Identification of a gene regulatory network associated with prion replication

Prions consist of aggregates of abnormal conformers of the cellular prion protein (PrPC). They propagate by recruiting host-encoded PrPC although the critical interacting proteins and the reasons for the differences in susceptibility of distinct cell lines and populations are unknown. We derived a lineage of cell lines with markedly differing susceptibilities, unexplained by PrPC expression differences, to identify such factors. Transcriptome analysis of prion-resistant revertants, isolated from highly susceptible cells, revealed a gene expression signature associated with susceptibility and modulated by differentiation. Several of these genes encode proteins with a role in extracellular matrix (ECM) remodelling, a compartment in which disease-related PrP is deposited. Silencing nine of these genes significantly increased susceptibility. Silencing of Papss2 led to undersulphated heparan sulphate and increased PrPC deposition at the ECM, concomitantly with increased prion propagation. Moreover, inhibition of fibronectin 1 binding to integrin α8 by RGD peptide inhibited metalloproteinases (MMP)-2/9 whilst increasing prion propagation. In summary, we have identified a gene regulatory network associated with prion propagation at the ECM and governed by the cellular differentiation state.


Introduction
Transmissible spongiform encephalopathies or prion diseases are a family of fatal neurodegenerative diseases and include scrapie in sheep, bovine spongiform encephalopathy (BSE) in cattle, and Creutzfeldt-Jakob disease in humans. Prions, the transmissible agents, consist of aggregates of abnormal conformers of the cellular prion protein (PrP C ), generally referred to as PrP Sc , and replicate in a self-perpetuating manner by conversion of host-encoded PrP C . Whilst the physiological role of PrP, a cell surface protein highly expressed in the central nervous system, is unclear, recent reports suggest that it may act as a receptor for amyloid beta (Ab) in Alzheimer's disease (Lauren et al, 2009;Freir et al, 2011). To better understand molecular events that lead to prion neurodegeneration, it is critical to identify genetic factors that facilitate or impede prion replication. Coding polymorphisms within Prnp, the gene encoding PrP, are known to affect disease incubation times and susceptibility in human, mouse, and sheep (Hunter, 1997;Collinge, 2001). The most prominent example, codon 129 polymorphism in humans, has major disease-modifying effects (Collinge, 2001) and homozygosity for methionine at codon 129 confers susceptibility to variant CJD (vCJD) (Collinge et al, 1996;Collinge, 2005). However, significant differences in incubation times for scrapie in mice with the same Prnp genotype indicate a major role of PrP-independent genetic factors, and several genetic loci have been identified on different chromosomes (Carlson et al, 1993;Lloyd et al, 2001Lloyd et al, , 2009b. A number of knockout mice with disruptions in specific genes that were believed to affect prion replication did not show any discernible effect on the pathogenesis of prion disease (Tamguney et al, 2008). However, ablation of two genes, amyloid beta precursor protein (App) and interleukin-1 receptor type I (Il1r1), and transgenic overexpression of human superoxide dismutase 1 (SOD1) prolonged incubation times by 13, 16, and 19%, respectively (Tamguney et al, 2008). Our recent genome-wide association study identified two new common variants, the retinoic acid receptor beta (RARB) and stathmin-like 2 (STMN2) that are associated with risk of vCJD (Mead et al, 2009). An E3 ubiquitin ligase, HECTD2, was found to be associated with susceptibility to mouse and human prion disease (Lloyd et al, 2009a).
Mammalian cell lines have proven invaluable to investigate aspects of prion pathogenesis in vitro, such as infection and propagation (Race et al, 1987;Krammer et al, 2009;Marijanovic et al, 2009;Goold et al, 2011), prion strain selection (Li et al, 2010;Weissmann et al, 2011), and prion dissemination (Fevrier et al, 2004;Gousset et al, 2009). However, most PrP-expressing cell lines are resistant to prion infection, indicating that factors in addition to PrP are required to initiate and/or maintain chronic propagation of prions. To better understand the molecular underpinnings of neurodegeneration in prion diseases, we sought to study cognate susceptible and resistant cells, an approach that provides a unique opportunity to identify genetic factors that modulate prion replication.
We isolated rare prion-resistant revertants from highly susceptible mouse neuroblastoma N 2 a cells, determined the expression differences between resistant and susceptible cells, and identified a gene signature that was associated with inhibition of prion replication. Validation by RNA interference confirmed the inhibitory activities on prion replication of nine genes, most of which encode proteins expressed at plasma membrane level or at the ECM, a compartment where disease-associated PrP accumulates. Here, we use the term 'disease-associated PrP' (PrP d ), rather than PrP Sc as the latter is defined biochemically as proteinase K (PK)-resistant PrP, and it is now established that there are important PK-sensitive forms of disease-related PrP as well (Safar et al, 1998). We suggest that fibronectin, which is highly expressed in prion-resistant revertants, activates ECM-resident metalloproteinases in an integrin a8-dependent manner. Notably, inhibition of integrin a8 signalling by the fibronectin fragment inhibitor RGD increased prion susceptibility and inhibited metalloproteinase activation. We furthermore show that silencing of Papss2, a gene expressed in revertants, led to undersulfation of heparan sulphate, increased PrP C deposition at the ECM and an increase in prion replication rates. Although the ECM has previously been implicated in modulating prion propagation (Caughey & Raymond, 1993b;Gabizon et al, 1993;Caughey et al, 1994), here we identify key genes involved in this process. The differential susceptibility of cell lines and different neuronal populations to prion infection has hitherto been unexplained, and these findings may be critical to understanding prion pathogenesis and selective vulnerability of different cell types to prion infection.

Isolation of cognate prion-resistant revertants from highly susceptible cells
Whilst most PrP-expressing neuronal cell lines are resistant to prions, subclones of otherwise poorly permissive cell lines showed marked differences in susceptibility to prion propagation (Bosque & Prusiner, 2000;Enari et al, 2001;Klohn et al, 2003;Mahal et al, 2007). After extensive subcloning, we derived PK1 cells, a mouse neuroblastoma cell line highly permissive to mouse RML prions (Fig 1A and C), which we used to develop a sensitive cell-based prion bioassay, the Scrapie Cell Assay (SCA) (Klohn et al, 2003). We reasoned that the isolation of prion-resistant revertants from highly susceptible PK1 cells may allow the identification of genes associated with prion propagation by analysis of their respective transcriptomes. By determining prion propagation rates of a thousand PK1 subclones, three revertant clones (R2, R5, and R7) showed markedly reduced prion propagation rates when compared to susceptible PK1 cells (Fig 1B). To further characterise the degree of kinship between cognate cell clones, we determined the global gene expression profiles of individual N 2 a clones depicted in Fig 1C and subsequently reduced the complexity of data sets using principal component analysis (PCA) (Fig 1D). When mapped onto a 3D transcript profile space, all PK1-derived subclones clustered around PK1 cells and were more distant from the parental N 2 a cells and their prion-resistant progeny, R33 and NN 2 a (Fig 1C and D). Given the close kinship between PK1 and its progeny, we reasoned that gene expression analysis of cognate cell clones may be favourable to reduce the number of false-positive calls, that is, expression differences unrelated to the phenotype of prion susceptibility. We therefore excluded N 2 a, NN 2 a, and R33 cells and selected the six closely related cell clones (PD88, PK1, S7, R2, R5, and R7) for further analyses.

