# How to Read Cell Line Expression Annotations in UniProt By the end of this guide you will be able to open any UniProt entry, locate and interpret every expression related annotation on it, judge how strongly each annotation is supported by evidence, download the same fields in bulk for a whole gene panel, and place those annotations side by side with expression changes you measure in your own cell line experiments. This is the practical core of knowing how to use UniProt for cell line work, and it takes about an hour the first time you run it. You need a web browser, a terminal with `curl`, and Python 3 for the scripted steps. Every command below works on macOS, Linux, and Windows with WSL. ## What UniProt Is, and What Its Expression Annotations Are and Are Not UniProt, the Universal Protein Resource, is a curated protein knowledgebase maintained as a collaboration between groups at EMBL-EBI, the SIB Swiss Institute of Bioinformatics, and PIR. It is a protein centric database. Each entry describes one protein, such as EGFR (epidermal growth factor receptor) or CASP3 (caspase 3), and collects what is known about that protein from literature, sequence analysis, and large scale experiments. The Expression section of an entry contains three main annotation types: - **Tissue specificity**, which describes where in the organism the protein is normally found or at what abundance, sometimes with notes on specific biological compartments or states. - **Developmental stage**, which captures changes across development or life cycle stages when those have been documented. - **Induction**, which records agents or conditions that change the protein's abundance, such as hormones, cytokines, drugs, or stress conditions. Here is the key framing for anyone working with cell lines. A UniProt expression annotation is a statement about the protein, usually aggregated across tissues, organisms, and experimental systems, not a measurement of a specific gene in a specific cell line under specific culture conditions. Studies that quantify expression in cell lines, for example qRT-PCR measurements of Bax and Bcl-2 in treated HT-29 colorectal cancer cells [1], or of Aquaporin 5 and EGFR in HCT-116 cells [2], answer a different question than the annotation does. The annotation tells you the baseline consensus. The study tells you what happens when you perturb the system. Reading them together, with an understanding of what each one can and cannot say, is the skill this guide teaches. If you want cell line specific expression data, UniProt is your starting index rather than your final destination. Human entries carry cross references to expression databases, and those links take you to where per tissue and per condition measurements live. A recent multi center effort to build multi omics reference materials for breast cancer cell lines, combining ATAC-seq, Methyl-seq, RNA-seq, and proteomics into community benchmarks, illustrates how much cell line specific expression data now exists outside any single annotation entry [3]. ## Step 1: Resolve Your Gene to a Reviewed UniProt Entry ![Four step flow from gene symbol search to a stable UniProt accession, ending with the EGFR example P00533.](/article-images/illustrations/uniprot-expression-annotation-cell-line-resolve-entry.webp) This part of the uniprot expression annotation cell line step by step workflow starts at the search bar on uniprot.org. Type a gene symbol, for example `EGFR`, and you will see a results list that mixes two entry classes. - **Reviewed (Swiss-Prot)** entries are manually curated by UniProt biocurators with literature support. - **Unreviewed (TrEMBL)** entries are computationally annotated and awaiting full curation. For human genes there is almost always exactly one reviewed entry per canonical protein. For EGFR that entry has accession **P00533** and entry name **EGFR_HUMAN**. Human BCL2 is **P10415**, CASP3 is **P42574**, AQP5 is **P55064**, CXCR4 is **P61073**, and BAX is **Q07812**. Learn to work with accessions, not gene symbols, because accessions are stable identifiers you can cite, share, and program against. You can do the same resolution from the command line with the UniProt REST API: ```bash curl "https://rest.uniprot.org/uniprotkb/search?query=gene:EGFR AND organism_id:9606 AND reviewed:true&fields=accession,id,gene_primary,protein_name,protein_existence&format=tsv" ``` Breaking down each parameter: | Field or option | What it does | Notes | |--|--|--| | `query=gene:EGFR` | Searches the gene name field | May return entries from multiple species unless filtered | | `organism_id:9606` | Restricts to human (NCBI taxonomy ID 9606) | Use the taxid of your organism | | `reviewed:true` | Returns only manually curated entries | Remove it to see TrEMBL entries too | | `fields=accession,id,gene_primary,...