Dataset DOI: 10.5061/dryad.stqjq2cjc

Description of the data and file structure

This dataset contains imaging, sequencing, and cytometry data, as well as code used to generate, analyze, and plot these data, published in two trained immunity manuscripts. There are three folders: "Extraction_Scripts", "Data", and "Plot_Scripts".

"Extraction_Scripts" contains R scripts and markdown files detailing data cleanup and extraction for ATAC, imaging, and scRNAseq data. See details below concerning individual files.

"Data" contains subfolders organized by data type: chromVAR, DEseq, and TOBIAS subfolders relate to ATACseq data analyses, scRNAseq subfolder contains single-cell RNAseq data from GEO:184241 (originally published in Kong et al 2021 Cell Reports) in a ggplot-friendly format. Flow subfolder contains MFI of targets of interest analyzed by FlowJo. FISH and IF folders contain imaging data analyzed on NimbusImage. See details below concerning individual files.

"Plot_Scripts" contains code used to generate each plot in the associated manuscript, organized by subfolders Specificity_Paper for the manuscript Human Macrophages Encode Stimulus-Specific Information of Prior Exposures Through Trained Immunity, and SingleCell_Paper for the manuscript Quantitative Cytokine Profiling of Primary Human Macrophages Reveals Distinct Single-Cell Modes of Trained Immunity. Scripts within each subfolder are organized by figure ("Fig1.Rmd" , "SupFig1.Rmd" , etc). See details below concerning individual files.

Files and variables

File: Plot_Scripts.zip

Description: Code used to generate figures are found in "Plot_scripts." Uploaded files contains data used in both manuscripts; plot scripts in the folder "Specificity_Paper" are used to generate figures for the manuscript Human Macrophages Encode Stimulus-Specific Information of Prior Exposures Through Trained Immunity, while those in the folder "SingleCell_Paper" are used to generate figures for the manuscript Quantitative Cytokine Profiling of Primary Human Macrophages Reveals Distinct Single-Cell Modes of Trained Immunity. Scripts are organized by figure ("Fig1.Rmd" , "SupFig1.Rmd" , etc). Below is a brief description of the findings of the plots generated by each file: 

Within Specificity_Paper

• Fig1.Rmd: Here we introduce that trained immunity in human monocyte-derived macrophages can generate stimulus-specific phenotypes, by comparing untrained cells, cells trained with fungal beta-glucan (BG), and cells trained with bacterial muramyl dipeptide (MDP). We show that cells do not express proinflammatory cytokines before restimulation by RNA FISH, but rapidly mount a response to restimulation in a manner dependent upon both the inducing stimulus and the secondary stimulus. 
• Fig2.Rmd: We next generate ATACseq data on human monocyte-derived macrophages that are untrained, BG-trained, or MDP-trained. We show specificity in differentially accessible regions by DESeq and GO analysis, and perform motif analysis with chromVAR and TOBIAS.
• Fig3.Rmd: We next perform immunofluorescence for pp65 and pSTAT1 on these populations before and after restimulation, demonstrating that trained cells retain residual active signaling factors and mount a stronger signaling response to restimulation compared to untrained cells.
• Fig4.Rmd: We compare ATACseq samples taken 6 days or 11 days after training, to demonstrate that memory of BG training leaves more durable epigenetic changes than memory of MDP training. 
• Fig5.Rmd: We extend the analysis of specificity beyond BG and MDP, comparing those memory states to that induced by interferon gamma (IFNg), comparing their transcriptional, functional, and epigenetic characteristics. 
• SupFig1.Rmd: Statistical analyses for the data shown in Fig1.Rmd, including statistical tests for significance for each donor. 
• SupFig2.Rmd: ATACseq analysis from Fig2.Rmd split by donor, as well as additional GO analyses on the ATACseq dataset.
• SupFig3.Rmd: Comparison of untrained cells on Day 6 vs Day 11 of culture by ATACseq, and immunofluorescence analyses of pSTAT1 and pp65 on Day 11. 
• SupFig4.Rmd: Statistical analyses for the data shown in Fig5.Rmd, including statistical tests for significance for each donor, as well as immunofluorescence analyses of pSTAT1 and pp65 in IFNg-trained cells, and motif analysis using chromVAR of ATACseq data taken on Day 6 vs Day 11 in IFNg-trained cells. 
• SupFig5.Rmd: Motif enrichment for ATACseq data from a large panel of training stimuli (BG, MDP, CpG, oxLDL, PolyIC, IFNg, TGFb), and a sample batching analysis demonstrating minimal batch effects across ATACseq samples collected on different dates.

