SUPPLEMENTARY INFORMATION doi:10.1038/nature11981
SUPPLEMENTARY NOTES
Our validated model highlights at least 12 novel regulators that either positively or
negatively impact the Th17 program (Fig. 4 and 5). Remarkably, we found that these and
known regulators are organized in two tightly coupled, self reinforcing and mutually
antagonistic modules, whose coordinated action may explain how the balance between
Th17, Treg, and other effector T cells is maintained, and how progressive directional
differentiation of Th17 cells is achieved while repressing differentiation of other T cell
subsets. We further validated and characterized the function of four of the 12 regulators –
Mina, Fas, Pou2af1, and Tsc22d3 – by undertaking Th17 differentiation of T cells from
corresponding knockout mice or with ChIP-Seq binding profiles.
Other regulators also highlight exciting predictions. For example, among the novel ‘Th17
positive’ factors is the zinc finger E-box binding homeobox 1 Zeb1, which is early-
induced and sustained in the Th17 time course (Supplementary Fig. 13a), analogous to
the expression of many known key Th17 factors. Zeb1 knockdown decreases the
expression of Th17 signature cytokines (including IL-17A, IL-17F, and IL-21) and TFs
(including Rbpj, Maff, and Mina) and of late induced cytokine and cytokine receptor
genes (p<10-4, cluster C19, Supplementary Tables 5 and 7). It is bound in Th17 cells by
ROR-γt, Batf and Stat3, and is down-regulated in cells from Stat3 knockout mice
(Supplementary Fig. 13a). Interestingly, Zeb1 is known to interact with the chromatin
factor Smarca4/Brg1 to repress the E-cadherin promoter in epithelial cells and induce an
epithelial-mesenchymal transition1. Smarca4 is a regulator in all three network models
(Fig. 2d) and a member of the ‘positive module’ (Fig. 4b). Although it is not
WWW.NATURE.COM/NATURE | 1
2 RESEARCH SUPPLEMENTARY INFORMATION
differentially expressed in the Th17 time course, it is bound by Batf, Irf4 and Stat3
(positive regulators of Th17), but also by Gata3 and Stat5 (positive regulators of other
lineages, Supplementary Fig. 13a). Chromatin remodeling complexes that contain
Smarca4 are known to displace nucleosomes and remodel chromatin at the IFN-γ
promoter and promote its expression in Th1 cells2. There are also potential Smarca4
binding DNA sequences within the vicinity of the IL-17a promoter3. Taken together, this
suggests a model where chromatin remodeling by Smarca4, possibly in interaction with
Zeb1, positive regulates Th17 cells and is essential for IL-17 expression.
Conversely, among the novel ‘Th17 negative’ factors is Sp4, an early-induced gene,
predicted in our model as a regulator of ROR-γt and as a target of ROR-γt, Batf, Irf4,
Stat3 and Smarca4 (Supplementary Fig. 13b). Sp4 knockdown results in an increase in
ROR-γt expression at 48h, and an overall stronger and “cleaner” Th17 differentiation as
reflected by an increase in the expression of Th17 signature genes, including IL-17, IL-21
and Irf4, and decrease in the expression of signature genes of other CD4+ cells, including
Gata3, Foxp3 and Stat4 (Supplementary Tables 5 and 7).
These novel and known regulatory factors act coordinately to orchestrate intra- and inter-
modules interactions and to promote progressive differentiation of Th17 cells, while
limiting modules that inhibit directional differentiation of this subset and promote
differentiation of T cells into other T cell subsets. For instance, knockdown of Smarca4
and Zeb1 leads to decrease in Mina (due to all-positive interactions between Th17
‘positive regulators’), while knockdown of Smarca4 or Mina leads to increase in Tsc22d3
2 | WWW.NATURE.COM/NATURE 3 SUPPLEMENTARY INFORMATION RESEARCH
expression, due to negative cross-module interactions. As we have shown using RNA-
seq, these effects extend beyond the expression of regulatory factors in the network and
globally affect the Th17 transcriptional program: e.g. knock-down of Mina has
substantial effects on the progression of the Th17 differentiation network from the
intermediate to the late phase, as some of its affected down-regulated genes significantly
overlap the respective temporal clusters (p<10-5, e.g., clusters C9, C19; Supplementary
Table 7). An opposite trend is observed for the negative regulators Tsc22d3 and Sp4. For example, the transcriptional regulator Sp4 represses differentiating Th17 cells from entering into the late phase of differentiation by inhibiting the cytokine signaling (C19; p<10-7) and heamatopoesis (C20; p<10-3) clusters, which include Ahr, Batf, ROR-γt, etc.
These findings emphasize the power of large-scale functional perturbation studies in
understanding the action of complex molecular circuits that govern Th17 differentiation.
SUPPLEMENTARY METHODS
mRNA measurements on Nanostring nCounter
Details on the nCounter system are presented in full in Geiss et al.4. We used a custom
CodeSet constructed to detect a total of 293 genes, selected as described above, including
18 control genes whose expression remain unaffected during the time course. Given the scarcity of input mRNA derived from each NW knockdown, a Nanostring-CodeSet specific, 14 cycle Specific Target Amplification (STA) protocol was performed according to the manufacturer’s recommendations by adding 5 µL of TaqMan PreAmp
Master Mix (Invitrogen) and 1 µL of pooled mixed primers (500 nM each, see
Supplementary Table 6 for primer sequences) to 5 µL of cDNA from a validated
4 WWW.NATURE.COM/NATURE | 3 RESEARCH SUPPLEMENTARY INFORMATION
knockdown. After amplification, 5 µL of the amplified cDNA product was melted at
95oC for 2 minutes, snap cooled on ice, and then hybridized with the CodeSet at 65oC for
16 hours. Finally, the hybridized samples were loaded into the nCounter prep station and
product counts were quantified using the nCounter Digital Analyzer following the
manufacturer’s instructions. Samples that were too concentrated after amplification were
diluted and rerun. Serial dilutions (1:1, 1:4, 1:16, & 1:64, pre-STA) of whole spleen and
Th17 polarized cDNAs were used to both control for the effects of different amounts of
starting input material and check for biases in sample amplification.
mRNA measurements on the Fluidigm BioMark HD
cDNA from validated knockdowns was prepared for quantification on the Fluidigm
BioMark HD. Briefly, 5 µL of TaqMan PreAmp Master Mix (Invitrogen), 1 µL of
pooled mixed primers (500 nM each, see Supplementary Table 6 for primers), and 1.5
µL of water were added to 2.5 µL of knockdown validated cDNA and 14 cycles of STA
were performed according to the manufacturer’s recommendations. After the STA, an
Exonuclease I digestion (New England Biosystems) was performed to remove
unincorporated primers by adding 0.8 µL Exonuclease I, 0.4 µL Exonuclease I Reaction
Buffer and 2.8 µL water to each sample, followed by vortexing, centrifuging and heating
the sample to 37°C for 30 minutes. After a 15 minute 80°C heat inactivation, the
amplified sample was diluted 1:5 in Buffer TE. Amplified validated knockdowns and
whole spleen and Th17 serial dilution controls (1:1, 1:4, 1:16, & 1:64, pre-STA) were
then analyzed using EvaGreen and 96x96 gene expression chips (Fluidigm BioMark
HD)5.
5 4 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
mRNA measurements using RNA-Seq
Validated single stranded cDNAs from the NW-mediated knockdowns were converted to double stranded DNA using the NEBNext mRNA Second Strand Synthesis Module (New
England BioLabs) according to the manufacturer’s recommendations. The samples were then cleaned using 0.9x SPRI beads (Beckman Coulter). Libraries were prepared using the Nextera XT DNA Sample Prep Kit (Illumina), quantified, pooled, and then sequenced on the HiSeq 2500 (Illumnia) to an average depth 20M reads.
Nanostring nCounter data analysis
For each sample, we divided the count values by the sum of counts that were assigned to a set of control genes that showed no change (in time or between treatments) in the microarray data (18 genes altogether). For each condition, we computed a change fold ratio, comparing to at least three different control samples treated with non-targeting
(NT) siRNAs. We then pooled together the results of all pairwise comparisons (i.e. AxB pairs for A repeats of the condition and B control (NT) samples): we required a substantial fold change (above a threshold value t) in the same direction (up/ down regulation) in more than half of the pairwise comparisons. The threshold t was determined as max {d1, d2}, where d1 is the mean+std in the absolute log fold change between all pairs of matching NT samples (i.e., form the same batch and the same time point; d1=1.66), and where d2 is the mean + 1.645 times the standard deviation in the absolute log fold change shown by the 18 control genes (determined separately for every
6
WWW.NATURE.COM/NATURE | 5 RESEARCH SUPPLEMENTARY INFORMATION
comparison by taking all the 18xAxB values; corresponding to p=0.05, under assumption
of normality). We ignored all pairwise comparisons in which both NT and knockdown
samples had low counts before normalization (<100).