` | Selects the columns you want back | Expression comments use the `cc_` prefix shown below | | `format=tsv` | Tab separated output, easy to open in a spreadsheet | Use `format=json` for full [structured data](/blog/data-analysis/what-is-structured-data) | | `cc_tissue_specificity` | Returns the Tissue specificity annotation text | The single most useful expression field | | `cc_induction` | Returns the Induction annotation text | Records regulators of abundance | | `cc_developmental_stage` | Returns developmental expression notes | Often empty for adult tissue proteins | | `cc_subcellular_location` | Returns where the protein localizes in the cell | Paired naturally with expression reading | | `xref_bgee` | Returns the entry's cross reference to Bgee expression data | Your bridge to per tissue expression calls | The expected output for the EGFR query is a single row, which looks like this in structure (your exact wording will follow the current UniProt release): ``` Entry Entry Name Gene Names (primary) Protein Names Protein existence P00533 EGFR_HUMAN EGFR Epidermal growth factor receptor Experimental evidence at protein level ``` ## Step 2: Read the Expression Section Field by Field Open the entry page for P00533 and scroll to the Expression section. Read it in this order, because each field answers a different question. **1. Tissue specificity answers "where is this protein normally found?"** The text is a condensed literature summary with inline evidence links. Each sentence carries a small colored evidence tag. Hover over the tag to see the exact source, which may be a single paper, a proteomics dataset, or a curators' inference. Some entries even mention specific cell lines in this field when a well characterized line contributed the key evidence, for example lines used in oncology research such as the colorectal carcinoma lines HT-29 [1] and HCT-116 [2] or the breast cancer line MCF-7 [4]. **2. Induction answers "what turns this protein up or down?"** For EGFR, induction annotations typically note transcriptional regulation by ligands and other signals. This field matters enormously for cell line experimentalists. If you plan to perturb a pathway, read this field first, because the annotation may already record that your protein of interest responds to that class of signal. **3. Developmental stage answers "does expression change over development?"** For many adult housekeeping type proteins this field is absent. Its absence is information, not an error. A concrete cautionary example of why you read these fields rather than assume them comes from a study of a human fallopian tube epithelial cell line. The investigators measured Toll like receptor expression in fallopian tube tissue and in the OE-E6/E7 cell line, and found TLR1 through TLR10 expressed in the tissue but TLR1 through TLR6 expressed in the cell line [5]. If you had assumed the cell line fully represents the tissue, you would have designed experiments around receptors the line does not express. The UniProt tissue specificity field, read carefully, plus cross references to measured expression data, is what lets you catch this class of mismatch before it costs you a month of culture. ## Step 3: Judge the Evidence Before You Trust the Annotation ![Triage path that checks entry review status, protein existence level and evidence tag before an expression annotation is cited.](/article-images/illustrations/uniprot-expression-annotation-cell-line-evidence-triage.webp) Every annotation line on a UniProt entry carries an evidence tag, and learning to read these tags is the difference between citing an annotation and overciting it. - **Experimental evidence tags** mean direct measurements, such as mass spectrometry detection or immunohistochemistry, support the statement. - **"By similarity" tags** mean the statement was transferred from a homologous protein in another organism. It is plausible but not demonstrated in your species. - **Curator inference tags** mean a human curator assembled the statement from indirect evidence. - **Automatic annotation tags** on TrEMBL entries mean no curator evaluated the claim at all. Check the **Protein existence (PE) level** in the entry header. UniProt assigns five levels: 1. **PE Level 1, Experimental evidence at protein level.** The protein itself has been observed, for example by mass spectrometry or purified protein characterization. 2. **PE Level 2, Experimental evidence at transcript level.** The mRNA is demonstrated but the protein has not been directly detected. 3. **PE Level 3, Inference from homology.** Existence is inferred from a related protein. 4. **PE Level 4, Predicted.** Existence is predicted from the genome sequence alone. 5. **PE Level 5, Uncertain.