disadvantage is the requirement for rapid processing and the risk of dissociation-induced transcriptional artifacts.

For blood and immune cells, scRNA-seq is the standard because cells are already in suspension. For solid tissues, the decision depends on whether dissociation preserves the populations of interest. The tumor toolbox study showed that scRNA-seq and snRNA-seq recover the same cell types from matched samples but at different proportions, so you need to know which proportions are biologically meaningful for your question.

Frozen Tissue With snRNA-seq

snRNA-seq on frozen tissue enables analysis of biobanked samples and large clinical cohorts. The method avoids the need for fresh tissue processing and reduces batch effects from variable dissociation times. The main limitation is that nuclear RNA may not fully represent cytoplasmic mRNA abundance, particularly for genes with rapid mRNA turnover.

A study of normal-appearing brain regions in multiple sclerosis used snRNA-seq to generate datasets of 33,197 nuclei from 8 brains and identified divergent cell type-specific changes between relapsing-remitting and secondary-progressive disease. The authors found that secondary-progressive brains downregulated astrocytic sphingosine kinases, the enzymes required to activate the drug fingolimod. This finding would have been difficult to obtain with scRNA-seq because the brain tissue was archived and the relevant cell types are vulnerable to dissociation stress.

Hard-to-Dissociate Tissues With snRNA-seq

Brain, kidney, heart, and muscle tissues present dissociation challenges that favor snRNA-seq. The extracellular matrix is dense, and enzymatic digestion can damage cells or activate stress responses. snRNA-seq bypasses these problems by isolating nuclei directly.

A study of the macaque claustrum used snRNA-seq of 227,750 cells to identify 48 transcriptome-defined cell types and compared them across macaque, marmoset, and mouse. The authors found macaque-specific cell types and showed that distinct claustral zones containing different glutamatergic cell types connected to specific brain regions. This level of cellular resolution in a deep brain structure required a method that could handle frozen or fixed tissue and preserve nuclear integrity.

FFPE Tissue With Specialized snRNA-seq

FFPE tissue is the most challenging input for single-cell transcriptomics because formalin cross-links RNA and proteins. The cryogenic enzymatic dissociation method developed for FFPE samples addresses this problem by using cold temperatures to reduce RNA degradation and enzymatic digestion to release nuclei. The method produced more nuclei, higher gene detection sensitivity, and better RNA coverage than traditional homogenate methods.

This capability opens archived clinical specimens to single-cell analysis. If you have FFPE blocks with associated clinical outcomes, snRNA-seq with specialized protocols may allow you to profile cellular heterogeneity and identify biomarkers. The method has been applied to Alzheimer's disease specimens and detected scarce cell populations that are difficult to recover with other approaches.

Plant Tissue With Either Method

Plant cells have rigid cell walls that complicate protoplast preparation for scRNA-seq. Protoplast-based methods require enzymatic digestion of the cell wall, which can induce stress responses and alter gene expression. snRNA-seq avoids this problem by isolating nuclei directly from plant tissue.

A study of hawthorn leaves integrated protoplast-based scRNA-seq and nucleus-based snRNA-seq from two species and profiled 32,292 high-quality cells across nine canonical cell types. The authors identified 642 differentially expressed genes between species and demonstrated spatial partitioning of flavonoid biosynthetic genes across multiple cell populations. This study shows that both methods can work for plant tissue, but the choice depends on whether protoplast preparation is feasible for your species and tissue.

Observations and Measurements

Gene Detection Sensitivity

Gene detection sensitivity differs between methods. scRNA-seq generally detects more genes per cell because cytoplasmic mRNA contributes to the library. snRNA-seq detects fewer genes per cell but can still provide sufficient depth for cell-type identification and differential expression analysis.

The cryogenic enzymatic dissociation study reported 1.5 to 2 times higher gene and UMI numbers per nucleus compared with traditional methods, with a minor rate of mitochondrial and ribosomal genes. This improvement in sensitivity makes snRNA-seq more competitive with scRNA-seq for FFPE and frozen samples.

Cell-Type Proportion Estimates

Cell-type proportions differ between methods even when the same cell types are recovered. The tumor toolbox study found that scRNA-seq and snRNA-seq from matched samples recovered the same cell types but at different proportions. This discrepancy can arise from differential cell loss during dissociation, differential nuclear recovery, or differences in RNA capture efficiency across cell types.

For studies that depend on accurate cell-type proportions, such as deconvolution of bulk RNA-seq data, you need to validate your method choice. The deconvolution comparison study showed that one method achieved moderate concordance with independent snRNA-seq astrocyte data while another showed substantially lower concordance. This finding demonstrates that computational methods vary in their ability to recover cell-type proportions, and biological validation is essential.

