Oncogene (2014) 33, 5151–5162 & 2014 Macmillan Publishers Limited All rights reserved 0950-9232/14 www.nature.com/onc

ORIGINAL ARTICLE High mobility group box 1 promotes tumor cell migration through epigenetic silencing of 3A

M Nehil1, J Paquette2, T Tokuyasu2 and F McCormick2

High mobility group box 1 (HMGB1) is a 25-kDa chromatin-associated that aids in transcription and DNA repair by directly binding to DNA and altering its conformation. Additionally, HMGB1 can act as an extracellular ligand. When released from dying or stressed cells, HMGB1 binds to the RAGE and activates the p42/44 MAP kinase (MAPK) cascade. HMGB1 is overexpressed in many types of cancer and frequently associated with tumor stage and metastasis. This has predominantly been attributed to an autocrine function that drives MAPK pathway activity. However, by using tumor cells with activating MAPK pathway , we have identified a role for HMGB1 in promoting metastasis and tumor growth that is independent of this pathway. In the absence of HMGB1, these tumor cells show defective in vitro migration as well as reduced metastasis and tumor growth in vivo despite high p42/44 phosphorylation. We found that semaphorin 3A (SEMA3A), previously shown to act as a suppressor of angiogenesis and migration, was highly increased during expression in the absence of HMGB1. SEMA3A/HMGB1 double knockdown rescued the migration defect in HMGB1 single knockdown cells. HMGB1 bound at the semaphorin 3A genomic locus, promoted hetrochromatin formation, and decreased occupancy of acetylated . Based on human tumor expression databases, HMGB1 was significantly inversely correlated with SEMA3A, suggesting that this mechanism may be more widely relevant in different cancer types.

Oncogene (2014) 33, 5151–5162; doi:10.1038/onc.2013.459; published online 11 November 2013 Keywords: HMGB1; SEMA3A; migration; chromatin; metastasis

INTRODUCTION toll-like receptors (TLR) 2/4, which are expressed on many types of High mobility group box 1 (HMGB1) is a 25-kDa DNA-binding immune and tumor cells. HMGB1 binding to RAGE results in protein that is overexpressed in various cancers. Its expression is activation of the Ras MAP kinase pathway and increased levels of often associated with poor prognosis, higher tumor grade and phosphorylated p42/44 (ERK1/2).15 metastasis.1 In a colon cancer study, HMGB1 was shown to be The pro-tumor effects of HMGB1 are often attributed to its overexpressed in tumor tissue compared with adjacent normal extracellular role, through stimulation of the MAPK pathway via tissue in 90% of patients.2 In addition, a causative role for HMGB1 RAGE binding.16 Functionally, preventing extracellular release of in promoting colon carcinogenesis was shown experimentally.3 HMGB1 by drug treatment or blocking HMGB1 signaling with Similar results have been reported for hepatocellular carcinoma,4 antibodies can reduce tumor growth and metastasis in some nasopharyngeal carcinoma,5 prostate cancer,6 melanoma7 and experimental models of cancer.3,17,18 In addition, RAGE-deficient breast cancer.8 The high prevalence of HMGB1 upregulation in mice are resistant to tumor formation in chemical carcinogen- diverse tumor types suggests a critical role in carcinogenesis; induced skin cancer models,19 and RAGE expression is also often however, despite a strong correlation between HMGB1 associated with poor clinical outcome and metastasis. The expression and negative clinical outcome, the mechanism by HMGB1–RAGE–MAPK signaling axis thus appears to be important which HMGB1 promotes tumor growth and metastasis remains in many cancers. However, tumor types with frequent activating unclear. mutations in the MAPK pathway, such as colon cancer and HMGB1 binds to DNA without sequence specificity and aids melanoma, still are often found to overexpress HMGB1, indicating in distorting the DNA structure to allow access for repair a possible MAPK-independent function. and transcription .9,10 HMGB1-deficient mice die within Here, we have tested the effect of depleting HMGB1 in cell lines 24 h after birth from hypoglycemia as a result of defective with constitutive MAPK activity. We found that HMGB1-deficient glucocorticoid-mediated transcription.11 HMGB1 interacts with cells were impaired in migration in vitro and metastasis in vivo, various transcription factors, including NFkB members,12 p5313 despite maintaining high levels of phosphorylated p42/44. Further and the TATA-binding protein.14 Interactions with these factors experiments revealed a role for HMGB1 in transcriptionally can promote or repress transcription depending on the cellular repressing the semaphorin 3A (SEMA3A) locus, a gene previously context. described to suppress cell migration and angiogenesis. A In addition to its roles in the nucleus, HMGB1 can also act as a functional connection between HMGB1 and SEMA3A was shown ligand when released from dying or stressed cells. HMGB1 binds by a partial rescue of the migration defect in HMGB1 knockdown to the receptor for advanced glycation end products (RAGE) or cells on dual knockdown of SEMA3A.