Overexpression of PrP does not render revertants susceptible
Whilst accelerated disease progression was observed in prioninfected Tga20 mice which express PrP at about 10 times the wildtype level (Fischer et al, 1996), overexpression of PrP in a range of mouse N 2 a sublines did not increase susceptibility to mouse prions (Enari et al, 2001). To investigate whether the rate of prion propagation is a function of PrP expression, we stably overexpressed PrP in a variety of cell clones and determined their susceptibility to prions (Supplementary Table S1). To confirm that the expressed Prnp is functional, we used it to stably reconstitute Prnp-silenced PK1 cells (PK1 Prnp-kd) and challenged a heterogeneous pool of these cells with mouse RML prions. Whilst prion susceptibility was recovered by reconstituting PK1 Prnp-kd cells, revertants remained nonpermissive to mouse RML prions after PrP overexpression. In addition, no significant increase in susceptibility of prion-permissive clones was observed at elevated PrP expression levels (Supplementary Table S1). To exclude the possibility that revertants express polymorphic Prnp and thus inhibit prion propagation by interference with the expressed Prnp transgene, we sequenced Prnp from representative PK1 clones. However, all PK1 subclones expressed Prnp allotype A (Prnp a ), the allotype of the transgene. This indicates that PrP expression is necessary, but not sufficient to confer susceptibility to prion propagation.

Differential gene expression between prion-resistant revertants and susceptible cells
We next determined differentially expressed genes between prionresistant revertants (R2, R5, R7) and susceptible cells (PK1, S7, PD88) by non-parametric statistics using 'Significance Analysis of Microarrays' (SAM) (Tusher et al, 2001) and corrected raw values for multiple testing at high stringency with a false discovery rate (FDR) (Benjamini & Hochberg, 1995) < 0.01. Genes significantly expressed in prion-susceptible cells, and revertants are listed in Supplementary Table S2. Unsupervised hierarchical clustering clearly segregated genes from revertant and susceptible cells and revealed gene clusters with similar expression patterns as depicted A Schematic for the isolation of prion-resistant revertants. B Susceptible cells (S7, PK1, PD88) propagate prions 2-3 orders of magnitude faster than revertants. Prion propagation rates of cells infected with mouse RML prions are expressed as tissue culture infectious units (TCIU)/day. C Lineage of susceptible and resistant cell clones isolated from parental N 2 a cells (grey: resistant, red: susceptible, blue: revertant resistant). D Gene expression profiles of N 2 a cell clones were mapped onto a 3D transcript profile space after reducing dimensionality to three principal components. PK1-derived susceptible and revertant clones cluster around PK1 cells. E Hierarchical clustering of genes differentially expressed between prion-resistant revertants and susceptible cells. Genes with a fold discovery rate (FDR) < 0.01 and a fold difference of at least two were included.  1E). Functional annotation clustering was used to infer whether gene sets, annotated by gene ontology (GO) terms, were overrepresented in the set of differentially expressed genes, when compared to their representation in the whole mouse genome. Two highly enriched gene sets, cellular differentiation (18 genes) and development (16 genes), were identified in a list of 100 differentially expressed genes (Supplementary Table S3). Notably, a set of five genes with a role in negative regulation of differentiation were expressed in revertants (Supplementary Table S3). Consistent with this notion, revertant cells showed a less differentiated morphology than prion-susceptible cells ( Supplementary Fig S1).
A phenotypic switch from prion-resistant to susceptible cells reveals putative prion susceptibility genes The enrichment of genes with a role in negative regulation of cell differentiation prompted us to test whether preincubation of revertants with retinoic acid (RA), a well-characterised differentiation agent, affected the rate of prion replication. Remarkably, preincubation of revertant clones with a single dose of 0.5 lM RA augmented the rate of prion replication by up to 40-fold as compared to vehicle alone (Supplementary Table S4). Under these conditions, the cellular morphology and the cell doubling rates of revertants were unaffected (Supplementary Fig S2). In contrast to this marked increase in susceptibility, the rate of prion propagation only doubled for the weakly susceptible clone PD88 and decreased for highly susceptible PK1 cells with a concomitant decrease in cell doubling ( Supplementary Fig S2). These results suggest that cellular processes associated with the differentiation state of cells modulate susceptibility to prion propagation, in agreement with a nerve growth factor (NGF)-mediated increase in prion susceptibility of PC12 cells (Rubenstein et al, 1990). This RA-mediated phenotypic switch from prion-resistant to prion-susceptible cells (Supplementary Table S4) provided us with an experimental approach to identify gene candidates associated with a gain of prion susceptibility. We therefore determined genes that were differentially expressed in revertants in the presence and absence of RA (Supplementary Table S5) and compared this set of genes with the candidate list of previously identified differentially expressed genes between revertant and susceptible cells (Supplementary Table S2). Remarkably, eighteen of the previously identified genes were also differentially expressed upon RA treatment (Fig 2B and C): sixteen genes expressed in revertants, but not in susceptible cells, were downregulated upon RA treatment, whereas two genes, Nckap1 l and Tshz1, downregulated in revertants, but expressed in susceptible cells, were induced in revertants under these conditions.
To validate the microarray data, we determined gene expression levels by quantitative real-time PCR (qPCR) using dual-labelled probes. Qualitative changes in gene expression values were fully confirmed with minor differences in gene expression levels ( Fig 2C). Together these data provide evidence for the identification of differentially expressed genes that are associated with prion susceptibility.

