Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression

Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression Étienne Becht Nicolás A. Giraldo Laetitia Lacroix Bénédicte Buttard Nabila Elarouci Florent Petitprez Janick Sèlves Pierre Laurent‐Puig Catherine Sautès‐Fridman Wolf H. Fridman Aurélien de Reyniès

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression 2016 Genome biology 10.1186/s13059-016-1070-5 becht2016-estimating-the-population-abundanc becht2016-estimating-the-population-abundanc.txt, the project's staged copy ed2eeb4668d42992701ec9858c0b4a37e6d88139 cc-by genomics

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them how much of each immune and stromal cell population is present in a tissue sample, read out of a bulk transcriptome - and what kind of comparison the resulting number licenses Quotation withheld because this record is not cleared for redistribution. whether markers exist that are expressed in one and only one cell population, and whether their average expression tracks that population's abundance well enough to compare samples Quotation withheld because this record is not cleared for redistribution. MCP-counter - ten populations scored from data-driven transcriptomic markers selected against curated negative controls, validated against in vitro RNA mixtures and against immunohistochemical cell densities, and explicitly scoped to inter-sample comparison in arbitrary units Quotation withheld because this record is not cleared for redistribution. method Quotation withheld because this record is not cleared for redistribution. a bulk transcriptome averaging signals from every cell type present, so that the composition of the tissue is recoverable only through markers that belong to one population and nothing else Quotation withheld because this record is not cleared for redistribution. system-or-tool-building system-or-tool-building Quotation withheld because this record is not cleared for redistribution. if a gene is expressed in exactly one cell type and at a stable level within it, then its expression in a mixture is proportional to how much of that cell type is there - no unmixing, no reference matrix, no dependence on what else is in the sample. The whole method is that one property, and the paper says such genes are not guaranteed to exist a marker that belongs to one population and nothing else Quotation withheld because this record is not cleared for redistribution. a marker qualifies by separating positive from negative samples, so a population with no purified counter-example in a public repository cannot be given markers at all. Effector memory CD4 T cells are named as a casualty: they would need central memory CD4 and effector memory CD8 controls, and those were not available the controls decide the vocabulary Quotation withheld because this record is not cleared for redistribution. 21 cancer types among the negative controls make the markers portable across tissues and remove any marker that any of those tissues expresses. NCAM1/CD56, a standard NK marker, is discarded because nervous malignant cells carry it - which protects every other tissue and gives up NK quantification in brain breadth of controls costing specificity Quotation withheld because this record is not cleared for redistribution. the score is a log2 mean of marker expression, so it moves with abundance and has no absolute meaning. Two consequences the paper states: the units depend on the measurement platform, and mapping them onto percentages would require calibrating samples of known composition - which is the step aran2017 later takes a scale with no zero and no unit Quotation withheld because this record is not cleared for redistribution. the scores are in arbitrary units that depend on the gene expression platform, so only samples produced on the same platform can be compared Quotation withheld because this record is not cleared for redistribution. transcriptomic markers are not guaranteed to exist for a given population, because the definition requires expression in one and only one cell type Quotation withheld because this record is not cleared for redistribution. populations without publicly available negative controls were dropped from the method regardless of their interest Quotation withheld because this record is not cleared for redistribution. the breadth of the control set discards markers that would work in a restricted setting - NCAM1 is lost because nervous malignant cells express it Quotation withheld because this record is not cleared for redistribution. spatial localisation is lost relative to immunohistochemistry, and histological confirmation may be needed where contamination by surrounding tissue is unavoidable Quotation withheld because this record is not cleared for redistribution. accuracy depends on the quality of the expression summary, and the populations estimated are usually rare, so high sequencing depth may be required Quotation withheld because this record is not cleared for redistribution. the prognostic meta-analysis is univariate and unadjusted for stage or treatment Quotation withheld because this record is not cleared for redistribution. surprising tissue-level results are referred back to histopathological knowledge rather than defended - the authors say the method should be compared against it Quotation withheld because this record is not cleared for redistribution. the cytotoxic lymphocyte population's mRNA was not controlled in the mixture experiment, so its validation is a linear model on the other populations rather than a direct calibration Quotation withheld because this record is not cleared for redistribution. the latin square, carried from agricultural experimental design into the layout of an RNA mixture so that no two populations' concentrations are collinear Quotation withheld because this record is not cleared for redistribution. the ROC curve and the signal-to-noise ratio, carried from signal detection into the selection criteria for a gene marker Quotation withheld because this record is not cleared for redistribution. 