1Department of Medical Oncology, Dana-Farber Cancer Institute, Harvard Medical Scool, Boston, MA, USA and 2Biomedical Sciences and Bioinformatics Programs, Helen Diller Family Comprehensive Cancer Center, University of California, San Francisco, CA, USA. Correspondence: Dr F McCormick, Helen Diller Comprehensive Cancer Center, University of California, San Francisco, 1450 3rd Street, Rm. HD-371 UCSF Box 0128, San Francisco 94158-9001, CA, USA. E-mail: [email protected] Received 1 April 2013; revised 29 August 2013; accepted 9 September 2013; published online 11 November 2013 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5152 RESULTS preference for sites of metastases between cell groups, with bone, HMGB1 knockdown alters cell morphology and decreases in vitro kidney and lung constituting the majority in both groups (data not migration in cells with an activated MAPK pathway shown). To identify MAPK-independent functions of HMGB1, we used a lentiviral short hairpin RNA (shRNA) system to stably knock-down HMGB1 is required for tumor growth in vivo HMGB1 in cancer cells with activating mutations in the MAPK Previous models have shown that HMGB1 is required for in vivo pathway. The MDA-MB-231 breast cancer cell line had mutations tumor growth owing to its ability to drive MAPK activation. Using in K-Ras and B-Raf resulting in high basal MAPK activity, while the our system, we tested whether HMGB1 was still required for tumor MDA-MB-435 cells were derived from a melanoma with an 20 growth in the presence of a constitutively active MAPK pathway. activating B-Raf . Stable populations of MDA-MB-231 cells carrying a luciferase HMGB1 protein expression was reduced by 70-92% as reporter gene and HMGB1 knockdown or control shRNA were compared with cells with non-silencing shRNA (Figure 1a) after a orthotopically injected into the mammary fat pads of 6-week-old minimum of 1 week growing in culture with puromycin selection. SCID mice and monitored for growth up to 8 weeks. Cells HMGB1 knockdown did not reduce basal MAPK signaling in these expressing HMGB1 shRNA showed a significantly reduced growth cell lines as judged by levels of phosphorylated ERK1/2 (Figure 1a), rate as measured by bioluminescent signal (Figure 2b). Consistent and in vitro proliferation was not affected (Figure 1b). However, with the in vitro data, HMGB1-deficient tumors did not have depletion of HMGB1 caused the cells to adopt a more epithelial significantly reduced levels of phosphorylated ERK1/2 as judged morphology with less extended processes in two-dimensional by western blot (Figure 2c). tissue culture (Figure 1c). Cells with diminished HMGB1 expression Because HMGB1 has previously been shown to contribute to also showed reduced F-actin foci at the cell periphery and angiogenesis in other systems,23 we next investigated the state of significantly reduced cell scattering in a colony formation assay tumor-associated endothelial cells in HMGB1-deficient tumors. In (Figure 1c), but did not affect colony-forming ability (data not the absence of HMGB1, tumors showed decreased CD31 cell shown). HMGB1-depleted cells were also deficient in migration as þ infiltration (Figure 2d). Macroscopically and by hematoxylin and measured in a scratch assay (Figure 1d). eosin staining, it was observed HMGB1 knockdown tumors had a more disorganized structure and increased necrosis (data not shown). HMGB1 is required for tumor cell colonization from the circulatory system in vivo Tumor metastasis is thought to occur in several steps. Initially, the HMGB1 knockdown alters the expression of 122 cells must detach from the primary tumor and degrade the stroma Given that the phenotypes we observed could not be explained to gain access to the circulatory system. Once in circulation, the by reduced MAPK signaling, we sought to investigate the cells must adhere to the endothelial cell wall and migrate through transcriptional functions of HMGB1. HMGB1 is known to cooperate the surrounding tissue to arrive at and populate new organ sites. with various transcription factors to promote or repress gene HMGB1 expression is correlated with metastasis in human expression depending on the cellular context. We performed a tumors21 and has been shown to be required for aspects of the transcriptional microarray in triplicate using HMGB1 knockdown or metastatic process in some experimental models.16 Because our control cell populations in log phase growth. We chose to use cells in vitro results suggested a deficiency in migration, we tested in log phase as opposed to resting to mimic conditions present whether HMGB1 knockdown also affected in vivo metastatic when cells had pronounced morphological differences and capacity. To investigate metastatic potential, we used a system decreased scattering. Using a cutoff of 1.5-fold difference in that has been described previously for MDA-MB-231 cells.22 In expression and a q-value o0.05 (Po0.001), we identified 25 genes this colonization/metastasis model, the cells are injected into that were downregulated by HMGB1 knockdown and 97 genes the circulatory system of an immune-compromised mouse and that were upregulated (Table 1). Among the top upregulated allowed to invade and populate new locations, therefore genes were two members of the class 3 semaphorin family, recapitulating the later steps of metastasis. semaphorin 3A (SEMA3A) and semaphorin 3E (SEMA3E). MDA-MB-231 luciferase-expressing cells with HMGB1 knock- Class 3 are secreted proteins that act as inhibitors down or control shRNA were injected intracardially into nude of angiogenesis and cellular migration.24 They are often mice. The resulting tumors were allowed to grow for 10–12 weeks downregulated in more aggressive tumors25 and suppress and counted in each mouse using bioluminescence. To account tumor growth and metastasis in experimental models.26 The loss for differences in rates of tumor growth and thus detection of their expression is increasingly being recognized as a critical between HMGB1 knockdown and control tumors, metastases step in the evolution of more aggressive tumors. SEMA3A is the were counted at a time point normalized to equal biolumines- best-characterized member of this family. SEMA3A binds to cence of the largest tumor. HMGB1 knockdown cells formed 40% neuropilin 1 and promotes GAP activity of A1, which fewer tumors in this assay (P ¼ 0.01) (Figure 2a). There was no through R-Ras leads to inactivation of Beta1 integrin.27

Figure 1. HMGB1 depletion causes loss of F-actin processes and migratory behavior without reducing basal P-ERK levels in MDA-MB-231 and MDA-MB-435 cells. (a) HMGB1 can be depleted without decreasing P-ERK levels. MDA-MD-231 and MDA-MB-435 cells were stably depleted of HMGB1 with lentiviral shRNA using control or two independent HMGB1 targeting shRNAs. After 1–3 weeks of selection, cells were analyzed by western blot for HMGB1 and P-ERK levels using Licor Odyssey software. (b) HMGB1 knockdown does not affect proliferation in vitro. MDA-MB- 231 cells were transduced with control shRNA or two independent HMGB1 shRNAs. After 1–3 weeks of selection, cells were seeded into six- well plates and counted with a Coulter counter in triplicate. Media was changed at 3 days. (c) HMGB1 knockdown changes cell morphology, reduces F-actin foci and diminishes scattering behavior. (Top) Control and HMGB1 knockdown cells were imaged by phase contrast microscopy to illustrate morphological differences. (Middle) Cells were stained with rhodamine-phalloidin to outline F-actin. (Bottom) Cells were seeded by serial dilution to B100 cells/well in six-well plates. Single cells were allowed to grow into colonies over 2–4 weeks and fixed with crystal violet stain. Colonies were then analyzed for dispersion during growth or ‘scattering’. For each condition, 10 different fields of view containing 4–9 colonies each were sorted into ‘scattered’ or ‘non-scattered’ groups and quantified. (d) HMGB1 knockdown decreases in vitro migration in a scratch assay. Control or HMGB1 knockdown cells were seeded into six-well plates overnight and scratched the next day. Images were collected at the time points listed. Relative migration was quantified by measuring the distance between scratch edges for 10 different points and compared with time 0.

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5153 Additionally, SEMA3A can initiate F-actin depolymerization HMGB1 silences SEMA3A expression through Plexin-A1–MICAL interactions.28 Because of this close To validate our array data, semaphorin during link between the functions of semaphorins and the phenotype of HMGB1 knockdown was examined with real-time PCR in different the HMGB1 knockdown, we chose to further investigate this tumor cell lines, including one melanoma, one lung and two potential connection. breast cancer cell lines in addition to MDA-MB-231. These cell lines

MDA-MB-435 MDA-MB-231

control #1 #2 control #1 #2 ControlshRNA #1 shRNA #2

HMGB1 1.00 0.30 0.19 1.00 0.26 0.08

P-p42/p44

1.00 1.30 0.88 1.00 2.96 0.97

p42/p44

actin

MDA-MB-231

3.0 control 2.5 sh1

^6) 2.0 sh2 1.5 1.0

cells (x10 0.5 0 Day 1 Day 2 Day 4 Day 6

0 hrs 6 hrs 10 hrs

control 231

435

#1 MDA-MB-

MDA-MB- #2 0 hrs 10 hrs 24 hrs

control 435

#1 MDA-MB- 90 #2 MDA-MB-435 80 MDA-MB-231 70

150 es MDA-MB-435 60 MDA-MB-231 50 100

ered coloni 40

gration 30 20 % Scatt 50

(arbitrary units) 10

Relative mi 0 control #1 #2 0 control #1 #2

& 2014 Macmillan Publishers Limited Oncogene (2014) 5151 – 5162 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5154 a

