Introduction

Experiments using conditional knockout mice have revealed a critical role for stress signaling-mediated transcriptional responses in determining the fates of neurons after axon injury. Prominent among these, neuronal knockout of the stress-responsive Dual leucine-zipper kinase (Dlk) suppresses the stimulation by injury of transcription factors of the bZIP family, including c-Jun and the Activating transcription factor-3 (Atf3), that substantially alter the transcriptomes of distressed neurons (Asghari Adib et al. 2018; Farley & Watkins 2018; Le Pichon et al. 2017; Tran et al. 2019; Welsbie et al. 2013, 2017). As is common in cellular stress signaling, this injury-induced MAP kinase (MAPK) signaling cascade couples the activation of transcriptional programs that prime damaged neurons for repair with those that prime those same neurons for apoptosis, the latter a feature that may facilitate the elimination of irreparable cells (Watkins et al. 2013).

These and other essential insights have been enabled by mouse optic nerve crush, an exceptionally tractable model for CNS axon injury that has been valuable for understanding axonopathy-driven CNS neurodegeneration, particularly glaucoma, and the neuron-intrinsic and extrinsic factors that limit CNS axon regeneration (Lindborg et al. 2021; Tran et al. 2019; Vidal-Sanz et al. 2017). Following crush injury, axotomized retinal ganglion cells (RGCs) face both intrinsic and environmental barriers to axon regeneration, and the pro-apoptotic aspects of this response ultimately result in extensive RGC neurodegeneration over subsequent weeks (Syc-Mazurek et al. 2017a; Watkins et al. 2013; Welsbie et al. 2013, 2017). Though neuronal knockout of Dlk, c-Jun, or Atf3 each confer considerable RGC neuroprotection, disrupting these also limits the upregulation of regeneration-associated genes (RAGs) and the success of regenerative interventions, such as knockout of the tumor suppressor Pten, that offer promise for enabling CNS repair (Jacobi et al. 2022; Watkins et al. 2013). Devising strategies to suppress neuronal loss without restricting intrinsic programs for repair will therefore require improving our understanding of which pathways and effectors within the injury response control axon regenerative and neurodegenerative programs and how these overlap (Patel et al. 2020).

Of particular interest is deciphering the contributions of the Integrated Stress Response (ISR). The ISR, activated in parallel to c-Jun as part of the Dlk signaling cascade (Larhammar et al. 2017), results in the suppression of general translation but enhanced transcription and translation of select mRNAs, including Atf4, encoding the Activating transcription factor-4, and Ddit3, encoding the C/ebp homologous protein (Chop) (Pitale et al. 2017). Both Atf4 and Chop have the potential to regulate transcription of Atf3 and/or heterodimerize with c-Jun, Atf3, other transcription factors of the C/ebp family, or each other, raising the possibility of crosstalk between the ISR and MAPK arms of the Dlk response (Pakos-Zebrucka et al. 2016). Genetic disruption of ISR activation or of neuronal Eif2ak3, the gene encoding the ISR-activating kinase Perk, confers substantial RGC neuroprotection after optic nerve crush (Larhammar et al. 2017; Yang et al. 2016). These findings demonstrate that the neuronal ISR represents a functional component of the intrinsic neurodegenerative response but have provided limited mechanistic insight.

Recent studies emphasize the importance of defining the downstream effectors of the ISR, their relationships to c-Jun, and their impacts on RGC axon regenerative potential. Based on CRISPR screening, RNA-seq, and ATAC-seq, Tian et al. (2022) proposed that Atf4 serves, along with C/ebpψ, as a critical component of a core neurodegenerative program, with Chop and Atf3 together mediating a second core program (Tian et al. 2022). Injured RGCs expressing Cas9 and a gRNA pool against Atf4 exhibit improved survival and modulation of genes primarily associated with intrinsic neuronal stressors (e.g., DNA damage response, autophagy, and the NAD/p53 pathway) distinct from canonical Atf4 target genes for adapting to amino acid deprivation, endoplasmic reticulum stress, and other ISR-activating insults. Additive to this, partial neuroprotection could also be achieved by gRNA pools targeting either Ddit3 or Atf3, either of which resulted in extensive, highly overlapping, transcriptional consequences related to cytokines and innate immunity (Tian et al. 2022). Importantly, a companion report shows that, of these same gRNA pools, only targeting Atf3 reduced RGC axon regeneration enabled by knockout of the tumor suppressor Pten (Jacobi et al. 2022). Together, these findings led to the proposal that Atf4 and Chop function in parallel neurodegenerative programs and, unlike Atf3 or c-Jun, do so without influencing RGC axon regenerative potential.