Identification of a gene regulatory network associated with prion propagation
We next examined in a systematic gene silencing approach whether the loss of function of single candidates of the gene signature could recapitulate the gain of susceptibility observed upon RA differentiation of revertants (Fig 2 and Supplementary Table S4).
Due to the substantial number of gene candidates, we decided to transiently co-express short hairpin RNAs (shRNAs) alongside with green fluorescent protein (GFP) using an internal ribosomal entry (IRES)-based bicistronic vector (pGIPZ, Fig 3B) and to enrich for GFP-expressing cells by fluorescent-activated cell sorting (FACS,Fig 3A,C and D). To validate this assay, we transfected susceptible PK1 cells with five distinct shRNAs against Prnp and enriched from a heterogeneous pool of fluorescent cells (Fig 3C) highly fluorescent cells in the 4th decade of the logarithmic fluorescence scale ( Fig 3D). As shown in cultured cells, the enrichment of GFP-fluorescent cells was associated with greatly reduced PrP expression levels ( Fig 3E). In a proof-of-concept experiment, we then demonstrated that transient Prnp silencing of prion-susceptible PK1 cells significantly reduced the rate of prion propagation ( Fig 3F). This enrichment procedure was used subsequently to examine whether gene silencing of each of our candidate genes affects prion replication rates.
Since the identified gene candidates were also expressed in susceptible cells, albeit at much lower expression levels, we examined whether gene knockdown in these cells might enhance their prion propagation kinetics further. Since S7, the cell clone with the fastest kinetics of prion propagation (Fig 1B), showed poor transfection efficiencies with the pGIPZ vector, we used instead small inhibitory RNAs (siRNAs) to transiently silence gene expression with no subsequent cell enrichment to address this question. Remarkably, gene silencing of Fn1, Micalcl and Papss2 significantly increased the rate of prion propagation by about twofold in S7 cells (Supplementary Table S7). Of note, knockdown of Nckap1l, a gene highly expressed in susceptible cells and overexpressed in revertants by fivefold after RA treatment (Fig 2), significantly reduced prion susceptibility of S7 cells. This result confirms Nckap1l, a gene that was shown to be differentially expressed in brains of prion-diseased mice (Hwang et al, 2009) as a prion susceptibility gene.
The increased susceptibility of revertants may be due to several factors, such as the uptake of prions, their transport to replication sites, and the steady-state rates of synthesis and degradation (Weissmann, 2004). To examine whether the identified genes affect the steady-state levels of prion turnover, we silenced gene candidates in chronically prion-infected cells (iS7) and determined relative changes of prion levels 3 days after transfection with siRNA or scrambled control RNA (Supplementary Table S8). Remarkably, a significant increase of prion conversion rates was determined for all genes.
To investigate whether the identified genes affect prion propagation in a strain-specific manner, we challenged R7 cells with 22L and determined changes in susceptibility (Supplementary Table S9).
Whilst a trend to increased prion propagation rates was observed for all genes studied, except for Galt, statistically significant results were obtained for more than half of the genes, including Fn1, Itga8, Papss2, Chga, Il11ra1, and Lrrn4. We conclude that some of the identified genes may control prion susceptibility in a strainindependent manner. To examine whether the increase in prion susceptibility by loss of gene function is restricted to N 2 a-derived cells, we silenced a selection of candidate genes in CAD5 cells, a cell line derived from CNS catecholaminergic-differentiated (CAD) cells (Mahal et al, 2007) prior to RML infection. Knockdown of four out of eight candidate genes (Fn1, Galt, Il11ra1, and Itga8) resulted in a significant increase in susceptibility (Supplementary Table S10), indicating that control of prion propagation by the identified genes is not restricted to N 2 a cells.

Prion modifiers are expressed at the extracellular matrix and plasma membrane level
To characterise the subcellular location of prion modifier proteins, we sourced suitable commercial anti-rabbit antibodies and Masue M Marbiah et al A gene regulatory network associated with prion replication The EMBO Journal co-immunolabelled candidate proteins and PrP (ICSM18) in fixed and permeabilised S7 and R7 cells (Fig 4). All co-labelling studies were conducted with highly cross-absorbed secondary antibodies to exclude cross-reactivity. Protein expression levels of Fn1, Chga, Lrrn4, and Il11ra1 were elevated in prion-resistant R7 compared to susceptible S7 cells as anticipated from the corresponding gene expression data (Fig 2). Furthermore, the expression of candidate proteins in R7 was greatly reduced upon treatment with 0.5 lM RA (Fig 4A-E). Fn1, a protein expressed at the extracellular matrix (ECM) with a major role in cell adhesion, migration, and differentiation, showed punctate, but no fibrillar structures, reminiscent of cells with defects in matrix assembly (Yoneda et al, 2007) (Fig 4A). Similarly, Chga, a secretory protein with a role in regulation of secretory granule synthesis (Kim et al, 2001), was deposited at the ECM as depicted in R7 cells ( Fig 4B). Neither Fn1 nor Chga showed colocalisation with PrP at the ECM level as documented by their corresponding Pearson correlation coefficients (PCC, Fig 4A and B). In contrast, Lrrn4 and Il11ra1, which were expressed at the ECM and the membrane level, showed partial colocalisation with PrP with PCC values of 0.55 AE 0.09 and 0.36 AE 0.14, respectively (Fig 4C and D). An antibody against integrin a8 confirmed higher protein expression levels in R7 in comparison with S7 cells; however, no expression difference could be detected in presence and absence of RA ( Supplementary Fig S3A), in agreement with gene expression levels (Fig 2). Micalcl, a putative binding protein of extracellular signal-regulated kinase 2 (ERK2) (Miura & Imaki, 2008), was expressed at the ECM as shown by N-terminal fusion of Miclacl with YFP ( Supplementary Fig S3B). Iqgap2, a cytoskeletal scaffolding protein, was expressed at the membrane level ( Supplementary Fig S3C). Since antibodies against PrP and Iqgap2 were both raised in mice, double-labelling experiments could not be performed. The protein subcellular location of Rgs4, Fst, Papss2, and Galt could not be determined due to the lack of specificity of commercial antibodies.