8 * 7

6

5 6.7 r mouse r 4

3 3.9

2 tumors pe tumors 1

0 control HMGB1 k.d. control HMGB1 k.d. n=9 n=7 b c

p-ERK/ERK 4

p=0.89 3

2

1

0 Day: 7 28 42 7 28 42 control HMGB1 control HMGB1 k.d. k.d. d 7

p= 0.059 6 20 control shRNA 18 HMGB1 shRNA 5 16 14 4 12 * control shRNA 10 3 8 6 **

2 CD31+ cells/ F.O.V

photons/ tumor (x10^8) 4 * 2 1 0 n.s * control HMGB1 0 HMGB1 shRNA shRNA shRNA Day: 714284256 Figure 2. HMGB1 is required for in vivo metastatic colonization and xenograft tumor growth. (a) HMGB1 is required for tumor cell colonization from circulation. A total of 50 000 MDA-231 cells with a luciferase reporter and expressing either control or HMGB1 shRNA were injected intracardially into nude mice. Mice were then imaged weekly and tumors counted. Each mouse’s tumors were quantified at a time point with roughly equal intensity of the largest tumor, B8 weeks for control shRNA and 10 weeks for HMGB1 shRNA. (b) HMGB1 is required for xenograft tumor growth. A total of 1 Â 106 MDA-MB-231 cells with luciferase reporter and control or HMGB1 shRNA were injected into the mammary fat pad of SCID beige mice. Mice were imaged weekly and the signal was quantified with Xenogen imaging software. (c) HMGB1 knockdown (k.d.) tumors do not have reduced P-ERK levels. Tumors were removed after 8 weeks of growth and flash frozen. Lysates were prepared by dounce homogenization in RIPA buffer and cleared by centrifugation. P-ERK was quantified as described in Figure 1. (d) HMGB1- deficient tumors show reduced CD31 þ cell infiltration. Tumors were removed from mice at 8 weeks post implantation and were formalin- fixed. Tumors were then analyzed by immunohistochemistry for CD31 þ cells. CD31 þ cells were quantified by counting all CD31 þ cells in five fields of view per tumor.

showed significant upregulation of SEMA3A mRNA when these cell lines showed a reduction in basal P-ERK levels HMGB1 was depleted, by as much as fivefold, which correlated on knockdown (Figure 3b), consistent with the known role for with the level of HMGB1 knockdown (Figure 3a). SEMA3A protein HMGB1 in this pathway. In these cells, SEMA3A was still was also upregulated with HMGB1 knockdown in cell lines upregulated by HMGB1 knockdown, demonstrating that MAPK and tumors as visualized by western blot and immunohisto- pathway activation and SEMA3A repression by HMGB1 are chemistry, respectively (Figures 3c and 3d). Interestingly, two of independent functions.

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5155 Table 1. Top up- and downregulated genes with HMGB1 knockdown

Downregulated Upregulated With HMGB1 KD With HMGB1 KD

Gene symbol Fold change Gene symbol Fold change Gene symbol Fold change Gene symbol Fold change

HMGB1 À 4.76 PTX3 5.24 COL8A1 1.78 BTN2A1 1.58 MAMDC2 À 3.74 TNFSF18 2.77 ABI3BP 1.77 DCK 1.58 CD24 À 3.18 PTPN22 2.76 BTN3A2 1.75 FLVCR1 1.57 GKN2 À 3.16 LIPH 2.59 NAV3 1.73 PIK3R3 1.57 HMGB1L1 À 2.92 F2RL2 2.51 SYTL3 1.73 SATB1 1.56 CCDC86 À 2.14 C7orf58 2.47 ENDOD1 1.71 ZNF721 1.55 ID2 À 1.94 STC1 2.38 C9orf84 1.71 PRR16 1.54 KDELR3 À 1.9 SEMA3A 2.37 IL31RA 1.71 G0S2 1.54 CDH11 À 1.85 DSC2 2.31 TNFRSF1B 1.7 GRIK4 1.54 TSPAN31 À 1.82 RGS4 2.29 OAS1 1.7 GLCE 1.53 KRT81 À 1.8 ZNF175 2.19 SVIP 1.69 ZNF100 1.53 TNS4 À 1.79 CA12 2.15 CPM 1.68 GNPNAT1 1.53 ZNF542 À 1.66 SEMA3E 2.14 KRTAP2-4 1.67 ADAMTSL1 1.53 UCA1 À 1.65 CDADC1 2.08 VWDE 1.67 PRSS12 1.52 ARSJ À 1.63 AFF3 2.04 PLAT 1.67 LAMA4 1.52 CYP1A1 À 1.63 DRAM1 1.99 SHC3 1.67 PEAR1 1.51 AKR1B1 À 1.62 NEDD4 1.97 TMEM106B 1.66 PAG1 1.51 KLHL5 À 1.62 EHF 1.95 CHST11 1.66 HIST1H2BK 1.51 C19ORF21 À 1.61 SERPINB2 1.94 GPRC5B 1.65 C18orf54 1.51 KIAA1462 À 1.56 ABHD5 1.92 SOCS2 1.65 TGM2 1.51 FSTL3 -1.54 CCDC68 1.92 SHISA2 1.65 BUB1B 1.51 DSP À 1.53 GALNT1 1.91 B3GNT5 1.64 PROS1 1.51 P4HA1 À 1.52 CSGALNACT2 1.9 ADAMTS12 1.63 PRKCE 1.51 FOXP2 À 1.52 CCRL2 1.89 TM4SF18 1.62 IMPAD1 1.5 DMKN À 1.5 XRCC4 1.88 WDR17 1.62 SNTB1 1.5 NNMT 1.88 RNF144B 1.62 HLA-DRA 1.85 TMCC3 1.61 TAF9B 1.85 POMGNT1 1.6 FRRS1 1.84 DENND1B 1.6 STRADB 1.83 CLDN1 1.6 PRSS7 1.83 CCDC109A 1.6 RIMS1 1.83 CCDC80 1.59 SPOPL 1.82 ALS2CR4 1.59 CALCR 1.79 RPL22 1.59 GPR56 1.79 GRB2 1.59 SKP2 1.79 PDE3B 1.58 Abbreviation: KD, knockdown. MDA-MB-231 cells were lentivirally treated with control or HMGB1 shRNA and selected with puromycin for 2 weeks. 90% knockdown was confirmed by western blot and RNA was extracted and analyzed by microarray using the Affymetrix Human Gene 1.0 platform. Fold change was calculated relative to control shRNA-treated cells.

