Polyamine metabolism is a central determinant of helper T cell lineage fidelity

Abstract

Polyamine synthesis represents one of the most profound metabolic changes during T cell activation, but the biological implications of this are scarcely known. Here, we show that polyamine metabolism is a fundamental process governing the ability of CD4⁺ helper T cells (T_H) to polarize into different functional fates. Deficiency in ornithine decarboxylase, a crucial enzyme for polyamine synthesis, results in a severe failure of CD4⁺ T cells to adopt correct subset specification, underscored by ectopic expression of multiple cytokines and lineage-defining transcription factors across T_H cell subsets. Polyamines control T_H differentiation by providing substrates for deoxyhypusine synthase, which synthesizes the amino acid hypusine, and mice in which T cells are deficient for hypusine develop severe intestinal inflammatory disease. Polyamine-hypusine deficiency caused widespread epigenetic remodeling driven by alterations in histone acetylation and a re-wired tricarboxylic acid (TCA) cycle. Thus, polyamine metabolism is critical for maintaining the epigenome to focus T_H cell subset fidelity.

Keywords: T cells; eIF5A; hypusine; immunity; immunometabolism; metabolism; polyamines.

Graphical abstract

Figure 1

Polyamine metabolism is dynamically regulated in CD4⁺ T cells and directed by cytokines (A) Polyamine (PA) synthesis pathway. (B) Immunoblot of WT naive CD4⁺ T cells activated for indicated time or rested overnight in 10 ng/mL IL-7, representative of 3 biological replicates. (C) Mass spectrometry (MS) analysis of PA levels in CD4⁺ T cells post-activation. Fold increase versus CD4⁺ T cells rested overnight in IL-7. (D) Immunoblot of PA enzymes in CD4⁺ T cells under stated condition day 4 post-activation, representative of 3 biological replicates. (E) ODC and SMS levels in CD4⁺ T_H cells by flow cytometry (FC) 96 h post-activation. (F) Naive CD4⁺ T cells polarized for 72 h then exposed to 1.1 mM ¹³C arginine for 24 h. (G) MS analysis of PA levels in naive CD4⁺ T_H cells polarized for 96 h. (H) Naive CD4⁺ T cells differentiated under various T_H conditions. On day 3, cells were treated for 24 h in the labeled substrates stated and tracing into PAs was assessed by MS. (I) Naive CD4⁺ T cells polarized into T_H subsets. After 48 h, cells were exposed to ¹³C putrescine for a further 48 h. (J) Naive CD4⁺ T cells activated in the presence of the defined cytokine(s) and/or blocking Ab. After 72 h, cells were treated with 1.1 mM ¹³C arginine for 24 h. ¹³C arginine in PAs assessed by MS. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). Data are representative of 2 (C, D, F, and G), 3 (E), 1–2 (H), or 1 experiment (I and J). Asterisks without line bar denote statistical significance relative to T_H1 cells (F, G, and I) or to CD3/CD28 condition (J). See also Figure S1.