Within SingleCell_Paper

• Fig1.Rmd: We demonstrate single-cell properties of macrophage transcriptional response to restimulation that change with trained immunity. For some genes (TNF, IL1b), almost every cell in the population transcribes the gene 2 hours after restimulation with LPS, while only some cells transcribe other genes (IL6, CXCL10). Memory manifests as an increase in the number of transcripts per cell for TNF, IL1b, and CXCL10, and as an increase in the number of expressing cells for IL6 and CXCL10. 
• Fig2.Rmd: We compare the individual cells expressing IL6, CXCL10, neither, or both, 2 hours after restimulation with LPS. Because the odds ratio of expression of the two genes does not change with training, we conclude there is not an elite subpopulation of trainable cells.
• Fig3.Rmd: We now compare the transcriptional response to restimulation at early (2hrs) and late (6hrs) timepoints. Although trained cells have greater gene  expression at early timepoints, they have lower gene expression at later timepoints, suggesting training may increase the speed but not the magnitude of the transcriptional response. This finding is supported by scRNAseq analysis from cells trained in vivo with the BCG vaccine, originally published in Kong et al 2021. 
• Fig4.Rmd: We define a metric called the "training capacity" to define how much an individual donor's transcriptional response changes with memory. This value was highly correlated across stimuli and genes, and was inversely correlated with the donor's baseline (untrained) state. Capacity appears to be driven by transient properties, as the same donor shows different capacity at different collection dates.
• SupFig1.Rmd: TNF and IL1b expression is uniformly high across populations 30 minutes after stimulation, and cell morphology is linked to transcriptional state (cell area is correlated with spot count, and cells that are smaller and more elongated are more likely to express IL6 or CXCL10). 
• SupFig2.Rmd: Transcriptional data shown in Fig1.Rmd, split by individual donor, with statistical calculations. 
• SupFig3.Rmd: Expression across genes is uncorrelated in single cells, and the odds ratio between IL6 and CXCL10 expression in single cells at 6hrs remains the same between trained and untrained populations, reinforcing that there is not an elite subpopulation of trainable cells.
• SupFig4.Rmd: Transcriptional data shown in Fig3.Rmd, split by individual donor, with statistical calculations. Relationships between cell density and gene expression, and between TNF and IL6 expression, at 6hrs post restimulation. Additional genes from the scRNAseq analysis, in further support of claims made in Fig3.Rmd. 
• SupFig5.Rmd: Donor training capacity does not appear linked to age, sex, or time of year of sample collection. IL1b training capacity is negatively correlated with baseline expression, but TNF capacity is not. 

File: Extraction_Scripts.zip

Description: Code used to analyze data can be found in "Extraction_scripts". All scripts are in R. Below is a brief description of each file:

• ATAC_1_MapandClean.Rmd: ATACseq data was collected from human moncyte-derived macrophages that were either unstimulated, trained with beta-glucan, or trained with MDP, on the first day of culture then rested for five days, with samples collected on day 6. This file details FASTQ alignment, duplicate removal, merging of technical replicates, and filtering to short fragments, to aid in peak calling. Raw data can be found at GEO: GSE323311
• ATAC_2_PeakCalling.Rmd: Using the files generated in the above scripts, peaks were called using MACS2. A final nonredundant peaks list, with counts from non-replicate-merged, non-fragment-size-filtered files was generated for use in further analyses. 
• ATAC_3_DESeq.Rmd: Differentially accessible regions were identified by DESeq to determine specificity of open chromatin state across conditions. Once GO analysis on these significant regions was conducted using GREAT (great.stanford.edu), the files were read back in for further analysis, including region annotation.
• FISH_1_cleannimbusfiles.R: RNA FISH data of gene expression in human monocyte-derived macrophages were analyzed using NimbusImage (https://nimbusimage.com/), while the software was in beta testing version. As such, the individual data files contained inconsistent names and unnecessary columns. This file details the data cleaning procedures, as well as the addition of metadata associated with each experiment.
• FISH_2_mergefiles.R: RNA FISH data files cleaned in the above script were merged by experiment type, to generate the final merged files found in Data > FISH (details below). 
• FISH_3_calculatecapacity.R: We define a metric called the "donor capacity", relating to how much an individual donor's gene expression changed due to memory. This script details the calculation for this metric, and generates the capacity.csv file found in Data > FISH
• scRNAseq_GSE184241.Rmd: The raw data from GSE184241 contains scRNAseq data of monocytes from humans before and after receiving the BCG vaccine, stimulated ex vivo with LPS. We re-analyzed this data using the methods described in the original manuscript (Kong et al 2021 Cell Reports), and generated a filtered dataframe (GSE184241_toplot.csv, found in Data > scRNAseq) to aid plotting with ggplot2.  