Although we identified HMGB1 as a repressor for SEMA3A (Figure 4a). Double-knockdown cells largely regained the in cell lines, it was not clear if this function was relevant in invasive morphology of the parent cells. F-actin was restored primary human tumors. To investigate a potential connection to the leading edges of cell processes and scattering ability between HMGB1 and SEMA3A in primary tumors, we analyzed in vitro was partially restored (Figure 4b). As judged by scratch publicly available GEO profile transcriptional microarray assay, migration was also significantly rescued by knockdown of data from various tumor types. Using microarray data from 23 SEMA3A (Figure 4c). Although these in vitro characteristics were primary breast tumors, we observed a significant inverse rescuedbydoubleknockdown,wewereunabletoobservea correlation between HMGB1 and SEMA3A (R2 ¼ 0.23, P ¼ 0.02, difference in metastatic potential in vivo. This was largely due to Figure 3e). A similar relationship was observed in a cohort the fact that double-knockdown cells showed a reduced growth of bladder tumors, with the more invasive samples from the rate and poor adhesion. This is likely due to adaptations that bladder cancer data set showing high HMGB1 and low SEMA3A occur in response to loss of HMGB1 and widespread gene (Figure 3f). expression changes. Nonetheless, our data indicates that HMGB1-mediated silencing of SEMA3A partially contributes to SEMA3A knockdown restores F-actin localization, invasive the invasive phenotype of tumor cells. This observation is morphology and in vitro migration of HMGB1-deficient cells consistent with previous studies that show an essential role for HMGB1 knockdown caused a significant change in cell morphol- SEMA3A in suppressing experimental models of metastasis and ogy and migration and also altered the expression of 122 genes, invasive cell morphology.26 These data, in addition to the including SEMA3A. To determine the contribution of SEMA3A correlation between HMGB1 and SEMA3A in primary tumors, led upregulation to the phenotype of HMGB1 knockdown cells, us to further investigate the mechanism of HMGB1-mediated we performed double knockdowns of HMGB1 and SEMA3A SEMA3A silencing.

& 2014 Macmillan Publishers Limited Oncogene (2014) 5151 – 5162 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5156

control ** 6 HS578T SUM159PT #1 *** control #1 #2 control #1 #2 5 *** p-ERK #2 ERK 4 ** control #1 #2 3 *

SUM159PT SEMA3A relative to control 2 * * *

MDA-MB-231 Fold-change SEMA3A mRNA 1

0 MDA-MB-435 HS578T SUM159PT A549 MDA-MB-231 MDA-MB-435 HMGB1

control HMGB1 shRNA SEMA3A SEMA3A

GSE20194 - breast cancer GDS1479 - bladder cancer 9.0 4.5

8.5 4.0

8.0 3.5 2 R = 0.23 R2 = 0.45

SEMA3A SEMA3A 7.5 p = 0.021 p = 5.6E-9 3.0 7.0 normal urothelium superficial transitional carcinoma in situ cell carcinoma lesion muscle invasive normal mucosa carcinoma 6.5 2.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 HMGB1 HMGB1 Figure 3. HMGB1 is required for SEMA3A silencing. (a) SEMA3A mRNA is upregulated in cell lines depleted for HMGB1. Different cell lines were depleted for HMGB1 as previously described and assayed for SEMA3A mRNA by QPCR. HMGB1 levels were evaluated by western blot for each condition. (b) P-ERK levels decrease in Hs578t and SUM159PT cell lines treated with HMGB1 shRNA. Cell lines were selected for 1–2 weeks with HMGB1 shRNA as described earlier. P-ERK levels were analyzed by western blot. (c) SEMA3A protein is upregulated in cell lines depleted for HMGB1. MDA-MB-435, SUM159PT and MDA-MB-231 cells were depleted for HMGB1 as previously described. SEMA3A expression was then analyzed by western blot. (d) SEMA3A protein is upregulated in HMGB1 knockdown tumors in xenograft. Implanted MDA-MB-231 cells were removed from mice after 8 weeks of growth. The tumors were analyzed by immunohistochemistry for SEMA3A expression. (e) HMGB1 inversely correlates with SEMA3A in primary breast tumors. Publicly available transcriptional microarray data from a set of 23 primary breast tumors were analyzed for HMGB1 and SEMA3A expression. (f) HMGB1 inversely correlates with SEMA3A in bladder cancers. Publicly available transcriptional microarray data from a set of 60 bladder cancers was analyzed for HMGB1 and SEMA3A expression and charted with individual sample clinical classification.

Nuclear HMGB1 remodels chromatin and promotes binding of silencing in HMGB1 knockdown cells, despite treatment with acetylated histones at the SEMA3A promoter greater than 20-fold excess HMGB1 over control cells (Figure 5a). HMGB1 can influence gene expression directly in the nucleus via We also tested recombinant HMGB1 at a concentration previously interactions with transcription factors, or indirectly by binding shown to promote P-ERK-dependent cell migration in fibro- receptors and activating signaling pathways that ultimately affect blasts.29 Although in this experiment the level of SEMA3A transcription. Although reduction of HMGB1 does not reduce the upregulation was less, it was still significantly increased even in MAPK pathway in this system, extracellular HMGB1 can also signal the presence of recombinant HMGB1. These data suggest that through other receptors, such as TLR family members, to initiate a nuclear HMGB1 is responsible for SEMA3A silencing, and led us to transcriptional response. To discriminate between the extracellular further investigate the mechanism of silencing at the SEMA3A and the nuclear effects of HMGB1 on SEMA3A regulation, we genomic locus. applied conditioned media from HMGB1-expressing cells to The mechanism of nuclear HMGB1-mediated transcription HMGB1 knockdown cells. After 4 days of treatment, cells were stimulation has been well characterized and involves the intrinsic analyzed by RT–PCR for SEMA3A expression. Extracellular HMGB1 ability of HMGB1 to bend DNA, thus allowing greater access from control cell supernatant was not able to restore SEMA3A to transcription factors.30 In contrast, the mechanism of

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5157 A549 MDA-MB-231 1.2 HMGB1+ HMGB1 1.0 control HMGB1 k.d. SEMA3A k.d. SEMA3A

0.8 actin 0.6 0.4

mRNA / / mRNA 0.2 0 control HMGB1 HMGB1+ shRNA shRNA SEMA3A shRNA

HMGB1+SEMA3A k.d.