Figure S1

Odc^−/− CD4⁺ T cells exhibit dysregulated T_H lineage commitment and proliferation, related to Figures 1 and 2 (A) Expression of ODC and SMS in CD4⁺ T_H cells on day 4 post-activation by flow cytometry (FC). Asterixis without line bar denote statistical significance relative to T_H1 cells. (B) Naive CD4⁺ T cells isolated from spleen and polarized under different T_H conditions for 3 days and then re-plated in fresh T_H polarizing media containing 1.1 mM ¹³C arginine for 24 hours. Asterixis without line bar denote statistical significance relative to T_H1 cells. (C) Schematic depicting potential substrates for polyamine synthesis. (D) Naive CD4⁺ T cells were polarized in the stated cytokines and blocking antibodies. After 72h, cells were replated into fresh media containing fresh cytokines and blocking antibodies and cultured for 24 hours with 1.1 mM ¹³C arginine. Table indicates combinations of cytokines with IL-2 and IFN-γ/IL-4 blocking antibodies. (E) Frequency/absolute numbers of T cells and (F) frequency of naive and memory CD4⁺ T cells in the spleen of 8 week old Odc-ΔT mice and their WT littermates. (G) Naive WT and Odc^−/− CD4⁺ T cells were activated for 4 days in the indicated T_H polarizing condition. Cytokine expression after 5 hour PMA/iono restim. (H) Naive WT and Odc^−/− CD4⁺ T cells activated with anti-CD3/CD28 ± anti-IFN-γ/IL-4 (T_H0) were assayed for the indicated cytokine and transcription factor by FC 4 days post-activation. Representative contour plots shown. (I) Naive WT and Odc^−/− CD4⁺ T cells were polarized under T_H17 conditions ± varying concentrations (low, medium, high, see methods) of blocking antibodies. After 4 days, expression of the indicated cytokine or transcription factors was assessed by FC (NB = no blocking antibody). (J) Fold change in the number of WT and Odc^−/− T_H cells between day 0 and day 4 of in vitro culture. (K) Naive WT and Odc^−/− CD4⁺ T cells were stained with cell trace violet (CTV) proliferation dye and then polarized into different T_H subsets and assessed for CTV levels on the indicated day. (L) Naive WT and Odc^−/− CD4⁺ T cells stained with CTV, polarized and then assessed for IFN-γ after restimulation with PMA/ionomycin on day 4. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). Representative of 3 (A, E-F,I), 2 (B,H, J-L), or 5 (G) experiments.

Figure 2

Polyamine biosynthesis via Odc regulates CD4⁺ T helper subset fidelity (A) Immunoblot of naive CD4⁺ T cells isolated from WT and Odc-ΔT mice 48 h post-activation. (B) MS analysis of PAs in WT and Odc^−/− CD4⁺ T cells during activation. (C and D) Intracellular cytokines analyzed in WT and Odc^−/− naive CD4⁺ T cells activated for 96 h in indicated T_H condition. Representative contour plots shown. (E) WT and Odc^−/− CD4⁺ T cells assessed for TF expression on day 4. (F and G) WT and Odc^−/− naive CD4⁺ T cells activated under T_H17 cell conditions ± 250 μM putrescine for 96 h. Representative contour plots are shown. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (B, F, and G) Representative of 2, (C) of 3, and (D and E) of 5 experiments. See also Figures S1 and S2.

Figure S2

Odc^−/− CD4⁺ T cells display altered effector differentiation in vivo, related to Figures 2, 3, and 5 (A) Naive WT and Odc^−/− CD4⁺ T cells stained with CTV, polarized and then assessed for T-Bet expression on day 4. (B) WT and Odc^−/− naive CD4⁺ T cells activated under T_H17 cell polarizing conditions ± 250 μM putrescine for 96h. Intracellular cytokine was assessed post re-stimulation with PMA/ionomycin. (C, D) 4x10⁵ WT or Odc^−/− naive (CD45RB^hi CD25^- CD44^lo CD62L^hi) CD4⁺ T cells were adoptively transferred into Rag1^−/− mice. On day 37, the frequency and number of CD4⁺ T cells in the MLN expressing the indicated cytokine was analyzed following 4 hours ex vivo re-stimulation with PMA/ionomycin. (E) Frequency and number of T_regs in colon and (F) MLN (day 37 post-transfer) from Rag1^−/− mice that received either WT or Odc^−/− CD4⁺ T cells. (G, H) WT and Odc-ΔT mice were treated 50 μg anti-CD3 antibody by i.p injection on day 0, day 2, and day 4 and sacrificed 4h after the last injection. The frequency of CD4⁺ T cells expressing indicated (G) cytokine and (H) TF was assessed from the small intestine by FC. (I) Naive WT and Dohh^−/− CD4⁺ T cells activated with anti-CD3/28 for 48h were assessed for the indicated proteins by immunoblot. (J) Frequency and absolute numbers of T cells and (K) frequency of naive and memory CD4⁺ T cells in the spleen of 8 week old Dohh-ΔT mice and their WT littermates. (L, M) WT and Dohh^−/− naive CD4⁺ T cells were activated for 96h under T_H cell polarizing conditions. Culture supernatent was analyzed for presence of denoted cytokine by cytokine bead array. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A, C-F, L-M) is representative of 2 experiments, (B,G-H, J-K) Representative of 3 experiments.

