supplementary materials for - science · 2016-11-22 · supplementary materials for social status...
TRANSCRIPT
www.sciencemag.org/content/354/6315/1041/suppl/DC1
Supplementary Materials for
Social status alters immune regulation and response to infection in
macaques Noah Snyder-Mackler, Joaquín Sanz, Jordan N. Kohn, Jessica F. Brinkworth, Shauna
Morrow, Amanda O. Shaver, Jean-Christophe Grenier, Roger Pique-Regi, Zachary P.
Johnson, Mark E. Wilson, Luis B. Barreiro,* Jenny Tung*
*Corresponding author. Email: [email protected] (J.T.); [email protected] (L.B.B.)
Published 25 November 2016, Science 354, 1041 (2016)
DOI: 10.1126/science.aah3580
This PDF file includes:
Materials and Methods
Supplementary Text
Figs. S1 to S13
Captions for Tables S1 to S8
References
Other Supplementary Material for this manuscript includes the following:
(available at www.sciencemag.org/content/354/6315/1041/suppl/DC1)
Tables S1 to S8
SUPPLEMENTARY CONTENT
Author Contributions
Materials and Methods
Text S1. Study subjects and experimental design
Text S2. Blood sample collection
Text S3. Glucocorticoid sensitivity test
Text S4. RNA-seq data generation
Text S5. Read normalization and correction for batch effects
Text S6. Genotyping
Text S7. Modeling rank effects on gene expression
Text S8. Gene Ontology enrichment
Text S9. Meta-analysis for tissue specificity
Text S10. Behavioral and GC sensitivity mediation analysis
Text S11. Transcription factor binding site enrichment
Text S12. Deconvolution of LPS challenge data and „reconvolution” of PBMC gene
expression levels from cell type-specific data
Text S13. Effects of dominance rank on the magnitude of the response to LPS stimulation
Text S14. Rank effects in MyD88 vs TRIF induced genes
Figures S1-S13
Figure S1: Correlation between female dominance ranks across phases
Figure S2. Association between dominance rank and proportions of 11 white blood cell
populations
Figure S3: Cell sorting strategy
Figure S4: Gene Ontology enrichment in the cell type-specific analysis
Figure S5. Within-individual changes in dominance rank are reflected in within-
individual changes in the expression levels of rank-responsive genes
Figure S6: Distribution of the absolute difference in magnitude of rank effects between
LPS-stimulated and untreated samples
Figure S7. Glucocorticoid (GC) sensitivity mediation of rank effects in the LPS challenge
experiment
Figure S8: Transcription factors with putative binding sites enriched near rank-responsive
genes
Figure S9: Enrichment of LPS-responsive genes in deconvoluted cell type-specific gene
expression data.
Figure S10. Stability of body mass over time
Figure S11: Kinship among study subjects
Figure S12: LPS challenge experiment cell phenotyping strategy
Figure S13: TFBS-based prediction of rank-responsive genes
Tables S1-S8 (all Tables will be hosted online in a single *.xls file)
Table S1: Study subject information
Table S2: Antibodies and fluorescent labels used for cell sorting and phenotyping
Table S3: Number of RNA-seq samples passing quality control
Table S4: Cell type-specific rank effects for all genes
Table S5: Gene set enrichment results for cell type-specific analysis
Table S6: LPS challenge results for all genes
Table S7: Gene set enrichment results for LPS challenge experiment
Table S8: Transcription factor binding site enrichment results
AUTHOR CONTRIBUTIONS
JT, LBB, MEW, and ZPJ designed the study. NSM, JNK, AOS, SM, and JFB collected the data.
NSM, JS, JT, LBB, JCG, and RPR analyzed the data. JT, LBB, NSM, and JS wrote the paper,
with contributions and edits from all authors. JT, LBB, MEW, ZPJ, and NSM provided funding
support.
MATERIALS AND METHODS (1-25) (1-23)
Text S1. Study subjects and experimental design
Study subjects were 45 adult female rhesus macaques (ages 3 – 20; mean age at sampling
= 9.12 years ± 4.24 s.d.) housed in groups of five females each at the Yerkes National Primate
Research Center (YNPRC) Field Station (Table S1). All but three of our study subjects were
bred and reared at YNPRC as part of their large breeding colonies. Three study animals (WXC,
VBB, and 482) were transferred to YNPRC from Alpha Genesis (Yemassee, SC) as juveniles in
an effort to increase the genetic diversity of the specific pathogen free (SPF) colony at YNPRC.
Nine social groups were initially formed in January – June 2013 (Phase 1), with group
membership rearranged to form 9 new groups in March – June 2014 (Phase 2). These groups
represented the complete set of social groups we formed (i.e., no groups were initially formed
and then excluded from subsequent analysis, and all groups produced the stable, linear
dominance hierarchies expected for this species). Group construction followed a previously
established protocol (10) in which females were sequentially introduced into indoor-outdoor run
housing (25 m by 25 m for each area) over the course of 2 - 15 weeks and maintained on
unrestricted access to typical low-fat, high-fiber nonhuman primate diet. In both phases,
behavioral data collection for each group started after the fifth and final female was introduced
into the group; biological sample collection started 10-16 weeks thereafter (median=12 weeks) to
ensure that rank hierarchies were completely stabilized. We also monitored body mass for each
female on a weekly basis throughout the both phases, which remained stable for all animals (i.e.,
rank did not predict either weight gain or weight loss in either phase: Figure S10). To maximize
between-phase changes in dominance rank, each social group in Phase 2 consisted of females
who shared the same dominance rank or (due to the uneven number of social groups in our
study) were separated by a maximum of 1 ordinal rank difference in Phase 1 (Table S1).
Two major characteristics differentiated the social groups in our study from those found
in wild or free-ranging rhesus macaques. First, they were substantially smaller (5 individuals per
group versus 20 to 200 in wild/free-ranging groups) (26). Second, they were demographically
unusual: they contained no males, immature females, or infants, and no close kin (see fig. S11), a
deliberate feature of our study design that allowed us to distinguish social status effects from the
confounding effects of kinship. Nevertheless, social status dynamics in these groups are
generally consistent with those described for naturally occurring groups of rhesus macaques.
Specifically, in both our social groups and natural/free-ranging populations, (i) rank hierarchies,
once established, were linear and stable; (ii) the closest grooming relationships were formed
between adjacently ranked females, with grooming relationships more often directed up rather
than down the hierarchy; and (iii) rank relationships were regularly enforced through agonistic
interactions (e.g., threats and displacements). More information about the behavioral correlates
of dominance rank in our social groups can be found in (15).
To monitor dominance rank, we collected behavioral data using weekly focal
observations (345 total hours of observation, 223.5 hours in Phase 1 and 121.5 hours in Phase 2).
Except where noted, we assigned dominance rank using Elo ratings, a continuous measure of
rank in which higher scores correspond to higher status and ratings are updated following any
interaction in which a winner or loser can be scored (11, 27-29). Wins increase the winner‟s Elo
rating and losses decrease the loser‟s Elo rating, proportional to the expected outcome of the
interaction. Thus, when a higher ranking female wins an encounter against a low ranking female,
her Elo rating changes little because the result was expected based on her pre-interaction rating
relative to her opponent‟s. However, if the outcome were reversed, both females‟ Elo ratings
would change substantially to reflect the observed rank dynamics.
We calculated Elo ratings using all dominance interactions that occurred after the fifth
and final female was introduced into a social group. For all females, the initial Elo rating was set
to a value of 1000 and the baseline number of points a female could potentially gain or lose
during a dominance interaction (k) was set to 100. For each interaction, k was weighted by the
expected probability of an individual winning or losing, based on a logistic function that was
updated after each dominance interaction (29). In both phases, order of introduction predicted
subsequent dominance rank in all 9 groups (Phase 1: r=-0.57, p=4.1x10-5
, n=45 females; Phase 2:
r=-0.68, p=3.3x10-7
, n=45 females, based on values at the time of sampling for cell type-specific
expression analyses; Fig. 1B), and Elo ratings remained highly stable for the duration of each
phase (Elo stability index range: 0.995 – 1.00; where 1 represents a perfectly stable hierarchy in
which lower-ranking females never outcompete higher-ranking females (28)). Elo ratings were
also highly correlated with more traditional ordinal measures of dominance rank (Pearson‟s r=-
0.94, p = 6.1x10-40
(15)).
