Transcription factor network analysis identifies REST/NRSF as an intrinsic regulator of CNS regeneration

The inability of neurons to regenerate long axons within the CNS is a major impediment to improving outcome after spinal cord injury, stroke, and other CNS insults. Recent advances have uncovered an intrinsic program that involves coordinate regulation by multiple transcription factors that can be manipulated to enhance growth in the peripheral nervous system. Here, we used a system-genomics approach to characterize regulatory relationships of regeneration-associated transcription factors, identifying RE1-Silencing Transcription Factor (REST; Neuron-Restrictive Silencer Factor, NRSF) as a predicted upstream suppressor of a pro-regenerative gene program associated with axon regeneration in the CNS. We validate our predictions using multiple paradigms, showing that mature mice bearing cell type-specific deletions of REST or expressing dominant-negative mutant REST showed improved regeneration of the corticospinal tract and optic nerve, accompanied by upregulation of regeneration-associated genes in cortical motor neurons and retinal ganglion cells, respectively. These analyses identify a novel role for REST as an upstream suppressor of the intrinsic regenerative program in the CNS and demonstrate the power of a systems biology approach involving integrative genomics and bio-informatics to predict key regulators of CNS repair.


SUMMARY 28
The inability of neurons to regenerate long axons within the CNS is a major impediment 29 to improving outcome after spinal cord injury, stroke, and other CNS insults. Recent 30 advances have uncovered an intrinsic program that involves coordinate regulation by 31 multiple transcription factors that can be manipulated to enhance growth in the peripheral 32 nervous system. Here, we used a system-genomics approach to characterize regulatory 33 relationships of regeneration-associated transcription factors, identifying RE1-Silencing Injured axons in the adult mammalian central nervous system (CNS) generally cannot 46 regenerate over long distances, limiting functional recovery from CNS injury (1). Potential 47 12 neuronal projection, metabolism, or synaptic transmission ( Figure 4F). These analyses 302 support a model whereby inhibition of REST activates a core molecular program driven 303 by a tightly controlled TF network similar to the one activated during peripheral nerve 304 regeneration, along with other complementary pathways, to enable subsequent 305 regenerative processes ( Figure 4G). 306 307 REST is a transcriptional repressor negatively correlated with the regeneration 308

state of retinal ganglion cells 309
To assess the potential generalizability of the bio-informatic predictions derived 310 from spinal cord and peripheral nerve injury above, we extended the same TF regulatory 311 network analysis to another CNS neuronal population, injured retinal ganglion cells 312 the regenerative gene program (20) and, most effectively, by combining two or more of 322 these treatments (17,20,94,95). 323

From our initial bio-informatic predictions comparing PNS and CNS injured tissues, 324
we hypothesized that the disrupted TF network in the injured, non-growing RGCs, similar 325 to the CNS-injured spinal cord tissues ( Figure 2D), would re-gain substantial connectivity 326 in RGCs treated so as to be in a more regenerative state. Using mice that express cyan-327 fluorescent protein (CFP) in RGCs (96), we induced robust axon regeneration by 328 combining a strong genetic pro-regenerative manipulation, RGC-selective PTEN knock-329 down (AAV2-shPten.mCherry; Methods; (14, 97)), with intraocular injection of the 330 neutrophil-derived growth factor oncomodulin (Ocm: (86, 87) and the non-hydrolyzable, 331 membrane-permeable cAMP analog CPT-cAMP (a co-factor of Ocm) immediately after 332 13 nerve injury. This combination provides one of the strongest regenerative responses 333 described to date ( Figure 5A), while avoiding complications that might be introduced by 334 inducing intraocular inflammation (15,16,87). Controls received an intraocular injection 335 of AAV2 expressing shLuciferase.mCherry 2 weeks before surgery and saline 336 immediately afterwards. These mice did not exhibit axon regeneration ( Figure 5A-B; see 337 Methods). We dissected retinas and FACS-sorted RGCs from non-regenerating, control 338 treatment, or from RGCs exposed to the pro-regenerative combinatorial treatment 1, 3 or 339 5 days after optic nerve crush injury, followed by transcriptomic analysis via RNA-seq in 340 8-10 biological replicates for each condition ( Figure 5C; Methods). 