Other reports, however, seem to be inconsistent with these non-canonical roles for Atf4 and Chop. The Dlk-mediated stress response includes ISR-dependent upregulation of Chac1, Ddit3, and other target genes consistent with a typical Atf4-mediated transcriptional program, inclusive of Chop activation (Larhammar et al. 2017). In addition, the proposed role of Chop as a principal neuron-intrinsic effector of the response to optic nerve injury appears to be incongruent with the mild impact of germline Ddit3 knockout on that response. RNA-seq of retinas from these mice uncovered few injury-induced transcripts that were significantly suppressed relative to the extensive impact of neural knockout of c-Jun (Syc-Mazurek et al. 2022). Though these Chop-null mice reproducibly exhibit partial RGC neuroprotection (Hu et al. 2012; Syc-Mazurek et al. 2017b), the modest influence of Chop deficiency on the transcriptome suggests the need for further evaluation of whether its primary role in regulating RGC survival is neuron-autonomous, as it may also reflect developmental effects or influences on other cell types, perhaps outside of the retina.

Here we use previously validated conditional knockout (cKO) mice lines to directly address these and other questions raised by germline knockout and CRISPR studies. We find that a canonical Atf4 response functions as the principal mediator of a Perk-activated response that broadly contributes to both RGC apoptosis and axon regenerative potential. Unexpectedly, neuronal Chop, long considered to be a primary neuron-intrinsic effector of the ISR in injured RGCs, plays a relatively modest role within these Perk/Atf4-stimulated programs. These findings reveal an integral role for Perk-stimulated Atf4 in the response to optic axon damage, serving to link neurodegenerative and axon regenerative transcriptional programs that determine the fates of distressed neurons.

Results

Neuronal Atf4 knockout mimics the neuroprotection provided by Perk deletion

Deletion of neuronal Perk, a key mediator of the ISR after optic nerve crush, provides significant, though incomplete, neuroprotection to RGCs (Larhammar et al. 2017). To determine how the ISR effectors Atf4 and Chop contribute to neurodegeneration, we evaluated the survival of RGCs in cKO mice using immunohistochemistry for the pan-RGC marker RBPMS. Using AAV serotype 2 and the human syapsin-1 promoter (hSyn1) to primarily drive Cre expression in neurons of the ganglion cell layer (Supplemental Figure S1), we observed that targeting Atf4, but, surprisingly, not Ddit3, confers significant neuroprotection of RBPMS-expressing RGCs (Figure 1A). We next compared the neuroprotection measured by this method with that determined using phospho-c-Jun as a robust, easily quantifiable nuclear marker of injured RGCs, finding similar improvements to RGC survival upon Atf4 knockout using either method (Figure 1B). To determine if Chop disruption might augment the neuroprotection conferred by Atf4 disruption, we generated Atf4/Ddit3 double conditional knockout (dcKO) mice. However, these mice did not exhibit significantly greater RGC survival than Atf4 cKO mice (Figure 1C). Finally, we directly compared the effects of Perk knockout and Atf4 knockout, finding disruption of Atf4 is sufficient to afford a similar degree of neuroprotection to that of Perk (Figure 1D). Together with previous findings that Atf4 activation is Dlk- and Perk-dependent (Larhammar et al. 2017), these results suggest that Atf4 serves as the primary mediator of the pro-apoptotic effects of the ISR after optic nerve crush.

Figure 1.