Detection of aberrant PrP d deposits at the ECM after delipidation with acetone
To investigate how the expression of prion modifiers might interfere with prion formation on a subcellular level, we sought to determine PrP d by immunofluorescence (IF) on formaldehyde-fixed cells according to established protocols (Veith et al, 2008;Marijanovic et al, 2009;Goold et al, 2011). However, whilst >95% of chronically infected S7 cells (iS7) were PrP Sc -positive on SCA, the proportion of cells with aberrant PrP d deposits revealed by IF after guanidinium or formic acid treatment (Veith et al, 2008;Goold et al, 2011) did not exceed 20% in agreement with previous studies (Goold et al, 2011) (Supplementary Fig S4B). We reasoned that procedural differences between the two assay types might account for differences in the proportion of PrP d . To investigate whether heat-treatment of cells after transfer to Elispot plates during SCA (Klohn et al, 2003), a treatment known to cause membrane delipidation, may explain these inconsistencies, we treated fixed cells with delipidating solvents prior to immunolabelling with anti-PrP antibody ICSM18. Delipidation of fixed cells with acetone, a solvent that preferentially dissolves neutral lipids, such as triacylglycerols and cholesterol esters, but not with methanol a solvent that dissolves polar lipids, such as phospholipids and glycosphingolipids, quantitatively removed neutral lipids in cells, as evidenced by the loss of C1-BODIPY 500/510 fluorescence (Fig 5A). C1-BODIPY-500/510 is a fatty acid analogue, which is deposited in triacylglyceride-rich lipid droplets. Similar results were obtained with BODIPY-cholesterol ( Supplementary Fig S5). Strikingly, acetone pretreatment followed Table 1. Gene silencing of distinct gene candidates in revertants is associated with increased prion susceptibility. Revertant R7 cells transiently expressing shRNA against distinct gene candidates using the bicistronic vector pGIPZ were enriched for highly GFP-fluorescent cells and subsequently infected with a 2 × 10 À5 dilution of RML mouse prions. Rates of prion propagation were determined by SCA and normalised against cells transfected with non-silencing control vectors (NSC GIPZ). Relative rates of prion propagation expressed as fold change (FC) to controls (NSC) AE SD for at least three independent experiments are shown. The level of gene knockdown (% kd) was determined as described in Materials and Methods. The EMBO Journal A gene regulatory network associated with prion replication Masue M Marbiah et al by denaturation with guanidinium thiocyanate (GTC) revealed PrP d deposits at the basement membrane level of iS7 cells, a phenotype that was absent in uninfected cells (Fig 5B). Similar labelling patterns were shown with Fab fragments of ICSM18, thus excluding the possibility of PrP redistribution upon binding of a divalent antibody (Fig 5B). A colocalisation of PrP d with neural cell adhesion molecule (NCAM) confirmed the deposition of PrP d at the ECM (Fig 5C, Supplementary Video S1). In contrast to the detection of abundant PrP d deposits at the ECM following acetone and GTC treatment, PrP d was detected predominantly in endosomal and perinuclear areas following formic acid or methanol/GTC treatment ( Supplementary Fig S4A). Punctate PrP d deposits in iS7 cells at the ECM are visible upon treatment with acetone and GTC, but are absent in controls ( Supplementary Fig S4C). Abundant punctate PrP d -positive patches were also detected on membranes above ECM level in delipidated iS7 cells (Supplementary Fig S4D). The detection of PrP d deposits following acetone delipidation and GTC treatment is not restricted to N2a cells as shown for chronically infected prion-permissive CAD5 cells (Mahal et al, 2007) (Supplementary Fig S4E). In summary, our results suggest that the cryptic ICSM18 epitope in PrP d deposits at the ECM is masked by neutral lipids.

Distinct phenotypes of prion-modulatory proteins at the ECM of chronically infected cells
The detection of PrP d deposits at the ECM of chronically infected cells now enabled us to investigate the subcellular distribution of prion-modulatory proteins in relation to aberrant PrP d . Remarkably, cells expressing Fn1 at the ECM level were completely devoid of PrP d deposits (Fig 6A), implying that Fn1 expression is negatively correlated with PrP d deposition in susceptible iS7 cells. Similarly, Chga, which is poorly expressed in iS7 cells, does not colocalise with PrP ( Fig 6B). Lrrn4 was expressed in chronically infected cells, albeit with a low level of colocalisation with PrP at the plasma membrane ( Fig 6C). Of note, integrin a8 colocalised with aberrant PrP d deposits at the ECM (Fig 6D), but not at the plasma membrane ( Supplementary Fig S3A).

Disruption of integrin a8 signalling inhibits Fn1-mediated metalloproteinase activation and augments the rate of prion replication
The question remained how the expression and deposition of secreted proteins might inhibit prion replication and aggregate formation at the ECM. Of note, RA-mediated remodelling of the ECM mimicked a gain of susceptibility, suggesting that matrix homeostasis and prion replication may be affected by common signalling pathways. Cellular differentiation is associated with ECM remodelling, regulation of matrix metalloproteinases (MMPs), reorganisation of the actin cytoskeleton, and changes in cell shape. These changes affect integrin signalling and the integrin-mediated crosstalk with growth factors. In our study, gene silencing of Itga8 and Fn1 was associated with a gain of susceptibility (Table 1). To examine whether prion replication is affected by Fn1-mediated integrin signalling, we incubated revertants and susceptible cells with RGD, a peptide that blocks the interaction between Fn1 and integrins (Koivunen et al, 1995) (Fig 7A). Whilst two integrins of the b1 subunit family, integrin a5 and integrin a8, harbour an RGD domain (Margadant et al, 2011), only the latter is expressed in susceptible and revertant cells. Remarkably, incubation of R7 cells with RGD inhibited secretion of activated MMP2 and MMP9 ( Fig 7B) and significantly increased the susceptibility of R7, but not of S7 cells (Fig 7C). Furthermore, a loss of function of MMP2 and MMP9 significantly increased the number of PrP Sc -positive cells in R7 cells (Fig 7D), but not in S7 cells (Fig 7E). This argues that the integrin a8-dependent activation and secretion of MMP2/9 in revertant R7 cells are mediated by Fn1 and associated with an inhibitory environment for prion replication at the ECM.
Papss2 loss of function leads to undersulphation of heparan sulphate proteoglycans and augments prion susceptibility Papss2 (3 0 -phosphoadenosine-5 0 -phosphosulphate (PAPS) synthase 2), one of the principal enzymes required for the sulphation of extracellular matrix molecules (Wang et al, 2012), catalyses the synthesis of activated sulphate, PAPS, in cells. Papss2 is expressed in revertants, and loss of function is associated with increased susceptibility (Table 1, Supplementary Table S8). By using a sulphate-specific anti-heparan sulphate (HS) antibody (David et al, 1992), we show that loss of Papss2 function in prion-resistant revertants leads to undersulphation of heparan sulphate proteoglycans (HSPGs, Fig 8A). A similar effect was achieved by incubation of cells with sodium chlorate, an inhibitor of sulfurylase, required for the formation of PAPS ( Fig 8B). In agreement with loss of Papss2 function in chronically prion-infected cells (Supplementary Table S8), the number of PrP Sc -positive cells significantly increased at 3 mM chlorate ( Fig 8D). The dose-response curve is biphasic due to a loss of cell viability at concentrations higher than 3 mM chlorate. Treatment of chronically infected cells with 30 mM chlorate in a previous study led to an inhibition of PrP Sc accumulation (Ben Zaken et al, 2003). The discrepancy between this result and our study remain unexplained. To unambiguously examine whether the 10E4 epitope colocalises with PrP C , we covalently conjugated anti-HS (mouse IgM) and ICSM18 (mouse IgG1) with Alexa Fluor dyes ( Fig 8E). Our data suggest that the 10E4 epitope colocalises neither with PrP C in S7 and R2 cells, nor with PrP d deposits in chronically infected cells (Supplementary Table S11). In addition, a monoclonal antibody  Fig S6), was used to test whether PrP C colocalises with HSPGs. As shown in Fig 8F and Supplementary Table S11, no colocalisation was observed between JM403 epitope and PrP in uninfected and chronically infected cells. Heparan sulphate mimetics are potent inhibitors of prion propagation (Schonberger et al, 2003), and heparin was suggested to displace PrP C from lipid rafts (Taylor et al, 2009). We therefore examined whether Papss2 knockdown is associated with phenotypic changes in PrP C deposition in cells. Remarkably, Papss2 as well as Fn1 silencing markedly altered PrP C distribution at the ECM ( Fig 9A). Serial scans along the z-axis in knockdown cells showed a higher granularity and fluorescence intensity of PrP C at ECM, when compared to control (scrambled RNA) cells. In contrast, ectopic expression of Prnp (Prnp (pLNXC2)) led to increased fluorescence at plasma membrane, but not at the ECM level. To quantify these phenotypic alterations, we recorded serial z-stacks, determined the fluorescence intensity profiles of single cells, and computed the mean fluorescence intensities ( Supplementary Fig S7A-C). When plotted against the distance from substrate, the mean fluorescence intensities of PrP C in Papss2and Fn1-silenced cells increased by more than 2-fold ( Fig 9B). A shift of the maximal intensity to larger distances from substrate in Fn1-silenced cells indicates that the amount of PrP C deposited is increased. A significant increase in PrP C expression levels in Papss2and Fn1-silenced, but not in Itga8silenced cells was confirmed on Western Blot (Fig 9C and D).

