Multiplexed single-cell transcriptional response profiling to define cancer vulnerabilities and therapeutic mechanism of action

Assays to study cancer cell responses to pharmacologic or genetic perturbations are typically restricted to using simple phenotypic readouts such as proliferation rate. Information-rich assays, such as gene-expression profiling, have generally not permitted efficient profiling of a given perturbation across multiple cellular contexts. Here, we develop MIX-Seq, a method for multiplexed transcriptional profiling of post-perturbation responses across a mixture of samples with single-cell resolution, using SNP-based computational demultiplexing of single-cell RNA-sequencing data. We show that MIX-Seq can be used to profile responses to chemical or genetic perturbations across pools of 100 or more cancer cell lines. We combine it with Cell Hashing to further multiplex additional experimental conditions, such as post-treatment time points or drug doses. Analyzing the high-content readout of scRNA-seq reveals both shared and context-specific transcriptional response components that can identify drug mechanism of action and enable prediction of long-term cell viability from short-term transcriptional responses to treatment.


Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above.

Software and code
Policy information about availability of computer code Data collection

Data analysis
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability

James McFarland
Jun 23, 2020 No specialized software was used for data collection 10x single-cell RNA-Sequencing data were processed using Cell Ranger software (v2 and v3). Classification of single cell identities (based on SNPs and gene expression profiles) was done using custom analysis code available at: https://github.com/broadinstitute/ single_cell_classification. This analysis utilized the following R packages: glmnet v3.0.2, mclust v5.4.5, DropletUtils v1.6.1, vcfR v.1.9.0 Code to perform the remaining analysis in the paper, as well as generating the figures, is available at https://github.com/broadinstitute/ mix_seq_ms. This analysis used the following R packages: Seurat v3. Validation DepMap_19Q3_Public/9201770/2]. The cell line drug sensitivity data was taken from the Sanger GDSC dataset 4,7, which is available for download from depmap.org or https://www.cancerrxgene.org/, and data generated using the PRISM multiplexed drug screening platform 17,20, which is available for download from depmap.org. The L1000 gene expression signatures were taken from either the LINCS Phase 2 data (GEO accession GSE70138, downloaded from http:// amp.pharm.mssm.edu/Slicr) or LINCS Phase 1 data (GEO accession GSE92742, downloaded from clue.io).
We aimed to recover 100 or more single-cell profiles per cell line and condition in our initial perturbation-response experiments, reasoning that this would be sufficient for robustly estimating average transcriptional responses, as well as allowing analyses of population heterogeneity. The number of cell lines used in each experiment varied from 24 to 99, with larger numbers of cell lines increasing power for analyses of the variation in responses across cell lines. We apply downsampling analyses to show how both cell/cell line, and cell line sample sizes affect results.
Gene expression profiles of individual cells were excluded from analysis if they did not pass quality-control measures, as described in the manuscript. These were: cells where the proportion of UMIs from mitochondrial genes was < 0.01 or > 0.25; cells identified as putative empty droplets based on clustering analysis of the SNP model fit stats across cells; cells identified as doublets using the SNP model; and cells that did not have a sufficiently high confidence in the assignment to a particular cell line. Additionally, two experiments were excluded from analysis. One using a dose of everolimus that we determined was too low (everolimus treatment was repeated). The second treatment (with the CDK7i THZ-2-102-1) was too toxic to the cells at the dose and time point used to recover usable transcriptional profiles. These exclusion criteria were not pre-established.
Robustness of our results is facilitated by analyzing many individual cells per sample/condition, and also many cell lines per treatment. Additionally, transcriptional response to trametinib (24 hours post-treatment) were measured 3 times in separate experiments (in different cell line pools), and the results showed good agreement across experiments.
For analyses of perturbation responses (drug and CRISPR perturbations), treatment and control responses were measured for all cell lines, and there was no need to assign samples into control or treatment groups.
Blinding was not relevant to the study because all samples were measured in both treatment and control groups.