We used a permutation test to evaluate the overlap between our predicted network model
(Fig. 2) and the knockdown effects measured in the Nanostring nCounter (Fig. 4,
Supplementary Fig. 6). We computed two indices for every TF for which predicted
target were available: (i) specificity – the percentage of predicted targets that are affected
by the respective knockdown (considering only genes measured by nCounter), and (ii)
sensitivity – the percentage of genes affected by a given TF knockdown that are also its
predicted targets in the model. To avoid circularity, we exclude from this analysis target
genes predicted in the original network based on knockout alone. We combined the
resulting values (on average, 13.5% and 24.8%, respectively) into an F-score (the
harmonic mean of specificity and sensitivity). We then repeat the calculation of F-score
in 500 randomized datasets, where we shuffle the target gene labels in the knockdown
result matrix. The reported empirical p-value is:
P=(1+ #randomized datasets with equal of better F-score)/(1+ # randomized datasets)
Fluidigm data analysis
For each sample, we subtracted the Ct values from the geometric mean of the Ct values
assigned to a set of four housekeeping genes. For each condition, we computed a fold
change ratio, comparing to at least three different control samples treated with non-
targeting (NT) siRNAs. We then pooled together the results of all pairwise comparisons
7 6 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
(i.e. AxB pairs for A repeats of the condition and B control (NT) samples): we required a substantial difference between the normalized Ct values (above a threshold value) in the same direction (up/ down regulation) in more than half of the pairwise comparisons. The threshold t was determined as max {log2(1.5), d1(b), d2}, where d1(b) is the mean+std in the delta between all pairs of matching NT samples (i.e., from the same batch and the same time point), over all genes in expression quantile b (1<=b <=10). d2 is the mean +
1.645 times the standard deviation in the deltas shown by 10 control genes (the 4 housekeeping genes plus 6 control genes from the Nanostring signature); d2 is determined separately for each comparison by taking all the 10xAxB values; corresponding to p=0.05, under assumption of normality). We ignored all pairwise comparisons in which both NT and knockdown samples had low counts before normalization (Ct<21 [taking into account the amplification, this cutoff corresponds to a conventional Ct cuotff of 35]).
RNA-seq data analysis
We created a Bowtie index based on the UCSC known Gene transcriptome6, and aligned paired-end reads directly to this index using Bowtie7. Next, we ran RSEM v1.118 with default parameters on these alignments to estimate expression levels. RSEM’s gene level expression estimates (tau) were multiplied by 1,000,000 to obtain transcript per million
(TPM) estimates for each gene. We used quantile normalization to further normalize the
TPM values within each batch of samples. For each condition, we computed a fold change ratio, comparing to at least two different control samples treated with non- targeting (NT) siRNAs. We then pooled together the results of all pairwise comparisons
(i.e. AxB pairs for A repeats of the condition and B control (NT) samples): we required a
8 WWW.NATURE.COM/NATURE | 7 RESEARCH SUPPLEMENTARY INFORMATION
significant difference between the TPM values in the same direction (up/ down
regulation) in more than half of the pairwise comparisons. The significance cutoff t was
determined as max {log2(1.5), d1(b)}, where d1(b) is the mean+1.645*std in the log fold
ratio between all pairs of matching NT samples (i.e., from the same batch and the same
time point), over all genes in expression quantile b (1<=b <=20). We ignored all pairwise
comparisons in which both NT and knockdown samples had low counts (TPM<10). To
avoid spurious fold levels due to low expression values we add to the expression values a
small constant, set to the value of the 1st quantile [out of 10] of all TPM values in the
respective batch.
We use a hypergeometric test to evaluate the overlap between our predicted network
model (Fig. 2) and the knockdown effects measured by RNA-seq (Fig. 4d). As
background, we used all of the genes that appeared in the microarray data (and hence
have the potential to be included in the network). As an additional test, we used the
Wilcoxon-Mann-Whitney rank-sum test, comparing the absolute log fold-changes of
genes in the annotated set to the entire set of genes (using the same background as
before). The rank-sum test does not require setting a significance threshold; instead, it
considers the fold change values of all the genes. The p-values produced by the rank-sum
test were lower (i.e., more significant) than in the hypergeometric test, and therefore, in
Fig. 4c, we report only the more stringent (hypergeometric) p-values.
Analysis of Tsc22d3 ChIP-seq data
ChIP-seq reads were aligned to the NCBI Build 37 (UCSC mm9) of the mouse genome
9 8 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
using Bowtie9. Enriched binding regions (peaks) were detected using MACS10 with a p-
value cutoff of 10-8. We associate a peak with a gene if it falls in proximity to its 5’ end
(10kb upstream and 1kb downstream from transcription start site) or within the gene’s
body. We used the RefSeq transcript annotations for gene’s coordinates.
We assess the overlap of ChIP-seq peaks with annotated genomic regions
(Supplementary Table 8). We say that a region A overlap with a peak B if A is within a
distance of 50bp from B’s summit (as determined by MACS). The regions we used
include: (i) regulatory features annotations from the Ensemble database11; (ii) regulatory features found by the Oregano algorithm12; (iii) conserved regions annotated by the multiz30way algorithm (here we consider regions with multiz30way score>0.7); (iv) repeat regions annotated by RepeatMasker (http://www.repeatmasker.org); (v) putative promoter regions - taking 10kb upstream and 1kb downstream of transcripts annotated in
RefSeq13; (vi) gene body annotations in RefSeq; (vii) 3’ proximal regions (taking 1kb
upstream and 5kb downstream to 3’ end); (viii) regions enriched in histone marks
H3K4me3 and H3K27me3 in Th17 cells 14; (ix) regions enriched in binding of Stat3 and
Stat515, Irf4 and Batf16, and RORγt (Xiao et al unpublished) in Th17 cells, and Foxp3 in iTreg (Xiao et al., unpublished).
For each set of peaks “x” and each set of genomic regions “y”, we used a binomial p- value to assess their overlap in the genome as described in 17. The number of hits is defined as the number of x peaks that overlap with y. The background probability in sets
(i)—(vii) is set to the overall length of the region (in bp) divided by the overall length of
10 WWW.NATURE.COM/NATURE | 9 RESEARCH SUPPLEMENTARY INFORMATION
the genome. The background probability in sets (viii)—(ix) is set to the overall length of
the region divided by the overall length of annotated genomic regions: this includes
annotated regulatory regions (as defined in sets i, and ii), regions annotated as proximal
to genes (using the definitions from set v-vii), carry a histone mark in Th17 cells (using
the definition from set viii), or bound by a transcription factors in Th17 cells (using the
definitions from set ix).
For the transcription factors (set ix), we also include an additional “gene-level” test: here
we evaluated the overlap between the set of bound genes using a hypergeometric p-value.
We use a similar test to evaluate the overlap between the bound genes and genes that are
differentially expressed in Tsc22d3 knockdown.
We repeated the analysis with a second peak-calling software (Scripture18,19), and
obtained consistent results in all the above tests. Specifically, we saw similar levels of
overlap with the Th17 factors tested, both in terms of co-occupied binding sites and in
terms of common target genes.
Using literature microarray data for deriving a Th17 signature and for identifying
genes responsive to Th17-related perturbations
To define the Th17 signatures genes, we downloaded and analyzed the gene expression
data from Ref. 14 and preprocessed it using the RMA algorithm, followed by quantile
normalization using the default parameters in the ExpressionFileCreator module of the
11 10 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
GenePattern suite20. This data includes replicate microarray measurements from Th17,
Th1, Th2, iTreg, nTreg, and Naïve CD4+ T cells. For each gene, we evaluated whether it is over-expressed in Th17 cells compared to all other cell subsets using a one-sided t-test.
We retained all cases that had a p-value < 0.05. As an additional filtering step, we required that that the expression level of a gene in Th17 cells be at least 1.25 fold higher than its expression in all other cell subsets. To avoid spurious fold levels due to low expression values, we add a small constant (c=50) to the expression values.
To define genes responsive to published Th17-related perturbations, we downloaded and analyzed gene expression data from several sources that provided transcriptional profiles of Th17 cells under various conditions (listed above). These datasets were preprocessed as above. To find genes that were differentially expressed in a given condition (compared to their respective control), we computed the fold change between the expression levels of each probeset in the case and control conditions. To avoid spurious fold levels due to low expression values, we add to the expression values a small constant as above. We only reported cases where more than 50% of all of the possible case-control comparisons were above a cutoff of 1.5 fold change. As an additional filter, when duplicates are available, we computed a Z-score as above and only reported cases with a corresponding p-value < 0.05.
12 WWW.NATURE.COM/NATURE | 11 RESEARCH SUPPLEMENTARY INFORMATION
SUPPLEMENTARY FIGURES AND LEGENDS
0.5hrs, …, 72hrs, Naive,T#cells, a.& IL#6+TGF#β1! Th17, IL#17A! IL#6+TGF#β1+IL#23!