Differential Expression Results

Differential expression results can differ between methods because nuclear and cytoplasmic RNA pools are not identical. Genes with rapid mRNA turnover may show different expression levels in nuclear versus whole-cell RNA. Genes with extensive cytoplasmic localization may be underrepresented in snRNA-seq data.

A study of the dorsal lateral geniculate nucleus compared scRNA-seq and snRNA-seq profiles in mice, non-human primates, and humans. The authors found that transcriptomic differences between principal cell types were subtle relative to observed differences in morphology and cortical projection targets. This finding suggests that both methods can resolve major cell types, but the resolution of subtle transcriptional differences may depend on the method and tissue.

Records and Measurements

Quality Control Metrics

Record standard quality control metrics for every sample and method. These include the number of cells or nuclei captured, the number of genes detected per cell or nucleus, the fraction of mitochondrial reads, and the fraction of ribosomal reads. For snRNA-seq, mitochondrial reads should be low because mitochondria are excluded from nuclear preparations.

The heart failure study used genotype-based demultiplexing to assign pooled myocardial biopsies to individual patients, with more than 75 percent of droplets assigned successfully. This approach enables pooling of samples to reduce cost while maintaining patient-level resolution. If you plan to pool samples, record the demultiplexing success rate and validate that your genotyping approach works for your sample type.

Batch Effect Tracking

Record the batch, processing date, and operator for every sample. Batch effects can confound biological differences, especially in large cohorts processed over extended periods. The multiple sclerosis study processed 8 brains and identified divergent cell type-specific changes, but the authors needed careful batch handling to distinguish disease effects from technical variation.

For multi-batch studies, consider using snRNA-seq with frozen tissue because it allows you to process all samples at the end of the study instead of as they are collected. This design reduces batch effects from variable processing times and enables more consistent library preparation.

Sample Metadata

Record complete sample metadata, including tissue type, collection method, storage conditions, and clinical annotations. The kidney injury study detected the failed-repair proximal tubule cell state in other models of kidney injury and found it increased during aging and after transplantation. These findings required detailed metadata linking molecular states to clinical outcomes.

For human samples, follow the NIH Genomic Data Sharing Policy for data deposition and access. The policy governs how genomic data are shared, stored, and protected. Check the policy requirements before you begin data generation so that your consent forms and data management plans are compliant.

Common Failure Patterns

Dissociation-Induced Transcriptional Artifacts

scRNA-seq on fresh tissue can induce transcriptional stress responses during dissociation. This artifact appears as upregulation of immediate early genes, heat shock proteins, and other stress markers. If you see these genes enriched in your scRNA-seq data, consider switching to snRNA-seq or optimizing your dissociation protocol.

The tumor toolbox study addressed this issue by developing a systematic approach for profiling fresh and frozen tumors. The authors evaluated protocols by cell and nucleus quality, recovery rate, and cellular composition. If your pilot data show stress artifacts, test alternative dissociation conditions or switch to snRNA-seq.

Low Nuclei Yield From FFPE Tissue

FFPE tissue often yields low numbers of nuclei because formalin cross-links RNA and proteins. The cryogenic enzymatic dissociation method addressed this problem by using cold temperatures and enzymatic digestion to release nuclei. If your FFPE samples yield too few nuclei, test this method or alternative protocols designed for fixed tissue.

Cell-Type Proportion Distortion

Both methods can distort cell-type proportions relative to the original tissue. Large cells are lost during dissociation for scRNA-seq, and some nuclei are lost during isolation for snRNA-seq. If your cell-type proportions do not match histological expectations, validate your method with an independent approach such as immunohistochemistry or spatial transcriptomics.

Batch Effects in Large Cohorts

Large cohorts processed over extended periods accumulate batch effects that can obscure biological differences. If you see clustering by processing date instead of by biological condition, your batch handling needs improvement. Consider processing all samples at the end of the study with frozen tissue, or use computational batch correction methods with careful validation.

Limitations and Interpretation

Nuclear RNA Does Not Equal Cytoplasmic mRNA

snRNA-seq measures nuclear RNA, which includes precursor and nascent transcripts. This measurement is not identical to cytoplasmic mRNA abundance. For genes with rapid mRNA turnover or extensive cytoplasmic localization, snRNA-seq may underestimate expression. Interpret snRNA-seq results as measures of transcriptional activity instead of absolute mRNA abundance.

Cell-Type Proportions Are Method-Dependent

Cell-type proportions recovered by scRNA-seq and snRNA-seq differ even for matched samples. The tumor toolbox study demonstrated this discrepancy across eight tumor types. Do not assume that proportions from one method apply to another. Validate proportions with independent methods when they are central to your conclusions.