Figure 3

Naive Odc^−/− CD4⁺ T cells are highly inflammatory in a T cell transfer model of colitis (A) WT or Odc^−/− naive CD4⁺ T cells were adoptively transferred into Rag1^−/− mice and weight loss tracked. (B) Disease score (see STAR Methods) assessed in Rag1^−/− recipients after WT or Odc^−/− naive CD4⁺ T cell transfer. (C and D) Colon length (C) and representative colon images (D) from Rag1^−/− mice following WT or Odc^−/− naive CD4⁺ T cell transfer. (E) Number of CD4⁺ T cells expressing indicated cytokines from colon, representative contour plots shown. (F) Frequency of colonic CD4⁺ T cells from Rag1^−/− recipient mice expressing indicated TF(s). (G) WT and Odc-ΔT mice injected intraperitoneally with 50 μg CD3 monoclonal antibody on days 0, 2, and 4 and sacrificed 4 h after the last injection. Number of CD4⁺ T cells expressing indicated cytokine was assessed from the small intestine (SI) by FC, representative contour plots are shown. (H) CD4⁺ T cells from the SI of WT and Odc-ΔT mice, treated as in (G), assessed for the indicated TF. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A–G) Representative of 2 experiments. See also Figure S2.

Figure 4

Synthesis of the amino acid hypusine underlies the central requirement for polyamine synthesis in T_H lineage commitment (A) PA synthesis and its role in eIF5A hypusination. (B) Immunoblot of total EIF5A and hypusinated (EIF5A^H) in CD4⁺ T cells from WT mice activated for indicated time or rested overnight in 10 ng/mL IL-7 (naive). Representative of 3 biological replicates. (C) Immunoblots of WT and Odc^−/− naive CD4⁺ T cells activated for 48 h. Representative of 3 biological replicates. N.B., same loading control and samples were used as in Figure 2A. (D and E) Naive WT and Odc^−/− CD4⁺ T cells electroporated with gDhps or gCTRL and Cas9. Cells were then activated under T_H17 conditions ±250 μM putrescine. After 96 h, proteins were assessed by (D) immunoblot and (E) cytokines expression analyzed by FC. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (B–E) Representative of 2 experiments.

Figure 5

T cell-specific deletion of the hypusine-synthesizing enzyme Dohh leads to T cell dysregulation and colitis (A–C) Naive WT and Dohh^−/− CD4⁺ T cells polarized for 96 h and cytokine levels (A and B) and TF expression (C) analyzed by FC, representative contour plots shown. (D and E) Post-natal survival (D) and colon length (E) of WT and Dohh-ΔT mice. (F and G) H&E staining of mid and distal colon (F) and cecum (G) from 9-week-old WT or Dohh-ΔT mice. (H) CD4⁺ T cells harvested from the indicated organ of 8-week-old WT and Dohh-ΔT mice and cytokine expression assessed. (I) Naive CD45.1⁺CD4⁺ T cells ± WT or Dohh^−/− CD45.2⁺CD4⁺CD25⁺ T cells were sorted and transferred into Rag1^−/− mice. Clinical disease was tracked and mice sacrificed when pre-determined criteria were met (see STAR Methods). Statistics are Mantel-Cox tests. (J and K) Donor T cells were transplanted into Rag1^−/− recipient mice as in (I). Numbers of CD45.2⁺CD4⁺ T_reg cells (J) and CD45.1⁺CD4⁺ T cells (K) present in the colon of surviving mice on day 113 post-transfer. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A–C) Represents 5, (H) 3, and (I) 1 experiment(s). See also Figures S2, S3, S4, and S5.