No,cytokines! an#$CD3(+(an#$CD28( Th0,
b.& c.& 4 4 10 No&cytokines& 10 TGF#β&+&IL#6& 1 0.9 # DifferenIated&Th17& )&
3 3 2 10 10 0.8 # Naive&T#cells&
0.7
102 102 0.6 FL2-H FL2-H 0.5
CD4& 0.29 31.8 101 101 0.4 CorrelaIon&(r 0.3
0 0 0.2 10 10 0 10 20 30 40 50 60 70 80 100 101 102 103 104 100 101 102 103 104 FL1-H FL1-H Time&(hr)& IL#17& d.& Rorc& IL17A& Ifng&
700 1400 # Il6+Tgfβ& 6000 600 1200 # Th0& 5000 500 1000 4000 800 400 3000 600 300
2000 400 200 mRNA expression mRNA normalized) (RMA 1000 200 100
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 Time (hr) Time (hr) Time (hr)
Fold change (log2) –2 0 2 e.& Transcriptional regulators 0.5 1 2 4 6 8 10 12 16 20 24
30 (hr) Time 42 48 50 52 60 72 ID2 ID3 JUN AES PML AHR IRF7 IRF4 IRF1 IRF8 KLF7 KLF9 UTF1 ETS1 ETV6 BCL3 BCL6 BATF VAV1 JUNB DAXX MAFF RXRA HIF1A RORA CREM RORC SP110 TBX21 BATF3 STAT4 STAT3 SAP30 STAT1 SATB1 NR4A1 FOXP1 GATA3 RUNX1 CEBPB JARID2 SMAD7 TRIM25 EOMES ARID5A NFE2L2 STAT5B PLAGL1 POU2F2 NOTCH2 NOTCH1 SERTAD1
Supplementary Figure 1. Treatment of Naïve CD4+ T-cells with TGF-β1 and IL-6
for three days induces the differentiation of Th17 cells. (a) Overview of the time
course experiments. Naïve T cells were isolated from WT mice, and treated with IL-6
and TGF-β1. We then used microarrays to measure global mRNA levels at 18 different
time points (0.5hr-72hr, Methods). As a control, we used the same WT naïve T cells
under Th0 conditions harvested at the same 18 time points. For the last four time points
(48hr - 72hr), we also profiled cells treated with IL-6, TGF-β1, and IL-23. (b) Generation
of Th17 cells by IL-6 and TGF-β1 polarizing conditions. FACS analysis of naïve T cells
13 12 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
differentiated with TGF-β1 and IL-6 (right) shows enrichment for IL-17 producing Th17
T cells; these cells are not observed in the Th0 controls. (c) Comparison of the obtained
microarray profiles to published data from naïve T-cells and differentiated Th17 cells
(Wei et. al, 2009; Supplemental Ref. 9). Shown is the Pearson correlation coefficient (Y
axis) between each of our 18 profiles (ordered by time point, X axis) and either the naïve
T cell profiles (blue) or the differentiated Th17 profiles (green). Our expression profiles
gradually transition from a naïve-like state (at t=0.5hr, r2>0.8, p<10-10) to a Th17 differentiated state (at t=72hr, r2>0.65, p<10-10). (d) Expression of key cytokines. Shown are the mRNA levels (Y axis) as measured at each of the 18 time points (X axis) in the
Th17 polarizing (blue) and Th0 control (red) conditions for the key Th17 genes RORc
(left) and IL-17a (middle), both induced, and for the cytokine IFN-γ, unchanged in our time course. (e) Transcriptional profiles of key transcriptional regulators. Shown are the differential expression levels (log2(ratio)) for each gene (column) at each of 18 time points (rows) in Th17 polarizing conditions (TGF-β1 and IL-6; left panel, Z-normalized per row) vs. control activated Th0 cells.
14 WWW.NATURE.COM/NATURE | 13 RESEARCH SUPPLEMENTARY INFORMATION
Supplementary Figure 2. Clusters of differentially expressed genes in the Th17 time
course data. For each of the 20 clusters in Fig. 1b shown are the average expression
levels (Y axis, ± standard deviation, error bars) at each time point (X axis) under Th17
polarizing (blue) and Th0 (red) conditions. The cluster size (“n”), enriched functional
annotations (“F”), and representative member genes (“M”) are denoted on top.
15 14 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
Supplementary Figure 3. Transcriptional effects of IL-23.
The late phase response depends in part on IL-23, as observed when comparing temporal
transcriptional profiles between cells stimulated with TGF-β1+IL-6 versus TGF-β1+IL-
6+IL-23, or between WT and IL-23r-/- cells treated with TGF-β1+IL-6+IL-23. For instance, in IL-23r-deficient Th17 cells, the expression of IL-17ra, IL-1r1, IL-21r, ROR-
γt, and Hif1a is decreased, and IL-4 expression is increased. The up-regulated genes in the IL-23r-/- cells are enriched for other CD4+ T cell subsets, suggesting that, in the absence of IL-23 signaling, the cells start to de-differentiate, thus further supporting the
16
WWW.NATURE.COM/NATURE | 15 RESEARCH SUPPLEMENTARY INFORMATION
hypothesis that IL-23 may have a role in stabilizing the phenotype of differentiating Th17
cells (see also Supplementary Table 1).
Shown are: (a) Transcriptional profiles of key genes. Shown are the expression levels (Y
axis) of three key genes (IL-22, RORc, IL-4) at each time point (X axis) in Th17
polarizing conditions (blue), Th0 controls (red), and following the addition of IL-23
(beginning at 48hr post differentiation) to the Th17 polarizing conditions (green). (b) IL-
23-dependnet transcriptional clusters. Shown are clusters of differentially expressed
genes in the IL-23r-/- time course data (blue) compared to WT cells, both treated with
Th17 polarizing cytokines and IL23 (red). For each cluster, shown are the average
expression levels (Y axis, ± standard deviation, error bars) at each time point (X axis) in
the knockout (blue) and wildtype (red) cells. The cluster size (“n”), enriched functional
annotations (“F”), and representative member genes (“M”) are denoted on top.
17 16 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
Early network Intermediate network Late network
ETS2 SMARCA4 SMARCA4 ETS2 SMARCA4 SP1 CHD7 SP1 ETS2 CHD7 SP1 CHD7 EGR1 SMAD2 EGR1 SMAD2 EGR1 SMAD2 NR3C1 ZFP281 NR3C1 ZFP281 NR3C1 ZFP281 SP4 SUZ12 SP4 SUZ12 SP4 SUZ12
CBFB E2F1 CBFB E2F1 CBFB E2F1
TRP53 FOXJ2 TRP53 FOXJ2 TRP53 FOXJ2
NMI CDC5L NMI CDC5L NMI CDC5L
TRIM24 TRIM28 TRIM24 TRIM28 TRIM24 TRIM28
ZFP161 MAZ ZFP161 MAZ ZFP161 MAZ
CREB1 IRF4 CREB1 IRF4 CREB1 IRF4
EGR2 IRF1 EGR2 IRF1 EGR2 IRF1
GATA3 IRF7 GATA3 IRF7 GATA3 IRF7
TBX21 IRF8 TBX21 IRF8 TBX21 IRF8
REL IRF9 REL IRF9 REL IRF9
IFI35 IRF2 IFI35 IRF2 IFI35 IRF2
FLI1 IRF3 FLI1 IRF3 FLI1 IRF3
RORC STAT3 RORC STAT3 RORC STAT3
SATB1 STAT2 SATB1 STAT2 SATB1 STAT2
NFKB1 STAT1 NFKB1 STAT1 NFKB1 STAT1
RUNX1 STAT5B RUNX1 STAT5B RUNX1 STAT5B BATF STAT4 BATF STAT4 BATF STAT4 JUN STAT5A JUN STAT5A JUN STAT5A RELA STAT6 RELA STAT6 RELA STAT6
Shape Color Size Sustained regulator Up-regulated Active regulator Non-Sustained Not Up-regulated Target Only Not Implicated
Supplementary Figure 4. At the heart of each network is its ‘transcriptional circuit’, connecting active TFs to target genes that themselves encode TFs. The transcription factor circuits shown (in each of the 3 canonical networks ) are the portions of each of the inferred networks (Supplementary Table 3) associating transcription factors to targets that themselves encode transcription factors. Yellow nodes denote transcription factor genes that are over-expressed (compared to Th0) during the respective time segment.
Edge color reflects the data type supporting the regulatory interaction (legend).
The transcriptional circuit in the early-response network connects 48 factors that are
predicted to act as regulators to 72 factors whose own transcript is up- or down-regulated
during the first four hours (a subset of this model is shown in the left panel; see
Supplementary Table 3 for the complete model). The circuit automatically highlights
many TFs that were previously implicated in immune signaling and Th17 differentiation,
either as positive or negative regulators, including Stat family members, both negative
(Stat1, Stat5) and positive (Stat3), the pioneering factor Batf, TFs targeted by TGF-β
18
WWW.NATURE.COM/NATURE | 17 RESEARCH SUPPLEMENTARY INFORMATION
signaling (Smad2, Runx1, and Irf7), several TFs targeted by TCR signaling (Rel, Nfkb1,
and Jun), and several interferon regulatory factors (Irf4 and Irf1), positioned both as
regulators and as target genes that are strongly induced. In addition, 34 regulators that
were not previously described to have a role in Th17 differentiation were identified (e.g.,
Sp4, Egr2, and Smarca4). Overall, the circuit is densely intraconnected, with 16 of the 48
regulators themselves transcriptionally controlled (e.g., Stat1, Irf1, Irf4, Batf). This
suggests feedback circuitry, some of which may be auto-regulatory (e.g., for Irf4, Stat3
and Stat1). As in the early network, there is substantial cross-regulation between the 64
TFs in the intermediate and late transcriptional circuits (middle and right panels
respectively), which include major Th17 regulators such as ROR-γt, Irf4, Batf, Stat3, and
Hif1a (see Supplementary Table 3 for the complete models).
19 18 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
a.& 0
−0.2
−0.4
−0.6
−0.8
−1
−1.2 mRNA expression −1.4 Sigmoidal approximation of mRNA
−1.6 Predicted protein levels Log10&Expression&& (normalized&by&maximum)& −1.8 0 20 40 60 80 Time&(hr)& b.& 4h 12h 24h 48h Th17 Th0 t u n o C
Rorc
Supplementary Figure 5. Predicted and validated protein levels of ROR-γt during
Th17 differentiation. (a) Shown are RORγt mRNA levels along the original time course
under Th17 polarizing conditions, as measured with microarrays (blue). A sigmoidal fit
for the mRNA levels (green) is used as an input for a model (based on Ref. 21) that predicts the level of RORγt protein at each time point (red). (b) Distribution of measured
ROR-γt protein levels (x axis) as determined by FACS analysis in Th17 polarizing conditions (blue) and Th0 conditions (red) at 4, 12, 24, and 48hr post stimulation.
20
WWW.NATURE.COM/NATURE | 19 RESEARCH SUPPLEMENTARY INFORMATION
Supplementary Figure 6. Predictive features for ranking candidates for knockdown.
Shown is the fold enrichment (Y axis, in all cases, p<10-3, hypergeometric test) in a
curated list of known Th17 factors for different (a) network-based features and (b)
expression-base features (as used in Fig. 3a).