To test whether Elo ratings predicted social interactions, as expected, we assessed rates of
received harassment and grooming. Received harassment was defined as the per hour rate of
harassment received from a female‟s group mates, mean-centered across groups so that the
average rate of harassment received by females in a group was set equal to 0. Grooming rates
were defined as the amount of time per hour a female spent grooming with her group mates, also
mean-centered to 0 for each social group. We also used these values to test the mediating role of
agonistic (received harassment) and affiliative (grooming) behaviors on rank effects on gene
expression levels, for the cell type-specific gene expression analysis (see “Behavioral mediation
analysis” below).
Text S2. Blood sample collection
To measure gene expression levels in purified cell populations, we drew 12-20 mL of
blood from each female, purified the PBMC fraction using density gradient centrifugation, and
performed fluorescent activated cell sorting on a BD FACSAria IIu machine housed at the Duke
Human Vaccine Institute Flow Cytometry Core. We sorted cell types as follows: classical
monocytes (CD3-/CD20
-/HLA-DR
+/CD14
+), natural killer cells (CD3
-/CD20
-/HLA-DR
-/CD16
+),
B cells (CD3-/CD20
+/HLA-DR
+), helper T cells (CD3
+/CD8
-/CD4
+), cytotoxic T cells
(CD3+/CD8
+/CD4
-) (see fig. S3 for sorting strategy and Table S2 for antibody information).
Purified cells were immediately pelleted, lysed in buffer RLT, and frozen at -80°C until DNA
and RNA extraction using the Qiagen AllPrep extraction kit. Sorted cell populations (five
populations per blood sample) were obtained once for each study subject in each of the two study
phases (Table S3).
To test how dominance rank affects the LPS response, we drew 1 mL of whole blood
from each female (in Phase 2 only) directly into a TruCulture tube (Myriad RBM) that contained
cell culture media plus 1 μg/mL ultra-pure lipopolysaccharide (LPS) from the E. coli 0111:B4
strain (“LPS+”), and another 1 mL of blood from the same draw into a TruCulture tube that
contained cell culture media only (“control”). Samples were incubated in parallel for 4 hours at
37°C. We then separated the serum and cellular fractions, and lysed and discarded the red cells
from the remaining cell pellet by applying red blood cell lysis buffer (RBC lysis solution, 5
Prime Inc.) for 10 minutes followed by centrifugation. We lysed the remaining white blood cells
in Qiazol for storage at -80°C, and extracted total RNA from each sample using the Qiagen
miRNAEASY kit. To control for variation in cellular composition in downstream analyses, we
also used flow cytometry to quantify the proportion of 11 different cell types in blood samples
obtained in the same draw: polymorphonuclear (CD14dim
/SSC-A>100K/FSC-A>100K), classical
monocytes (CD14+/CD16
-), CD14
+ intermediate monocytes (CD14
+/CD16
+), CD14
- non-
classical monocytes (CD14-/CD16
+), helper T cells (CD3
+/CD4
+), cytotoxic T cells
(CD3+/CD8
+), double positive T cells (CD3
+/CD4
+/CD8
+), CD8
- B cells (CD3
-/CD20
+/CD8
-),
CD8+ B cells (CD3
-/CD20
+/CD8
+), natural killer T lymphocytes (CD3
+/CD16
+), and natural
killer cells (CD3-/CD16
+) (fig. S12).
Text S3. Glucocorticoid sensitivity test
To measure glucocorticoid sensitivity, we obtained two blood samples from all 45
subjects in Phase 2: one before and one after administration of the glucocorticoid receptor
agonist, Dexamethasone (Dex). The first sample (“baseline”) was collected at 0800 h
immediately followed by an intramuscular injection of Dex (0.125 mg/kg). The second sample
(“post-Dex”) was collected 4.5 h after the injection. All samples were collected in serum
separator tubes and cortisol levels were quantified at the YNPRC Biomarkers Core Laboratory
using liquid chromatography-mass spectrometry and comparison against a standard curve, as
reported in (30). Intra- and inter-assay coefficients of variation were 1.21% and 5.78%
respectively. For our measure of glucocorticoid sensitivity, we calculated the percent change in
circulating cortisol from the baseline to the post-Dex sample.
Text S4. RNA-seq data generation
Library construction. We constructed RNA-sequencing libraries for each sample
(purified cell populations in both phases and control and LPS+ samples in phase 2) using the
NEBNext Ultra RNA Library Prep Kit (New England Biolabs). Briefly, we purified the poly-
adenylated mRNA from 200 ng of total RNA using the NEBNext Poly(A) mRNA Magnetic
Isolation Module. The mRNA was then reverse transcribed into cDNA, ligated to Illumina
adapters, size-selected for a median size of ~350 bp, and amplified via PCR for 13 cycles. We
tagged each sample with a unique molecular barcode and pooled 10-12 samples per Illumina
HiSeq 2500 lane of single-end 100 bp sequencing.
Read alignment. Following sequencing, we trimmed Illumina adapter sequence from the
ends of the reads and removed bases with quality scores < 20 using Trim Galore! (v0.2.7),
mapped trimmed reads to the rhesus macaque genome (MacaM v7;(31)) using the STAR 2-pass
method (32), and collated the number of reads that mapped uniquely to each annotated MacaM
gene (v7.6.8) using HTSeq-count (v0.6.1) with the option “intersection-nonempty” (33). For the
LPS challenge experiment and each cell type-specific data set, this procedure resulted in a p x n
read-count matrix, where p is the number of genes measured and n is the number of samples in
the given data set. After excluding samples that did not produce sequenceable libraries and post-
sequencing quality control, we generated read counts for 83 samples in the LPS challenge
experiment (43 controls and 40 treatment), and 440 samples for the cell type-specific analysis
(89 in helper T cells, 89 in cytotoxic T cells, 86 in B cells, 88 in natural killer cells, and 88 in
monocytes; Table S3). We confirmed the identity of all samples based on genotyping from the
RNA-seq reads (see below for details).
Text S5. Read normalization and correction for batch effects
Prior to RNA-seq data analysis, we first filtered out genes that were very lowly or not
detectably expressed in our samples. For the cell type-specific data sets, we removed any gene
with a median RPKM ≤ 2 in that cell type. For the LPS challenge experiment, genes were
removed if the median RPKM was ≤ 2 in both control and LPS-stimulated samples. This
procedure resulted in a different number of genes that passed filter in each data set: 8,823 genes
in helper T cells, 8,802 genes in cytotoxic T cells, 8,388 genes in monocytes, 8,754 genes in
natural killer cells, 8,951 genes in B cells and 9,047 genes for the LPS experiment (control and
LPS+).
For each of the six data sets, we normalized the resulting read count matrix using the
function voom from the R package limma (34). We then modeled the normalized expression
values as a function of the sample donor‟s social group membership (9 social groups in each
phase for the cell type-specific data sets, for a total of 18 distinct groups; and the 9 social groups
in Phase 2 only for the LPS challenge data set). Because we sampled all females in each group
within a short time frame (median time frame for the complete group to be sampled was 3 days;
groups were entirely sampled for a given analysis within a 10 day window, with all but two
exceptions when the final female was sampled several weeks later), controlling for the identity of
the group removes biological variation related to differences in group dynamics, and most
technical batch effects related to sample collection and processing. Thus, we used the residuals
of the model relating normalized expression to social group identity as our primary outcome
measure in all subsequent analyses.
Text S6. Genotyping
We used genotype data to confirm sample identity across experiments and to control for
genetic relatedness among individuals in our analyses (pairwise genetic relatedness in our sample
was on average low, but several close kin were included in the data set, although never housed in
the same social group: fig. S8) To do so, we combined the RNA-seq reads for all five purified
cell types for each female and called variants using HaplotypeCaller from the Genome Analysis
Toolkit (GATK v3.3.0), following the Best Practices for variant calling in RNAseq
(https://www.broadinstitute.org/gatk/guide/article?id=3891). After genotyping we retained sites
that passed the following filters: quality score ≥100; QD < 2.0; MQ < 35.0; FS > 60.0;
HaplotypeScore >13.0; MQRankSum < −12.5; and ReadPosRankSum < −8.0. Finally, we
estimated kinship with the program lcMLkin (35) using single nucleotide variants that were
genotyped in all 45 females and that were thinned to be at least 10 kb apart (n=54,165 SNVs;
genotypes available on github:
https://github.com/nsmackler/status_genome_2016/tree/master/genotypes).