742 tumour cell line samples from 21 non-hematopoietic cancer types included as negative controls, with no hematopoietic line included and the reason stated eliminates: a marker that is merely abundant being taken for a marker that is specific - and the exclusion of hematopoietic lines removes a control that would have looked like the positive class · outcome: held Quotation withheld because this record is not cleared for redistribution. three simultaneous thresholds per candidate feature - a log2 expression difference, an AUC for identifying positive samples, and a signal-to-noise ratio eliminates: a feature qualifying on one favourable statistic; all three must be met at once, which is what makes the marker set conservative rather than large · outcome: held Quotation withheld because this record is not cleared for redistribution. two independent validation series on different array platforms, 4804 samples, with the selection criteria re-applied to see whether the same markers come out eliminates: the marker sets being an artefact of the discovery data or of one array platform - the overlap is tested for significance rather than eyeballed · outcome: held Quotation withheld because this record is not cleared for redistribution. an in vitro RNA mixture experiment with the populations laid out in two transposed latin squares eliminates: collinearity between the populations - without the latin square, a score could track its own population or any other that happened to co-vary with it · outcome: held Quotation withheld because this record is not cleared for redistribution. immunohistochemical digital quantification of three cell densities on tissue sections from 38 colorectal tumours, compared against the corresponding scores eliminates: in vitro mixtures standing in for tissue - purified RNA mixed in a tube is not a tumour, and the ex vivo check is on real infiltrate · outcome: held Quotation withheld because this record is not cleared for redistribution. a simulation in which the within-leukocyte proportions are held constant and equal while the tumour fraction varies, run through both methods eliminates: the two methods being compared on a case that cannot separate them - this one is built so that the correct answers differ, and both give theirs · outcome: held Quotation withheld because this record is not cleared for redistribution. marker sets from two previously published methods scored on the same three microenvironment series as this one eliminates: the comparison being against the competitors' published numbers rather than against their markers on identical data · outcome: held Quotation withheld because this record is not cleared for redistribution. a limit of detection established per population against non-hematopoietic control samples eliminates: a score being read as presence at any level - below the stated fraction the method cannot distinguish the population from a sample containing none · outcome: held Quotation withheld because this record is not cleared for redistribution. the same tumour analysis repeated on three transcriptomic platforms across more than 19,000 samples eliminates: a platform-specific artefact being reported as biology, within the constraint that the units themselves are not comparable across platforms · outcome: partial Quotation withheld because this record is not cleared for redistribution. the discovery set spanning 344 different culture conditions, purification methods and cell treatments eliminates: marker genes selected because of the conditions the reference cells were grown in rather than because of the cell type · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. MCP-counter MCP-counter Quotation withheld because this record is not cleared for redistribution. transcriptomic markers transcriptomic markers Quotation withheld because this record is not cleared for redistribution. cell population cell population Quotation withheld because this record is not cleared for redistribution. negative controls negative controls Quotation withheld because this record is not cleared for redistribution. arbitrary units arbitrary units Quotation withheld because this record is not cleared for redistribution. CIBERSORT CIBERSORT Quotation withheld because this record is not cleared for redistribution. flow cytometry flow cytometry Quotation withheld because this record is not cleared for redistribution. limit of detection limit of detection Quotation withheld because this record is not cleared for redistribution. immunohistochemical immunohistochemical Quotation withheld because this record is not cleared for redistribution. what an MCP-counter score is - IN THE ABSTRACT: a robust quantification of the ABSOLUTE abundance of ten cell populations Quotation withheld because this record is not cleared for redistribution. what an MCP-counter score is - IN THE RESULTS: a measure of the INTER-SAMPLE RELATIVE abundance of one population Quotation withheld because this record is not cleared for redistribution. what an MCP-counter score is - IN THE DISCUSSION: linearly correlated with abundance, in ARBITRARY UNITS that depend on the gene expression platform, comparable only between samples run on the same platform Quotation withheld because this record is not cleared for redistribution. what the score answers - MCP-COUNTER: is there more of this cell type in sample A than in sample B Quotation withheld because this record is not cleared for redistribution. what the score answers - CIBERSORT: what share of the leukocytes in THIS sample are of this cell type. The paper's own simulation shows both answering correctly at once Quotation withheld because this record is not cleared for redistribution. critiques → methods that characterise proportions of immune populations within the leukocyte fraction while omitting