21 20 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
a b 100 0.15 p > 0.05 10-hr knockdowns p < 0.05 75 p > 0.05 IL-17a 0.10 IL-17f Rorc 50
0.05 25 Normalized mRNA (%)
0 l t r r c c x 5 6 1 2 1 8 6 1 4 1 2 b 1 3 3 8 1 4 1 4 3 s l f r c f f h r r r y m x v c o s i p p d 2 2 g a 8 d r af1 t z 6 6
0 m c c I o o t t g I A l a 3 p g b Relative cytokine expression a d 2 r P r S M k m I m r a r t a E C C E R I i y I T V d Petri dish Nanowires Petri dish Nanowires X P r S T K N G fp 3 ou 2 T P M Notch1 Control sc 2 Z P No stimulation TGF-β + IL-6 T c
900%!
800%!
700%!
600%!
500%!
400%!
300%! Relative mRNA Expression ! Relative mRNA
200%!
100%!
0%! ! Average ! Average ! Average ! Average ! Average ! Average ! Average ! Average ! Average Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 ! Replicate 3 ! Replicate 1 ! Replicate 2 !
Ccr6! Gata3! Irf8! Mina! Smarca4! Sp4! Trim24! Tsc22d3! Zeb1!
Knockdown! Il17f! Il17a!
Supplementary Figure 7. Nanowire activation on T-cells, knockdown at 10h, and
consistency of NW-based knockdowns and resulting phenotypes. (a) Nanowires do
not activate T cells and do not interfere with physiological stimuli. Shown are the levels
of mRNA (mean ± standard error, n = 3) for key genes, measured 48hr after activation by
qPCR (Y axis, mean and standard error of the mean), in T cells grown in petri dishes
(left) or on silicon nanowires (right) without polarizing cytokines (‘no cytokines’) or in
the presence of Th17 polarizing cytokines (‘TGF-β1 + IL6’). (b) Effective knockdown by
siRNA delivered on nanowires. Shown is the % of mRNA remaining after knockdown
(by qPCR, Y axis: mean ± standard error relative to non-targeting siRNA control, n = 12,
black bar on left) at 10 hours after introduction of polarizing cytokines. The genes 22
presented are a superset of the 39 genes selected for transcriptional profiling.
(c) Shown are average target transcript reductions and phenotypic changes (as measured
by IL-17f and IL-17a expression) for three different experiments of NW-based WWW.NATURE.COM/NATURE | 21 knockdown (from at least 2 different cultures) of 9 genes at 48 hours post stimulation.
Light blue bars: knockdown level (%remaining relative to siRNA controls); dark grey
and light green bars: mRNAs of IL-17f and IL-17a, respectively, relative to siRNA
controls.
23 (by qPCR, Y axis: mean ± standard error relative to non-targeting siRNA control, n = 12,
RESEARCH SUPPLEMENTARYblack bar on left)INFORMATION at 10 hours after introduction of polarizing cytokines. The genes
presented are a superset of the 39 genes selected for transcriptional profiling.
(c) Shown are average target transcript reductions and phenotypic changes (as measured
by IL-17f and IL-17a expression) for three different experiments of NW-based
knockdown (from at least 2 different cultures) of 9 genes at 48 hours post stimulation.
Light blue bars: knockdown level (%remaining relative to siRNA controls); dark grey
and light green bars: mRNAs of IL-17f and IL-17a, respectively, relative to siRNA
controls.
23
22 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
a.* b.* r _48h r _48h r B 48 h X_48h r 4_48h r 1_48h r 8_48h r IM 24_48h r F F F BPJ _48h r R D2_48h r L17R_48h r L21R_48h r L17R A R R R MINA_48h r I CCR6_48h r PROCR_48h r MYC_48h r FAS _ NFKB1_48h r PML_48h r NR3C1_48h r POU2AF1_48h r SMARCA4_48h r ITGA3_48h r R IL6ST_48h r ZEB1_48h r EGR2_48h r KAT 2 TRPS1_48hr I IL12RB1_48h r I IKZF4_48h r I ETS2_48h r NOTCH1_48h r SP 4_48h r TSC22D3_48h r SM O T ZFP36L1_48h r I I MYD88_48h r CCR5_48h r TGIF 1_48h r EGR1_48h r ETV6_48h r BATF_48h r I 20 2 Gene data 1 CSF2 r =60.78* linear IFNG CASP4 15 ID2 * SMOX CCL4 LIF LTA 10 IL2RA IL10 CD5L * Fluidigm CD24A 5 IL21 IRF4 EGR2 ITGA3 MAFF 0 IL23R IL17A PROCR CCR6 −5 AHR −12 −10 −8 −6 −4 −2 0 2 RORA TRPS1 BATF MINA Nanostring*read*count*(log2)* TGIF1 TBX21 ETV6 NOTCH1 CD80 CEBPB HIF1A ARID5A MAF REL NFE2L2 KLRD1 IL17F IL9 IL2 INHBA CCL20 RBPJ RORC IRF8 LOC100046232;NFIL3 ICOS MYC ZFP36L1 CCR5 CASP6 ELK3 PML IL22;ILTIFB SKAP2 SOCS3 CXCR3 CTLA2B TGFBR1 KLF6 FOXP3 ID3 PSMB9 GATA3 IL7R TNFSF8 CLCF1 IL10RA TSC22D3 STAT4 FAS IL21R NUDT4 SKI NFATC2 IRF1 IL17RA SGK1 STAT1 SERPINB1A POU2AF1 TRIM24
Fold Change (Log2) - 1 1
Supplementary Figure 8. Cross-validation of the Nanostring expression profiles for each nanowire-delivered knockdown using Fluidigm 96x96 gene expression chips.
(a) Comparison of expression levels measured by Fluidigm (Y axis) and Nanostring (X axis) for the same gene under the same perturbation. Expression values were normalized to control genes as described in Methods. (b) Analysis of Fluidigm data recapitulates the partitioning of targeted factors into two modules of positive and negative Th17
24
WWW.NATURE.COM/NATURE | 23 RESEARCH SUPPLEMENTARY INFORMATION
regulators. Shown are the changes in transcription of the 86 gene signature (rows)
following knockdown of each factor (columns).
25 24 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
1
0.9
0.8
0.7
0.6
0.5
0.4
% of interactions 0.3
0.2 Sustained 0.1 Flipped Transient 0 ID2 SP4 PML IRF1 IRF8 IRF4 MYC MINA IL6ST IL21R IKZF4 CCR5 CCR6 EGR1 TGIF1 SMOX ITGA3 KAT2B TRPS1 NR3C1 MYD88 TRIM24 PROCR IL12RB1 NOTCH1 ZFP36L1 TSC22D3 POU2AF1
Supplementary Figure 9. Rewiring of the Th17 “functional” network between 10hr to 48hr post stimulation. For each regulator that was profiled at 10hr and 48hr, we calculate the percentage of “edges” (i.e., gene A is affected by perturbation of gene B) that either appear in the two time points with the same activation/repression logic
(Sustained); appear only in one time point (Transient); or appear in both networks but with a different activation/repression logic (Flipped). In the sustained edges, the perturbation effect (fold change) has to be significant in at least one of the time point
(Methods), and consistent (in terms of activation/repression) in the other time point
(using a more permissive cutoff of 1.25 fold).
26 WWW.NATURE.COM/NATURE | 25 RESEARCH SUPPLEMENTARY INFORMATION
Z=2.0 Z=1.8
Z=1.8 Z=1.8
Supplementary Figure 10. “Chromatic” network motifs. We used a ‘chromatic’
network motif analysis to find recurring sub networks with the same topology and the
same node and edge colors. Shown are the four significantly enriched motifs (p<0.05).
Red nodes: positive regulators; blue nodes: negative regulator; red edges from A to B:
knockdown of A downregulates B; blue edge: knockdown of A upregulates B. Motifs
were found using the FANMOD software 22.