Text S7. Modeling rank effects on gene expression
To identify genes that were significantly affected by dominance rank, we used a linear
mixed effects model that controls for relatedness within the sample (36-38). We analyzed each of
the six data sets (five purified cell types plus the LPS challenge data) separately using the R
package EMMREML (39), in each case working with gene expression values after controlling
for social group/batch effects, as described above.
For each gene in the cell type-specific data sets, we first estimated the effect of
dominance rank on gene expression levels across phases using the following model:
(1)
where y is the n by 1 vector of residual gene expression levels for the n samples collected in
Phase 1 and Phase 2; μ is the intercept; r is an n by 1 vector of Elo ratings and β is its effect size;
and a is an n by 1 vector of female age in years at the time of sample collection and γ is its effect
size. The m by 1 vector u is a random effects term to control for kinship and other sources of
genetic structure. Here, m is the number of unique females in the analysis (m=45), the m by m
matrix K contains estimates of pairwise relatedness derived from a 45 x 54,165 genotype data
set, is the genetic variance component (0 for a non-heritable trait, but most gene expression
levels are heritable: (40, 41)), and Z is an incidence matrix of 1‟s and 0‟s that maps
measurements in Phase 1 and Phase 2 to individuals in the random effects term (thus controlling
for repeated measurements for the same individual across phases). Residual errors are
represented by ε, an n by 1 vector, where represents the environmental variance component
(unstructured by genetic relatedness), I is the identity matrix, and MVN denotes the multivariate
normal distribution. For each data set and gene, we tested the null hypothesis that β = 0 versus
the alternative hypothesis, β ≠ 0.
To test for dominance rank effects on the cellular composition of our samples (based on
flow cytometry data), we used a parallel model replacing gene expression level (y) with the logit-
transformed proportional representation of each of 11 cell types and added a fixed effect term to
account for three rounds of repeated measures for the flow data (once in Phase 1 and twice in
Phase 2; fig. S2). As in eq (1), this model included a random effects term to control for repeated
measurements from the same genotype/individual.
To test for consistency and plasticity of rank effects on gene expression levels across
phases, we also ran a nested model that estimated the effects of rank specific to each phase:
(2)
where the notation is the same as above, but I is an indicator variable for phase, p (phase 1 or
phase 2). To produce an empirical null distribution for rank effects in these two models, we
performed 1000 permutations of female dominance rank and re-ran the analyses using
randomized rank values.
To test for rank effects on gene expression in the LPS challenge experiment, we
estimated the effects of rank specific to the control and treatment conditions (all data were
collected for Phase 2 only):
(3)
where the notation for gene expression, dominance rank, the intercept, residuals, and random
effects are consistent with the models above, but I is an indicator variable for c, the condition
(control = 0, treatment = 1), rather than for phase, and an additional effect, , is fit for the binary
condition variable. Other fixed effects covariates, for mean-centered age and the first two
principal components from a PCA decomposition of the normalized cell type proportion data
(matrix X), are also estimated as nested effects within condition (vectors V1 and V2). To explicitly
extract interaction effects between dominance rank and condition, we reformulated model (3) as
follows:
(4)
where we tested the null hypothesis that βrxc = 0 versus the alternative hypothesis, βrxc ≠ 0. For
the LPS challenge experiment, we produced empirical null distributions for rank effects by
permuting the dominance rank values associated with pairs of samples (control and LPS-treated)
as a block, 1000 times, and re-running our analyses. To produce a corresponding empirical null
for the effect of LPS stimulation (i.e., condition) on gene expression, we randomized condition
within each pair of samples for each female. When only a treatment or control sample was
available for a female, we randomly assigned it to one of the two conditions with 50%
probability.
For all of our analyses, we controlled for multiple testing using a generalization of the
false discovery rate method of Storey and Tibshirani (42) to empirical null p-value distributions
generated via permutation.
Text S8. Gene Ontology enrichment
Gene ontology (GO) enrichment analyses were performed using the Cytoscape module
ClueGO (43). We conducted one-sided Fisher‟s Exact Tests for enrichment and corrected for
multiple tests using the Benjamini-Hochberg (B-H) method (44). To reduce the multiple testing
space and account for the nested nature of GO terms, we analyzed only terms that fell between
levels 3 and 8 of the GO tree in the term category of Biological Process, included at least 10
genes in our data set, and for which at least 5% of the total number of genes belonging to the GO
term were present in the test gene set. We report significant terms as those that were enriched in
the target gene set at a B-H corrected p-value ≤ 0.05 (Table S4). In the network diagrams, this
threshold was sometimes made more stringent for the sake of visualization (i.e., in Fig. 3D we
only show terms enriched in category I at FDR < 10-5
, while, in fig. S4A, we restrict the results
shown to those with a FDR < 0.01).
For the five cell type-specific data sets, we conducted enrichment analyses separately for
genes that were more highly expressed in low-ranking animals and genes that were more highly
expressed in high-ranking animals, in both cases against the background set of all measured
genes in that cell type. For the LPS challenge data, we conducted one enrichment analysis for
each of the four LPS-rank interaction categories (stratified by direction of the LPS response and
direction of the rank effect on gene expression levels in the LPS+ condition, as reported in the
main text; Fig. 3D) against the background set of all genes that responded significantly to LPS at
an FDR < 0.05.
Text S9. Meta-analysis for tissue specificity
True shared effects of rank across cell types could be missed in our above analyses if they
failed to survive FDR correction in some cell types. We therefore complemented our
independent analyses of purified cell populations with a meta-analytic approach, MeSH (Meta-
analysis with Subgroup Heterogeneity; (45)), focusing on the union set of 1,622 genes that were
detectably expressed in all five cell types and significantly associated with rank, in any cell type,
at an FDR<10%. We ran MeSH for each gene using the Exchangeable Effects (EE) model, using
β, the effect of rank on a gene‟s expression; se(β), the standard error on that coefficient; and a
grid file of discrete priors for the mean effect of rank across cell types (ω) and the heterogeneity
of effects across cell types (ψ). We constructed the grid file to span the effect sizes observed in
our data (ω = [0.00,0.01,0.02,0.03,0.04,0.05,0.06,0.07,0.08] and ψ =
[0.000,0.002,0.004,0.006,0.008,0.010,0.012,0.014,0.016,0.018,0.020]).
Text S10. Behavioral and GC sensitivity mediation analysis
To estimate the contribution of variation in behavior and GC sensitivity to the effect of
social status on gene expression, we conducted three sets of mediation analyses. First, to
determine how agonistic and affiliative behaviors mediate the effect of social status on gene
expression levels, we conducted two analyses: one investigating the mediating effects of
received harassment on natural killer and helper T cell expression, and the second testing the
mediating effects of grooming rates on natural killer and helper T cell expression. We
investigated only genes for which we detected a significant rank-gene expression relationship in
the cell type-specific analyses, focusing on the two cell types (natural killer and helper T cells)
for which this signal was strongest. To test the contribution of glucocorticoid physiology in the
immune challenge experiment, we performed parallel analyses for rank-responsive genes in the
control condition and rank-responsive genes in the LPS-stimulated condition, in this case using
the results of the glucocorticoid sensitivity test (see Text S3) as the candidate mediating variable.
For each gene, we were interested in estimating the indirect effect of dominance rank on
gene expression levels through the mediating variable (grooming, received harassment, or
glucocorticoid sensitivity). The strength of the indirect effect was estimated as the difference
between the effect of rank in two models: the “unadjusted” model that did not account for the
mediator (equivalent to in equation 1 above), and the effect of rank in an “adjusted” model (
in equation 5 below) that incorporated the mediator, t.
(5)
where notations are the same as in equation (1), but with the addition of an effect for the
mediating variable, To assess the significance of the indirect effect, , we performed
1000 iterations of bootstrap resampling to calculate 95% confidence intervals for each mediator.
We deemed an indirect effect significant if the lower bound of the 95% confidence interval did
not overlap with 0.
Text S11. Transcription factor binding site enrichment
To investigate transcription factors involved in rank effects on gene expression, we
combined data on the location of predicted transcription factor binding sites in the rhesus
macaque genome with data on open chromatin in PBMCs collected from our study population.