Figure S3

Dohh^−/− CD4⁺ T cells exhibit a broad pattern of dysregulated cytokine and transcription factor expression in vitro, related to Figure 5 (A) Naive WT and Dohh^−/− CD4⁺ T cells polarized for 96h and assayed for the expression of designated cytokine by FC after 5h of PMA/ionomycin re-stimulation. (B) Naive CD4⁺ T cells from WT and Dohh-ΔT-Great mice polarized for 96h and examined for YFP expression. Representative contour plots are shown. (C) Naive WT and Dohh^−/− CD4⁺ T cells activated with anti-CD3/CD28 ± anti-IFN-γ/IL-4 (T_H0) assayed for the indicated cytokine and transcription factor 4 days post-activation. (D) Naive WT and Dohh^−/− CD4⁺ T cells polarized under T_H17 conditions ± varying concentrations (low, medium, high, or no blocking - see methods) of blocking antibodies. After 96h, expression of the indicated cytokine or transcription factors was assessed. (E) Fold change in the number of WT and Dohh^−/− T_H cells between day 0 and day 4 of in vitro culture. (F) Naive WT and Dohh^−/− CD4⁺ T cells were stained with cell trace violet (CTV) proliferation dye, polarized into T_H subsets, and assessed for CTV levels by FC on the indicated day. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005. (A, D) Representative of 3 experiments, (D-G, I-N) (B,C, E,F) Representative of 2 experiments.

Figure S4

T cells from Dohh-ΔT mice exhibit perturbed phenotypes across multiple organs, related to Figure 5 Naive WT and Dohh^−/− CD4⁺ T cells stained with CTV, polarized, and then assessed for (A) IFN-γ after restimulation or (B) T-Bet expression on day 4 of culture. (C,D) Serum from 6-10 week old WT and Dohh-ΔT mice assayed by cytokine bead array. (E) CD4⁺ T cells harvested from designated organs of WT and Dohh-ΔT and cytokine expression analyzed by FC following 4h re-stim (SI = small intestine). (F) CD4⁺ T cells harvested from designated organs of 7 week old Dohh-ΔT-Great mice and their WT littermate controls and YFP expression analyzed. Representative dot plots are shown and gated on CD45⁺ TCRβ⁺ cells. (G,H) CD4⁺ T cells harvested from designated organs of WT and Dohh-ΔT mice and expression of indicated transcription factor analyzed by FC. Representative dot plots in (G) gated on Foxp3^- CD4⁺ T cells. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A, B, F) is representative of two experiments, (C-E, G,H) representative of 3 experiments.

Figure S5

T cell-specific deletion of Dhps confirms a role for hypusine synthesis in T_H lineage fidelity, related to Figure 5 (A) Immunoblot of WT and Dhps^−/− CD4⁺ T cells activated for 48h with anti-CD3/CD28. (B) Frequency/absolute numbers of T cells and (C) frequency of naive/memory CD4⁺ T cells in the spleen of 7 week old Dhps-ΔT mice and their WT littermates. (D) WT and Dhps^−/− naive CD4⁺ T cells polarized for 96h. Cytokine expression was assessed by FC 5 hours after re-stimulation, representative contour plots are shown. (E) WT and Dhps^−/− naive CD4⁺ T cells polarized as in (D) and labeled transcription factor assessed by FC. (F) Post-natal survival of WT and Dhps-ΔT mice from Freiburg or Memphis facilities. (G) Colon length and representative images from 6 week old WT and Dhps-ΔT mice. (H) CD4⁺ T cells harvested from the designated organ of WT and Dhps-ΔT and cytokine expression analyzed by FC. (I,J) CD4⁺ T cells harvested from indicated organ of WT and Dhps-ΔT mice and the frequency of CD4⁺ T cells expressing labeled transcription factor assessed. Analysis is gated on Foxp3^- CD4⁺ T cells. Representative contour plots are shown. (K) Frequency of T_regs among CD4⁺ T cells in colon and spleen of WT and Dhps-ΔT. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A) Representative of 1 experiment, (B-F) Representative of 3 experiments, (H-J) Representative of 4 experiments.