341 To quantitatively determine a TFs' association with RGCs' regeneration state, we 342 first performed gene set enrichment analysis (GSEA) to compare a gene expression 343 signature correlated with the RGC axon regenerative state against 'tag gene sets' with 344 known binding sites for TFs (98). GSEA returns an enrichment score (ES) of this 345 comparison to determine whether the gene set represented by regeneration-associated 346 genes is enriched in targets of any TFs and if it is a positive or negative regulator of the 347 genes associated with regeneration phenotype ( Figure 5D; (99). Among the ~1000 TF-348 target gene sets unbiasedly tested, REST is ranked as the top negative regulator of the 349 RGC regeneration state-associated gene set at day 1 following injury, which is attenuated 350 on days 3 and 5 after injury ( Figure 5D), consistent with REST being an early, upstream 351 event in the regulatory cascade. 352 We next performed a complementary analysis using the same ARACNe-based 353 pipeline as used in our initial analysis of published PNS and CNS microarray datasets to 354 construct a data-driven, unsupervised, hierarchical network of the regenerative TFs within 355 this new RNA-seq dataset. Similar to CNS injured tissues in the first analysis ( Figure 2D), 356 non-regenerative RGCs with control treatment adopt a simpler, less inter-connected, and 357 less structured TF network. This unsupervised analysis showed again that REST appears 358 at the top-layer of the non-regenerating network ( Figure 5E, Control), and is negatively 359 correlated with other lower-layer TFs ( Figure 5F, Control). By contrast, pro-regenerative 360 treatments re-established a more complex, multi-layered network with higher connectivity 361 ( Figure 5E, global clustering coefficient in Control = 0.25, versus the pro-regenerative 362 treatment = 0.54), in which REST is dissociated and the key regenerative TFs (ATF3, Jun, 363 Sox11, Stat3) are more connected ( Figure 5F), similar to the microarray data from PNS 364 ( Figure 2). Other commonly used statistics for network connectivity such as local 365 clustering coefficient, betweenness centrality, and in-and out-degree (Methods), further 366 revealed significantly higher connectivity for the RAG TFs in the regenerating versus non-367 regenerating group ( Figure S5A). These results from independent datasets and different 368 tissues further support our original bio-informatic predictions that neurons displaying 369 regenerative potential are associated with a highly inter-connected, structured TF-370 regulatory network. Further, these analyses (e.g., Figure 2 and 5) show that REST 371 appears as an inhibitory TF at the apex of a dismantled TF network in the non-372 regenerating CNS neurons, but is not associated with the highly interacting TF network 373 present in neurons in a regenerating state. 374 These multiple analyses of independent data suggested that REST is an upstream 375 transcriptional repressor potentially limiting the interactions between lower-level TFs and 376 the expression of regeneration-associated genes. One prediction of this model is that 377 REST target genes should be enriched in RAGs and RAG-associated processes, parallel 378 with GSEA ( Figure 5F). We observed 630 transcriptional interactions with REST predicted 379 by ARACNe, including 339 positively regulated (activated) genes and 321 negatively 380 regulated (repressed) genes ( Figure S5B, Supplemental Table 4; Methods). Enriched GO 381 terms for genes predicted to be activated by REST include metabolic processes, 382 response to endoplasmic reticulum (ER) stress, and RNA binding and transport ( Figure  383 S5C), whereas genes predicted to be repressed by REST are indeed implicated in 384 processes or pathways associated with axon regeneration (18), including calcium ion 385 transport, axon guidance, synaptogenesis, CREB-and cAMP-mediated signaling ( Figure  386 S5C). The REST-repressed, regeneration-associated gene set was enriched with down-387 regulated genes at early stages (day 1), which were up-regulated in the later stages of 388 regeneration (day 3 and 5) ( Figure 5G, GSEA), suggesting a release of the transcriptional 389 brake by REST on these genes. Altogether, two independent analyses of data from 390 different sources that were focused on identifying key upstream TFs regulating CNS 391 regeneration using unsupervised methods revealed REST to be a key transcriptional 392 upstream repressor of a RAG program, suggesting that it would be a potential novel 393 suppressor of regeneration. Conversely, since REST is a repressor of a pro-regenerative 394 program, these analyses predict that counteracting REST would enhance regeneration 395 after injury. To formally test this model, we next performed several experiments both in 396 vitro and in vivo, using dorsal root ganglia (DRGs) cultured on a growth-suppressive 397 substrate to model CNS-injured environment and two different in vivo models of CNS 398 injury -complete spinal cord injury (SCI) and optic nerve crush. 399 400