27
26 | WWW.NATURE.COM/NATURE global global Cluster -0.50-0.50 0 0.500 C1 C5 globalSUPPLEMENTARY INFORMATION RESEARCH -1 0 1
Th17/posi>ve% Th17/nega>ve% Th17/posi>ve% Th17/nega>ve% a.% c.% A 8 U2 AF 1 S F O A R EGR2 FAS MINA POUF1A PROCR ETV6 IRF8 ZEB1 SMARCA4 IKZF4 SP4 TSC22D3 EGR 2 F MI N P PROC R ETV 6 I ZEB 1 SMARCA 4 IKZF 4 SP 4 TSC22D 3 Gene Gene EGR2 EGR2 ETV6 IRF8 EGR2 FAS MINA ZEB1 PROCR POU2AF1 SMARCA4 IKZF4 SP4 TSC22D3 Cell-small Gene Cluster FAS FAS HelperCell iTreg MINA MINA NEB C5 Th1 Th17 POUF1A POU2AF1 DYNLT3 C5 ADAM19 C5 Th2 PROCR PROCR ETV6 SLC12A2 C5 global ETV6 PMAIP1 C5 Cluster IRF8 IRF8 TNFRSF9 C5 - 1 C1 C2 ZEB1 0 ZEB1 1 LPXN C5 Cluster%5%(Hill%et%al)% SMARCA4 SMARCA4 FRMD4B C5 C3 C4 TGM2 C5 C5 C6 IKZF4 IKZF4 AHR C5 SP4 SP4 C7 IRF4 C5 TSC22D3 TSC22D3 LXN C5 global GADD45G C5 /0.5% 0.5% 2010111I01RIK C5 -2 0 2 HOMER1 C5 Correla>on%(Spearman)% IL2RB C5 TNFRSF4 C5 STX11 C5 EEA1 C5 b.% Th17/posi>ve% Th17/nega>ve% BCL2L1 C5 SLC4A7 C5 ZFP52 C5 1 DUSP4 C5 MAF C5 A 8 U2A F
S CSDA C5 F O A
R ALCAM C5 EGR 2 F MI N ETV 6 I PROC R P ZEB 1 SMARCA 4 IKZF 4 SP 4 TSC22D 3 CTLA4 C5 Gene Cell-small Cell-large Cluster TF Cyokine-activity Cell-surface Immu DE
Helper%cells% CST7 C5 TNFRSF8 HelperCell 0 0 1 CD810 1 C5 IL2 HelperCell 0 1 0 PAFAH1B31 1 C5 TNFSF4 HelperCell 0 1 1 NFIL31 1 C5 CTLA4 HelperCell C5 0 0 1 SAMSN11 1 C5 MAF HelperCell C5 1 0 0 1 1 IRF4 HelperCell Treg C5 1 0 0 8430427H17RIK1 1 C1 IL10 HelperCell 0 1 0 CORO2A1 1 C1 ICOS HelperCell 0 0 1 RABGAP1L1 1 C1
BACH2 C1 Cluster%1%(Hill%et%al)% EGR2 FAS ETV6 IRF8 MINA PROCR POUF1A ZEB1 SMARCA4 IKZF4 SP4 TSC22D3 IL7R HelperCell Th17 0 1 1 SLAMF11 1 C1 Gene Cell-smallCell-largeClusterTF Cyokine-activityCell-surfaceImmuDE GATA3 Th2 1 0 0 SPIN21 1 C1 GSTO1 Treg C6 0 0 0 0 1 Th2% ATP6V0A1 C1 RGS16 Treg C6 0IL4RA0 0 0 1 Th2 0 1 1 1 1 PHLDA1 Treg C6 0 0 0 0 1 CD38 C1 AHNAK Treg C6 0STAT0 0 6 0 1 Th2 1 0 0 1 1 EPHX1 Treg C6 0 0 0 0 1 NIPA2 C1 GPRASP2 Treg C6 0CCR0 0 8 0 1 Th2 0 1 0 TRIM590 1 C1 MATN2 Treg C6 0IL0 9 0 0 1 Th2 0 1 0 1 1 HIVEP3 Treg C6 1 0 0 0 1 D18ERTD653E C1 RORA C6 1IL20 40 1 1 Th2 0 1 0 2810474O19RIK0 1 C1 EGR2 Treg C6 1 0 0 0 1 PIP5K1A Treg C6 0 0 0 0 1 SAMHD1 C1 CD200 Treg C6 0ROR0 1C 0 1 Th17 Th17 1 0 0 1 1 GPR174 Treg C6 0 0 0 0 1 Th17% ECM1 C1 PTGER4 Treg C6 0IL230 0R 0 1 Th17 0 0 0 CERK 1 1 C1 ZC3H12D C6 0IL1R0 0 1 0 1 Th17 Th17 0 1 1 1 1 SLC35D1 Treg C6 0 0 0 0 1 CAPG C1 TNFRSF25 Treg C6 0CCR0 0 6 0 1 Th17 Treg C6 0 1 1 4833442J19RIK0 1 C1 IFT80 Treg C6 0 0 0 0 1 IGF1R C6 0STAT0 1 3 0 1 Th17 1 0 0 GVIN11 1 C1 PRG4 Treg C6 0IL20 10 1 1 Th17 Th17 0 1 0 1 1 BID Treg C6 0 0 0 0 1 NT5E C1 FOLR4 Treg C6 0IL12RB0 0 0 11 Th17 0 1 1 P2RY101 1 C1 SYT11 Treg C6 0 0 0 0 1 SLC22A5 Treg C6 0IL170 0 A 0 1 Th17 Th17 0 1 0 RRAGD1 1 C1 FRMD6 Treg C6 0IL170 0F 0 1 Th17 0 1 0 HIPK21 1 C1 CCR6 Th17 Treg C6 0CCL21 1 0 0 1 Th17 Cluster 0 1 0 FOXP31 1 C1
NEB Treg C5 0 0 0 0 1 Th1% PRR13 C1 DYNLT3 Treg C5 0 0 0 0 1 C1 C2 ADAM19 Treg C5 0IL12RB0 0 0 21 Th1 0 1 0 TMEM641 1 C1 SLC12A2 Treg C5 0IL180 R0 A0P1 Th1 C3 C4 0 1 1 GPR831 1 C1 PMAIP1 Treg C5 0 0 0 0 1 TNFRSF9 Treg C5 0CXCR0 1 30 1 Th1 0 1 1 MGAT50 1 C1 LPXN Treg C5 0 0 0 0 1 C5 C6 FRMD4B Treg C5 0IL18R0 0 10 1 Th1 0 1 1 1 1 TGM2 C5 0IL27R0 0 A0 1 Th1 C7 0 1 0 1 1 AHR Treg C5 1 0 0 1 1 IRF4 HelperCell Treg C5 1STAT0 0 1 1 1 Th1 iTreg % 1 0 global 0 1% 1 1 LXN Treg C5 0 0 0 0 1 /1% GADD45G Treg C5 0 0 0 1 1 2010111I01RIK Treg C5 0TGF0 B0 1 0 1 iTreg 0 -2 0 0 0 2 1 1 HOMER1 Treg C5 0 0 0 0 1 IL2RB Treg C5 0GPR81 1 3 1 1 iTreg Treg C1 0Fold%change%(log2)%0 0 0 1 TNFRSF4 Treg C5 0 0 1 1 1 STX11 Treg C5 FOXP0 0 0 3 0 1 iTreg Treg C1 1 0 0 1 1 EEA1 Treg C5 0 0 0 0 1 BCL2L1 Treg C5 0 0 0 0 1 SLC4A7 Treg C5 0 0 0 0 1 ZFP52 Treg C5 0 0 0 0 1 DUSP4 Treg C5 0 0 0 0 1 MAF HelperCell C5 1 0 0 1 1 CSDA Treg C5 1 0 0 0 1 ALCAM Treg C5 0 0 1 0 1 CTLA4 HelperCell C5 0 0 1 1 1 CST7 Treg C5 0 0 0 0 1 Supplementary FigureCD81 Treg C5 0 0 1 011 . RNA-seq analysis of nanowire-delivered knockdowns PAFAH1B3 C5 0 0 0 0 1 NFIL3 Treg C5 1 1 0 1 1 SAMSN1 Treg C5 0 0 0 0 1
8430427H17RIK Treg C1 0 0 0 0 1 CORO2A Treg C1 0 0 0 0 1 RABGAP1L Treg C1 0 0 0 0 1 BACH2 C1 1 0 0 0 1 (a) Correlation matrixSLAMF1 Treg C1 0 of0 1 knockdown0 1 profiles. Shown is the Spearman rank correlation SPIN2 Treg C1 0 0 0 0 1 ATP6V0A1 Treg C1 0 0 0 0 1 CD38 Treg C1 0 0 1 0 1 NIPA2 Treg C1 0 0 0 0 1 TRIM59 Treg C1 0 0 0 0 1 D18ERTD653E Treg C1 0 0 0 0 1 2810474O19RIK Treg C1 0 0 0 0 1 SAMHD1 Treg C1 0 0 0 1 1 coefficient betweenECM1 theTreg C1 0 0 RNA0 0 1 -Seq profiles (fold change relative to NT siRNA controls) of CERK Treg C1 0 0 0 0 1 CAPG Treg C1 0 0 0 0 1 4833442J19RIK Treg C1 0 0 0 0 1 EGR2 FAS ETV6 IRF8 MINA PROCR POUF1A ZEB1 SMARCA4 IKZF4 SP4 TSC22D3 GVIN1 Treg C1 0 0 0 0 1 NT5E Treg C1 0 0 1 0 1 Gene Cell-smallCell-largeClusterTF Cyokine-activityCell-surfaceImmuDE P2RY10 Treg C1 0 0 0 0 1 GSTO1 Treg C6 0 0 0 0 1 RRAGD Treg C1 0 0 0 0 1 RGS16 Treg C6 0 0 0 0 1 HIPK2 C1 0 0 0 0 1 PHLDA1 Treg C6 0 0 0 0 1 regulators perturbedFOXP3 iTreg TregbyC1 1 0 knockdowns0 1 1 . Genes that were not significantly differentially AHNAK Treg C6 0 0 0 0 1 PRR13 Treg C1 0 0 0 0 1 EPHX1 Treg C6 0 0 0 0 1 TMEM64 Treg C1 0 0 0 0 1 GPRASP2 Treg C6 0 0 0 0 1 GPR83 iTreg Treg C1 0 0 0 0 1 MATN2 Treg C6 0 0 0 0 1 MGAT5 Treg C1 0 0 0 0 1 HIVEP3 Treg C6 1 0 0 0 1 RORA C6 1 0 0 1 1 TWSG1 Treg C4 0 0 0 0 1 EGR2 Treg C6 1 0 0 0 1 PLSCR1 C4 0 0 0 1 1 PIP5K1A Treg C6 0 0 0 0 1 expressed in any ofIL2RA theTreg C4 0 1 samples1 1 1 were excluded fromCD200 theTreg C6 0 profiles0 1 0 1 . (b) Knockdown effects LTA Treg C4 0 1 0 1 1 GPR174 Treg C6 0 0 0 0 1 OSBPL3 Treg C4 0 0 0 0 1 PTGER4 Treg C6 0 0 0 0 1 BMPR2 Treg C4 0 0 0 0 1 ZC3H12D C6 0 0 0 0 1 LONRF1 Treg C4 0 0 0 0 1 SLC35D1 Treg C6 0 0 0 0 1 SH3BGRL2 Treg C4 0 0 0 0 1 TNFRSF25 Treg C6 0 0 0 0 1 TNFSF11 Treg C4 0 1 1 1 1 IFT80 Treg C6 0 0 0 0 1 SESTD1 C4 0 0 0 0 1 IGF1R C6 0 0 1 0 1 PRNP Treg C4 0 0 1 0 1 PRG4 Treg C6 0 0 0 1 1 on known markerENDOD1 genesTreg C4 0 0 0of0 1 different CD4+ T cell lineagesBID Treg C6 0 0 . 