** The existence of the protein is questioned. For a cell line expression project this has a direct consequence. A PE Level 1 protein whose tissue specificity is experimentally annotated is on solid ground. A PE Level 2 protein may be present in your cell line at the mRNA level but undetectable as protein, which is exactly the discrepancy that qRT-PCR studies can reveal while Western blots or proteomics cannot. When you interpret a study such as the caspase 3 induction work in MCF-7 breast cancer cells, where expression of CASP3 was quantified by real-time PCR across extract concentrations and time points, you are working at the transcript level [4]. UniProt tells you the protein is known to exist and describes its function. Your qPCR tells you what the treated cells are doing with the transcript. Both are legitimate. They are different layers. ## Step 4: Follow Cross-References to Cell Line Expression Data On any reviewed human entry, scroll to the Cross References section and find the expression databases. Three are most useful. **Bgee** provides curated reports of [gene expression](/blog/guides/gene-expression) across tissues, developmental stages, and health states, with quality scores attached to each call. It is the fastest way to see whether a gene is considered expressed in a given normal tissue. **Expression Atlas**, from EMBL-EBI, aggregates RNA-seq and proteomics experiments, including baseline tissue expression and differential expression under conditions. When your question is "does this gene change under perturbation in some system," Expression Atlas is where the cross reference leads. **[Human Protein Atlas](/blog/guides/human-protein-atlas)** maps protein localization and abundance across normal tissues, cancers, and cell lines, with antibody based evidence you can inspect image by image. The workflow habit worth building is this. UniProt gives you the curated sentence. Cross references give you the raw measurements behind similar sentences for your exact tissue or cell type. Then your own experiments give you the perturbation answer. A useful mental model for integrating all three comes from modern annotation thinking itself. Rather than relying on any single annotation, curated gene expression programs are now used to assign cell identities and activation states reproducibly across datasets [6], and single cell tools such as 3DCellAtlas formalize how expression and geometry annotations are assigned and quantified at cellular resolution [7]. The principle transfers to cell line work. An annotation you read as a program of evidence, with sources inspected, is far more useful than an annotation skimmed as a fact. ## Step 5: Retrieve Expression Fields in Bulk with the REST API Real projects never involve one gene. A typical apoptosis panel in a colorectal or breast cancer study covers BAX, BCL2, CASP3, CASP8, CASP9, TP53, and signaling receptors. UniProt lets you pull all their expression annotations in one request. ```bash curl "https://rest.uniprot.org/uniprotkb/search?query=(gene_primary:BAX OR gene_primary:BCL2 OR gene_primary:CASP3) AND organism_id:9606 AND reviewed:true&fields=accession,id,gene_primary,protein_name,cc_tissue_specificity,cc_induction&format=tsv" -o apoptosis_panel.tsv ``` Expected output shape, abbreviated: ``` Entry Entry Name Gene Protein names Tissue specificity Q07812 BAX_HUMAN BAX Apoptosis regulator BAX ... P10415 BCL2_HUMAN BCL2 Apoptosis regulator BCL-2 ... P42574 CASP3_HUMAN CASP3 Caspase-3 ... ``` Each tissue specificity cell can be long. Open `apoptosis_panel.tsv` in a text editor or spreadsheet and inspect one row at a time. To work with the structured version in Python: ```bash curl "https://rest.uniprot.org/uniprotkb/search?query=(gene_primary:BAX OR gene_primary:BCL2 OR gene_primary:CASP3) AND organism_id:9606 AND reviewed:true&format=json" -o panel.json ``` ```python import json with open("panel.json") as handle: data = json.load(handle) for entry in data["results"]: accession = entry["primaryAccession"] for comment in entry.get("comments", []): if comment["commentType"] in ("TISSUE SPECIFICITY", "INDUCTION"): for text in comment.get("texts", []): print(accession, comment["commentType"], text["value"]) ``` Expected output is one line per annotation block, for example an accession followed by `TISSUE SPECIFICITY` and the annotation sentence. If a gene has no induction annotation, nothing prints for that type, which is the correct behavior, not a bug. Common errors and their exact fixes at this step: - **HTTP 400 with a query syntax error message.** Your parentheses or Boolean operators are unbalanced. UniProt's query parser is strict about `AND` and `OR` being uppercase. - **HTTP 404, entry not found.** You asked for an accession that does not exist or was merged. Merged accessions usually redirect, but deleted ones do not. Verify the accession on the website. - **An empty tsv column.** You misspelled a field name, for example `cc_tissuespecificity` instead of `cc_tissue_specificity`. Field names use underscores and the `cc_` prefix for comment types. - **Zero rows returned.