REST deletion facilitates, and over-expression inhibits, neurite growth in vitro 401
We first tested the consequences of gain-and loss-of function of REST in 402 dissociated adult dorsal root ganglion (DRG) neurons in vitro. We hypothesized that if 403 REST were indeed inhibitory, its depletion should be permissive, whereas its over-404 expression would inhibit the normal ability of PNS neurons to extend processes. REST  Notably, REST deletion did not affect neurite extension of DRG neurons when cultured 423 on laminin ( Figure 6A-B, laminin group), suggesting that REST-mediated inhibition of 424 growth processes may be activated by a growth-suppressive environment that mimics the 425 injured CNS, such as the presence of CSPG, and is not present in the presence of 426 permissive substrates that support peripheral axonal growth. 427 We further hypothesized that REST over-expression might inhibit the ability of 428 DRG neurons to extend processes following a PNS injury. To test this hypothesis, we 429 over-expressed REST in cultured DRG neurons for seven days using lentiviral constructs, 430 followed by re-plating, a process to remove existing DRG neurites in vitro. This model 431 recapitulates many biochemical and morphological features of an in vivo pre-conditioning 432 peripheral nerve injury (Methods) (104-106). The efficiency of REST over-expression was 433 confirmed by qPCR ( Figure S6D). We observed that increasing REST construct 434 concentration dose-dependently inhibited neurite extension, particularly at the highest 435 concentration ( Figure 6D). To distinguish between these potential mechanisms, we first examined CST axons 449 3 days post-injury. Apparent dieback and large numbers of retraction bulbs were 450 observed at this early time point in both control and REST-deleted axons ( Figure S7A). 451 We then measured branching of CST axons at 4 weeks post injury which, when increased, 452 is considered to be strong evidence of regenerative growth (68, 110) ( Figure 7D; To validate these observations in vivo, we used two independent methods to 476 counteract REST ( Figure S8A). In the first of these approaches, we examined whether 477 AAV2-d/n REST was sufficient to induce optic nerve regeneration and/or promote RGC 478 survival. Two weeks after optic nerve injury, expression of d/n REST was sufficient to 479 stimulate 43% of the level of axon regeneration ( Figure 8D, E) that was achieved with the 480 powerful combinatorial treatment (pten deletion, rOcm, CPT-cAMP) subsequently used 481 to generate the transcriptome dataset (c.f. Figure 5A, B). In addition, d/n REST 482 expression more than doubled RGC survival at two weeks post-optic nerve injury 483 (compared to mice injected with AAV2-GFP: Figure 8F), an effect that fully recapitulated 484 the strong neuroprotection afforded by the combination of pten deletion, rOcm, and CPT-485 cAMP ( Figure 5B). In parallel to our cell culture studies ( Figure 8A-C), we also examined 486 the effect of combining d/n REST expression with Ocm plus cAMP in vivo. Whereas a 487 single injection of rOcm + cAMP alone induced little regeneration and no increase in RGC 488 survival relative to untreated controls, combining rOcm + cAMP with the expression of d/n 489 REST increased axon regeneration 55% above the level achieved with d/n REST 490 expression alone ( Figure 8D, E). RGC survival was elevated to the same extent as with 491 d/n REST expression alone ( Figure 8F). 492 As an alternative approach ( Figure S8A in these studies based on the lack of significant differences in outcomes among controls 505 for AAV2-Cre plus REST fl/fl (strain C57/B6, Mean ± SEM: 71.07 ± 14.65) and for AAV2-506 d/nREST injections in wild-type 129S1 mice (Mean ± SEM: 41.57 ± 13.65: P = 0.09; see 507 legend for Figure 8). In addition, as observed with d/n REST expression, deletion of REST 508 in RGCs doubled the level of RGC survival above that seen in control retinas two weeks 509 after optic nerve injury ( Figure 8F), an effect similar to that achieved with the combinatorial 510 treatment used to generate the transcriptional dataset. 