1

MDA-MB-23

MDA-MB-435 0.8 0 hrs 24hrs 0.7 control 0.6 HMGB1 k.d. 0.5 HMGB1+ 0.4 SEMA3A #1 0.3 HMGB1+ SEMA3A #2 0.2

% scattered colonies scattered % 0.1 250 0 !0.1 control HMGB1 HMGB1+ HMGB1+ 200 k.d. SEMA3A #1 SEMA3A #2

150

100

(arbitrary units)

Relative migration 50

0 control HMGB1 HMGB1+ HMGB1+ k.d. SEMA3A #1 SEMA3A #2 Figure 4. Depletion of SEMA3A in addition to HMGB1 restores the migratory phenotype of cells in vitro.(a) HMGB1 and SEMA3A can be co-depleted by shRNA. (Top) Lentiviral shRNA targeting HMGB1 or HMGB1 and SEMA3A was used as previously described. After 1–2 weeks of selection in puromycin, the levels of SEMA3A or HMGB1 mRNA was analyzed by RT–PCR. (Bottom) Double knockdown of HMGB1 and SEMA3A causes MDA-MB-231 cells to adopt an invasive morphology and actin staining pattern similar to the parental MDA-MB-231 cells. (b) A549 lung cancer cells show altered migration and morphology after HMGB1 knockdown that is restored to the parental phenotype by the additional knockdown of SEMA3A. As in panel a, HMGB1 was depleted singly or in combination with SEMA3A. Migration was analyzed by scattering assay, and F-actin localization was visualized with phalloidin-rhodamine staining. Scattering was quantified as described previously. (c) Reduced MDA-MB-435 cell migration with HMGB1 knockdown can be rescued by additional SEMA3A knockdown. Cells were treated as described above and migration was analyzed by a scratch assay as in Figure 1.

HMGB1-mediated gene silencing is much less understood. HMGB1 To investigate the mechanism of HMGB1-mediated gene has previously been reported to associate with silenced chromatin silencing further, we focused on a region of SEMA3A between at the TNFa promoter in models of endotoxin tolerance.31 the transcription start site and 2 kb upstream (Figure 6b). As a

& 2014 Macmillan Publishers Limited Oncogene (2014) 5151 – 5162 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5158

Figure 5. HMGB1 from conditioned media is unable to restore SEMA3A silencing in HMGB1 knockdown cells. (a) MDA-MB-231 cells were seeded at 300 000 per well in six-well plates with 1.5 ml of 10% fetal bovine serum–Dulbecco’s modified Eagle media. Cells were grown for 1 week with no media change. One-week-conditioned media was applied to control shRNA or HMGB1 knockdown cells each day for 4 days. On day 4, supernatant from the cells was collected and analyzed by western blot. RNA was extracted to measure SEMA3A expression by RT–PCR. (b) MDA-MB-231 cells were grown as in panel a, but instead of conditioned media, 10 nM recombinant HMGB1 was applied on day 1 and day 3. On day 4, RNA was extracted and analyzed by RT–PCR.

Figure 6. HMGB1 is required for heterochromatin formation and reduces acetylated H4 and H3 binding at the SEMA3A locus. (a) 5-Azacytidine and trichostatin A treatment increases SEMA3A mRNA expression in MDA-MB-231 cells. Cells were treated with 5 mM 5-azacytidine for 3 days with or without the addition of 100 nm trichostatin A during the last day. RNA was measured by RT–PCR. All experiments were conducted in triplicate. (b) Genomic regions of SEMA3A and SEMA4F being analyzed. From the Santa Cruz Genome Browser entry for SEMA3A and SEMA4F, regions are displayed that were analyzed by specific primer sets. (c) HMGB1 is not enriched for binding at SEMA3A over the SEMA4F locus. ChIPs were conducted with cells stably expressing -HMGB1. Anti-Myc antibody or IgG (mock) was used and binding was quantified with RT–PCR for the regions illustrated in panel b.(d) HMGB1 depletion increases the sensitivity of the SEMA3A locus to DNAse digestion. DNAse sensitivity was analyzed using the EpiQ chromatin analysis kit, as per manufacturer’s directions. Sensitivity was measured in control or HMGB1 shRNA-expressing MDA-MB-231 cells. All experiments were conducted a minimum of six times. (e) HMGB1 depletion increases binding of acetylated histone H4 and H3 at the SEMA3A locus. ChIP was conducted on MDA-MB-231 cells with or without HMGB1-targeted shRNA, as in panel d. All experiments were performed in biological triplicate to derive s.d. Regions that were bound by Ac-H4 or Ac-H3 were analyzed with RT–PCR and quantified by enrichment analysis (see Materials and methods).

control, we analyzed the homologous region in the SEMA4F gene, HMGB1-dependent SEMA3A and HMGB1-independent SEMA4F,as whose expression was not affected by HMGB1 knockdown in our revealed by chromatin immunoprecipitation (ChIP) experiments microarray analysis. HMGB1 bound at roughly equal levels to both and QPCR analysis (Figure 6c). This suggested that promoter

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5159 binding alone is not sufficient to explain the locus-specific SEMA3A promoter (Figure 7a). Although we found no difference in transcriptional repression by HMGB1 and is consistent with CpG methylation between control and HMGB1 knockdown (data previous studies showing HMGB1 binding to DNA without gene not shown), we found significantly reduced non-CpG cytosine specificity.32 We next investigated whether HMGB1 altered methylation at two different residues in MDA-MB-231 and chromatin structure in a locus-specific manner. To study this, we SUM159PT cells (Figure 7a). Strikingly, both of these cytosines examined chromatin sensitivity to DNAse digestion in control or occurred in almost the same relative position downstream of a HMGB1 shRNA-expressing cells. Open chromatin or euchromatin CCCTC-binding factor (CTCF) putative recognition sequence. To is characterized as being more sensitive to nuclease digestion and test if CTCF binding was affected by HMGB1 knockdown, we more transcriptionally active, while closed chromatin or performed a ChIP, as in Figure 6. Interestingly, although CTCF is heterochromatin is resistant to digestion and associated with often thought of as a transcriptional repressor, HMGB1 knock- silenced genes. As shown in Figure 6d, in the absence of HMGB1, down enhanced the binding of CTCF upstream of SEMA3A the SEMA3A promoter became sensitized to DNAse digestion (Figure 7b). while the SEMA4F promoter was not significantly affected. Heterochromatin is often associated with several types of epigenetic modification, including histone hypoacetylation and DISCUSSION DNA methylation. To analyze the potential epigenetic regulation In this work, we identified a role for HMGB1 in promoting tumor of SEM3A expression, we first used chemical inhibitors of histone growth, cell migration and metastasis that is independent of deacetylases and DNA methyltransferases, trichostatin A and MAPK activity but at least partly mediated by transcriptional 5-Azacytidine, respectively. 5-Azacytidine treatment increased repression of SEMA3A. SEMA3A, and generally the class 3 expression of SEMA3A mRNA significantly and showed an even semaphorins are increasingly being recognized as key suppressors greater increase when used in combination with trichostatin A of tumor growth and metastasis. Our findings suggest that (Figure 6a). overexpression of HMGB1 is one mechanism that contributes to To test if HMGB1 altered the presence of acetylated histones in a semaphorin silencing in cancer. locus-specific manner, we used ChIP with an antibody to acetylated In vivo, we showed a significantly reduced growth rate of histone H4 or H3 with or without HMGB1 knockdown. As shown in HMGB1 knockdown tumors despite no difference in proliferation Figure 6e, loss of HMGB1 increased acetylated histone H4 (Ac-H4) in vitro. This reduction in growth correlated with increased occupancy at the SEMA3A promoter by ninefold compared with IgG necrosis and reduced angiogenesis in the HMGB1-deficient control immunoprecipitation. In contrast, binding of Ac-H4 tumors. These tumors also showed higher levels of SEMA3A increased to a lesser degree at the SEMA4F promoter at about in vivo. SEMA3A is a well-established inhibitor of angiogenesis in threefold. A similar SEMA3A-specific enrichment was observed for various experimental systems. Although we did not formally acetylated histone H3 in the absence of HMGB1 (Figure 6f). investigate the possibility here, it is plausible that increased To further investigate the impact of HMGB1 on DNA methyla- SEMA3A expression contributed to the reduced angiogenesis and tion, we performed bisulfite sequencing on a region near the defective growth of HMGB1 knockdown tumors.