Computational Methods Require Biological Validation

Reference-free deconvolution methods can estimate cell-type proportions from bulk RNA-seq data, but their accuracy varies. The deconvolution comparison study found that one method achieved moderate concordance with independent snRNA-seq astrocyte data while another showed substantially lower concordance. Validate computational predictions against direct single-cell measurements before drawing biological conclusions.

Species Differences Affect Method Choice

Species differences in cell size, nuclear density, and tissue architecture affect method performance. The dorsal lateral geniculate nucleus study compared mice, non-human primates, and humans and found expanded diversity of GABAergic neurons in primate versus mouse. If you work with non-model organisms, pilot test both methods on your species before committing to a full study.

Safety and Regulatory Context

Human Sample Handling

Human tissue samples require institutional review board approval and informed consent. The NIH Genomic Data Sharing Policy governs data deposition and access for human genomic data. Review the policy requirements before you begin data generation to ensure your consent forms and data management plans are compliant.

Data Sharing and Reproducibility

Deposit your data in public repositories such as the NCBI Data Resources to enable reproducibility and secondary analysis. The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable. Follow these principles when you organize your data files, metadata, and analysis code.

Computational Infrastructure

Large-scale single-cell and single-nucleus RNA-seq datasets require substantial computational resources. The Cumulus framework provides cloud-based analysis for large-scale datasets, combining cloud computing with algorithmic improvements to achieve high scalability and low cost. If your institution lacks local high-performance computing, consider cloud-based analysis platforms.

Professional Escalation Criteria

When to Consult a Bioinformatics Specialist

Consult a bioinformatics specialist if your pilot data show unexpected clustering patterns, low gene detection, or poor cell-type recovery. These issues may indicate technical problems that require specialized expertise to diagnose and fix. The deconvolution comparison study showed that computational methods vary in accuracy, so specialist input can prevent incorrect biological conclusions.

When to Seek Protocol Optimization

Seek protocol optimization if your FFPE samples yield too few nuclei, your fresh tissue shows dissociation artifacts, or your cell-type proportions do not match histological expectations. The cryogenic enzymatic dissociation method and the tumor toolbox approach provide starting points, but your tissue may require custom optimization.

When to Reconsider Your Method Choice

Reconsider your method choice if your research question depends on cytoplasmic mRNA abundance but you are using snRNA-seq, or if your samples are frozen but you are attempting scRNA-seq. The decision tree based on sample type and research question should guide your initial choice, but pilot data may reveal that a different method is more appropriate.

Frequently Asked Questions

What is the main difference between single-cell and single-nucleus RNA sequencing?

Single-cell RNA sequencing profiles RNA from intact cells dissociated from fresh tissue, capturing both nuclear and cytoplasmic mRNA. Single-nucleus RNA sequencing profiles RNA from isolated nuclei, which can be obtained from fresh, frozen, or fixed tissue. The main difference is the input material and the RNA pool measured.

When should I choose single-nucleus RNA sequencing over single-cell RNA sequencing?

Choose single-nucleus RNA sequencing when your samples are frozen or fixed, when your tissue is hard to dissociate, or when you want to avoid dissociation-induced transcriptional artifacts. The tumor toolbox study showed that snRNA-seq is needed to profile frozen or hard-to-dissociate tumors, while scRNA-seq profiles RNA from cells dissociated from fresh tumors.

Can I use frozen tissue for single-cell RNA sequencing?

Frozen tissue is generally not suitable for scRNA-seq because the freeze-thaw process damages cells and reduces viability. snRNA-seq is the standard choice for frozen tissue because nuclei survive freezing and can be isolated without requiring viable cells. The tumor toolbox study demonstrated that snRNA-seq profiles frozen tumors effectively.

Do single-cell and single-nucleus RNA sequencing recover the same cell types?

Both methods recover the same cell types from matched samples, but at different proportions. The tumor toolbox study found this pattern across eight tumor types. If your research question depends on accurate cell-type proportions, you need to validate that your chosen method preserves the populations of interest.

What are the advantages of single-nucleus RNA sequencing for brain tissue?

snRNA-seq avoids the dissociation-induced transcriptional artifacts that can occur with scRNA-seq in neural tissue. It also works with frozen or archived brain samples. Studies of the macaque claustrum and human multiple sclerosis brains used snRNA-seq to profile deep brain structures and archived specimens successfully.

Can single-nucleus RNA sequencing be used on FFPE tissue?

Yes, but specialized protocols are required. The cryogenic enzymatic dissociation method was developed for FFPE samples and produced more nuclei, higher gene detection sensitivity, and better RNA coverage than traditional methods. This approach enables profiling of archived clinical specimens.