To identify these regions, we generated ATAC-seq libraries from three mid-ranking females
sampled during Phase 2 (using 50,000 cells per female and following (46), but skipping the cell
lysis step). Libraries were barcoded, multiplexed, and sequenced on an Illumina NextSeq 500
using paired-end, 38 bp reads. Reads were mapped to MacaM using the bwa-mem algorithm with
default settings (47), and exhibited the typical periodic pattern of insert sizes associated with
nucleosomes. Open chromatin regions were identified using MACS2 (48) (--nomodel --keep-dup
all -q 0.05 -f BAMPE) after merging data from all three libraries into a single alignment file, at a
5% FDR threshold.
To identify putative TFBSs associated with open chromatin and rank-responsive genes,
we scanned the rhesus macaque genome for matches to 1900 position weight matrices (PWM)
obtained from TRANSFAC and JASPAR (49, 50), using a PWM model as in (51) and a
threshold of 13 (i.e., a 213
-fold increase in the likelihood that a motif is a binding site relative to
background). Because multiple PWMs are often associated with the same transcription factor
(TF), we clustered them by calculating the Jaccard distance between genomic locations
associated with each pair of PWMs with the bedtools function jaccard (52). After hierarchical
clustering of the resulting dissimilarity matrix, we considered all PWMs with a dissimilarity ≤
0.2 (i.e., a Jaccard statistic ≥ 0.8) as members of the same “TF cluster”, producing 913 non-
redundant TF clusters. We also filtered out all TF clusters that were rarely found in open
chromatin near annotated genes (<100 cases genome-wide), resulting in a final set of 460 motif
clusters. For each gene in our data set, we then noted whether each TF cluster fell within 5 kb
upstream or downstream of the gene transcription start site.
Using this information, we performed two analyses. First, we used Fisher‟s Exact Tests to
ask whether rank-responsive genes were more likely to be associated with a given TF cluster
than genes with no evidence for rank effects (p > 0.3). We tested for overrepresentation of TF
clusters separately for genes that were more highly expressed in high-ranking females versus
those that were more lowly expressed, and performed this analysis separately for the natural
killer cell data set, the helper T cell data set, and the LPS challenge data set. To control for
multiple hypothesis testing, we used the false discovery rate approach of Benjamini and
Hochberg (44). Second, we used an elastic net logistic regression approach to ask about the
predictive power of TFBS locations in open chromatin regions to discriminate between (i) genes
upregulated in high ranking individuals versus genes unaffected by rank; (ii) genes
downregulated in high ranking individuals versus genes unaffected by rank; and (iii) genes that
were upregulated versus downregulated with higher rank. To avoid biases caused by unbalanced
outcome classes, we randomly subsampled genes from the majority class (without replacement)
equal to the number of genes in the minority class, calculated prediction accuracy (1 – the model
misclassification rate) using the function “cv.glmnet” in the R package glmnet (53), and repeated
this procedure 1000 times In each iteration, we extracted the betas for each TFBS from the
model with the best prediction accuracy (because the elastic net approach induces sparsity, most
betas equal 0). We considered TFBS clusters to have the best evidence for association with
rank-responsive genes if they (i) had a non-zero beta in more than half of the 1000 elastic net
iterations, and (ii) a Fisher‟s Exact Test B-H corrected p-value < 0.05.
Using these criteria, we identified multiple enriched TFBS that predicted whether a gene
was significantly positively or negatively associated with dominance rank with 62.5% (range:
60.8-64.5%) and 58.8% (range: 52.5%-60.5%) accuracy in the natural killer cells and helper T
cells, respectively (fig. S13). In the NK cells, many of the transcription factors that were most
predictive of a gene being more highly expressed in low status individuals are linked to
inflammatory processes, including NFkB (Fisher‟s Test OR = 4.01, p= 3.3x10-5
), GLI-3, and
(Fisher‟s Test OR = 2.44, p= 1.3x10-5
), and SRF (Fisher‟s Test OR = 8.91, p= 5.3x10-4
; fig. S6
and Table S7)
Text S12. Deconvolution of LPS challenge data and ‘reconvolution” of PBMC gene
expression levels from cell type-specific data
To test that the cell types responsible for gene expression responses to LPS stimulation
conformed with expectations (i.e., were largely driven by monocytes and polymorphonuclear
granulocytes), we performed a deconvolution analysis of the gene expression data collected in
the LPS stimulation experiment. To do so, we drew on the cell type composition data measured
via flow cytometry for each sample (fig S9), and grouped them into six main cell populations:
(1) CD14+ monocytes; (2) CD16
+ NKs; (3) B cell; (4) helper T cells; (5) cytotoxic T cells; and
(6) granulocytes. We then applied the least-squares, partial deconvolution approach implemented
in R package csSAM (54). We ran csSAM separately for the gene expression data from control
versus LPS+ samples (in both cases after controlling for batch effects). We then tested for
differential gene expression between control and LPS+ conditions in each of the six deconvolved
datasets using a Welch-Satterthwaite t-test with a Benjamini-Hochberg FDR correction for
multiple testing. Finally, we performed Fisher‟s Exact Tests to test whether LPS-responsive
genes were enriched within the set of LPS-responsive genes identified for each de-convolved
gene expression data set (fig S7: cell lines with median proportion >0.05 in whole blood are
shown). Only granulocytes (log2(OR)=1.15, p<10-63
) and monocytes (log2(OR)=0.46, p=5.7x10-
12) were significantly enriched for the LPS responsive genes identified in the full sample (fig.
S7).
To test whether the control samples in the LPS challenge experiment were consistent
with our results for the purified, cell type-specific gene expression data, we also performed an in
silico “reconvolution” analysis for each individual. This analysis was based on the gene
expression data for each purified cell type and the flow cytometry-based estimates of cell type
proportions. Specifically, we calculated the reconvoluted expression level for each gene-
individual combination as the mean of the expression level of the gene across purified cell types
(helper T, cytotoxic T, monocyte, NK, and B), weighted by the proportion of each cell type in
the PBMC pool for a given sample. For each female/phase combination:
(6)
where p is a 1 by 5 vector of the proportion of each of the 5 cell types; X is a 5 by n gene matrix
of the gene expression values, for the n genes measured (in counts per million, CPM) in the cell
type-specific gene expression data (one column for each cell type); and Y is a 1 by n gene vector
of the “reconvoluted” gene expression values for the PBMC pool. We calculated PBMC gene
expression level estimates for all 45 females for each phase separately and modeled the
reconvoluted expression values as a function of rank and age using the same mixed effects
models as in the cell type-specific analysis. Note that our reconvoluted estimates differ
somewhat from the control samples in the LPS challenge experiment because we included only
PBMC cell types in these estimates, whereas in the LPS challenge data, we profiled gene
expression levels for all white blood cells, including polymorphonuclear cells.
We found that the effect of rank on the reconstituted gene expression was significantly
positively correlated with the effect of rank on gene expression in white blood cells in the control
condition of the LPS experiment (Pearson‟s r = 0.35, p<2.50x10-248
). This relationship was
stronger when we only considered rank-responsive genes detected in the control condition of the
LPS experiment at an FDR < 1% (Pearson‟s r = 0.59, p=7.14x10-165
).
Text S13. Effects of dominance rank on the magnitude of the response to LPS stimulation
As described in equations (3) and (4), rank effects on gene expression in the LPS
challenge data are estimated separately for each condition (control versus LPS+). This
formulation allowed us to estimate the LPS effect on gene expression for high-ranking and low-
ranking females separately as:
(7)
where is the effect of the LPS treatment for a given Elo rating r, is the LPS treatment
effect estimated in equation (4), and is the interaction effect between rank and condition,
also estimated in equation (4). In the main text, we calculated these values, across all LPS-
responsive genes, for r equal to the mean Elo rating for the lowest ranking females across groups
and for r equal to the mean Elo rating for the highest ranking females across groups. This
procedure produced a distribution of LPS treatment effects for low ranking and high ranking
females, showing that low-ranking females exhibit exaggerated responses to LPS treatment
overall (Fig. 4C, Mann-Whitney test p<10-20
).
Text S14. Rank effects in MyD88 vs TRIF induced genes.