** You combined filters that exclude everything, such as `reviewed:true` for a gene that only has TrEMBL entries in your organism. Drop `reviewed:true` and re-run. - **Slow or failed requests when looping.** Do not query one gene at a time in a loop. Combine genes into a single OR query or download by organism. Thousands of individual rapid requests can be throttled. ## Step 6: Put Annotations Side by Side with Your Own Cell Line Data This is where the uniprot expression annotation cell line workflow earns its keep. Suppose you are following up on work showing that mesenchymal stem cell derived exosomes reduce expression of Aquaporin 5 and EGFR in HCT-116 colorectal carcinoma cells, measured by real-time PCR with relative expression calculated by the 2^-ΔΔCt method [2]. You want to plan a similar experiment and interpret it against what UniProt already knows about EGFR. The standard calculation works like this, with illustrative numbers: ``` Delta Ct (target) = Ct(target) - Ct(reference gene) Delta delta Ct = Delta Ct(treated) - Delta Ct(control) Fold change = 2^(-Delta delta Ct) ``` Worked example, clearly illustrative: ``` Control: Ct(EGFR) = 25.0, Ct(ACTB) = 18.0 -> Delta Ct = 7.0 Treated: Ct(EGFR) = 26.5, Ct(ACTB) = 18.0 -> Delta Ct = 8.5 Delta delta Ct = 8.5 - 7.0 = 1.5 Fold change = 2^(-1.5) = 0.35, roughly a 65 percent reduction ``` Now the reconciliation with UniProt. The entry for human EGFR, P00533, annotates the protein's normal tissue distribution, its function, its signaling interactions, and regulators of its abundance. Your fold change of 0.35 under exosome treatment is a condition specific measurement that no baseline annotation would predict. What UniProt gives you is the frame around the result: which domains the receptor uses, what pathways it feeds, and which of its known regulators resemble the perturbation you applied. The same pattern runs through the cell line expression literature. In the HT-29 study, treatment shifted Bax to 1.48 and Bcl-2 to 0.45 relative to controls, an increase in the pro-apoptotic gene and a decrease in the anti-apoptotic one [1]. In AGS gastric adenocarcinoma cells, saffron extract treatment was used to probe Bcl-2 and Bax expression by real-time PCR alongside MTT cytotoxicity testing [8]. In Caco-2 cells, cerium oxide nanoparticles combined with fluorouracil lowered the 5-FU IC50 from 101 to 71 micrograms per milliliter after 72 hours and shifted a panel of apoptosis and chemokine genes by qRT-PCR [9]. In NALM6 acute lymphoblastic leukemia cells, a bioinformatics guided workflow combined Cytoscape network analysis, RNAhybrid target prediction, and pathway analysis to interpret microRNA changes under parthenolide and vincristine treatment [10]. None of those condition specific numbers belong in UniProt. What UniProt contributes to each project is the verified identity, function, and baseline biology of every gene in the panel, which is exactly what you need before designing primers, choosing reference genes, or interpreting fold changes. ## How Published Cell Line Studies Relate to UniProt Annotations Three research level points deserve their own space because they change how you read every expression annotation in a cell line context. **Culture conditions change expression, so baseline annotations are not transferable defaults.** A study of MDA-MB-231 triple negative breast cancer cells showed that moving cells into three dimensional collagen matrices changed cell state and migration related gene expression, driving a mesenchymal to amoeboid transition [11]. In CHO cell lines, recombinant protein expression and production stability shift over long-term culture, which is a central concern in bioprocess work [12]. Transfection method itself changes what you can measure. In Vero cells, comparison of chemical transfection, electroporation, and lentiviral transduction identified very different efficiency profiles, with a cationic polymer reagent performing best under specific DNA and reagent amounts [13]. Every one of these variables sits upstream of the numbers you measure, while UniProt annotations sit above all of them. **Cell lines are not tissues, and different lines are not each other.** The TLR mismatch between fallopian tube tissue and the OE-E6/E7 line [5] makes the point within one study. Across lines, multi omics characterization of breast cancer and B cell reference materials found cell line specific expression patterns tied to chromatin accessibility and methylation landscape [3]. Marker based subpopulation work in HepG2, isolating CD133 positive cancer stem-like cells by magnetic-activated cell sorting, spheroid culture, or drug resistant selection, shows that even within one line the population you select determines the expression profile you measure [14]. When you cite a UniProt tissue specificity annotation in a cell line paper, cite it as context, never as a prediction of what your line expresses. **Downstream analysis pipelines need exactly the identifiers UniProt provides.