0 Shown0 1 are the expression SOCS2 Treg C4 0 0 0 0 1 FOLR4 Treg C6 0 0 0 0 1 RCN1 Treg C4 0 0 0 0 1 SYT11 Treg C6 0 0 0 0 1 ADAMTS6 Treg C4 0 0 0 0 1 SLC22A5 Treg C6 0 0 0 0 1 MBOAT1 C4 0 0 0 0 1 FRMD6 Treg C6 0 0 0 0 1 FABP5 Treg C4 0 0 0 0 1 CCR6 Th17 Treg C6 0 1 1 0 1 GATA1 Treg C4 1 0 0 0 1 P4HA1 Treg C4 0 0 0 0 1 NEB Treg C5 0 0 0 0 1 MAP3K8 Treg C4 0 0 0 0 1 DYNLT3 Treg C5 0 0 0 0 1 levels for canonicalRGS1 genesTreg C4 0 0 0 0 (rows)1 of different T cell ADAM19lineagesTreg C5 0 0 0 0 1 (labeled on right) following TNFRSF18 Treg C4 0 0 0 0 1 SLC12A2 Treg C5 0 0 0 0 1 MXD1 Treg C4 1 0 0 0 1 PMAIP1 Treg C5 0 0 0 0 1 NDRG1 Treg C4 0 0 0 0 1 TNFRSF9 Treg C5 0 0 1 0 1 MELK Treg C4 0 0 0 0 1 LPXN Treg C5 0 0 0 0 1 CISH Treg C4 0 0 0 0 1 FRMD4B Treg C5 0 0 0 0 1 AI480653 Treg C4 0 0 0 0 1 TGM2 C5 0 0 0 0 1 SEC24A Treg C4 0 0 0 0 1 AHR Treg C5 1 0 0 1 1 PROS1 Treg C4 0 0 0 0 1 IRF4 HelperCell Treg C5 1 0 0 1 1 knockdown of each of 12 regulators (columns). Red/Blue:LXN Treg C5 0 0 0 increase/decrease0 1 in gene GMFG Th17 C7 0 0 0 0 1 GADD45G Treg C5 0 0 0 1 1 PHXR4 C7 0 0 0 0 1 2010111I01RIK Treg C5 0 0 0 0 1 PDK1 C7 0 0 0 0 1 HOMER1 Treg C5 0 0 0 0 1 PIK3R5 C7 0 0 0 0 1 IL2RB Treg C5 0 1 1 1 1 ATP1B1 C7 0 0 0 0 1 TNFRSF4 Treg C5 0 0 1 1 1 AMIGO2 C7 0 0 0 0 1 STX11 Treg C5 0 0 0 0 1 SCML4 C7 0 0 0 0 1 EEA1 Treg C5 0 0 0 0 1 RASGRP2 C7 0 0 0 0 1 BCL2L1 Treg C5 0 0 0 0 1 ST8SIA1 C7 0 0 0 0 1 SLC4A7 Treg C5 0 0 0 0 1 SPNB2 C7 0 0 0 0 1 ZFP52 Treg C5 0 0 0 0 1 F2RL1 C7 0 0 0 0 1 DUSP4 Treg C5 0 0 0 0 1 ALS2CL C7 0 0 0 0 1 MAF HelperCell C5 1 0 0 1 1 TGFBR2 C7 0 0 0 1 1 CSDA Treg C5 1 0 0 0 1 TCF7 C7 1 0 0 0 1 ALCAM Treg C5 0 0 1 0 1 1190002H23RIK C7 0 0 0 0 1 CTLA4 HelperCell C5 0 0 1 1 1 GRAMD1A C7 0 0 0 0 1 CST7 Treg C5 0 0 0 0 1 POU2AF1 C7 1 0 0 0 1 CD81 Treg C5 0 0 1 0 1 XKRX C7 0 0 0 0 1 PAFAH1B3 C5 0 0 0 0 1 CCDC22 C7 0 0 0 0 1 NFIL3 Treg C5 1 1 0 1 1 SLC12A7 C7 0 0 0 0 1 SAMSN1 Treg C5 0 0 0 0 1 28
AW112010 Treg C3 0 0 0 0 1 8430427H17RIK Treg C1 0 0 0 0 1 GPR171 C3 0 0 0 0 1 CORO2A Treg C1 0 0 0 0 1 WWW.NATURE.COM/NATURE | 27 ID2 C3 1 0 0 1 1 RABGAP1L Treg C1 0 0 0 0 1 SLA Treg C3 0 0 0 0 1 BACH2 C1 1 0 0 0 1 KLRD1 C3 0 0 1 0 1 SLAMF1 Treg C1 0 0 1 0 1 UGCG C3 0 0 0 0 1 SPIN2 Treg C1 0 0 0 0 1 VPS54 Treg C3 0 0 0 0 1 ATP6V0A1 Treg C1 0 0 0 0 1 TMEM9B C3 0 0 0 0 1 CD38 Treg C1 0 0 1 0 1 GPR15 C3 0 0 0 0 1 NIPA2 Treg C1 0 0 0 0 1 TRIM59 Treg C1 0 0 0 0 1 D18ERTD653E Treg C1 0 0 0 0 1 1110038D17RIK C2 0 0 0 0 1 2810474O19RIK Treg C1 0 0 0 0 1 METTL9 C2 0 0 0 0 1 SAMHD1 Treg C1 0 0 0 1 1 SEMA4A C2 0 0 0 0 1 ECM1 Treg C1 0 0 0 0 1 ITK C2 0 0 0 0 1 CERK Treg C1 0 0 0 0 1 PDE3B C2 0 0 0 0 1 CAPG Treg C1 0 0 0 0 1 RANBP10 C2 0 0 0 0 1 4833442J19RIK Treg C1 0 0 0 0 1 TMCC3 C2 0 0 0 0 1 GVIN1 Treg C1 0 0 0 0 1 DOPEY1 C2 0 0 0 0 1 NT5E Treg C1 0 0 1 0 1 LEF1 C2 1 0 0 0 1 P2RY10 Treg C1 0 0 0 0 1 PDLIM4 C2 0 0 0 0 1 RRAGD Treg C1 0 0 0 0 1 ADD3 C2 0 0 0 0 1 HIPK2 C1 0 0 0 0 1 TMEM71 C2 0 0 0 0 1 FOXP3 iTreg Treg C1 1 0 0 1 1 IGFBP4 C2 0 0 0 0 1 PRR13 Treg C1 0 0 0 0 1 ENTPD5 C2 0 0 1 0 1 TMEM64 Treg C1 0 0 0 0 1 ITGB3 C2 0 0 1 0 1 GPR83 iTreg Treg C1 0 0 0 0 1 ACPL2 C2 0 0 0 0 1 MGAT5 Treg C1 0 0 0 0 1 GPD2 C2 0 0 0 0 1 CHD7 C2 1 0 0 0 1 KLHDC2 C2 0 0 0 0 1 TWSG1 Treg C4 0 0 0 0 1 LYST C2 0 0 0 1 1 PLSCR1 C4 0 0 0 1 1 EMB Th17 C2 0 0 0 0 1 IL2RA Treg C4 0 1 1 1 1 ENC1 C2 0 0 0 0 1 LTA Treg C4 0 1 0 1 1 OSBPL3 Treg C4 0 0 0 0 1 BMPR2 Treg C4 0 0 0 0 1 LONRF1 Treg C4 0 0 0 0 1 SH3BGRL2 Treg C4 0 0 0 0 1 TNFSF11 Treg C4 0 1 1 1 1 SESTD1 C4 0 0 0 0 1 PRNP Treg C4 0 0 1 0 1 ENDOD1 Treg C4 0 0 0 0 1 SOCS2 Treg C4 0 0 0 0 1 RCN1 Treg C4 0 0 0 0 1 ADAMTS6 Treg C4 0 0 0 0 1 MBOAT1 C4 0 0 0 0 1 FABP5 Treg C4 0 0 0 0 1 GATA1 Treg C4 1 0 0 0 1 P4HA1 Treg C4 0 0 0 0 1 MAP3K8 Treg C4 0 0 0 0 1 RGS1 Treg C4 0 0 0 0 1 TNFRSF18 Treg C4 0 0 0 0 1 MXD1 Treg C4 1 0 0 0 1 NDRG1 Treg C4 0 0 0 0 1 MELK Treg C4 0 0 0 0 1 CISH Treg C4 0 0 0 0 1 AI480653 Treg C4 0 0 0 0 1 SEC24A Treg C4 0 0 0 0 1 PROS1 Treg C4 0 0 0 0 1
GMFG Th17 C7 0 0 0 0 1 PHXR4 C7 0 0 0 0 1 PDK1 C7 0 0 0 0 1 PIK3R5 C7 0 0 0 0 1 ATP1B1 C7 0 0 0 0 1 AMIGO2 C7 0 0 0 0 1 SCML4 C7 0 0 0 0 1 RASGRP2 C7 0 0 0 0 1 ST8SIA1 C7 0 0 0 0 1 SPNB2 C7 0 0 0 0 1 F2RL1 C7 0 0 0 0 1 ALS2CL C7 0 0 0 0 1 TGFBR2 C7 0 0 0 1 1 TCF7 C7 1 0 0 0 1 1190002H23RIK C7 0 0 0 0 1 GRAMD1A C7 0 0 0 0 1 POU2AF1 C7 1 0 0 0 1 XKRX C7 0 0 0 0 1 CCDC22 C7 0 0 0 0 1 SLC12A7 C7 0 0 0 0 1
AW112010 Treg C3 0 0 0 0 1 GPR171 C3 0 0 0 0 1 ID2 C3 1 0 0 1 1 SLA Treg C3 0 0 0 0 1 KLRD1 C3 0 0 1 0 1 UGCG C3 0 0 0 0 1 VPS54 Treg C3 0 0 0 0 1 TMEM9B C3 0 0 0 0 1 GPR15 C3 0 0 0 0 1
1110038D17RIK C2 0 0 0 0 1 METTL9 C2 0 0 0 0 1 SEMA4A C2 0 0 0 0 1 ITK C2 0 0 0 0 1 PDE3B C2 0 0 0 0 1 RANBP10 C2 0 0 0 0 1 TMCC3 C2 0 0 0 0 1 DOPEY1 C2 0 0 0 0 1 LEF1 C2 1 0 0 0 1 PDLIM4 C2 0 0 0 0 1 ADD3 C2 0 0 0 0 1 TMEM71 C2 0 0 0 0 1 IGFBP4 C2 0 0 0 0 1 ENTPD5 C2 0 0 1 0 1 ITGB3 C2 0 0 1 0 1 ACPL2 C2 0 0 0 0 1 GPD2 C2 0 0 0 0 1 CHD7 C2 1 0 0 0 1 KLHDC2 C2 0 0 0 0 1 LYST C2 0 0 0 1 1 EMB Th17 C2 0 0 0 0 1 ENC1 C2 0 0 0 0 1 RESEARCH SUPPLEMENTARY INFORMATION
expression in knockdown compared to non-targeting control (Methods). Shown are only
genes that are significantly differentially expressed in at least one knockdown condition.