Discussion
We here present the first evidence for a gene regulatory network associated with susceptibility to prion propagation and modulated by the differentiation state of cells. Our data suggest that prion conversion is controlled by expression of genes with a role in the homeostasis of the ECM, a compartment characterised by abundant deposition of aberrant PrP d in susceptible cells. Silencing of nine gene candidates expressed in prion-resistant revertants, Fn1, Itga8, Chga, Iqgap2, Il11ra1, Micalcl, Papss2, Galt, and Rgs4, significantly increased the rate of prion propagation. The RA-mediated downregulation of prion modifier genes led to a marked gain of susceptibility, suggesting that the rate of prion propagation is associated with transcriptional regulation of gene candidates. Loss of Papss2 function led to undersulphation of HSPGs, increased deposition of PrP C at the ECM, and a concomitant increase in prion conversion, indicating that HSPG sulphation is negatively correlated with susceptibility. Furthermore, inhibition of Fn1 binding to integrin a8 by the RGD peptide inhibited secretion of MMP2/9 and was associated with an increase in prion susceptibility. These data provide evidence that the ECM plays a critical role in the control of prion conversion. Prion diseases are a group of fatal infectious neuronal disorders that are associated with the conversion of host-encoded PrP to misfolded pathogenic conformers and neurotoxicity (Prusiner & DeArmond, 1987). We here present the first in vitro evidence of an aberrant deposition of PrP d at the ECM, a compartment that is deemed critical for prion replication (Caughey & Raymond, 1993b;Gabizon et al, 1993;Caughey et al, 1994). Delipidation with acetone and denaturation are critical to reveal PrP d deposits at ECM and membrane level of cells, suggesting that in their aggregated state, the epitope is masked by lipids.
The ECM not only provides structural support for force transmission and tissue structure maintenance, but also plays a critical role in the regulation of physiological processes such as differentiation, migration, and intercellular communication. Transmembrane receptor proteins of the integrin family mechanically couple the actin cytoskeleton to the ECM by binding to Fn1 and other adhesion molecules, such as collagen and vitronectin. A convincing body of evidence suggests that MMP activation and expression is triggered by interaction of the cell adhesion molecule Fn1 or its proteolytic fragments with integrins (Esparza et al, 1999;Schedin et al, 2000;Yan et al, 2000;Thant et al, 2001;Forsyth et al, 2002;Loeser et al, 2003;Jin et al, 2011). This interaction can be inhibited by the RGD peptide (Koivunen et al, 1995), a tripeptide domain located in the 10th type-III module of Fn1 and site of cell attachment via b1 integrins. Integrin a8 regulates cell adhesion and migration by binding to Fn1, vitronectin, and tenascin-C in an RGD-dependent manner (Muller et al, 1995;Schnapp et al, 1995;Denda et al, 1998;Benoit et al, 2009). Remarkably, disruption of Fn1 binding to integrin a8 with the RGD peptide increased susceptibility and inhibited secretion of activated MMP2/9. Whilst expression of nine prion modulators in revertants is negatively correlated with prion susceptibility, incubation of revertants with RA downregulated 18 gene candidates and led to a significant increase in prion propagation. In accord with this study, RA was shown, in human arterial smooth muscle cells, to inhibit the expression of Fn1 and MMPs, a phenotype that was associated with inhibition of migration and cell invasion (Axel et al, 2001;Scarpa et al, 2002).
A large body of evidence suggests that sulphated glycans and heparan sulphate mimetics act as potent inhibitors of prion propagation (Caughey & Race, 1992;Caughey et al, 1993aCaughey et al, , 1994Schonberger et al, 2003). Conversely, association of endogenous sulphated glycosaminoglycans (GAGs) with PrP d deposits in vivo was taken as evidence that they may facilitate PrP d formation (McBride et al, 1998). Inhibition of PAPS formation by two distinct approaches in this study, Papss2 silencing and inhibition of sulfurylase, was associated with undersulphation of HSPGs and increased prion susceptibility. Our study does not support a direct interaction of sulphated HS chains with PrP C in vitro, as shown with an anti-HS antibody, 10E4. In addition, no colocalisation between PrP C and HSPGs was observed with the anti-HS antibody JM403. Heparin, a highly A Susceptible S7 cells were labelled with 1 lM C1-BODIPY 500/510 (green label) for 12 h, fixed, and treated for 1 min with ice-cold acetone, methanol or PBS. Cells were counter-stained with ICSM18 (red label). B Chronically infected iS7 cells and uninfected S7 cells were fixed, delipidated with acetone or treated with PBS. All samples were denatured with 3 M GTC and washed at least five times with PBS before labelling with anti-PrP antibody ICSM18 or ICSM18 Fab fragment (Fab). C Infected iS7 and uninfected S7 cells were fixed, delipidated with acetone, and denatured with 3 M GTC prior to labelling with anti-NCAM and ICSM18 (anti-PrP). The degree of colocalisation, expressed as PCC values are 0.42 AE 0.08 for iS7 cells and 0.56 AE 0.08 for uninfected S7 cells.
Data information: Scale bar, 20 lm. Masue M Marbiah et al A gene regulatory network associated with prion replication The EMBO Journal sulphated glycosaminoglycan was shown to displace PrP C from rafts in a previous study and to promote its endocytosis (Taylor et al, 2009).
To address the question of how the sulphation status of HSPGs may affect prion propagation, we examined the distribution of PrP C in Papss2-silenced R7 cells. Significantly more PrP C was deposited in Papss2 knockdown cells compared to controls, a phenotype that was also observed upon Fn1 silencing. Notably, whilst ectopic expression of PrP C failed to increase the rate of prion replication in revertants (Supplementary Table S1), silencing of Papss2 and Fn1 led to increased ECM deposition of PrP C (Fig 9) with a concomitant increase in conversion rates (Table 1 and Supplementary Table S8). This implies that the subcellular deposition, rather than the relative levels of PrP C expression, correlates with the corresponding prion conversion rates and may be due to a decrease in ECM proteolysis upon Fn1 downregulation ( (Axel et al, 2001) and this study) and/or a disruption of FN matrix assembly by undersulphated HSPGs (Galante & Schwarzbauer, 2007). We therefore suggest that PrP C , deposited as a result of perturbed ECM homeostasis, may be a suitable substrate for prion conversion. A perturbation of matrix assembly has previously been reported in association with mutant PAPSS2. Chondrodysplasias, severe bone disorders, are associated with mutations in PAPSS2 and SLC26A2 and lead to reduced sulphate uptake, undersulphation of GAGs, and defective FN matrix assembly in cells (Ikeda et al, 2001;Galante & Schwarzbauer, 2007). Strikingly, a member of the Slc26a family of sulphate transporters, Slc26a4, is present in the gene signature of prion modifiers in our study, but its loss of function did not affect prion susceptibility, most likely due to the functional redundancy of this protein family, since Slc26a2, Slc26a4, Slc26a6, Slc26a8, and Slc26a11 are highly expressed in revertants. A loss of ECM proteins in gamma-aminobutyric acid (GABA)interneurons of Creutzfeldt-Jakob patients was suggested to precede extracellular prion deposition (Belichenko et al, 1999).
Other prion modifiers identified in this study are associated directly or indirectly to ECM homeostasis and remodelling. Chga, a member of the granin family of neuroendocrine secretory proteins, is located in secretory vesicles of neurons and neuroendocrine cells with a suggested role as a modulator of cell adhesion. Proteolytic processing of Chga to several bioactive peptides by plasmin (Metzboutigue et al, 1993;Colombo et al, 2002) and other proteases has been linked to modulation of cell adhesion (Metzboutigue et al, 1993) and negative regulation of angiogenesis (Belloni et al, 2007). Lrrn4, a transmembrane protein expressed in the hippocampus and cortex, contains leucine-rich repeat (LRR) motifs and fibronectin type-III-like repeats and is covalently linked to glycosaminoglycan side chains in its extracellular region (Bando et al, 2012). Lrrn4 was suggested to play an important role in hippocampus-dependent long-lasting memory (Bando et al, 2005). Il11ra1 is a type I cytokine receptor which contains two fibronectin type-III domains and one Ig-like C2-type (immunoglobulin-like) domain. Ablation of Il11ra1 is associated with defective decidualisation in the uterus of mice and leads to alterations in ECM components (White et al, 2004), but its role in ECM regulation is unknown.
In summary, we identified a gene regulatory network associated with prion replication and present evidence for the control of prion conversion at the ECM by an integrin-dependent activation of MMPs and the sulphation state of HSPGs. That genes involved in ECM homeostasis affect the kinetics of prion replication might help us to better understand the selective vulnerability of different neuronal populations during neurodegeneration (Guentchev et al, 1999;Siskova et al, 2013).