To investigate rank-dependent polarization of the TLR4 signaling pathways through
TRIF versus MyD88-dependent arms, we used the results from (23), which identified sets of
mouse genes for which a normal response to antigen induction was either MyD88 or TRIF-
dependent. These gene lists were obtained by comparing the gene expression response of
macrophages from wild-type mice to MyD88 or TRIF knock-out mice, using six purified TLR
agonists. Ramsey et al reported 334 genes that were up-regulated upon stimulation via the
MyD88-dependent pathway and 274 by that were up-regulated via the TRIF-dependent pathway,
among which 238 and 168 rhesus macaque orthologues were included in our LPS challenge data
set, respectively.
Using these annotations, we performed three analyses. First, we tested for a difference in
the distributions of dominance rank effect sizes between MyD88-induced versus TRIF-induced
genes (conditional on being rank-responsive in the LPS+ condition), using a Mann-Whitney test
(Fig. 4B). Second, we tested whether MyD88- induced or TRIF- induced genes were enriched
among Category I (upregulated after LPS stimulation and more highly expressed in low-ranking
females) or Category II (upregulated after LPS stimulation and more highly expressed in high-
ranking females) genes, using a Fisher‟s Exact Test (Fig. 4C). Third, we asked whether females
exhibited systematically different gene expression levels, depending on their social status, in the
MyD88 versus TRIF-induced gene sets. For this last analysis, we extracted, for each female, the
median gene expression level for all genes in the LPS challenge data set that are up-regulated
upon stimulation via the MyD88-dependent pathway (based on Ramsay et al), after controlling
for batch effects and other covariates (age, cell composition). We then performed linear
regression of Elo rating versus median MyD88-dependent gene expression levels across females
(Fig. 4D). We repeated the same analysis for genes up-regulated via the TRIF-dependent
pathway.
Figure S1. Correlation between female dominance ranks across phases. Solid lines connect
the same female in the two phases of the study, with dots colored by ordinal rank in each phase.
There was no significant correlation between female ranks across Phases (Pearson‟s r=0.06, p =
0.68 between phases).
Figure S2. Association between dominance rank and proportions of 11 white blood cell
populations. Each plot shows the relationship between Elo rating and the logit-transformed
residual proportions of 11 blood cell populations after controlling for subject age and
measurement round (proportion data was collected once in Phase 1 and twice in Phase 2, for
three repeated measures per individual). (A) polymorphonuclear (CD14dim
/SSC-A>100K/FSC-
A>100K), (B) classical monocytes (CD14+/CD16
-), (C) CD14
+ intermediate monocytes
(CD14+/CD16
+), (D) CD14
- activated non-classical monocytes (CD14
-/CD16
+), (E) helper T
cells (CD3+/CD4
+), (F) cytotoxic T cells (CD3
+/CD8
+), (G) double positive T cells
(CD3+/CD4
+/CD8
+), (H) CD8
- B cells (CD3
-/CD20
+/CD8
-), (I) CD8
+ B cells (CD3
-
/CD20+/CD8
+), (J) natural killer T lymphocytes (CD3
+/CD16
+), and (K) natural killer cells
(CD3-/CD16
+)
Figure S3. Cell sorting strategy. (A) Schematic of the cell sorting strategy for purifying five
populations of PBMCs prior to gene expression analysis. We first removed all dead cells and
then sorted the cells using 7 cell surface markers (detailed in the Materials and Methods and
Table S2). (B) Example of the sorting strategy for one sample visualized using FlowJo (FlowJo,
LLC, Ashland, OR).
A
B
20160601 Workspace.jo Layout
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
FSC-H
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD3
0
102
103
104
105
Live/Dead
0 102
103
104
105
CD4
0
102
103
104
105
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
SSC-A
SSC-A
SSC-A
FSC-H
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD3
0
102
103
104
105
Live/Dead
0 102
103
104
105
CD4
0
102
103
104
105
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A
Cytotoxic T
Helper T
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
FSC-H
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD3
0
102
103
104
105
Live/Dead
0 102
103
104
105
CD4
0
102
103
104
105
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD16
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A
FSC-H
SSC-A
Live/Dead
0 102
103
104
105
CD4
0
102
103
104
105
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD16
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A
B
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
FSC-H
0 50K 100K 150K 200K 250K
FSC-A
0
50K
100K
150K
200K
250K
SSC-A
Live/Dead
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
SSC-A
0 102
103
104
105
CD16
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A
FSC-H
SSC-A
Live/Dead
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
SSC-A
0 102
103
104
105
CD16
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A
NK
FSC-H
SSC-A
Live/Dead
0 102
103
104
105
CD4
0
102
103
104
105
CD8
0 102
103
104
105
CD20
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
HLA-DR
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD16
0
50K
100K
150K
200K
250K
SSC-A
0 102
103
104
105
CD14
0
50K
100K
150K
200K
250K
SSC-A Monocytes
CD3+CD3-
CD20+CD20-
HLA-DR+HLA-DR-
Live/dead (7-AAD)
CD3+CD3-
HLA-DR+
CD8+CD4+CD20+ CD20-
HLA-DR- HLA-DR+
CD14+CD16+
B
natural killer monocyte
helper T cytotoxic T
Figure S4. Gene Ontology enrichment in the cell type-specific analysis. (A) Categories
enriched in genes more highly expressed in low-ranking animals in NK cells. (B) Categories
enriched in genes more highly expressed in high-ranking animals in NK cells. (C) Categories
enriched in genes more highly expressed in low ranking animals in helper T cells. We detected
only two significant terms for genes that were more highly expressed in high-ranking individuals
in helper T cells, so have not plotted them here (regulation of nervous system development,
FDR-corrected p=7.0x10-4
; regulation of vesicle-mediated transport, FDR-corrected p=2.4x10-3
).
Figure S5. Within-individual changes in dominance rank are reflected in within-individual
changes in the expression levels of rank-responsive genes. Change in dominance rank from
Phase 1 to Phase 2 is shown on the x-axis for all 45 study subjects, plotted against the mean
change in normalized gene expression levels for rank-responsive genes. Females that increased
rank across phases exhibited a corresponding increase in mean gene expression levels for genes
more highly expressed with high rank (orange points; helper T cells: Pearson‟s r=0.66, p=1.1x10-
6, n=488 genes; natural killer cells: r=0.57, p=5.8x10
-5, n=714 genes). Similarly, females that
decreased rank across phases exhibited a corresponding increase in mean gene expression levels
for genes more highly expressed with low rank (blue points; helper T cells: r=-0.65, p=1.5x10-6
,
n=335 genes; natural killer cells: r=-0.62, p=1.0x10-5
, n=449 genes).
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
downreg
upreg
helper T
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
natural killer
increase
in rank
decrease
in rank
no change
in rank
increase
in rank
decrease
in rank
no change
in rank
change in dominance rank (Elo)
ch
ang
e in g
ene
exp
ressio
n le
ve
ls
ch
ang
e in g
ene
exp
ressio
n le
ve
ls
A Bhelper T natural killer
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
downreg
upreg
helper T
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
natural killer
high expression with low rank
high expression with high rank
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
downreg
upreg
helper T
-1.0
-0.5
0.0
0.5
1.0
-1000 0 1000
change in rank
change in e
xpre
ssio
n
natural killer
high expression with low rank
high expression with high rank
Figure S6. Distribution of the absolute difference in magnitude of rank effects between
LPS-stimulated and untreated samples. Gray: genome wide distribution. Blue: genes where
we detected a significant interaction effect between condition (control versus LPS+) and
dominance rank (FDR < 0.05). Interaction effects are biased towards systematically larger effects
of rank after LPS stimulation (Mann-Whitney test, p=4.2x10-86
).
0
1
2
3
−0.5 0.0 0.5 1.0
abs(Rank effect at LPS) − abs(Rank effect at NC)
De
nsity
Background
Interactions
background
interaction
-0.5 0.0 0.5 1.0
0
1
2
3
abs(rank effect in LPS+) - abs(rank effect in LPS-)
den
sity
Figure S7. Glucocorticoid (GC) sensitivity mediation of rank effects in the LPS challenge
experiment. Points depict the proportion of the rank effect on gene expression levels mediated
by GC sensitivity for rank-responsive genes. 42 genes were significantly mediated by GC
sensitivity (blue points) in the LPS- control, whereas 155 genes were significantly mediated by
GC sensitivity in the LPS+ stimulated condition. The mediation effects by GC sensitivity were
also significantly larger in the LPS condition than in the control condition, consistent with the
known immunomodulatory effects of GC signaling (Mann-Whitney test p=1.5x10-4
). For both
control and LPS+ conditions, observed mediation effects were substantially larger than expected
by chance based on permuting the GC sensitivity data (median proportion mediation was 0.021
in the control condition and 0.022 in the LPS+ condition, compared to 0.081 and 0.089 in the
real data, respectively; Mann-Whitney test p < 10-15
in both cases).