511 Deletion of pten is perhaps the most effective single treatment described to date 512 for inducing optic nerve regeneration (14, 17). On average, counteracting REST captured 513 ~ 2/3 of the effect of pten deletion on axon regeneration ( Figure 8E) and the full effect of 514 pten deletion on RGC survival ( Figure 8F). Thus, REST can be considered a major 515 suppressor of RGC survival and optic nerve regeneration in mature mice. We also 516 investigated whether pten deletion would occlude the effects of counteracting REST, 517 which would suggest that the two share common effector pathways, or whether they might 518 show some degree of additivity. Our results point to partially additive effects on axon 519 regeneration ( Figure 8E), suggesting at least some independence of effector pathways. 520 Accompanying its effects on RGC survival and axon regeneration, expression of 521 d/n REST increased expression of several regenerative TFs (ATF3, SOX11, pSTAT3, 522 pCREB) in the TF regulatory network in RGCs, as assessed by immunostaining retinal 523 sections 1 day after optic nerve injury ( Figure 8G We used a stepwise, systems genomics approach to identify upstream 537 transcriptional regulators of intrinsic regeneration-associated gene expression programs 538 in the nervous system. Multiple independent bio-informatic analyses were used to 539 evaluate existing and newly produced gene expression datasets, all of which converged 540 on the transcriptional repressor, REST, as a potential upstream negative regulator of a 541 regenerative gene expression program in the CNS ( Figure 1A). We then experimentally 542 demonstrated that disrupting REST activates a core molecular program driven by a tightly 543 controlled TF network similar to the one activated during peripheral regeneration ( Figure  544 1B). This would also predict that counteracting REST would substantially improve 545 regeneration, which was supported in two well established models of CNS injury, the optic 546 nerve and the corticospinal tract (CST) ( Figure 1C). These data are consistent with a 547 model whereby REST may act by suppressing the interaction and the expression of pro-548 regenerative TFs within the RAG network, consistent with its known function as a 549 transcriptional suppressor. Perhaps most importantly, these results firmly demonstrate for 550 the first time that REST represses CNS regeneration in vivo, and conversely that its TFs are the key factors upstream of regeneration. TF binding is a dynamic process, and 559 a TF can be present or absent from its target loci at different time points and/or under 560 different conditions. In addition, TFs act in a combinatorial manner, forming tiered 561 regulatory networks to drive gene expressions. Therefore, experiments like gain-or loss-562 of-function of a single or a few TFs at one time is unlikely to recapitulate these TF 563 regulatory events. Here, we used an unsupervised, step-wise bio-informatic approach to 564 characterize the regulatory network structure of regeneration-associated TFs ( Figure 2A). 565 By leveraging existing and new gene expression datasets generated in multiple labs and 566 in PNS and CNS injury models at different timescales, we identified a core set of five TFs 567 (Jun, SMAD1, Sox11, STAT3 and ATF3) that occupied a standard, three-tiered core 568 regulatory network (30, 33, 34) that was conserved across all PNS datasets ( Figure 2C). 569 Each of these core pro-regenerative TFs is increased early after PNS injury ( Figure S1A), 570 in agreement with previous findings of their essential role during PNS regeneration (20, 571 57-66) and each connection of TF pairs is experimentally supported (55, 56), adding 572 confidence to our bio-informatic predictions. 573 By contrast, in the non-regenerating CNS (spinal cord and optic nerve), this 574 network loses its three-tiered structure, and instead adopts a simpler, less inter-575 connected, dismantled structure (spinal cord: Figure 2D; optic nerve: Figure 5E-F). 576 Remarkably, CNS neurons with enhanced regenerative capacity induced by combined 577 genetic and molecular manipulations re-gain the complex, multi-layer TF network with 578 higher inter-connectivity ( Figure 5E-F), similar to the TF network induced in the 579 regenerating PNS ( Figure 2C). In the dismantled CNS network, REST appears as a top-580 tier regulator, predicted to inhibit other lower-level TFs. The prediction of REST being a 581 transcriptional repressor was further supported by an independent, unbiased TF-582 screening approach that evaluated ~1000 TFs and their experimentally-proven target 583 genes, identifying REST as a top negative regulator of the gene set activated in 584 regenerating CNS neurons ( Figure 5D). 