Figure 7. HMGB1 depletion reduces non-CpG methylation and increases CTCF binding at the SEMA3A locus in SUM159 and MDA-MB-231 cells. (a) Region investigated by bisulfite sequencing. A minimum of 10 control shRNA and HMGB1 knockdown shRNA colonies derived from either SUM159PT or MDA-MB-231 DNA were analyzed and percentages of cytosine methylation calculated. The specific de-methylated cytosine in either cell type is highlighted and displayed as: (% methylated in control cells)–(% methylated in HMGB1 knockdown cells). (b) CTCF binding by ChIP. CTCF occupancy at the SEMA3A locus was investigated as described in Figure 6.

& 2014 Macmillan Publishers Limited Oncogene (2014) 5151 – 5162 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5160 Although SEMA3A upregulation contributed to the HMGB1 selected for by incubating in this media for a minimum of 1 week. Three knockdown phenotype, it was only one of 122 genes targeted by sequences that resulted in efficient knockdown of HMGB1 were used in HMGB1. It is likely that the other HMGB1 targets also contribute to subsequent experiments: the pro-tumor phenotype associated with HMGB1 overexpression. Sh3: 50-CCGGCCGTTATGAAAGAGAAATGAACTCGAGTTCATTTCTCTTTCAT It is of note that although HMGB1 has no intrinsic sequence AACGGTTTTT-30 specificity in binding DNA, we nonetheless observed altered Sh4: 50-CCGGGCAGATGACAAGCAGCCTTATCTCGAGATAAGGCTGCTTGTC expression at only specific gene targets during knockdown. We ATCTGCTTTTT-30 and others have shown that HMGB1 is not differentially enriched Sh5: 50-CCGGCCCAGATGCTTCAGTCAACTTCTCGAGAAGTTGACTGAAGCA 0 for binding at specific loci, regardless of its differential effect on TCTGGGTTTTT-3 specific gene expression. It is possible that HMGB1 achieves gene One shRNA that did not have any effect on HMGB1 expression was used specificity due to being a required cofactor only at certain loci, as a control vector: making these genes especially sensitive to HMGB1 expression Sh1: 50-CCGGCGAGACTTTCATTACAAGTATCTCGAGATACTTGTAATGAAAG levels. Our studies show that HMGB1 participates in gene silencing TCTCGTTTTT-30 through driving chromatin remodeling that is associated with decreased binding of acetylated histones H3 and H4. These findings suggest that those gene targets that are particularly Scattering assay B sensitive to histone deacetylase inhibitors may also be sensitive to Cells were serially diluted into six-well plates to a final concentration of 50– HMGB1 inhibition. We have also observed a decrease in non-CpG 100 cells per well. Cells were grown in 10% fetal bovine serum and Dulbecco’s modified Eagle medium for 3–4 weeks, with media being changed every 3 cytosine methylation on HMGB1 knockdown in two cell lines. This days.After3–4weeks,colonieswereimagedbylightmicroscopy. occurred in proximity to a CTCF-binding sequence in both cases and HMGB1 knockdown caused an increase in CTCF occupancy upstream of SEMA3A. Although CTCF has been described as a Scratch assay transcriptional repressor, in this case increased CTCF binding Equal numbers of cells were seeded into six-well plates overnight with full correlated with transcriptional activation of SEMA3A. Future media. Scratches were conducted with a 20-ml pipette tip and media was studies will be needed to examine the relationship between changed. Scratches were imaged every 3 h using a marked position to return to the same scratch on the plate over time. cytosine methylation, CTCF binding and SEMA3A expression. The inverse correlation between HMGB1 and SEMA3A that we observed in primary tumor transcriptional microarray data suggest F-actin staining that HMGB1-mediated repression of SEMA3A might occur in Cells were seeded overnight onto chamber slides and fixed the next day human tumors. Although this is only correlative data, in light of with 4% paraformaldehyde in PBS for 20 min. Cells were then washed and the causative and mechanistic role for HMGB1 in cell lines, this permeabilized with 1% Triton-X in PBS for 10 min. After washing, 5% goat correlation may be functionally relevant. serum was used to block for 1 h. Rhodamine conjugated to phalloidin was added at 1:500 in 5% goat serum in PBS. After 1-h incubation in the dark, HMGB1 has been recognized as a tumor-promoting factor for slides were washed and stained with Hoechst dye. many years, but the majority of work has focused on its role as an activator for the MAPK pathway. Several proposed therapies seek to neutralize, block release of or inhibit signaling by extracellular Metastatic colonization assay HMGB1. Our results shed light on the significant role nuclear A total of 50 000 luciferase-expressing MDA-MB-231 cells were intracar- HMGB1 has in promotion of tumor growth and metastasis, dially injected into 6- to 7-week-old nude-Foxn1nu mice (Harlan), as previously described.22 Mice were imaged weekly using Xenogen software. independently of its role as a ligand for MAPK activation. This suggests that future cancer therapies should consider targeting nuclear HMGB1 and emphasize the important pro-tumor functions Tumor growth in vivo of HMGB1 even in malignancies that harbor MAPK mutations. Two million luciferase-expressing MDA-MB-231 cells were injected into the mammary fat pads of SCID/Beige (C57BL/6 background) mice obtained from Charles River Laboratories (Wilmington, MA, USA). Tumors were MATERIALS AND METHODS imaged weekly. Cell lines MDA-MB-231-luciferase cells were a gift from Dr Byron Hann. SUM159PT Immunohistochemistry and HS578T cells were a gift from Dr Madhu Macrae. A549 cells were a gift Tumors were removed from mice and fixed overnight at 4 1C in Z-fix solution from Dr Tony Karnezis. MDA-MB-435 cells were a gift from Dr Amy Young. (Anatech, Battle Creek, MI, USA). After subsequent dehydration and sectioning, slides were incubated with biotinylated rat anti-mouse CD31 Western blotting and proteins (BD Biosciences, San Jose, CA, USA) overnight. Slides were developed using HRP conjugated to streptavidin (R&D Systems). Polyclonal rabbit anti-HMGB1 was obtained from Abcam (Cambridge, MA, USA). Anti-SEMA3A was from ECM Biosciences (Versailles, KY, USA). Phospho-p42/44 and p42/44 were purchased from Cell Signaling (Beverly, Microarray MA, USA). Images were scanned and quantified using the Odyssey system RNA was isolated from cells with the RNEasy kit (Qiagen). Transcriptional and Licor imaging software (Lincoln, NE, USA). Recombinant HMGB1 was microarray analysis was conducted with help from the David Gladstone obtained from R&D Systems (Minneapolis, MN, USA). Research Center Genome Core. Affymetrix Human Gene 1.0 chips were used in triplicate for each sample. Data was analyzed using the Exploratory 33 Lentiviral shRNA transduction Gene Association Network (EGAN) program (http://akt.ucsf.edu/EGAN/). Five bacterial clones carrying HMGB1-targeted shRNA sequences from the Mission shRNA consortium in the pLKO.1 vector were propagated Real-time PCR (RT–PCR) overnight. DNA was then isolated using the Qiagen Maxiprep kit as per RT–PCR for SEMA3A expression was performed with help from the UCSF manufacturer’s instructions (Qiagen, Hilden, Germany). DNA was trans- Genome Core using the Taqman probes to HMGB1 and SEMA3A (Applied fected with Fugene into 293T cells in the following amounts: 4.2 mg shRNA, Biosystems, Foster City, CA, USA). 0.4 mg VSVG, 3.7 mg GAG/POL, 3.7 mg REV. After 3 days of incubation at 37 1C, the virus-containing supernatant was removed and filtered through a 0.45-mm filter before being added to target cells. A volume of 4.5 mg/ml Chromatin immunoprecipitation polybrene was added to aid in the infection. After 24 h, media was ChIP was performed as previously described.34 Briefly, 500 mg of protein/ replaced with 6 mg/ml puromycin-containing media. Transduced cells were chromatin mixture was incubated with 4 mg antibody overnight at 4 1C.