Figure S8. Transcription factors with putative binding sites enriched near rank-responsive
genes. FET odds ratios for TFBS clusters in which (i) the FET results were significant at a B-H
FDR < 5%, and (ii) presence of the TFBS in open chromatin within 5 kb of gene transcription
start sites contributed to predicting whether genes were rank-responsive or more highly
expressed in high versus low-status females (non-zero betas in an elastic net logistic regression,
for at least 50% of test sets). Points and whiskers show FET OR ± 95% CI.
Figure S9. Enrichment of LPS-responsive genes in deconvoluted cell type-specific gene
expression data. We tested for enrichment of LPS-responsive genes identified in the
deconvoluted data among the set of LPS-responsive genes identified in the non-deconvoluted
empirical data. In both cases, LPS-responsive genes were defined as genes differentially
expressed between the control and LPS+ conditions at an FDR<0.01. As expected, granulocytes
(log2(OR)=1.15, p=8.7x10-64
) and monocytes (log2(OR)=0.46, p=5.7x10-12
) were most important
in mediating this response.
Figure S10. Stability of body mass over time. Body mass in our study subjects tended to
remain stable across the duration of each phase, and weight loss/gain was not predicted by
dominance rank (Phase 1: r=-0.02, p=0.91, n=45 females, green points; Phase 2: r=0.10, p=0.50,
n=45 females, blue points). Thus, rank-related differences in energy balance are unlikely to
account for the rank-responsive gene expression patterns we identified. Notably, body mass was
also strongly correlated with female age (Phase 1: r=0.80, p=2.90x10-11
, n=45 females; Phase 2:
r=0.65, p=1.12x10-6
, n=45 females), which was included as a covariate in all of our statistical
models.
Figure S11. Kinship among study subjects. No groups contained first-degree relatives. The x-
axis depicts the coefficient of relatedness (r), which ranges from 0-1 (r = 0.5 for full siblings and
parent-offspring dyads).
0
100
200
300
400
500
0.0 0.2 0.4
relatedness
num
ber of dyads
between group dyad
within group dyad
0
100
200
300
400
500
0.0 0.2 0.4
relatedness
num
ber of dyads
between group dyad
within group dyad
coefficient of relatedness
Figure S12. LPS challenge experiment cell phenotyping strategy. (A) Schematic of the gating
strategy for phenotyping the cells in the LPS challenge experiment. We first gated cells based on
forward and side scatter, and then classified them into 11 cell populations using a combination of
cell surface markers (Table S2). (B) Example of the phenotyping strategy for one sample
visualized using FlowJo (FlowJo, LLC, Ashland, OR).
Figure S13. TFBS-based prediction of rank-responsive genes. Using putative TF binding sites
(TFBS) in regions of open chromatin 5 kb upstream of the TSS of a gene, we could predict
whether that gene was significantly positively ( rank) or negatively ( rank) associated with
dominance rank with 62.5% (range: 60.8-64.5%) and 58.8% (range: 52.5%-60.5%) accuracy in
natural killer cells and helper T cells, respectively.
CD16_downreg
CD16_up_v_down
CD16_upreg
CD4_downreg
CD4_up_v_down
CD4_upreg
40% 50% 60%
accuracy
rank vs. no rank effect
rank vs. no rank effect
rank vs. rank
rank vs. no rank effect
rank vs. no rank effect
rank vs. rank
prediction accuracy
NK
helper T
Supplementary Tables (all Tables are hosted online as worksheets in a single Excel file)
Table S1. Study subject information. ID, sampling date, date of birth, sample tissue
composition information, and other metadata. Table S1 is hosted online as an Excel file.
Table S2. Antibodies and fluorescent labels used for cell sorting and phenotyping.
Antibodies for cell surface markers, including clone, product number, and fluorescent label.
Table S2 is hosted online as an Excel file.
Table S3. Number of RNA-seq samples passing quality control. RNA-seq samples used in all
analyses, by phase, cell type, or inclusion in the immune challenge experiment. Table S3 is
hosted online as an Excel file.
Table S4. Cell type-specific rank effects for all genes. Test statistics, p-values, and false
discovery rate-corrected q-values for all genes tested in all cell types. Table S4 contains 5
subsections, for helper T cells (S4a), cytotoxic T cells (S4b), monocytes (S4c), natural killer cells
(S4d), and B cells (S4e) respectively. Table S4 is hosted online as an Excel file.
Table S5. Gene set enrichment results for cell type-specific analysis. Categorical enrichment
results based on Gene Ontology categories, for dominance rank-associated genes in natural killer
cells and helper T cells. Table S5 contains 4 subsections, for genes more highly expressed in
high ranking females in NK cells (S5a), genes more highly expressed in low ranking females in
NK cells (S5b), genes more highly expressed in high ranking females in helper T cells (S5c), and
genes more highly expressed in low ranking females in helper T cells (S5d). Table S5 is hosted
online as an Excel file.
Table S6. LPS challenge results for all genes. Test statistics, p-values, and false discovery rate-
corrected q-values for all genes tested in the immune challenge experiment. Table S6 includes
results for rank effects in each condition (negative control and LPS), plus results for rank-
condition interactions and categorization of each gene, where appropriate, into Categories I – IV
(as defined in the main text). Table S6 is hosted online as an Excel file.
Table S7. Gene set enrichment results for LPS challenge experiment. Categorical enrichment
results based on Gene Ontology categories, for dominance rank-associated genes in the
TruCulture immune challenge data set. Table S7 contains 2 subsections, for genes in Category I
(S7a: more highly expressed after LPS challenge, with higher expression in low-ranking
females) and Category II (S7b: more highly expressed after LPS challenge, with higher
expression in high-ranking females), respectively. We did not identify significant enrichment for
genes in Categories III and IV. Table S7 is hosted online as an Excel file.
Table S8. Transcription factor binding site enrichment results. Enrichment of TFBS sites in
open chromatin regions, near rank-associated gene sets defined in our study. Table S8 shows
results for Fisher‟s Exact Test enrichments and elastic net regression analysis to differentiate
between gene sets. Table S8 contains 2 subsections, for comparisons based on the immune
challenge gene expression data set (S8a) and comparisons based on the cell type-specific gene
expression data set (S8b), and is hosted online as an Excel file.
References and Notes
1. M. Marmot, Social determinants of health inequalities. Lancet 365, 1099–1104
(2005).doi:10.1016/S0140-6736(05)74234-3 Medline
2. B. G. Link, J. Phelan, Social conditions as fundamental causes of disease. J. Health Soc.
Behav. 35, 80–94 (1995).doi:10.2307/2626958 Medline
3. R. Chetty, M. Stepner, S. Abraham, S. Lin, B. Scuderi, N. Turner, A. Bergeron, D. Cutler, The
association between income and life expectancy in the United States, 2001-2014. JAMA
315, 1750–1766 (2016).doi:10.1001/jama.2016.4226 Medline
4. R. M. Sapolsky, The influence of social hierarchy on primate health. Science 308, 648–652
(2005).doi:10.1126/science.1106477 Medline
5. J. Tung, L. B. Barreiro, Z. P. Johnson, K. D. Hansen, V. Michopoulos, D. Toufexis, K.
Michelini, M. E. Wilson, Y. Gilad, Social environment is associated with gene regulatory
variation in the rhesus macaque immune system. Proc. Natl. Acad. Sci. U.S.A. 109, 6490–
6495 (2012).doi:10.1073/pnas.1202734109 Medline
6. V. Michopoulos, K. M. Reding, M. E. Wilson, D. Toufexis, Social subordination impairs
hypothalamic-pituitary-adrenal function in female rhesus monkeys. Horm. Behav. 62,
389–399 (2012).doi:10.1016/j.yhbeh.2012.07.014 Medline
7. C. A. Shively, T. B. Clarkson, J. R. Kaplan, Social deprivation and coronary artery
atherosclerosis in female cynomolgus monkeys. Atherosclerosis 77, 69–76
(1989).doi:10.1016/0021-9150(89)90011-7 Medline
8. S. W. Cole, Human social genomics. PLOS Genet. 10, e1004601
(2014).doi:10.1371/journal.pgen.1004601 Medline
9. M. R. Irwin, S. W. Cole, Reciprocal regulation of the neural and innate immune systems. Nat.
Rev. Immunol. 11, 625–632 (2011).doi:10.1038/nri3042 Medline
10. H. Jarrell, J. B. Hoffman, J. R. Kaplan, S. Berga, B. Kinkead, M. E. Wilson, Polymorphisms
in the serotonin reuptake transporter gene modify the consequences of social status on
metabolic health in female rhesus monkeys. Physiol. Behav. 93, 807–819
(2008).doi:10.1016/j.physbeh.2007.11.042 Medline
11. C. Neumann, J. Duboscq, C. Dubuc, A. Ginting, A. M. Irwan, M. Agil, A. Widdig, A.
Engelhardt, Assessing dominance hierarchies: Validation and advantages of progressive
evaluation with Elo-rating. Anim. Behav. 82, 911–921 (2011).