File: Data.zip

Description: "Data" contains subfolders organized by data type. See details below concerning individual files.

Within chromVAR:

• Each file contains differential zscore calculations against the untrained control background for individual transcription factors. Each column is a transcription factor (defined by the column name), and each row is the z-score value for "BG" (beta-glucan trained sample), "NT" (untrained control sample), or "MDP" (muramyl-dipeptide trained sample). Files were generated using the code found in Plot_Scripts > Specificity_Paper > Fig2.Rmd, Fig4.Rmd, and Fig5.Rmd, and are titled by section. motif_zscores_a.csv, motif_zscores_b.csv, and  motif_zscores_c.csv reflect data shown in Fig2.Rmd for Donors ND410b, ND650, and ND500b respectively. motif_zscores_a_ifn.csv, motif_zscores_b_ifn.csv, and  motif_zscores_c_ifn.csv is inclusive of the IFNg condition as shown in Fig5.Rmd. motif_zscores_d6.csv and motif_zscores_d11.csv compare cells from day six vs day 11 of culture from Donor ND410b, as shown in Fig4.Rmd. 

Within DESeq:

• ReadCutoff20 reflects the data extracted in ATAC_3_DESeq.Rmd and plotted in Fig2.Rmd (a cutoff of 20 reads per peak was selected to avoid false positives at regions with low coverage). DESeq results files are named as "Condition_res_Donor.csv" (for example, bg_res_A.csv is the DESeq results file comparing BG-trained vs untrained cells, for Donor ND410b. These results were further filtered to generate CSV and BED files for regions shared across at least 2 of the 3 donors (bg_sig_two.bed, etc), and those which gained accessibility (bg_sig_two_UP.bed, etc) or lost accessibility (bg_sig_two_DOWN.bed, etc) due to training. Code used to generate these files is found in ATAC_3_DESeq.Rmd. These regions were then used for GO analysis using GREAT (great.stanford.edu), the results of which can be found in the GO_up and GO_down folders. Within GO_up and GO_down, files are labeled by condition (for example, GO_bg_only_up.csv contains GO hits for regions differentially opened only in BG-trained cells. GO_both_up.csv contains GO hits for regions differentially opened in both BG-trained and MDP-trained cells. "Labeled" CSVs contain manual category labels added to the dataset. 
• D6vD11 contains DESeq analyses from Fig4.Rmd, comparing cells trained 6 days or 11 days ago, for Donor ND410b. These files contain a similar naming structure: "Condition_res_Day.csv" (for example, bg_res_d11.csv is the DESeq results file comparing BG-trained vs untrained cells on day 11). These files were used to determine which regions of differentially accessible chromatin on day 6 remained differentially accessible on day 11, generating "Condition_retained.bed" files (for example, bg_retained.bed contains regions in BG-trained cells significantly differentially accessible on both Day 6 and Day 11 compared to untrained cells). 

Within FISH:

• These files all contain NimbusImage-analyzed RNA FISH data, cleaned using FISH_1_cleannimbusfiles.R and merged using FISH_2_mergefiles.R. 
• all_merged_2hr_LPSrestim.csv, all_merged_6hr_LPSrestim.csv, and all_merged_30min_LPSrestim.csv contain RNA FISH data for trained and untrained cells after restimulation with LPS. merged_time0.csv contains data for cells on day 6 before restimulation. merged_2hr_ZymovsLPS.csv contains RNA FISH data comparing restimulation with LPS vs zymosan, as shown in Plot_Scripts > SpecificityPaper > Fig1.Rmd. merged_IL1b_CXCL10_2hr_fullmorpho.csv and merged_TNF_IL6_2hr_fullmorpho.csv contain RNA FISH data with additional cellular morphology characteristics that are not present in other files, for morphology analysis in Plot_Scripts > SingleCellPaper > SupFig1.Rmd. merged_nd410_3dates.csv, capacity.csv, and donor_metadata.csv concern donor characteristics related to training behavior, shown in Plot_Scripts > SingleCellPaper > SupFig5.Rmd and Fig4.Rmd. merged_combo.csv contains analysis of cells given both MDP and BG during the first 24 hours of culture, shown in Plot_Scripts > SpecificityPaper > Fig1.Rmd. merged_d11.csv contains analysis of cells on day 6 vs day 11 of culture, shown in Plot_Scripts > SpecificityPaper > Fig4.Rmd. Finally, merged_timecourse_nd410a.csv and merged_timecourse_nd608.csv contain analysis of transcription across the full timecourse following restimulation (0-8 hours), shown in Plot_Scripts > SingleCellPaper > Fig3.Rmd
• Within the merged files, columns Donor, Condition, Time, and Stim detail the experimental conditions (for instance, data from a beta-glucan trained cell from Donor ND224 2hrs after restim with LPS would have "ND224" in the donor column, "BG" in the condition column, "2hr" in the time column, and "LPS" in the stim column). Columns Area, Perimeter, Eccentricity, Rectangularity, etc relate to the morphology of the cell annotation. XY refers to the image plane the cell came from (for instance, if 20 images of 20 different places in the well plate were taken, XY will range from 1-20 for that well). TNF, IL6, IL1b, and CXCL10 columns are for the four genes assessed in these experiments by RNA FISH, with values in the column reflecting the number of spots within the borders of the cell annotation, identified by the machine learning algorithm Piscis (these values are typically area-normalized within the plot scripts). If the value is NA, this probe was not included in the analysis of that particular cell. If the value is 0, the probe was included but there were no spots identified in that cell. 

Within Flow:

• Cells from 5 donors were assessed for Dectin1 and TLR4 expression by flow cytometry, with data analyzed in FlowJo. flow_mfi.csv contains MFI (mean fluorescence intensity) of the receptor-targeting antibody, across donor, train condition (including an unstained control sample without antibody), and receptor.

Within IF:

• Cells from 5 donors (3 on Day 6, 2 on Day 11 of culture) were assessed for nuclear pp65 and pSTAT1 fluorescence by IF. Images were analyzed on NimbusImage. Files are named as "Day_IF_Donor.csv" (for instance, d6_IF_nd410d.csv contains cells from Donor ND410d, analyzed after six days of culture. These files came directly from NimbusImage, meaning they still contain unnecessary columns including "Id" (annotation ID tag given by the software), "Channel", "Time" (both unused in this data type), "Tags", "Shape" (referring to the annotations, in this case nuclear annotations to assess nuclear fluorescence), and "Name" (unused). Plot_scripts > Specificity_Paper > Fig3.Rmd uses the median intensity measurement, found in column "Nuclei.Blob.intensity.measurements...MedianIntensity"

Within scRNAseq:

• GSE184241_toplot.csv contains the file generated in Extraction_scripts > scRNAseq_GSE184241.Rmd, generation detailed above. Each row is a cell and each column is a gene.

Within TOBIAS:

• This folder contains files generated from TOBIAS analysis of transcription factor occupancy (https://github.com/loosolab/TOBIAS). The output files are named as "Donor_Comparison.csv" (for example, A_BGvNT.csv contains transcription factor occupancy comparison for ATACseq data from BG trained cells vs untrained cells, for Donor ND410b. The output files contain columns "output_prefix", "name", "motif_id" and "cluster" that define the factor being assessed (matching to the JASPAR database), "total_tfbs" (total transcription factor binding sites for the motif), "mean_score" and "bound" for each condition, the "change" in bound factor across conditions, and the associated "pvalue". Factors are "highlighted" if they have -log10(p-value) above the 95% quantile and/or differential binding scores smaller/larger than the 5% and 95% quantiles (top 5% in each direction). From these files, transcription factors differentially bound in at least two of the three donors are found in files named as "comparison_direction_ two.csv" (for example, bgvnt_bgup_two.csv contains factors that, in the comparison between BG-trained and untrained cell chromatin, were differentially more bound in BG-trained cells in at least 2 of the three donors. 

Code/software

All code provided can be run in R, with required packages detailed at the beginning of each markdown file. 

Access information

Other publicly accessible locations of the data:

• Dropbox: https://www.dropbox.com/home/Aoife%20O'Farrell/TrainedImmunity_Papers

Data was derived from the following sources:

• For scRNAseq analysis in Quantitative Cytokine Profiling of Primary Human Macrophages Reveals Distinct Single-Cell Modes of Trained Immunity: GEO GSE184241

Human subjects data

All human participants provide explicit consent for publication of their de-identified data. Individual donors are provided an anonymous ID number during sample collection by the Human Immunology Core, and are only ever referenced by this ID number.