585 Independent analyses of data from different sources that were focused on 586 identifying key upstream TFs regulating CNS regeneration all pointed to REST as a key 587 transcriptional repressor upstream of the core pro-regenerative TFs driving RAG program 588 expression. This prediction was supported by the findings that Rest was specifically 589 upregulated across multiple CNS injury datasets ( Figure  animals with REST deletion showed substantially increased growth relative to wild-type 638 controls. We note that although these axons did not grow across an anatomically 639 complete SCI lesion ( Figure 7B-C), inability to cross the lesion boundary after complete 640 SCI was expected, as such growth is known to require both intrinsic growth cues and 641 external growth facilitators such as tissue or biomaterial bridges that provide growth-642 supportive molecules within the lesion site (8, 135-138). As a therapeutic strategy for 643 regenerating axons across a complete SCI, it will probably be necessary to augment 644 intrinsic growth capabilities such as REST or PTEN deletion, which activate regeneration- REST in RGCs led to considerably greater regeneration than either one alone. In addition, 665 expression of d/n REST or REST knock-down was sufficient to double levels of RGC 666 survival, affording the same level of neuroprotection as either combinatorial therapy or 667 PTEN deletion alone, which is notable, since to date, few factors other than PTEN deletion 668 enhance both RGC regeneration and survival. For example, ATF3 is pro-survival but has 669 no effect on RGC regeneration (141); Sox11 is pro-regenerative, but when overexpressed, 670 24 lead to the death of alpha RGCs (25); and STAT3 is pro-regenerative, but does not 671 increase survival (89). 672

Limitations and future directions 673
Here we demonstrate via several lines of experimental evidence that REST is an inhibitor 674 of CNS axon regeneration. Based on multiple forms of bio-informatic and experimental 675 analyses, we present a model whereby REST acts via repression of pro-regenerative 676 genes, whose regulatory elements it binds. Although we know that REST does repress 677 this regenerative program, and its reduction leads to regeneration, we cannot yet say with 678 certainty that its effects on regeneration are solely via this pathway. Thus, we view this 679 as a working model that warrants further testing. We also note that the genetic 680 manipulations required for direct testing of this model (e.g. simultaneous suppression of 681 multiple core regeneration-associated TFs in the context of REST deletion) are at the very 682 least daunting and at the limit of current experimental tractability. It is also plausible that 683 transcriptional regulation by REST is one of several mechanisms by which its deletion 684 promotes cell-intrinsic growth. From this perspective, it is likely that other key regulators 685 act synergistically with REST to control CNS regeneration. One potential REST-686 interacting factor could be PTEN, inhibition of which, plus Oncomodulin and cAMP 687 elevation up-regulates a regeneration-associated gene set that is predicted to be 688  Further studies will also be required to clarify the precise molecular mechanisms 702 by which REST acts on the core TF network in the RAG complex to regulate regeneration-703 associated pathways during CNS repair, and to explore other possible mechanisms. 704 REST may be recruited directly to the regulatory sites for repressing regeneration-705 associated transcription following CNS injury. ChIP-seq studies have shown that REST 706 can directly bind to regenerative TFs such as Sox11, KLF6, Jun and STAT3 (79, 81, 145). 707 Whether REST binds and represses additional regenerative factors in the context of 708 axonal injury needs to be further investigated. It is also possible that REST deploys 709 additional mechanisms of regulating CNS regeneration in addition to acting directly on 710 the core TFs. As a transcriptional regulator, REST can induce chromatin remodeling (46, 711 47), a process that rearranges the chromatin to facilitate or prevent gene transcription. 712 Overall, future studies on a genome-wide profiling of REST occupancy induced by CNS 713 injury or chromatin regulatory changes with and without REST inhibition in CNS neurons 714 will be necessary to identify how REST regulates regeneration-associated transcription 715 to enhance CNS repair. In addition, the mechanisms by which REST itself is regulated in 716 the context of CNS injury is unclear. Others have shown that REST can be regulated  The number of GAP43-or Synaptophysin-expressing axons co-labeled with BDA were 783 counted at 0.5 mm and 3 mm rostral to the SCI crush, and are expressed as percent of 784 BDA labeled axons at respective distances. We examined BDA labeling 3 mm caudal to 785 the lesion center to make sure the SCI lesions were complete. All axon counts were 786 carried out by an investigator blind to the identity of the cases. In some studies, 129S1 mice received an intraocular injection of AAV2-d/nREST or an 802 AAV2 control virus two weeks before the optic nerve crush and were euthanized at day 1 803 or day 7 after nerve injury. Retinas from these mice were prepared for immunostaining of 804 serial sections (details in Methods: Immunostaining of retinal sections and intensity 805 quantitation). 