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5161 Antibodies used were as follows: anti-Myc (9E10) (Santa Cruz REFERENCES Biotechnology, Santa Cruz, CA, USA), anti-CTCF (Abcam), mouse IgG 1 Ellerman JE, Brown CK, de Vera M, Zeh HJ, Billiar T, Rubartelli A et al. Masquerader: isotype (Abcam), normal rabbit serum (Sigma, St Louis, MO, USA) and high mobility group box-1 and cancer. Clin Cancer Res 2007; 13: 2836–2848. acetylated-H3 and H4 (Cell Signaling). After several washes, complexes 2 Volp K, Brezniceanu ML, Bosser S, Brabletz T, Kirchner T, Gottel D et al. Increased were eluted and DNA cross-linking reversed. DNA was purified either by expression of high mobility group box 1 (HMGB1) is associated with an elevated phenol–chloroform extraction or using the Bio-Rad ChIP DNA purification level of the antiapoptotic c-IAP2 protein in human colon carcinomas. Gut 2006; kit, per manufacturer’s instructions (Bio-Rad, Berkeley, CA, USA). 55: 234–242. RT–PCR for ChIP analysis was performed using SYBR green (Applied 3 Maeda S, Hikiba Y, Shibata W, Ohmae T, Yanai A, Ogura K et al. Essential roles of Biosystems). The amplification program used was as follows: high-mobility group box 1 in the development of murine and colitis- Step 1: 94 1C for 10 min associated cancer. Biochem Biophys Res Commun 2007; 360: 394–400. Step 2: 94 1C for 20 s 4 Cheng BQ, Jia CQ, Liu CT, Lu XF, Zhong N, Zhang ZL et al. Serum high mobility Step 3: 60 1C for 1 min group box chromosomal protein 1 is associated with clinicopathologic features in patients with hepatocellular carcinoma. Dig Liver Dis 2008; 40: 446–452. Steps 2 and 3 were repeated for 50 cycles. The amplification signal was 5 Wu D, Ding Y, Wang S, Zhang Q, Liu L. Increased expression of high mobility measured at Step 3, and dissociation curves were used to ensure single group box 1 (HMGB1) is associated with progression and poor prognosis in products. human nasopharyngeal carcinoma. J Pathol 2008; 216: 167–175. The primers used for amplification of the regions in SEMA3A and 6 Ishiguro H, Nakaigawa N, Miyoshi Y, Fujinami K, Kubota Y, Uemura H. Receptor for SEMA4F were as follows: advanced glycation end products (RAGE) and its ligand, amphoterin are over- 3A-1: 50-CCGGATAATGAGGCACAACT-30 expressed and associated with prostate cancer development. Prostate 2005; 64: 50-TAGAGACTGCCACCGGCTAT-30 92–100. 3A-2: 50-GTAGTTGGCTGTGGCCTCTC-30 7 Poser I, Golob M, Buettner R, Bosserhoff AK. Upregulation of HMG1 leads to 50-GGGGTAGGGCAGAATCATTT-30 melanoma inhibitory activity expression in malignant melanoma cells and 4A-1: 50-TAGGCAGCCGTCCTTAAATG-30 contributes to their malignancy phenotype. Mol Cell Biol 2003; 23: 2991–2998. 50-GGCCACTCCAAAAACTCAAA-30 8 Brezniceanu ML, Volp K, Bosser S, Solbach C, Lichter P, Joos S et al. HMGB1 inhibits cell death in yeast and mammalian cells and is abundantly expressed in human Fold-change enrichment was quantified using the formula: breast carcinoma. FASEB J 2003; 17: 1295–1297. Delta critical threshold target ¼ dCT.target region ¼ CT input–CT IP 9 Malarkey CS, Churchill ME. The high mobility group box: the ultimate utility player Delta critical threshold isotype ¼ dCT.isotype ¼ CT input–CT isotype of a cell. Trends Biochem Sci 2012; 37: 553–562. Delta s.d. ¼ dSD.target region ¼ sqrt( (SD.input)^2 þ (s.d. target region)^2) 10 Joshi SR, Sarpong YC, Peterson RC, Scovell WM. dynamics: HMGB1 /sqrt(n)) relaxes canonical nucleosome structure to facilitate binding. Delta delta critical threshold ¼ ddCT ¼ dCT.target region–dCT.isotype Nucleic Acids Res 2012; 40: 10161–10171. Delta delta s.d. ¼ ddSD ¼ sqrt((dSD.target region)^2 þ (dSD.isotype)^2) 11 Calogero S, Grassi F, Aguzzi A, Voigtlander T, Ferrier P, Ferrari S et al. The lack of Fold enrichment ¼ 2^(ddCT) chromosomal protein Hmg1 does not disrupt cell growth but causes lethal Fold-enrichment error ¼ ln(2) Â ddSD Â fold enrichment. hypoglycaemia in newborn mice. Nat Genet 1999; 22: 276–280. 12 Agresti A, Lupo R, Bianchi ME, Muller S. HMGB1 interacts differentially with members of the Rel family of transcription factors. Biochem Biophys Res Commun DNAse sensitivity assay 2003; 302: 421–426. 13 Jayaraman L, Moorthy NC, Murthy KG, Manley JL, Bustin M, Prives C. High mobility Analysis was performed using the Bio-Rad EpiQ kit as per manufacturer’s group protein-1 (HMG-1) is a unique activator of . Genes Dev 1998; 12: instructions. Briefly, cells were seeded into 24-well plates overnight. Cells 462–472. were permeabilized with a weak detergent and incubated with DNAse for 14 Ge H, Roeder RG. The high mobility group protein HMG1 can reversibly inhibit 1 1 h at 37 C. Genomic DNA was isolated with a kit (Qiagen) and analyzed by class II gene transcription by interaction with the TATA-binding protein. J Biol RT–PCR using the following program: Chem 1994; 269: 17136–17140. Step 1: 2 min at 94 1C 15 van Beijnum JR, Buurman WA, Griffioen AW. Convergence and amplification of Step 2: 20 s at 94 1C toll-like receptor (TLR) and receptor for advanced glycation end products (RAGE) Step 3: 30 s at 54 1C signaling pathways via high mobility group B1 (HMGB1). Angiogenesis 2008; 11: Step 4: 30 s at 70 1C 91–99. Steps 2–4 were repeated for 60 cycles 16 Taguchi A, Blood DC, del Toro G, Canet A, Lee DC, Qu W et al. Blockade of RAGE- amphoterin signalling suppresses tumour growth and metastases. Nature 2000; 405: 354–360. 17 Liang X, Chavez AR, Schapiro NE, Loughran P, Thorne SH, Amoscato AA et al. Ethyl Bisulfite sequencing pyruvate administration inhibits hepatic tumor growth. J Leukoc Biol 2009; 86: Genomic DNA was isolated with a kit (Qiagen). DNA was then subjected to 599–607. bisulfite conversion with the Epitect kit (Qiagen). Converted DNA was 18 Dave SH, Tilstra JS, Matsuoka K, Li F, DeMarco RA, Beer-Stolz D et al. Ethyl pyruvate amplified using the following primers: decreases HMGB1 release and ameliorates murine colitis. J Leukoc Biol 2009; 86: 633–643. 50-GTTTGTGGGATTTTGTGTTATTTTT-30 19 Gebhardt C, Riehl A, Durchdewald M, Nemeth J, Furstenberger G, Muller-Decker K 50-ACAATCTCTAATCCTACTCCCTACTTC-30 et al. RAGE signaling sustains inflammation and promotes tumor development. PCR-amplified DNA was then cloned into pGEM-T vectors (Invitrogen) J Exp Med 2008; 205: 275–285. and transformed into DH5alpha bacteria. Colonies were individually 20 Ikediobi ON, Davies H, Bignell G, Edkins S, Stevens C, O’Meara S et al. Mutation isolated and sequenced. analysis of 24 known cancer genes in the NCI-60 cell line set. Mol Cancer Ther 2006; 5: 2606–2612. 21 Yao X, Zhao G, Yang H, Hong X, Bie L, Liu G. Overexpression of high-mobility group box 1 correlates with tumor progression and poor prognosis in human CONFLICT OF INTEREST colorectal carcinoma. J Cancer Res Clin Oncol 2010; 136: 677–684. 22 Bos PD, Zhang XH, Nadal C, Shu W, Gomis RR, Nguyen DX et al. Genes that The authors declare no conflict of interest. mediate breast cancer metastasis to the brain. Nature 2009; 459: 1005–1009. 23 Lin Q, Yang XP, Fang D, Ren X, Zhou H, Fang J et al. High-mobility group box-1 mediates toll-like receptor 4-dependent angiogenesis. Arterioscler Thromb Vasc Biol 2011; 31: 1024–1032. ACKNOWLEDGEMENTS 24 Neufeld G, Kessler O. The semaphorins: versatile regulators of tumour progression We thank the UCSF Tetrad program for providing funding. We thank Dr Lewis Lanier and tumour angiogenesis. Nat Rev Cancer 2008; 8: 632–645. for all of his advice. We thank Dr Byron Hann, Dr Paul Phojanakong and the UCSF 25 Staton CA, Shaw LA, Valluru M, Hoh L, Koay I, Cross SS et al. Expression of class 3 Clinical Therapeutics core for all of their help with mouse xenograft experiments and semaphorins and their receptors in human breast neoplasia. Histopathology 2011; the UCSF Genome core for their help with gene expression studies. 59: 274–282.