Cell culture
N2a-derived cell lines were maintained in OptiMEM containing 10% foetal calf serum and 1% penicillin/streptomycin (OFCS). Cad5 cells The EMBO Journal A gene regulatory network associated with prion replication Masue M Marbiah et al were maintained in OptiMEM, supplemented with 10% bovine growth serum (HyClone) and 1% penicillin/streptomycin.

Quantification of prion infection and rates of prion replication
Differences in the kinetics of cellular prion propagation were determined by quantifying the number of prion-infected cells during the course of three cell passages after infection using the Scrapie Cell Assay (SCA) (Klohn et al, 2003). The SCA is based on the microscopic detection of PK-resistant PrP (PrP Sc ) in prion-permissive cells in an automated manner using spot detection software (Imaging Associates, UK) and Zeiss KS Elispot system. Cells were infected using serially diluted brain homogenates with known titres and the number of PrP Sc -positive cells expressed as Tissue Culture Infectious Units (TCIU). Briefly, 1.8 × 10 4 cells were plated into wells of a 96-well plate. After 16 h, cells were incubated with 300 ll aliquots of serially diluted RML brain homogenate for 3 days. Cells were then split 1:8 for three subsequent passages, and aliquots of 2.5 × 10 4 cells transferred onto Elispot plates (MultiScreen HTS-IP Filter Plate, Millipore). If not stated otherwise, the number of PrP Scpositive cells was determined after digestion with 2.2 mU (0.5 lg) recombinant proteinase K (Roche Diagnostics) per millilitre of lysis buffer. The specific activity of PK (44 U/ml) was determined with the Chromozyme assay (Roche Diagnostics). Infectious titres were quantified from dilution series of RML brain homogenate I2424 (8.4 log LD50 units/g brain) and are expressed as tissue culture infectious units (TCIU) (Klohn et al, 2003). To determine the rate of prion replication, cells were infected with concentrated supernatants from chronically prion-infected cells. Due to the small size of these prion-infected nanoparticles, the inoculum does not have to be diluted out by serial cell passages as compared to the standard protocol (Klohn et al, 2003).

RNA isolation and quality control
RNA from 8 × 10 6 cells was isolated using RNeasy plus mini kit (Qiagen) with an average yield of 100 lg. To remove DNA contaminations, aliquots of RNA (10-30 lg) were incubated with 7 U RNase-free DNase I (Qiagen) in 50 ll RDD buffer at RT for 10 min. This step was repeated twice followed by heat inactivation of DNase at 70°C for 5 min. Subsequently, RNA was purified using RNeasy MinElute cleanup kit (Qiagen) according to the instructions of the manufacturer, and the RNA concentration was determined using a NanoDrop spectrophotometer (Labtech International). The purity of RNA samples was assessed using quantitative real-time PCR. Briefly, GAPDH expression levels of RNA isolates from cells were determined using one-step RT-PCR master mix (Applied Biosystems) with a rodent GAPDH probe (Applied Biosystems) in presence and absence of reverse transcriptase (RT). Differences between cycle threshold (Ct) values of purified RNA samples in presence and absence of RT were typically > 20, corresponding to a 40-fold difference between RNA over contaminating DNA concentration, confirming the quality of RNA isolates. Subsequently, cDNA was synthesised from 200 ng of total RNA from three biological replicates per cell line with cDNA archive kit (Applied Biosystems) according to the manufacturer's instructions.

Microarray analysis
Relative gene expression levels were determined using GeneChip Mouse Genome 430 2.0 arrays (Affymetrix). Hybridisation, scanning, and microarray analysis were conducted at the Wolfson Institute for Biomedical Research and the UCL Cancer Institute. Microarrays were scanned using Genechip scanner 3000 and images analysed using GCOS version 1.2. Signal intensities were determined using one-step Tukey's biweight estimate with consideration of mismatch values to account for stray signal. Internal array controls were cross-checked for quality control. Log intensity ratios (M) versus average log intensity (A) were determined using R and Bioconductor and plotted before and after robust multi-array average (RMA)-normalisation. Box plots of array distributions are shown in Supplementary Fig S8. Pairwise significance analysis was performed using t-test with P-value cut-off of 0.01. The raw microarray data are deposited at NCBI Gene Expression Omnibus (GEO, accession number GSE56275).