Neuronal Perk mediates an extensive contribution to the injury response primarily through Atf4

The equivalent neuroprotection afforded by knockout of Perk and Atf4 raises the possibility that Atf4 may be the principal neuron-autonomous mediator of ISR-activated transcription influencing RGC survival. To better understand the contributions of Perk and its effectors to the transcriptional response, we performed expression profiling of retina by RNA-seq three days after optic nerve crush in wildtype mice and three cKO lines: Eif2ak3 cKO to disrupt Perk, Ddit3 cKO to disrupt Chop, and Atf4 cKO to disrupt Atf4 (Table S1).

We began by determining the contribution of Perk and its effectors to the transcriptional injury response. Using a stringent statistical assessment for differentially expressed genes (DEGs, false discovery rate FDR<0.05), we identified 282 transcripts modulated by injury in control wildtype mice injected with AAV2-hSyn1-Cre. Among these, 117, or 41.5%, reached this same strict threshold for differential expression in a comparison between Perk cKO retinas after injury wildtype retinas after injury (Table S1), with this number increased to 157, or 55.7%, using a moderately relaxed threshold of p<0.01 and FDR<0.2 (Figure 2A). All but three of these change in the opposite direction of their modulation by injury, suggesting a central role for the ISR in mediating these injury responses. Among the 154 transcripts regulated by injury in a Perk-dependent manner, 56 (36.4%) are also suppressed by Atf4 knockout, with a particularly pronounced contribution among those transcripts that are upregulated after injury (37 out of 73, or 50.7%), including seven of the nine transcripts that display Chop-dependence at a threshold of p<0.01 and FDR<0.2 (Figure 2A). Notably, among Atf4-regulated transcripts, 86.2% also exhibit Perk-dependence at this threshold, consistent with Atf4 activation after optic nerve crush being mediated predominantly by Perk and not subject to compensation by other eIF2α kinases (Larhammar et al. 2017). This analysis, though limited by its arbitrary thresholds for defining factor-dependence, implies an essential role of the Perk-activated ISR in the transcriptional injury response, with Chop serving as a relatively minor contributor within a program primarily mediated by Atf4.

Figure 2.

To evaluate these relationships in a threshold-independent manner, we performed linear regression of the 282 injury-regulated genes. We first compared the impact of Perk cKO on the regulation of these transcripts by injury with that of Atf4 cKO. This analysis reveals a strong correlation (R=0.87) that, along with a best-fit slope of 0.75, suggests that the role of Atf4 in mediating the effects of Perk is more comprehensive than suggested by threshold-dependent analyses alone (Figure 2B). A similar assessment uncovers a milder (R=0.59), but still highly significant, correlation between Perk-dependent and Chop-dependent injury-responsive transcripts (Figure 2C). However, the shallow slope of 0.23 for the best-fit line reveals that, though Chop influences similar genes as Perk, it does so relatively mildly. Consistent with these findings, we observe a significant correlation between Atf4-dependent and Chop-dependent transcripts, again with a shallow slope that indicates a more potent influence of Atf4 knockout (Figure 2D). These results imply that Chop serves as a potentiator of the ISR after optic nerve crush rather than a primary mediator. Together, these findings are consistent with Atf4 exerting the dominant role in the transcriptional response and promotion of RGC apoptosis.

Perk signaling regulates canonical Atf4 and c-Jun transcriptional programs

To explore the transcriptional programs downstream of Perk, we next applied the Upstream Regulator tool of Ingenuity Pathway Analysis (IPA), which includes determination of potential transcription factors likely to be mediators of the observed patterns of transcriptional changes and known target genes derived from extensive curation of the literature. Consistent with previous studies (Syc-Mazurek et al. 2022; Yasuda et al. 2016), we find that Atf4, along with c-Jun, emerges among the top five candidates for a positive transcriptional regulator (p<10^-13 and Activation Z-score>2.5), based on the upregulation of many genes previously demonstrated to be direct targets of Atf4, including Atf3, Mthfd2, Stc2, and Phgdh (Ben-Sahra et al. 2016; Han et al. 2013; Pan et al. 2007; Zhao et al. 2016) (Figure 3A). IPA also suggested PGC1α (gene name: Ppargc1a), which can be negatively regulated by Atf4 activation (Montori-Grau et al. 2022; Wang et al. 2013), as the most influential negative transcriptional regulator (Table S2). Chop (Ddit3) was implicated among a cluster of dozens of less influential positive regulators (p<10^-5 and Activation Z-score>1.25) (Figure 3A; Table S2). Comparisons between wildtype and Perk (Eif2ak3) cKO retinas 3 days after injury suggest that disruption of Perk abrogates Atf4 activation most strongly, with c-Jun and PGC1α also affected (Figure 3B; Table S3). This analysis suggests the activation of a canonical Perk/Atf4-mediated transcriptional response (Figure 3C) after RGC axon injury and raised the unexpected possibility that Perk signaling might also influence c-Jun-regulated target genes.