& 2014 Macmillan Publishers Limited Oncogene (2014) 5151 – 5162 HMGB1 promotes cell migration by silencing SEMA3A M Nehil et al 5162 26 Casazza A, Fu X, Johansson I, Capparuccia L, Andersson F, Giustacchini A et al. 30 McCauley MJ, Zimmerman J, Maher 3rd LJ, Williams MC. HMGB binding to DNA: Systemic and targeted delivery of semaphorin 3A inhibits tumor angiogenesis single and double box motifs. J Mol Biol 2007; 374: 993–1004. and progression in mouse tumor models. Arterioscler Thromb Vasc Biol 2011; 31: 31 El Gazzar M, Yoza BK, Chen X, Garcia BA, Young NL, McCall CE. Chromatin-specific 741–749. remodeling by HMGB1 and linker histone H1 silences proinflammatory genes 27 Schmidt EF, Strittmatter SM. The CRMP family of proteins and their role in during endotoxin tolerance. Mol Cell Biol 2009; 29: 1959–1971. Sema3A signaling. Adv Exp Med Biol 2007; 600: 1–11. 32 Sapojnikova N, Maman J, Myers FA, Thorne AW, Vorobyev VI, Crane-Robinson C. 28 Schmidt EF, Shim SO, Strittmatter SM. Release of MICAL autoinhibition by Biochemical observation of the rapid mobility of nuclear HMGB1. Biochim Biophys semaphorin-plexin signaling promotes interaction with collapsin response med- Acta 2005; 1729: 57–63. iator protein. J Neurosci 2008; 28: 2287–2297. 33 Paquette J, Tokuyasu T. EGAN: exploratory gene association networks. Bioinfor- 29 Ranzato E, Patrone M, Pedrazzi M, Burlando B. Hmgb1 promotes wound healing matics 2010; 26: 285–286. of 3T3 mouse fibroblasts via RAGE-dependent ERK1/2 activation. Cell Biochem 34 Szak ST, Mays D, Pietenpol JA. Kinetics of p53 binding to promoter sites in vivo. Biophys 2010; 57: 9–17. Mol Cell Biol 2001; 21: 3375–3386.

Oncogene (2014) 5151 – 5162 & 2014 Macmillan Publishers Limited