Quantitative real-time PCR
Relative gene expression levels were estimated by quantitative realtime PCR (qPCR) using 7500 Fast Real-Time PCR System (Applied Biosystems). Gene-specific dual-labelled probes (5 0 -FAM/3 0 -TAMRA) and primers were designed within the target sequence of the corresponding Affymetix probes using Primer Express software (Applied Biosystems) and synthesised by Eurofins MWG Operon (Ebersberg, Germany). Duplex PCR was carried out using TaqMan Gene Expression Master Mix (Applied Biosystems) in presence of VIC-conjugated mouse actin B (Actb) or Gapdh (Applied Biosystems). Cycling was at 50°C for 2 min, 95°C for 10 min, followed by 40-45 cycles of 95°C for 15 s, and 60°C for 1 min. Relative expression levels were calculated from serially diluted cDNA and were normalised to Actb or Gapdh.

Overexpression of PrP c
To overexpress PrP c in PK1 subclones, 1.2 × 10 5 cells were plated per well of a 6-well plate and transfected with 2 lg of Quantitative analysis of fluorescence intensities using Volocity. D Dose-response effects of sodium chlorate on the number of PrP Sc -positive cells in chronically infected cells (iS7) and cell viability, assessed by quantifying changes in cellular ATP levels using Ultra-Glo luciferase assay (Promega) in parallel experiments. Statistically significant differences (P < 0.001) between control and chlorate treatments are denoted (**). E, F No colocalisation between PrP and the 10E4 epitope (E) and between PrP and the JM403 epitope (F) was observed using covalently Alexa Fluor-conjugating antibodies ICSM18 and anti-HS in the cell types specified. 19-nt oligonucleotides were extended to 22nts using shRNA retriever. Constructs were amplified using Vent polymerase, the forward primer 5 0 -cagaaggctcgagaaggtatattgct gttgacagtgagcg-3 0 and the reverse primer 5 0 -ctaaagtagccccttgaattccg aggcagtaggca-3 0 , containing XhoI and EcoRI restriction sites, respectively, according to the specification of the manufacturer Thermo Scientific. All 22-nt sense shRNA sequences used for gene silencing of target genes in this study are listed in Supplementary Table S12.
To silence Prnp expression in N 2 a cells a 19-mer, 5 0 -taggagatcttgactctga-3 0 , targeting the 3 0 UTR of Prnp, was annealed and ligated into BglII and HinDIII sites of pRetroSuper vector and packaged into retroviral particles using Phoenix eco cells. Prion-susceptible PK1-10 cells were infected with viral supernatants in presence of 8 lg polybrene per ml OFCS, and antibiotic-resistant clones were selected with 4 lg puromycin/ml medium. Clone PK1-10/Si8kd (PK1 Prnp-RNAi) failed to propagate prions. Gene silencing by targeting of the 3 0 UTR region of Prnp enabled the reconstitution of PrP c expression with a construct harbouring the Prnp open-reading frame.

Generation of a Micalcl-YFP fusion protein
Micalcl (NM_027587) is poorly characterised, and Mammalian Gene Collection (MGC) clones are unavailable. We therefore assembled the full-length 2-kb coding sequence from synthetic gene fragments using gBlocks (Integrated DNA Technologies, IDT). Briefly, five gBlocks with a size ranging from 300 to 500 bp with flanking 20 bp complementary overhangs were designed and synthesised by IDT. Two and three gBlocks were assembled, respectively, using Gibson assembly master mix (New England Biolabs) according to the specification of the manufacturer. The product was amplified with forward primer 5 0 -atgaaccaaagagcaccatcgcctccaaagg-3 0 , reverse primer 5 0 -tcaagtcctgctgagctgacagcctctgg-3 0 and AccuPrime Taq DNA polymerase high fidelity (Life Technologies). Micalcl was then cloned into the Gateway pCR8/GW/TOPO vector (Life Technologies) and transformed into TOP10 chemically competent E. coli and grown at 37°C overnight on agar plates, containing 100 lg spectinomycin per ml agar. Five colonies were picked and expanded, and plasmid DNA was isolated and purified using Qiagen spin miniprep kit. Plasmid-containing colonies were identified by EcoRI restriction digestion and sequenced to confirm the correct orientation of the insert. Micalcl was then transferred into a Gateway vector, pEYFP_C1 DEST, kindly provided by Prof. Stefan Wiemann (DKFZ, Germany) by recombinational cloning using LR Clonase II (Life Technologies).

Gene silencing in cells and FACS enrichment
For gene silencing experiments, 1.1 × 10 6 R7 cells were plated into 6-cm dishes and grown for 16 h. Five lg of pGIPZ vector, harbouring shRNA of the gene target, and 5 ll PLUS (Life Technologies) were added to 500 ll FBS-free OptiMEM and incubated for 5 min at RT. Seven microlitres of LTX (Life Technologies) were added, and the DNA-lipid mix incubated for 20 min. Cell medium was then replaced with DNA-lipid complex in a total of 3 ml OFCS. After 24 h, fluorescent cells were resuspended and enriched using a MoFlo (Beckman Coulter) according to the instructions of the manufacturer. Briefly, shRNA-expressing cells were enriched from a heterogeneous pool of fluorescent cells by gating GFP-positive cells in the 4th decade of the logarithmic fluorescence scale (see Fig 3D). FACS-enriched cells were counted using a Coulter Counter Z2 (Beckman Coulter) at an upper threshold of 36 and a lower threshold of 12 and plated into wells of a 96-well plate at 1.8 × 10 4 cells per well. After 16 h, cells were infected with 2 × 10 À5 RML and processed according to the SCA protocol above.
Transient knockdown of gene candidates in cells without enrichment was conducted with double-stranded RNA dicer substrates (siRNAs) obtained from IDT. Briefly, 2 nmoles of siRNA was reconstituted in 200 ll duplex resuspension buffer (IDT). RNA-lipid complex formation was performed by adding 4 ll siRNA and 5.3 ll DharmaFect 3 (Thermo Scientific) to 100 ll FBS-free OptiMEM. After 20 min, RNA-lipid complexes were added to a total volume of 2 ml OFCS and combined with 2 ml of cell suspension at a concentration of 1 × 10 5 cells per ml. Aliquots of 300 ll were then transferred out into 12 wells of a 96-well plate. After 2 days, cells were infected with RML and split the following day at a split ratio of 1:8. Cells were then processed according to the SCA protocol above.