The experiments are hierarchically clustered, forming distinct clusters for Th17-positive
regulators (left) and Th17-neagtive regulators (right). (c) Knockdown effects on two sub-
clusters of the T-regulatory cell signature, as defined by Hill et al23. Each cluster
(annotated in Hill et al as Clusters 1 and 5) includes genes that are over expressed in T-
regs cells compared to conventional T cells. However, genes in Cluster 1 are more
correlated to Foxp3 and responsive to Foxp3 transduction. Conversely, genes in cluster 1
are more directly responsive to TCR and IL-2 and less responsive to Foxp3 in Treg cells.
Knockdown of Th17-positive regulators strongly induces the expression of genes in the
‘Foxp3’ Cluster 1. The knockdown profiles are hierarchically clustered, forming distinct
clusters for Th17-positive regulators (left) and Th17-neagtive regulators (right).
Red/Blue: increase/decrease in gene expression in knockdown compared to non-targeting
control (Methods). Shown are only genes that are significantly differentially expressed in
at least one knockdown condition.
29 28 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
WT MINA -/- *** *** WT POU2AF1 -/- a 4000 b 3.71 0.019 3.38 0.023 3.25 0 1.07 7.54e-3 3000 ) l m
pg / 2000 F (
No cytokines N No cytokines T 1000
0 7 7 0 0 96.1 0.17 96.5 0.082 h 96.6 0.17 98.5 0.44 IFN- Ȗ Wt T Wt Th1
MinA KO Th 0.16 0.047 IL-17a MinA KO Th1 4.62e-3 0
0.8 0.49 TGF-β1+ IL-6 No Cytokines
No cytokines 82.9 16.9 90.9 9.12 IFN- Ȗ
CD4 IL-17a FOXP3
600 5000 WT FAS -/- 4000 WT OCT1 def. *** c d 4000 )
l
4.34 0.069 8.27 0.053 l 0.89 0.048 5.58 0.042 3000 400 m )
l 3000 g ng/ m m p ( pg / 17
( 2000 2000
200 IF N IL - 2 - L I No cytokines 1000 1000 0 0 7 7 0 0 7 7 0 0 h h No Cytokines 0 h1 h1 Wt T Wt T 7 7 0 0 Wt Th1 Wt Th1 h def. T def. T 1 1 Oct1 def. Th Oct1 def. Th Wt T 94.8 0.83 91.2 0.46 Oct Oct 98.6 0.43 94.2 0.21 Wt Th1 Fas KO Th Fas KO Th1 ns 0.083 4.15e-3 0.072 0.016 4000 ns 4000 0.7 0.31 1.57 0.51 5000
3000 3000 ) ) l 4000 l m m ) l pg / pg / m ( 2000 2000 3000 F ( 2 -
pg / TGF-β1+ IL-6 N L I T F ( 2000 1000 1000 N T
TGF-`1 + IL-6 1000 0 0 7 7 0 0 7 7 0 0 h h 0 77 22.9 88.7 11.2 h1 IFN- Ȗ Wt T Wt T 7 0 Wt Th1 Wt Th1 59.2 39.8 72.7 25.2 7 0 def. T FOXP3 h 1 Oct1 def. Th Oct1 def, Th Wt T IL-17a OCT1 def. Th1 Oct IL-17a Wt Th1 Fas KO Th Fas KO Th1
Supplementary Figure 12. Quantification of cytokine production in knockout cells
at 72h of in-vitro differentiation using Flow cytometry and Enzyme-linked
immunosorbent assay (ELISA).
All flow cytometry figures shown, except for Oct1, are representative of at least 3
repeats, and all ELISA data has at least 3 replicates. For Oct1, only a limited amount of
cells were available from reconstituted mice, allowing for only 2 repeats of the Oct1
deficient mouse for flow cytometry and ELISA. (a, left) Mina-/- T cells activated under
Th0 controls are controls for the graphs shown in Fig. 5a. (a, right) TNF secretion by
Mina-/- and WT cells, as measured by cytometric bead assay showing that Mina-/- T cells
produce more TNF when compared to control. (b) Intracellular cytokine staining of
Pou2af1-/- and WT cells for IFN-γ and IL-17a as measured by flowcytometry. (c, left)
Flow cytometric analysis of Fas-/- and WT cells for Foxp3 and Il-17 expression. (c, right) IL-2 and Tnf secretion by Fas-/- and WT cells, as measured by a cytokine bead assay ELISA. (d, left). Flow cytometry on Oct1-/- and WT cells for IFN-γ and IL-17a,
30
WWW.NATURE.COM/NATURE | 29 RESEARCH SUPPLEMENTARY INFORMATION
showing an increase in IFN-γ positive cells in the Th0 condition for the Oct1 deficient
mouse. (d, right) Il-17a, IFN-γ, IL-2 and TNF production by Oct1-/- and WT cells, as
measured by cytokine ELISA and cytometric bead assay. Statistical significance in the
ELISA figures is denoted by: * p < 0.05, ** p < 0.01, and *** p <0.001.
31 30 | WWW.NATURE.COM/NATURE Figure 5 a WT MINA -/- 0.021 6.02e-3 0.045 2.99e-3 *** * 175 1500 IL23R TGF-`1 + IL-6 150 1000 BATF 125 RORC 500 80.1 19.8 91.6 8.38 100 IFN- a (ng/ml) 75 IL-17a 60 50 IL-17 MYC MINA IFN- g (ng/ml) 40 25 20 2.39 10 0 0 TGF-`1 + IL-6 7 7 0 0 7 7 0 0 IL17A FOXP3 1 h O Th Wt T WT Th Wt Th1 WT Th K A in MinA KO Th
CD4 M MinA KO Th1 MinA KO Th1 Foxp3 b WT FAS -/- 3.04 0.15 16.2 0.32 400 3000 *** *** *** STAT3 BATF IL23R No Cytokines 300 2000
(ng/ml) 200 96.4 0.37 83.2 0.3 FAS RBPJ IL-17 0.27 1.26 1.12 1.85 IFN- g (ng/ml) 1000 100
TGF-`1 + IL-6 IL17A 0 0 7 7 0 0 7 7 0 0 h h
59 39.4 72.6 24.4 Wt T Wt T
IFN- a Wt Th1 Wt Th1 Fas KO Th Fas KO Th IL-17a Fas KO Th1 Fas KO Th1 c WT POU2AF1 -/- 0.81 4.35e-3 13.2 0.083 150 *** SUPPLEMENTARY INFORMATION RESEARCH BATF STAT3
No Cytokines 100
POU2F1A 99 a.#0.22 86.2 0.45 (ng/ml) b.# 50 BATF IL17A RORC BATF STAT3 0.55 0.31 0.1 0.058 IL-17 IRF4 IRF4 RORC
0 TGF-`1 + IL-6 7 7 0 0 1 ZEB1 SMARCA4 WT Th GATA3 WT Th SP4 STAT3 STAT4 FOXP3 IL17A RORC IRF4 81.2 18 89.7 10.1 L- 2 I Pou2af1 KO Th IL-17a Pou2af1 KO Th1
GATA3 FOXP3 TSC22D3 IL17A MINA GATA3 d IL17 module Genomic regions Bound genes 70 RORC BATF IL17A IRF4 60 Down-regulation 50 Up-regulation Only binding 40 Only perturbation effect TSC22D3 30 Pivot node % Overlap 20 Down-regulated in KD of pivot nodes Up-regulated in KD of pivot nodes 10 0 FOXP3 IL21 CEBPB AHR MT1 IRF4 Stat3 Stat5 BATF SupplementaryRORC Figure 13. Zeb1, Smarca4, and Sp4 are key novel regulators
IL10 module Foxp3 (Treg) affecting the Th17 differentiation programs. Shown are regulatory network models
centered on different pivotal regulators (square nodes): (a) Zeb1 and Smarca4, and (b)
Sp4. In each network, shown are the targets and regulators (round nodes) connected to
the pivotal nodes based on perturbation (red and blue dashed edges), TF binding (black
solid edges), or both (red and blue solid edges). Genes affected by perturbing the pivotal
nodes are colored (red: target is up-regulated by knockdown of pivotal node; blue: target
is down-regulated).
