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: jt5@duke.edu (J.T.); luis.barreiro@umontreal.ca (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