Figure 6

The polyamine-hypusine axis regulates the T cell epigenome to enforce appropriate T_H cell differentiation and function (A) RNA-seq of naive WT, Odc^−/−, and Dohh^−/− CD4⁺ T cells polarized for 96 h. Effect of Odc and Dohh deficiency is shown via principal component analysis, indicating the percentage of variance allocated to each component in parenthesis. (B) Venn diagram depicting number of differentially regulated genes disparate, or shared, between Odc^−/− and Dohh^−/− CD4⁺ T cells of indicated T_H lineage. Dashed circle indicates number of differentially expressed genes common to all T_H cell subsets. (C) ATAC-seq on CD4⁺ T cells treated as in (A). (D) Venn diagram depicting number of differentially regulated regions of open chromatin disparate or shared between Odc^−/− and Dohh^−/− CD4⁺ T cells. Dashed circle indicates number of differentially regulated regions of open chromatin common to all T_H cell subsets. (E) WT, Odc^−/−, or Dohh^−/− CD4⁺ T cells assayed for chromatin modifications by FC on day 4. Analysis performed on Ki-67⁺, diploid cells with “single” DNA content based on FxCycle (DAPI) staining in live cell gate. (F) ATAC-seq on naive CD44^loCD62L⁺ CD4⁺ T cells from spleens of WT and Dohh-ΔT mice. Volcano plots depict all differentially regulated regions of open chromatin with immunologically relevant loci in orange. (G) YFP expression assessed in CD4⁺ T cells from 7-week-old WT and Dohh-ΔT-Great mice. Bar graphs depict % of cells YFP⁺ in CD44^lo CD4⁺ T cells, representative contour plots are shown. (H and I) Naive WT and Odc^−/− CD4⁺ T cells (H) and naive WT and Dohh^−/− CD4⁺ T cells (I) electroporated with gTbx21 or gCTRL with Cas9 and activated under T_H1 conditions. IFN-γ assessed by FC on day 4. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). Representative of 1–3 (E), and of 2 (G–I) experiments. See also Figures S6 and S7.

Figure S6

RNA-seq and ATAC-seq data from in vitro activated WT, Odc^−/−, and Dohh^−/− CD4⁺ T_H cell subsets, related to Figure 6 Naive CD4⁺ T cells from WT, Odc-ΔT, and Dohh-ΔT mice polarized in indicated T_H conditions for 96h were assessed by (A) RNA-seq or (B) ATAC-Seq. Volcano plots depict differentially regulated genes (A) or differentially regulated regions of open chromatin (B). (C) WT and Dohh^−/− CD4⁺ T cells assayed for chromatin modifications by FC in indicated T_H cell subset 96h post-activation/polarization. Analysis was performed on Ki-67⁺ cells and diploid cells with ‘single’ DNA content based on FxCycle (dapi) staining in the live cell gate. (D-F) Naive WT, Odc^−/− and Dohh^−/− CD4⁺ T cells electroporated with guide RNAs specific for Tbx21 (gTbx21) or control guide (gCTRL) with Cas9 nuclease and activated under T_H17 conditions. After 4 days, levels of indicated cytokine or transcription factor assessed by FC. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (C) is representative of 1-3 experiments, (D-F) is representative of 2 experiments.