32
WWW.NATURE.COM/NATURE | 31 RESEARCH SUPPLEMENTARY INFORMATION
12 ChIP seq ChIP seq + knockout effects 10
8
6
4 Fold enrichment
2
0 IRF4_mid IRF4_late BATF_mid BATF_late IRF4_early STAT3_mid STAT3_late RORC_late BATF_early STAT3_early
Supplementary Figure 14. Overlap with ChIP-seq and RNA-seq data from Ciofani
et al (Cell, 2012). Fold enrichment is shown for the four TF that were studied by Ciofani
et al using ChIP-seq and RNA-seq and are predicted as regulators in our three network
models (early, intermediate [denoted as “mid”], and late). We compare to the ChIP-seq
based network of Ciofani et al. (blue) and to their combined ChIP-seq/RNA-seq network
(taking a score cutoff of 1.5, as described by the authors; red). In all cases the p-value of
the overlap (with ChIP-seq only or with the combined ChIP-seq/ RNA-seq network) is
below 10-10 (using Fisher exact test), but the fold enrichment is particularly high in genes
that are both bound by a factor and affected by its knockout, the most functional edges.
33
32 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
Supplementary Table Legends
Supplementary Table 1. List of microarray probesets that were differentially
expressed in the TGF-β1+Il6 microarray data. Columns: DE score and DE score IL-
23r-/- vs. WT: number of differential expression methods that identify a particular probe in the original experiment and the experiment with IL-23r-/- cells, respectively; Columns
Cluster and IL-23r-/- cluster ID: cluster assignment in each experiment (Fig. 1b,
Supplementary Fig. 2 and 3); Columns {TF or chromatin modifier, Cell surface,
Cytokine activity, In Codeset}: 1 indicates inclusion, 0 exclusion.
Supplementary Table 2. Functional enrichments for expression clusters. Tabs present enrichments for: (1) clusters for the combined Th0 and Th17 microarray data, (2) genes up or down-regulated in the Th17 microarray data relative to Th0, (3) genes up or down regulated in time, (4) genes differentially expressed upon the addition of IL-23, and
(5) genes differentially expressed in IL-23r-/- cells relative to their WT counterparts.
Supplementary Table 3. Regulatory interactions in the three canonical temporal
networks (Early, Intermediate, and Late). For each of the three consecutive networks,
listed are the TFs active in the network, their regulatory targets, supporting evidence and
its source for regulatory interactions, p-value (defined by the statistical significance of the
overlap between the putative targets of the TF [according to the respective data source]
and the gene group to which the target gene belongs), and the time stamp of regulation
(Methods).
34 WWW.NATURE.COM/NATURE | 33 RESEARCH SUPPLEMENTARY INFORMATION
Supplementary Table 4. Table of ranked regulators of Th17 differentiation. Sheet 1:
Presented is a ranked list of the regulators (Methods), including those displayed in Fig.
3a. For columns {Knocked down, Attempted knockdown, Known, Th17 microarray
signature, Targets key molecules, ChIP and phenotype (IPA), Predicted Regulator,
Induced, Is overexpressed in Th17 time course}: 1 denotes inclusion, 0 exclusion; for
columns {DE score in Th17 time course, DE score in IL23-/-}: number of methods (out
of 4 used for the Th17 time course and 3 for the IL23-/- data) by which the gene was
reported as differentially expressed. Column ‘Expression changes in key perturbations in
Th17 cells’: the number of corroborating pieces of evidence, 1- between 2 to 3 sources; 2
– more than 3 sources; 0 –otherwise (Methods). The “Comments” column presents
rationale for selecting low-ranking genes that were successfully knocked down. Sheet 2:
a similar ranked list for receptor genes.
Supplementary Table 5. Results of Nanostring nCounter and Fluidigm analysis.
Sheet 1: List of signature genes (including control genes); Sheets 2-3: Expression fold
changes (log2) at the 10hr (Fluidigm) and 48hr (Nanostring) time points. Presented are
only fold changes above the inference cutoff (Methods).
Supplementary Table 6. Primers for Nanostring STA and qRT-PCR/Fluidigm and
siRNA sequences. Sheet 1: Presented are the sequences for each forward and reverse
primer used in the Fluidigm/qRT-PCR experiments and Nanostring nCounter gene
expression profiling. Sheet 2: Sequences for RNAi used for knockdown analysis.
35
34 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
Supplementary Table 7. RNA-seq data analysis.
Sheet 1: Differentially expressed genes. Values shown are log2 of fold change (compared
to a non-targeting siRNA control). Displayed are all genes with a significant fold change
in at least one condition; comparisons that did not pass our differential expression criteria
are marked as “0”. Sheet 2: Overlap with gene signatures. For each annotated set of
genes (e.g., a temporal cluster from our data), we test for its enrichment with genes differentially expressed following knockdown. We test for the significance of the enrichment using two different tests (and hence present two p-values): (i) a hypergeometric test, evaluating the overlap between the differentially expressed genes and the genes in the cluster; (ii) A Wilcoxon-Mann-Whitney rank-sum test, comparing the absolute fold-change values of genes in the annotated set to the entire set of genes;
Sheet 3: TF binding enrichment analysis of differentially expressed genes. We used a hypergeometric p-value to evaluate the overlap between sets of differentially expressed genes and sets of bound genes as determined by ChIP-seq. The ChIP-Seq data is from
Th17 cells unless indicated otherwise on the second column.
Supplementary Table 8. Tsc22d3 ChIP-seq data analysis.
Sheet 1: overlaps with annotated regions. Two types of analysis are presented: (i) a genomic-region based score – evaluating the overlap between the sites bound by Tsc22d3
(as determined using MACS10) and annotated genomic regions. We used a binomial p-
value17 to assess the significance of the overlap. (ii) a gene-based analysis, evaluating the
overlap between the set of genes that are proximal to sites bound by Tsc22d3 and the
36
WWW.NATURE.COM/NATURE | 35 RESEARCH SUPPLEMENTARY INFORMATION
indicated sets of genes, using a hypergeometric p-value. Sheet 2: overlaps– list of all the
Tsc22d3 binding regions and their overlaps with annotated genomic regions.
37 36 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH
Supplementary References
1 Sánchez-Tilló, E. et al. ZEB1 represses E-cadherin and induces an EMT by recruiting the SWI/SNF chromatin-remodeling protein BRG1. Oncogene 29, 3490-3500, doi:10.1038/onc.2010.102 (2010). 2 Zhang, F. & Boothby, M. T helper type 1-specific Brg1 recruitment and remodeling of nucleosomes positioned at the IFN-gamma promoter are Stat4 dependent. J. Exp. Med. 203, 1493-1505, doi:10.1084/jem.20060066 (2006). 3 Matys, V. et al. TRANSFAC: transcriptional regulation, from patterns to profiles. Nucleic Acids Res. 31, 374-378 (2003). 4 Geiss, G. K. et al. Direct multiplexed measurement of gene expression with color- coded probe pairs. SI. Nature Biotechnology 26, 317-325, doi:10.1038/nbt1385 (2008). 5 Dalerba, P. et al. Single-cell dissection of transcriptional heterogeneity in human colon tumors. Nat Biotechnol 29, 1120-1127, doi:10.1038/nbt.2038 (2011). 6 Fujita, P. A. et al. The UCSC Genome Browser database: update 2011. Nucleic Acids Res. 39, D876-882, doi:10.1093/nar/gkq963 (2011). 7 Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. Ultrafast and memory- efficient alignment of short DNA sequences to the human genome. Genome Biol 10, R25, doi:10.1186/gb-2009-10-3-r25 (2009). 8 Li, B. & Dewey, C. N. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 12, 323, doi:10.1186/1471-2105-12-323 (2011). 9 Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. in Genome Biol Vol. 10 R25 (2009). 10 Zhang, Y. et al. in Genome Biol Vol. 9 R137 (2008). 11 Flicek, P. et al. Ensembl 2011. Nucleic Acids Res. 39, D800-806, doi:10.1093/nar/gkq1064 (2011). 12 Smith, R. L. et al. Polymorphisms in the IL-12beta and IL-23R genes are associated with psoriasis of early onset in a UK cohort. J Invest Dermatol 128, 1325-1327, doi:5701140 [pii] 10.1038/sj.jid.5701140 (2008).
38 WWW.NATURE.COM/NATURE | 37 RESEARCH SUPPLEMENTARY INFORMATION
13 Pruitt, K. D., Tatusova, T. & Maglott, D. R. NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res. 35, D61-65, doi:10.1093/nar/gkl842 (2007). 14 Wei, G. et al. in Immunity Vol. 30 155-167 (2009). 15 Yang, X. P. et al. Opposing regulation of the locus encoding IL-17 through direct, reciprocal actions of STAT3 and STAT5. Nat. Immunol. 12, 247-254, doi:10.1038/ni.1995 (2011). 16 Glasmacher, E. et al. A Genomic Regulatory Element That Directs Assembly and Function of Immune-Specific AP-1-IRF Complexes. Science, doi:10.1126/science.1228309 (2012). 17 Mclean, C. Y. et al. in Nature biotechnology Vol. 28 nbt.1630-1639 (2010). 18 Guttman, M. et al. in Nature biotechnology Vol. 28 503-510 (2010). 19 Garber, M. et al. A High-Throughput Chromatin Immunoprecipitation Approach Reveals Principles of Dynamic Gene Regulation in Mammals. Molecular cell, doi:10.1016/j.molcel.2012.07.030 (2012). 20 Reich, M. et al. GenePattern 2.0. Nat. Genet. 38, 500-501, doi:10.1038/ng0506- 500 (2006). 21 Schwanhäusser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337-342, doi:10.1038/nature10098 (2011). 22 Wernicke, S. & Rasche, F. FANMOD: a tool for fast network motif detection. Bioinformatics (Oxford, England) 22, 1152-1153, doi:10.1093/bioinformatics/btl038 (2006). 23 Hill, J. A. et al. Foxp3 transcription-factor-dependent and -independent regulation of the regulatory T cell transcriptional signature. Immunity 27, 786-800, doi:S1074-7613(07)00492-X [pii] 10.1016/j.immuni.2007.09.010 (2007).
39 38 | WWW.NATURE.COM/NATURE