doi:10.1016/j.anbehav.2011.07.016
12. J. Sprent, Lifespans of naive, memory and effector lymphocytes. Curr. Opin. Immunol. 5,
433–438 (1993).doi:10.1016/0952-7915(93)90065-Z Medline
13. T. Flutre, X. Wen, J. Pritchard, M. Stephens, A statistical framework for joint eQTL analysis
in multiple tissues. PLOS Genet. 9, e1003486 (2013).doi:10.1371/journal.pgen.1003486
Medline
14. K. G. Ardlie, D. S. Deluca, A. V. Segre, T. J. Sullivan, T. R. Young, E. T. Gelfand, C. A.
Trowbridge, J. B. Maller, T. Tukiainen, M. Lek, L. D. Ward, P. Kheradpour, B. Iriarte,
Y. Meng, C. D. Palmer, T. Esko, W. Winckler, J. N. Hirschhorn, M. Kellis, D. G.
MacArthur, G. Getz, A. A. Shabalin, G. Li, Y.-H. Zhou, A. B. Nobel, I. Rusyn, F. A.
Wright, T. Lappalainen, P. G. Ferreira, H. Ongen, M. A. Rivas, A. Battle, S. Mostafavi, J.
Monlong, M. Sammeth, M. Mele, F. Reverter, J. M. Goldmann, D. Koller, R. Guigo, M.
I. McCarthy, E. T. Dermitzakis, E. R. Gamazon, H. K. Im, A. Konkashbaev, D. L.
Nicolae, N. J. Cox, T. Flutre, X. Wen, M. Stephens, J. K. Pritchard, Z. Tu, B. Zhang, T.
Huang, Q. Long, L. Lin, J. Yang, J. Zhu, J. Liu, A. Brown, B. Mestichelli, D. Tidwell, E.
Lo, M. Salvatore, S. Shad, J. A. Thomas, J. T. Lonsdale, M. T. Moser, B. M. Gillard, E.
Karasik, K. Ramsey, C. Choi, B. A. Foster, J. Syron, J. Fleming, H. Magazine, R. Hasz,
G. D. Walters, J. P. Bridge, M. Miklos, S. Sullivan, L. K. Barker, H. M. Traino, M.
Mosavel, L. A. Siminoff, D. R. Valley, D. C. Rohrer, S. D. Jewell, P. A. Branton, L. H.
Sobin, M. Barcus, L. Qi, J. McLean, P. Hariharan, K. S. Um, S. Wu, D. Tabor, C. Shive,
A. M. Smith, S. A. Buia, A. H. Undale, K. L. Robinson, N. Roche, K. M. Valentino, A.
Britton, R. Burges, D. Bradbury, K. W. Hambright, J. Seleski, G. E. Korzeniewski, K.
Erickson, Y. Marcus, J. Tejada, M. Taherian, C. Lu, M. Basile, D. C. Mash, S. Volpi, J.
P. Struewing, G. F. Temple, J. Boyer, D. Colantuoni, R. Little, S. Koester, L. J. Carithers,
H. M. Moore, P. Guan, C. Compton, S. J. Sawyer, J. P. Demchok, J. B. Vaught, C. A.
Rabiner, N. C. Lockhart, K. G. Ardlie, G. Getz, F. A. Wright, M. Kellis, S. Volpi, E. T.
Dermitzakis; The GTEx Consortium, The Genotype-Tissue Expression (GTEx) pilot
analysis: Multitissue gene regulation in humans. Science 348, 648–660
(2015).doi:10.1126/science.1262110 Medline
15. N. Snyder-Mackler, J. N. Kohn, L. B. Barreiro, Z. P. Johnson, M. E. Wilson, J. Tung, Social
status drives social relationships in groups of unrelated female rhesus macaques. Anim.
Behav. 111, 307–317 (2016).doi:10.1016/j.anbehav.2015.10.033 Medline
16. E. A. Archie, J. Tung, M. Clark, J. Altmann, S. C. Alberts, Social affiliation matters: Both
same-sex and opposite-sex relationships predict survival in wild female baboons. Proc.
Biol. Sci. 281, 20141261 (2014).doi:10.1098/rspb.2014.1261 Medline
17. J. B. Silk, J. C. Beehner, T. J. Bergman, C. Crockford, A. L. Engh, L. R. Moscovice, R. M.
Wittig, R. M. Seyfarth, D. L. Cheney, Strong and consistent social bonds enhance the
longevity of female baboons. Curr. Biol. 20, 1359–1361
(2010).doi:10.1016/j.cub.2010.05.067 Medline
18. J. Holt-Lunstad, T. B. Smith, J. B. Layton, Social relationships and mortality risk: A meta-
analytic review. PLOS Med. 7, e1000316 (2010).doi:10.1371/journal.pmed.1000316
Medline
19. J. N. Kohn, N. Snyder-Mackler, L. B. Barreiro, Z. P. Johnson, J. Tung, M. E. Wilson,
Dominance rank causally affects personality and glucocorticoid regulation in female
rhesus macaques. Psychoneuroendocrinology 74, 179–188
(2016).doi:10.1016/j.psyneuen.2016.09.005 Medline
20. T. Lawrence, The nuclear factor NF-κB pathway in inflammation. Cold Spring Harb.
Perspect. Biol. 1, a001651 (2009).doi:10.1101/cshperspect.a001651 Medline
21. J. J. O’Shea, P. J. Murray, Cytokine signaling modules in inflammatory responses. Immunity
28, 477–487 (2008).doi:10.1016/j.immuni.2008.03.002 Medline
22. S. Akira, K. Takeda, Toll-like receptor signalling. Nat. Rev. Immunol. 4, 499–511
(2004).doi:10.1038/nri1391 Medline
23. S. A. Ramsey, S. L. Klemm, D. E. Zak, K. A. Kennedy, V. Thorsson, B. Li, M. Gilchrist, E.
S. Gold, C. D. Johnson, V. Litvak, G. Navarro, J. C. Roach, C. M. Rosenberger, A. G.
Rust, N. Yudkovsky, A. Aderem, I. Shmulevich, Uncovering a macrophage
transcriptional program by integrating evidence from motif scanning and expression
dynamics. PLOS Comput. Biol. 4, e1000021 (2008).doi:10.1371/journal.pcbi.1000021
Medline
24. M. G. Marmot, R. Sapolsky, in Sociality, Hierarchy, Health: Comparative Biodemography,
M. Weinstein, M. A. Lane, Eds. (National Academies Press, 2014), pp. 365–388.
25. R. Hamilton, M. Siva-Jothy, M. Boots, Two arms are better than one: Parasite variation leads
to combined inducible and constitutive innate immune responses. Proc. Biol. Sci. 275,
937–945 (2008).doi:10.1098/rspb.2007.1574 Medline
26. J. Teas, T. Richie, H. Taylor, C. Southwick, in The Macaques: Studies in Ecology, Behavior
and Evolution (Van Nostrand Reinhold Company, 1980), pp. 247–262.
27. A. E. Elo, The Rating of Chess Players, Past and Present (Arco, 1978).
28. C. Neumann, L. Kulik, Package ‘EloRating’: Animal Dominance Hierarchies by Elo Rating,
version 0.43 (2014); https://cran.r-project.org/web/packages/EloRating/EloRating.pdf.