the overall abundance of leukocytes in the sample Quotation withheld because this record is not cleared for redistribution. qualifies → CIBERSORT - not as wrong but as answering a different question; the paper's own simulation has it estimating stable within-leukocyte proportions correctly while total abundance changes Quotation withheld because this record is not cleared for redistribution. qualifies → its own headline - the abstract says absolute abundance and the discussion says arbitrary, platform-dependent units comparable only within a platform Quotation withheld because this record is not cleared for redistribution. extends → immunohistochemistry, which MCP-counter is said to be conceptually closest to - ten populations from one experiment, at the cost of all spatial information Quotation withheld because this record is not cleared for redistribution. MCP-counter —distinct-from→ CIBERSORT MCP-counter CIBERSORT differs in: whether the comparison licensed is between cell types inside one sample or between samples for one cell type · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether the comparison licensed is between cell types inside one sample or between samples for one cell type · the relation stated: one measures proportions within a sample, the other abundance of one population across samples, and the second pays for it in arbitrary units Quotation withheld because this record is not cleared for redistribution. MCP-counter —distinct-from→ flow cytometry MCP-counter flow cytometry differs in: which axis of comparison the measurement is built for · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): which axis of comparison the measurement is built for · the relation stated: flow-cytometry-inspired methods describe the relative proportions of populations within one sample; this one is designed for the other axis Quotation withheld because this record is not cleared for redistribution. in a simulation where within-leukocyte proportions are constant and tumour content varies, MCP-counter tracks abundance and CIBERSORT accurately reports stable proportions - both are right about different questions in a simulation where within-leukocyte proportions are constant and tumour content varies, MCP-counter tracks abundance and CIBERSORT accurately reports stable proportions - both are right about different questions Quotation withheld because this record is not cleared for redistribution. transcriptomic markers are not guaranteed to exist for a population, because the definition demands expression in one and only one cell type transcriptomic markers are not guaranteed to exist for a population, because the definition demands expression in one and only one cell type Quotation withheld because this record is not cleared for redistribution. Populations lacking publicly available negative controls, with few positive samples, or with no identified markers were discarded. Populations lacking publicly available negative controls, with few positive samples, or with no identified markers were discarded. Quotation withheld because this record is not cleared for redistribution. The screening discarded NCAM1 (CD56), a common NK-cell marker, because nervous malignant cells also express it. The screening discarded NCAM1 (CD56), a common NK-cell marker, because nervous malignant cells also express it. Quotation withheld because this record is not cleared for redistribution. Scores correlated with known cell proportions in the RNA mixtures at Pearson 0.94 to 0.99. Scores correlated with known cell proportions in the RNA mixtures at Pearson 0.94 to 0.99. Quotation withheld because this record is not cleared for redistribution. the limit of detection is below 2% of total sample RNA for every population assayed the limit of detection is below 2% of total sample RNA for every population assayed Quotation withheld because this record is not cleared for redistribution. mapping the scores onto percentages would require calibrating samples of known composition, and setting-specific tuning may still be needed mapping the scores onto percentages would require calibrating samples of known composition, and setting-specific tuning may still be needed Quotation withheld because this record is not cleared for redistribution. gene expression features expressed in one and only one cell population and whose expression shows little variation within that population - a definition the authors note is restrictive enough that such markers may not exist for a given population transcriptomic markers Quotation withheld because this record is not cleared for redistribution. the scale of an MCP-counter score: linearly correlated with abundance, dependent on the gene expression platform, and comparable only between samples run on the same platform arbitrary units Quotation withheld because this record is not cleared for redistribution. the Pearson correlations between scores and known RNA proportions in the in vitro mixtures 0.94 to 0.99 Quotation withheld because this record is not cleared for redistribution. the AUC floor for every signature across 4804 validation samples 0.994 Quotation withheld because this record is not cleared for redistribution. the upper bound on the limit of detection for every population assayed - from 1/950 to 1/50 of total RNA depending on which 2 % Quotation withheld because this record is not cleared for redistribution. tumour cell line samples included as negative controls, from 21 non-hematopoietic cancer types 742 Quotation withheld because this record is not cleared for redistribution. AUC AUC Quotation withheld because this record is not cleared for redistribution. Pearson Pearson Quotation withheld because this record is not cleared for redistribution. arbitrary units arbitrary units Quotation withheld because this record is not cleared for redistribution. limit of detection limit of detection Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1186/s13059-016-1070-5.