** Modern drug synergy prediction models project cell line expression profiles into a latent space to pair them with drug representations [15]. Those pipelines are only as good as their gene identifiers, and UniProt accessions are the stable backbone that keeps a 2026 analysis comparable to a 2019 dataset. Any time your pipeline ingests expression matrices, mapping every gene to a UniProt accession before analysis is cheap insurance against silent ID mismatches. ## Common Mistakes and Limitations These are the mistakes that actually show up in manuscripts, reviews, and reproducibility checks, phrased so you can check yourself against each one. 1. **Reading Tissue specificity as a cell line truth.** The annotation describes the protein across the organism. It does not certify what your particular line expresses under your particular conditions. 2. **Confusing transcript level measurements with protein level annotations.** qRT-PCR fold changes measure mRNA. UniProt PE Level 1 evidence and subcellular location annotations concern protein. The two layers can disagree, and the disagreement is often the interesting biology. 3. **Trusting unreviewed entries as established fact.** TrEMBL records are computational. For expression claims, use reviewed entries or traced evidence. 4. **Ignoring evidence tags.** A tissue specificity sentence tagged "By similarity" describes a homolog, not your species. A sentence tagged experimental evidence can be traced to a paper you can read. 5. **Treating missing annotation as absent expression.** UniProt annotates what is documented. A quiet Expression section means the literature has not established a statement, not that the protein is absent from your cells. 6. **Using gene symbols as identifiers.** Symbols are recycled and ambiguous across species. Resolve to accessions first, every time. 7. **Forgetting isoforms.** Many entries describe multiple isoforms with different expression patterns. If your primers or antibody target a specific isoform, read the isoform specific annotations, not only the canonical block. 8. **Assuming cell line annotations are stable over time.** Lines drift with passage number and adaptation. Long-term culture effects on expression and stability are documented enough that lab notebooks tracking passage number are standard practice [12]. 9. **Citing UniProt without a version or date.** Annotation changes between releases. Record the release version and access date when you cite an entry. The standing limitation of this whole workflow is that annotations and databases describe populations of evidence, while your bench results are particular to your cells, your passage, and your treatment. Individual experimental cases need judgment from someone with hands on the system. ## Practical Tips for Faster, Cleaner Work - **Save your queries as URLs.** A REST search URL is both a command and a bookmark. Store the panel query next to your analysis code so the exact fields you retrieved are documented. - **Add `reviewed:true` reflexively** for human work, then relax it only when you deliberately want predicted entries. - **Retrieve `cc_subcellular_location` alongside expression fields.** If a qPCR positive gene encodes a protein annotated to the plasma membrane, that predicts surface protein assays are feasible, and marker based sorting strategies such as those used for CD133 in HepG2 depend on exactly that [14]. - **Check the Induction field before perturbation experiments.** It frequently names signaling classes relevant to treatments like the hormone combinations studied in the fallopian tube line [5], saving you from redundant literature searches. - **Use cross references as your expansion pack.** From Bgee and Expression Atlas you get quantitative calls. From there, expression program based annotation approaches [6] give you a framework for assigning states to your own single cell or bulk data. - **Record the UniProt release version** shown at the bottom of the entry page whenever you screenshot or cite an annotation. ## Frequently Asked Questions ### Does UniProt annotate expression separately for each cell line? No. UniProt annotates protein level knowledge such as tissue specificity, developmental stage, and induction, aggregated across documented evidence. Cell line specific measurements live in linked expression databases and in the primary literature, and cell line mentions can appear inside annotation text when a well known line provided the evidence. ### What does the Tissue specificity field actually mean? It is a curated summary of where in the organism the protein is normally found, with