806 To investigate the transcriptome of RGCs during optic nerve regeneration or after 807 counteracting REST, we carried out optic nerve crush surgery with different intraocular 808 treatments in vivo, then used FACS to isolate RGCs for subsequent analyses (details in 809 with 4% paraformaldehyde and blocked for one hour at room temperature in PBS with 875 0.05% Tween-20 + 0.01% Triton-X + 1% BSA + 5% goat serum, followed by primary 876 antibody incubation with ß-III-tubulin (Biolegend, 1:500) overnight at 4 °C in blocking 877 solutions and secondary antibody (Invitrogen, 1:500) for 1-2 hr at room temperature. For 878 quantification of DRG neurites, at least 9 images were randomly taken from each replicate 879 using a Zeiss Confocal Microscope at 20x. Neurites were counted using Imaris Surface 880

Quantitation of optic nerve regeneration and RGC survival.
Rendering function, and the average neurite surface per neuron was quantified. proteins were transferred to PVDF membranes that were incubated with antibodies to 899 REST (Abcam, 1:1000), using anti-b-actin as a loading control. Quantitation of western 900 blot results was carried out with ImageJ software. 901

FACS isolation of adult cortical motor neurons. Surgeries and AAV injections were 902
carried out in the same way as described in the Methods section "Spinal cord injury and 903 corticospinal tract (CST) injections". In order to induce neuron-specific REST depletion, 904 we used AAVs expressing GFP or Cre recombinase under the human synapsin promoter. 905 Adult mouse brain tissue was dissociated as previously described (151). Briefly, 906 sensorimotor cortex injected with AAV-Syn-GFP or AAV-Syn-CRE to induce tdTomato 907 expression from REST flx/flx ; tdTomato mice was immediately dissected into ice-cold 908 Step 3: To define the hierarchical structure of the directed TF network, we used a 965 graph-theoretical algorithm called vertex-sort (33), which identifies strongly connected 966 components and applies the leaf removal algorithm on the graph and on its transpose 967 which can identify the precise topological ordering of members in any directed network 968 based on the number of connections that start from or end at each TF, indicating whether 969 a TF is more regulating or more regulated. This allows for an approximate stratification of 970 TFs within each dataset. Edges and nodes in the network were visualized by igraph R 971 package (https://igraph.org/r/). Centrality statistics of each TF node was calculated using 972 qgraph R package centrality_auto () function.  Table S1. Differentially expressed genes (DEGs) comparing wild-type and REST 1101 knockout cortical motor neurons at 0, 1, 3, 7 days after SCI. 1102 Table S2. Annotation of molecules in the regeneration-associated protein-protein 1103 interaction network in Figure 4D. 1104 Table S3. Module eigengenes (MEs) of co-expression gene networks and module 1105 membership of each gene in RNA-seq of wild-type or REST knockout cortical motor 1106 neurons in sham or SCI conditions. 1107 Table S4. Expression level changes of REST-repressed genes predicted by 1108 ARACNe comparing RGCs sorted at 1, 3, 5 days after optic nerve crush with pro-1109 regenerative treatment to non-regenerative RGCs with control treatment.    Multiple independent functional genomics analyses of distinct injury models were analyzed to computationally identify upstream TFs associated with CNS regeneration. In the first set of analysis (A, left), we performed a mutual information-based network analysis using ARACNe to characterize the transcriptional regulatory network formed by regenerationassociated TFs in multiple independent data sets from spinal cord and peripheral nerve injury. The hierarchical structure of the TF regulatory network was further characterized, so as to identify potential upstream regulators. This step-wise analysis predicted REST, a transcriptional repressor, as an upstream negative regulator inhibiting the core pro-regenerative TFs to drive the expression of regeneration-associated genes (RAGs). In parallel (A, right), we performed an additional unbiased genome-wide screen in another CNS tissue, optic nerve, under pro-growth and native conditions