Detection of PrP d deposits
Aliquots of 5 × 10 4 chronically infected or uninfected N 2 a cells were plated into wells of 8-well chamber slides (Thermo Scientific) and cultured for three to four days. Cells were fixed with 4% formaldehyde/PBS for 12 min and washed once with PBS. Prolonged fixation greatly impedes PrP d detection and has to be optimised for each cell line. To remove lipids, cells were incubated for 30 s with chilled

A
Gene expression of Papss2 and Fn1 was silenced in R7 cells prior to immunolabelling with ICSM18. In addition, Prnp was overexpressed using pLNCX2 vector. Representative images of serial confocal z-stacks, acquired at identical laser setting, are shown. B Mean fluorescence intensities of Papss2-and Fn1-silenced cells labelled with ICSM18 and anti-mouse Alexa Fluor 488 antibodies were acquired according to the procedure described in Supplementary Fig S7. Mean fluorescence intensities of at least 40 intensity profiles AE standard error of the means, as well as upper and lower limits of 95% confidence intervals are shown. C, D PrP C protein expression levels in R7 cells at 3 days after transfection with siRNA against Papss2, Fn1, Itga8, and scrambled RNA control are shown (C) and quantified for three biological repeats with b-actin as an endogenous control (D; anti-actin, clone ACTN05 (C4), Abcam). Significant changes (P < 0.05) between gene silencing of candidates and scrambled RNA control are shown (*). Masue M Marbiah et al A gene regulatory network associated with prion replication The EMBO Journal acetone, methanol, or PBS and washed with PBS. Subsequently, cells were incubated with 3M guanidinium thiocyanate (Sigma) for 10 min and washed at least five times with PBS. Cells were then incubated with primary antibody in a 1:4 dilution of Superblock (Pierce) solution/PBS (v/v) for 1 h at RT or overnight at 4°C. Cells were rinsed with PBS twice and incubated with a 1:10,000 dilution of 4 0 ,6-diamidino-2-phenylindole dihydrochloride (DAPI, 2 mg/ml DMSO) and Alexa Fluor-conjugated secondary antibodies (Life Technologies) at a dilution of 1:500 to 1:2,000 for 1 h. After two washes with PBS, cells were stored at 4°C until further processing. To exclude cross-reactivity of secondary antibodies during co-labelling experiments, batches of highly cross-adsorbed secondary antibodies were routinely tested using antibodies against distinct targets, that is, a rabbit-anti-EEA1 antibody (CST, #3288) or a rat anti-Lamp antibody (Santa Cruz, Cat# sc19992) and mouse anti-PrP antibody (ICSM18), followed by incubation with secondary antibodies. Immunofluorescence was analysed with a Zeiss LSM 710 confocal microscope and Zen imaging software (Carl Zeiss).

Determination of PrP surface expression levels
Relative PrP surface expression levels of PK1 clones were determined by FACS. Briefly, 1 × 10 6 cells were pelleted at 300 × g for 4 min and washed with PBS. Cell pellets were then fixed on ice with 4% paraformaldehyde/PBS for 30 min. After washing with PBS, cells were incubated with 5 lg of ICSM 18 in PBS/0.1% bovine serum albumin (BSA) for 30 min. Cells were washed again with PBS/0.1% BSA, spun, and incubated with a 1:200 dilution of Alexa Fluor 488-goat anti-mouse IgG (H+L) antibody (Invitrogen, Paisley, UK) in PBS/0.1% BSA for 30 min. After washing, PrP surface expression levels were determined using a FACS Calibur flow cytometer (BD Biosciences). Background levels of fluorescence were determined by labelling cells with secondary antibody only.

Determination of cell population doubling times
Effects of RA treatment on cell doubling times of revertants and susceptible cells were determined using an automated cell counter, Z2 coulter counter (Beckman Coulter). In preliminary experiments, cell counting was compared to photometric determination with dyes WS1 and MTT. Owing to a high dynamic range of up to 3 orders of magnitude, cell doubling time was determined by automated cell counting. Briefly, cells were plated into 96-well plates at a concentration of 1.8 × 10 4 cells per well and 16 h later incubated with various RA concentrations. Cells were harvested at 12, 24, 48, and 72 h after RA incubation by resuspending cell layers with multichannel pipettes and combined aliquots of 3 to 4 wells were counted.

Analysis of confocal images
To collect confocal images of proteins expressed at the ECM, serial scans along the z-axis were conducted. A marked increase in fluorescence intensities at the level of the substrate delimits the ECM as depicted in Fig 9. To quantify expression levels of candidate proteins, confocal images were collected at a 630-fold magnification (1.4 oil, Plan-Apochromat) and analysed using Volocity (Volocity, version 6.1.1). Where proteins were expressed at the ECM with no cell boundaries, data were analysed as follows. Nuclei were identified by DAPI in channel Ch-TS1 with an average size of 25 lm 3 . A dilation of 10 iterations from the nucleus was taken to represent the cell soma. Intensity values of candidate proteins, labelled with Alexa Fluor 488-conjugated secondary antibodies, were detected in channel ChS1-T2 on a single cell level and average intensities calculated with a cut-off of 5,000 units. The degree of colocalisation was assessed using Volocity by computing background-corrected threshold PCC values.

PrP C expression profiling at ECM
To assess the relative levels of PrP C expressed at the ECM, cells were labelled with anti-PrP antibody ICSM18 and Alexa Fluor 488conjugated secondary antibody and profiles of single cells from sequential z-stacks were analysed by Zen 2011 software (Zeiss, Cambridge, UK). Briefly, serial z-stacks of 0.2 lm were acquired for silenced and control cells at identical confocal settings. For image processing, fluorescence intensity profiles of single cells were acquired using Zeiss Zen software and incremental mean fluorescence intensities of sequential focal planes computed.

MMP zymography
To determine activities of MMP2 and MMP9, 3 × 10 6 cells were plated into 15-cm dishes in OFCS. After 16 h, cells were incubated with 1 lM RGD peptide (Santa Cruz Biotechnology) or vehicle (DMSO). After 72 h, supernatants were collected and cleared of cells and debris at 500 g for 10 min and 5,500 g for 20 min, respectively, using an Allegra 25R centrifuge (Beckman Coulter). Subsequently, supernatants were concentrated at 5,500 g on VivaSpin 20 columns (GE Healthcare Life Sciences) for 30 min. Concentrated samples were diluted 1:1 in Tris-glycine SDS gel loading buffer (Novex, Invitrogen) and separated by electrophoresis on 10% Tris-glycine gels containing 0.1% gelatine (Novex, Invitrogen) at 125 Volts for 90 min. Gels were incubated in 1× Zymogram renaturation and developing buffer (Novex, Invitrogen) according to the manufacturer's specification. Subsequently, gels were stained with SimplyBlue SafeStain (Novex, Invitrogen) for 6-12 h with 3-5 changes until protein bands were clearly visible.

Statistical analysis
Statistical significance of differential gene expression was computed by non-parametric statistics using 'Significance Analysis of Microarrays' (SAM) (Tusher et al, 2001), and raw values corrected for multiple testing and expressed as false discovery rate (FDR) values (Benjamini & Hochberg, 1995). All other data are expressed as mean AE standard deviation (SD), unless otherwise stated. Comparisons of mean values were conducted by Student's t-test.
Supplementary information for this article is available online: http://emboj.embopress.org