How do I decide between scRNA-seq and snRNA-seq for my study?

Assess your sample type, define your biological question, check published precedents for your tissue, and run a pilot comparison if feasible. The decision tree based on sample type and research question should guide your initial choice. Fresh soft tissue suits scRNA-seq, while frozen, fixed, or hard-to-dissociate tissue favors snRNA-seq.

What quality control metrics should I record for single-cell and single-nucleus RNA sequencing?

Record the number of cells or nuclei captured, the number of genes detected per cell or nucleus, the fraction of mitochondrial reads, and the fraction of ribosomal reads. For snRNA-seq, mitochondrial reads should be low because mitochondria are excluded from nuclear preparations. Track batch, processing date, and operator for every sample.

Related Bioinformatics Guides

• Single-Cell RNA-seq Clustering and Cell-Type Annotation Pipelines
• Single-Cell RNA Sequencing: From Bulk to Resolution
• Master Guide: Single-Cell RNA Sequencing Bioinformatics Workflows
• Single-Cell RNA-Seq Analysis Pipelines for Veterinary Immunology
• Single-cell RNA-seq Trajectory Inference and Cell Lineage Tracing

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Cell profiling of mouse acute kidney injury reveals conserved cellular responses to injury.. Proceedings of the National Academy of Sciences of the United States of America, 2020.
• Single cell transcriptomic analyses of human heart failure with preserved ejection fraction.. bioRxiv : the preprint server for biology, 2025.
• Single-cell spatial transcriptome atlas and whole-brain connectivity of the macaque claustrum.. Cell, 2025.
• Single-Nucleus RNA-seq of Normal-Appearing Brain Regions in Relapsing-Remitting vs. Secondary Progressive Multiple Sclerosis: Implications for the Efficacy of Fingolimod.. Frontiers in cellular neuroscience, 2022.
• Single-cell and single-nucleus RNA-seq uncovers shared and distinct axes of variation in dorsal LGN neurons in mice, non-human primates, and humans.. eLife, 2021.
• Transcriptomic analysis of the ocular posterior segment completes a cell atlas of the human eye.. Proceedings of the National Academy of Sciences of the United States of America, 2023.
• Single nuclei RNA-sequencing unveils alveolar macrophages as drivers of endothelial damage in obese HFpEF-related pulmonary hypertension.. Cardiovascular diabetology, 2025.
• Single-Nucleus RNA Sequencing Reveals Cellular Transcriptome Features at Different Growth Stages in Porcine Skeletal Muscle.. Cells, 2025.
• Single cell and single nucleus RNA sequencing in liver tissues: applications and prospects in model and non-model organisms.. 2026.
• Comparative evaluation of reference-free transcriptomic deconvolution highlights the importance of biological validation in astrocytes across Alzheimer's disease.. 2026.
• Single-cell RNA-sequencing profiles reveal the developmental landscape of hawthorn leaves.. 2026.
• Progress in the Application of Machine Learning in the Field of Single-Cell and Spatial Transcriptomics. 2026.
• Decoding cardiac homeostasis and injury: the evolving landscape of spatial transcriptomics
• The applications of single-cell and spatial transcriptomics in neuroscience and brain disorders.. 2026.
• Mosaic Loss of Y Chromosome in Proximal Tubular Cells is Associated with Recovery from DGF Post DCD Kidney Transplantation: A New Single Nucleus RNA-Seq Signature of Organ Quality. American Journal of Transplantation, 2025.
• Cumulus provides cloud-based data analysis for large-scale single-cell and single-nucleus RNA-seq. Nature Methods, 2020.
• A single-cell and single-nucleus RNA-Seq toolbox for fresh and frozen human tumors. Nature Medicine, 2019.
• snCED-seq: high-fidelity cryogenic enzymatic dissociation of nuclei for single-nucleus RNA-seq of FFPE tissues. bioRxiv, 2024.
• Single-Nucleus RNA-Seq Reveals Spermatogonial Stem Cell Developmental Pattern in Shaziling Pigs. Biomolecules, 2024.
• Review: Challenges and perspectives in applying single nuclei RNA-seq technology in plant biology. Plant Science, 2022.
• In silico Single-Cell Analysis of Steroid-Responsive Gene Targets in the Mammalian Cochlea. Frontiers in Neurology, 2022.
• The Long Noncoding RNA Cardiac Mesoderm Enhancer-Associated Noncoding RNA (Carmn) Is a Critical Regulator of Gastrointestinal Smooth Muscle Contractile Function and Motility. Gastroenterology, 2023.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.