to identify TF regulators of regeneration. Among the ~1000 TF-target gene sets unbiasedly tested via Gene Set Enrichment Analysis, REST was ranked as the top negative regulator of the RGC regeneration state-associated gene set. Multiple independent bio-informatic analyses of external data sets confirmed and converged on our model (B), by which REST is activated by CNS injury and acts as a potential upstream negative regulator of the core regenerative TFs. To test this, we performed gene expression analysis in the injured CNS with REST and after REST depletion, showing REST increases following CNS injury, while the core pro-regenerative TFs and genes remain suppressed. Depleting REST activates a core molecular program driven by a tightly controlled TF network similar to the one activated during regeneration. These results predicted that REST depletion would improve regeneration, which we directly tested in two different, well-established models of regeneration in vivo (C), confirming REST's functional effect as a suppressor of regeneration. In the case of optic nerve injury, REST depletion or inhibition enhanced both RGC regeneration and survival. These analyses identify a novel role for REST as an upstream suppressor of the intrinsic regenerative program in the CNS and demonstrate the power of a systems biology approach involving integrative genomics and bio-informatics to predict key regulators of CNS repair.  Step2

Validation of a data-driven model shows that REST represses pro-regerative TFs and the RAG program
Step3  Schematic diagram illustrating step-wise approaches employed to infer hierarchical TF regulatory networks from (B) time-course microarray datasets.
Step 1: First, ARACNe was applied to each dataset to find TF-target pairs that display correlated transcriptional responses by measuring mutual information (MI) of their mRNA expression profiles (Methods). The sign (+/-) of MI scores indicates the predicted mode of action based on the Pearson's correlation between the TF and its targets. A positive MI suggests activation of this TF on its targets, while a negative MI score suggests repression. All nonsignificant associations were removed by permutation analysis. Second, ARACNe eliminates indirect interactions, such as two genes connected by intermediate steps, through applying a well-known property of MI called data-processing inequality (DPI).
Step 2: To determine the direction of regulation between each TF interactions, ChIP-datasets from ENCODE and previously published ChIP-ChIP and ChIP-seq datasets were integrated to compile a list of all observed physical TF-target binding interactions.
Step 3: To identify the hierarchical structure within directed TF networks, we used graphtheoretical algorithms to determine precise topological ordering of directed networks based on the number of connections that start from or end at each TF, indicating whether a TF is more regulating or more regulated.    Given an a priori gene set known to be targeted by a TF, the goal of GSEA is to determine whether this TF's targets are randomly distributed throughout genes of interest, or primarily found at the top or bottom. An enrichment at the bottom suggests that the TF down-regulates genes of interest, and is thus a negative regulator of the regenerative state (ES <0; TF2 as an example), while an enrichment at the top suggests this TF is a positive regulator of regeneration (ES >0; TF1 as an example). Bottom panel: A total of 1137 TF targeted gene sets were screened and the top 10 negative TF regulators of RGCs' regeneration state were shown in the heatmap by their normalized enrichment scores (NES). (E) Transcriptional regulatory networks comparing RGCs in non-regenerating (control) and regenerating state (pro-regenerative). The networks were constructed using the unbiased, stepwise pipeline described in Figure 2A.  Bars represent mean ± SEM; Asterisks denote statistical significance assessed by two-way ANOVA with Bonferroni post-hoc test (*p < 0.05). (C) Representative western blot and quantitation of REST levels in DRG cells transduced with AAV-GFP or AAV-CRE. (D) Volcano plot showing the mean neurite outgrowth of re-plated DRG neurons infected with lentiviral constructs expressing either REST (Lv135-REST) or humanized luciferase protein (Lv135-hLuc) as a control driven by the CMV promoter at indicated genome copies per cell for 7 days. Neurite extension was quantified 24 hr following re-plating. Each dot represents the mean neurite outgrowth from 6 wells from a replicate experiment normalized to control at indicated viral doses. Asterisks denote statistical significance assed by Student's t-test (* p-value < 0.05; ** p-value < 0.01)