29. P. C. H. Albers, H. de Vries, Elo-rating as a tool in the sequential estimation of dominance
strengths. Anim. Behav. 61, 489–495 (2001). doi:10.1006/anbe.2000.1571
30. V. Michopoulos, M. P. Diaz, M. E. Wilson, Social change and access to a palatable diet
produces differences in reward neurochemistry and appetite in female monkeys. Physiol.
Behav. 162, 102–111 (2016).doi:10.1016/j.physbeh.2016.04.023 Medline
31. A. V. Zimin, A. S. Cornish, M. D. Maudhoo, R. M. Gibbs, X. Zhang, S. Pandey, D. T.
Meehan, K. Wipfler, S. E. Bosinger, Z. P. Johnson, G. K. Tharp, G. Marçais, M. Roberts,
B. Ferguson, H. S. Fox, T. Treangen, S. L. Salzberg, J. A. Yorke, R. B. Norgren Jr., A
new rhesus macaque assembly and annotation for next-generation sequencing analyses.
Biol. Direct 9, 20 (2014).doi:10.1186/1745-6150-9-20 Medline
32. A. Dobin, T. R. Gingeras, Mapping RNA-seq Reads with STAR. Curr. Protoc.
Bioinformatics 51, 11.14.1–11.14.19 (2015).doi:10.1002/0471250953.bi1114s51 Medline
33. S. Anders, P. T. Pyl, W. Huber, HTSeq—a Python framework to work with high-throughput
sequencing data. Bioinformatics 31, 166–169 (2014). Medline
34. M. E. Ritchie, B. Phipson, D. Wu, Y. Hu, C. W. Law, W. Shi, G. K. Smyth, limma powers
differential expression analyses for RNA-sequencing and microarray studies. Nucleic
Acids Res. 43, e47 (2015).doi:10.1093/nar/gkv007 Medline
35. M. Lipatov, K. Sanjeev, R. Patro, K. Veeramah,
http://biorxiv.org/content/early/2015/07/29/023374 (2015).
36. J. Yu, G. Pressoir, W. H. Briggs, I. Vroh Bi, M. Yamasaki, J. F. Doebley, M. D. McMullen,
B. S. Gaut, D. M. Nielsen, J. B. Holland, S. Kresovich, E. S. Buckler, A unified mixed-
model method for association mapping that accounts for multiple levels of relatedness.
Nat. Genet. 38, 203–208 (2006).doi:10.1038/ng1702 Medline
37. X. Zhou, M. Stephens, Genome-wide efficient mixed-model analysis for association studies.
Nat. Genet. 44, 821–824 (2012).doi:10.1038/ng.2310 Medline
38. H. M. Kang, N. A. Zaitlen, C. M. Wade, A. Kirby, D. Heckerman, M. J. Daly, E. Eskin,
Efficient control of population structure in model organism association mapping.
Genetics 178, 1709–1723 (2008).doi:10.1534/genetics.107.080101 Medline
39. D. Akdemir, O. U. Godfrey, Package ‘EMMREML’: Fitting Mixed Models with Known
Covariance Structures, version 3.1 (2015); https://cran.r-
project.org/web/packages/EMMREML/EMMREML.pdf.
40. F. A. Wright, P. F. Sullivan, A. I. Brooks, F. Zou, W. Sun, K. Xia, V. Madar, R. Jansen, W.
Chung, Y.-H. Zhou, A. Abdellaoui, S. Batista, C. Butler, G. Chen, T.-H. Chen, D.
D’Ambrosio, P. Gallins, M. J. Ha, J. J. Hottenga, S. Huang, M. Kattenberg, J. Kochar, C.
M. Middeldorp, A. Qu, A. Shabalin, J. Tischfield, L. Todd, J.-Y. Tzeng, G. van
Grootheest, J. M. Vink, Q. Wang, W. Wang, W. Wang, G. Willemsen, J. H. Smit, E. J. de
Geus, Z. Yin, B. W. J. H. Penninx, D. I. Boomsma, Heritability and genomics of gene
expression in peripheral blood. Nat. Genet. 46, 430–437 (2014).doi:10.1038/ng.2951
Medline
41. J. Tung, X. Zhou, S. C. Alberts, M. Stephens, Y. Gilad, The genetic architecture of gene
expression levels in wild baboons. eLife 4, e04729 (2015).doi:10.7554/eLife.04729
Medline
42. J. D. Storey, R. Tibshirani, Statistical significance for genomewide studies. Proc. Natl. Acad.
Sci. U.S.A. 100, 9440–9445 (2003).doi:10.1073/pnas.1530509100 Medline
43. G. Bindea, B. Mlecnik, H. Hackl, P. Charoentong, M. Tosolini, A. Kirilovsky, W.-H.
Fridman, F. Pagès, Z. Trajanoski, J. Galon, ClueGO: A Cytoscape plug-in to decipher
functionally grouped gene ontology and pathway annotation networks. Bioinformatics 25,
1091–1093 (2009).doi:10.1093/bioinformatics/btp101 Medline
44. Y. Benjamini, Y. Hochberg, Controlling the false discovery rate: A practical and powerful
approach to multiple testing. J. R. Stat. Soc. B 57, 289–300 (1995).
45. X. Wen, M. Stephens, Bayesian Methods for genetic association analysis with heterogeneous
subgroups: From meta-analyses to gene-environment interactions. Ann. Appl. Stat. 8,
176–203 (2014).doi:10.1214/13-AOAS695 Medline
46. J. D. Buenrostro, B. Wu, H. Y. Chang, W. J. Greenleaf, Atac-seq: A method for assaying
chromatin accessibility genome-wide. Curr. Protoc. Mol. Biol. 109, 21.29.1–21.29.9
(2015).doi:10.1002/0471142727.mb2129s109 Medline
47. H. Li, https://arxiv.org/abs/1303.3997 (2013).
48. Y. Zhang, T. Liu, C. A. Meyer, J. Eeckhoute, D. S. Johnson, B. E. Bernstein, C. Nusbaum, R.
M. Myers, M. Brown, W. Li, X. S. Liu, Model-based analysis of ChIP-Seq (MACS).
Genome Biol. 9, R137 (2008).doi:10.1186/gb-2008-9-9-r137 Medline
49. A. Sandelin, W. Alkema, P. Engström, W. W. Wasserman, B. Lenhard, JASPAR: An open-
access database for eukaryotic transcription factor binding profiles. Nucleic Acids Res. 32
(suppl. 1), D91–D94 (2004).doi:10.1093/nar/gkh012 Medline
50. V. Matys, O. V. Kel-Margoulis, E. Fricke, I. Liebich, S. Land, A. Barre-Dirrie, I. Reuter, D.
Chekmenev, M. Krull, K. Hornischer, N. Voss, P. Stegmaier, B. Lewicki-Potapov, H.
Saxel, A. E. Kel, E. Wingender, TRANSFAC® and its module TRANSCompel
®:
Transcriptional gene regulation in eukaryotes. Nucleic Acids Res. 34 (suppl. 1), D108–
D110 (2006).doi:10.1093/nar/gkj143 Medline
51. R. Pique-Regi, J. F. Degner, A. A. Pai, D. J. Gaffney, Y. Gilad, J. K. Pritchard, Accurate
inference of transcription factor binding from DNA sequence and chromatin accessibility
data. Genome Res. 21, 447–455 (2011).doi:10.1101/gr.112623.110 Medline
52. A. R. Quinlan, I. M. Hall, BEDTools: A flexible suite of utilities for comparing genomic
features. Bioinformatics 26, 841–842 (2010).doi:10.1093/bioinformatics/btq033 Medline
53. J. Friedman, T. Hastie, R. Tibshirani, Regularization paths for generalized linear models via
coordinate descent. J. Stat. Softw. 33, 1–22 (2010).doi:10.18637/jss.v033.i01 Medline
54. S. S. Shen-Orr, R. Tibshirani, P. Khatri, D. L. Bodian, F. Staedtler, N. M. Perry, T. Hastie,
M. M. Sarwal, M. M. Davis, A. J. Butte, Cell type–specific gene expression differences
in complex tissues. Nat. Methods 7, 287–289 (2010).doi:10.1038/nmeth.1439 Medline