evidence tags you can click to reach the underlying source. It describes the species level baseline, not any particular cell line or condition. ### What is the difference between reviewed and unreviewed UniProt entries? Reviewed entries, the Swiss-Prot set, are manually curated with literature based evidence and evidence tags. Unreviewed TrEMBL entries carry computational annotations. For expression decisions, use reviewed entries unless you have a specific reason not to. ### What are UniProt protein existence levels? They rate how well the protein's existence is established, from Level 1 with direct protein level experimental evidence, through Level 2 transcript level evidence and Level 3 homology inference, down to Level 4 predicted and Level 5 uncertain. The level tells you what kind of data you can legitimately expect. ### How do I download expression annotations for many genes at once? Use the REST search endpoint with an OR query across `gene_primary` fields, set `fields` to include `cc_tissue_specificity` and related comment fields, and request `format=tsv`. One request per panel beats hundreds of single gene requests. ### My qPCR shows expression in my cell line but the UniProt entry has no expression annotation. What does that mean? The two findings are compatible. UniProt annotates documented baseline knowledge, while your measurement is a condition specific observation. An absent annotation reflects the state of published evidence, not the state of your cells. ### Why can mRNA expression differ from protein annotation in UniProt? Transcript abundance and protein abundance are governed by different rates, including translation efficiency and protein degradation. A gene can be transcriptionally induced while its protein stays stable or vice versa, which is why UniProt separates transcript level and protein level evidence explicitly. ### How should I cite a UniProt expression annotation? Cite the entry by accession, record the UniProt release version and the date you accessed it, and where possible trace the annotation's evidence tag back to the primary source it points to before you rely on the claim. ## Sources 1. [The Effect of Artemisia sieberi (Dermeneh Dashti) Plant Extract on the Expression of Genes Involved in the Apoptosis Signaling Pathway in the HT_29 Cell Line](https://doi.org/10.5812/gct-145547) 2. [Mesenchymal stem cells- derived exosomes inhibit the expression of Aquaporin-5 and EGFR in HCT-116 human colorectal carcinoma cell line](https://pubmed.ncbi.nlm.nih.gov/36114463/) 3. [Epigenomic, transcriptomic, and proteomic characterization of breast cancer cell line reference samples](https://pubmed.ncbi.nlm.nih.gov/41539304/) 4. [Effect of Hydroalcoholic Extract of Ephedra major, Momordica charantia, and Resveratrol on Cytotoxicity and Caspase-3 Genes Expression Level in MCF-7 Breast Cancer Cell Line](https://doi.org/10.5812/GCT.110658) 5. [The Effect of Estradiol and Progesterone on Toll Like Receptor Gene Expression in A Human Fallopian Tube Epithelial Cell Line](https://pubmed.ncbi.nlm.nih.gov/26862527/) 6. [Reproducible annotation of T cell subsets and activation states with gene expression programs](https://pubmed.ncbi.nlm.nih.gov/40903642/) 7. [In Silico Methods for Cell Annotation, Quantification of Gene Expression, and Cell Geometry at Single-Cell Resolution Using 3DCellAtlas.](https://pubmed.ncbi.nlm.nih.gov/27864762/) 8. [Effect of Saffron Extract on Expression of Bax and Bcl-2 Genes in Gastric Adenocarcinoma Cell Line (AGS)](https://doi.org/10.5812/GCT.63608) 9. [Anticancer Effect of Fluorouracil and Gum-Based Cerium Oxide Nanoparticles on Human Malignant Colon Carcinoma Cell Line (Caco2)](https://pubmed.ncbi.nlm.nih.gov/37038699/) 10. [Bioinformatics-Guided Discovery of miRNAs Involved in Apoptosis Modulated by Parthenolide Combined with Vincristine in The NALM6 Cell Line](https://pubmed.ncbi.nlm.nih.gov/38459731/) 11. [Three-dimensional cell culture conditions promoted the Mesenchymal-Amoeboid Transition in the Triple-Negative Breast Cancer cell line MDA-MB-231](https://pubmed.ncbi.nlm.nih.gov/39156975/) 12. [Improved CHO Cell Line Stability and Recombinant Protein Expression During Long-Term Culture.](https://pubmed.ncbi.nlm.nih.gov/28493127/) 13. [Comparing chemical transfection, electroporation, and lentiviral vector transduction to achieve optimal transfection conditions in the Vero cell line](https://pubmed.ncbi.nlm.nih.gov/38741034/) 14. [Isolation of cancer stem-like cells from hepatocellular carcinoma cell line HepG2 by methods of magnetic-activated cell sorting, spheroid culture, and anti-tumor drug-resistant selection: A primary evaluation](https://doi.org/10.15419/psc.v7i1.407) 15. [Predicting synergistic drug combinations via hierarchical molecular representation and cell line latent space fusion](https://pubmed.ncbi.nlm.nih.gov/40714415/)