Figure S7

HAT inhibition or deletion restores efficient T_H differentiation in Odc^−/− and Dohh^−/− CD4⁺ T cells, related to Figures 6 and 7 (A,B) Naive WT, Odc^−/− and Dohh^−/− CD4⁺ T cells deleted for Tbx21 using CRISPR Cas9 and polarized in T_H1 conditions. Levels of labeled protein was assessed by FC after 96h. (C) Naive CD4⁺ T cells from WT and Dohh-ΔT-Great mice deleted for Tbx21 as described in (A) and polarized for 96h in T_H1 conditions. YFP expression analyzed by FC, representative contour plots are shown. (D, E) Naive WT and Dohh^−/− CD4⁺ T cells polarized in indicated condition for 72 hours. Cells were treated with 20 μM C646 for the final 48h of culture and assayed for (D) IFN-γ and T-bet or (E) H3k27 acetylation levels by FC. Analysis in (E) performed on Ki-67⁺ cells and diploid cells with ‘single’ DNA content based on FxCycle (dapi) staining in live cell gate. (F) Naive WT, Odc^−/− and Dohh^−/− CD4⁺ T cells electroporated with gP300 or a gCTRL with Cas9 nuclease and activated under T_H17 conditions. Protein expression was assessed after 96h. (G) Naive WT and Odc^−/− CD4⁺ T cells were treated as in (F), after 96h H3k27 acetylation assessed by FC. (H) Naive WT and Odc^−/− CD4⁺ T cells polarized for 72. Cells were treated with 20 μM CPTH2 for final 48h of culture and assayed for IFN-γ after 5h restimulation. (I) Naive WT and Odc^−/− CD4⁺ T cells electroporated with gKat2a or gCTRL with Cas9 nuclease and activated under T_H17 conditions. Indicated protein was assessed on day 4. (J) Naive WT and Odc^−/− CD4⁺ T cells cultured for 72h under different polarizing conditions then exposed to 4 mM ¹³C glutamine for 24h. Tracing of glutamine into stated metabolites was performed by mass spectrometry. (K) Naive WT and Dohh^−/− CD4⁺ T cells activated with anti-CD3/CD28 for 72h then treated as in (J) for the flux of glutamine carbons into TCA cycle metabolites. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). (A-C, F-I) Representative of two experiments, (D) Representative of 3 experiments, (E) is representative of one experiment.

Figure 7

Histone acetyltransferases and a rewired TCA cycle govern the remodeled epigenome in Odc^−/− and Dohh^−/− CD4⁺ T cells (A) Naive CD4⁺ T cells from Odc-ΔT and Dohh-ΔT mice and their littermate controls were electroporated with gP300 or gCTRL with Cas9 and activated under T_H17 conditions for 96 h. Representative histograms of P300 levels are shown. (B and C) T-bet and IFN-γ expression (B) and H3k27 acetylation levels (C) in WT, Odc^−/−, or Dohh^−/− T cells as treated in (A). (D–F) Naive WT and Odc^−/− CD4⁺ T cells electroporated with gKat2a or gCTRL with Cas9 and activated under T_H17 conditions. After 96 h, proteins were assessed by immunoblot (D) or FC (E and F). (G) Naive WT and Odc^−/− CD4⁺ T cells were polarized and after 72 h re-plated in 11 mM ¹³C glucose for 24 h. Glucose tracing performed by MS. (H) MS analysis of ¹³C glucose tracing in WT and Dohh^−/− T_H cells treated as in (G). (I) Naive WT and Odc^−/− CD4⁺ T cells polarized in T_H17 conditions for 72 h. Cells were treated with 30 μM BMS303141 for the final 48 h of culture. (J) Naive WT or Odc^−/− CD4⁺ T cells electroporated with gAcly or gCTRL with Cas9 and activated under T_H17 conditions. After 96 h, levels of the indicated protein were assessed by immunoblot. (K) Naive WT or Odc^−/− CD4⁺ T cells treated as in (J). After 96 h, T-bet and IFN-γ were assessed by FC. All data are mean ± SEM (p^∗ < 0.05, p^∗∗ < 0.005, p^∗∗∗ < 0.0005, p^∗∗∗∗ < 0.00005). Representative of 2 (A–F, I, and J), or 1 experiment (G and H). See also Figure S7.