Figure 3.

To further investigate a potential interaction between c-Jun- and Perk-regulated transcripts, we leveraged RNA-seq datasets from a separate, similarly designed, study that included mice lacking c-Jun in the majority of neural retina and Chop-null mice (Syc-Mazurek et al. 2022). First, we evaluated the expression profiles of the wildtype injury conditions between these independent studies, finding an exceptional correlation (R=0.95, Slope = 1.07) among 282 injury-regulated genes (Figure 3D). That robust relationship between the control conditions provided a basis for further comparisons. We therefore proceeded with a cross-study analysis comparing the impact of Perk cKO with that of c-Jun cKO. That analysis revealed a significant correlation (R=0.49; Slope = 0.44) among c-Jun- and Perk-modulated transcripts (Figure 3E). Nevertheless, some injury-modulated transcripts are regulated exclusively by c-Jun (e.g., Sox11 and Arid5a) or Perk (e.g., Chac1 and Phgdh). Together, these findings suggest that, despite a lack of influence of Jun knockout on Eif2ak3, Atf4, or Ddit3 transcripts and vice versa, these parallel pathways participate in substantially overlapping transcriptional programs.

We next evaluated the similarity between germline and neuronal knockout of Ddit3 between these two studies. Though each uncovered only a small number of Chop-dependent, injury-responsive transcripts, those few exhibit considerable overlap. Just four transcripts passed a stringent test for Chop-dependence in retinas of Ddit3^-/- mice (Syc-Mazurek et al. 2022), three of which – Stbd1, Gm13889, and Avil (also known as Doc6 for “dependent on Chop-6”) – have been demonstrated to be upregulated neuron-autonomously by injury within RGCs by scRNA-seq (Tran et al. 2019). Using similar stringency (FDR<0.05), these three were among only five genes that we independently detected as injury-upregulated in a neuronally Chop-dependent manner using Ddit3 cKO mice (Figure 3F). Additional injury-induced transcripts suppressed by greater than 50% in both Ddit3 cKO and Ddit3 KO (though without surpassing the FDR<0.05 threshold) include Fibin (p<0.0005 in both studies) and Cdsn (p<0.001 in both studies), genes for which ChIP-seq has previously identified Chop binding near their transcriptional start sites under stress conditions (Han et al. 2013). These results show that, despite differences in neuroprotection between Ddit3^-/- mice (Hu et al. 2012; Syc-Mazurek et al. 2017b) and neuronal targeting of Ddit3 in cKO retinas, their transcriptional consequences within the retina closely resemble one another and align with expected Chop target genes.

Together, these expression profiles suggest that the Perk-mediated ISR is an integral contributor to the retinal transcriptional response after optic nerve injury, acting primarily through canonical Atf4 targets and interacting with c-Jun-mediated transcriptional programs.

RGC-autonomous Atf4- and Chop-dependent transcriptional changes are prominently represented by whole retina transcriptomics

Our targeted expression of Cre recombinase in cKO mice argues for a central role for the neuronal ISR in the response to optic axon injury, with whole retina transcriptomics likely reporting both these neuron-autonomous effects and secondary consequences to other retinal cells. To determine the extent to which ISR-dependent transcription uncovered by retinal RNA-seq represents RGC-autonomous expression changes, we next leveraged an independent single-cell RNA-seq (scRNA-seq) data set that details the injury-induces expression changes within subtypes of RGCs at 2 and 4 days after optic nerve crush (Tran et al. 2019).

We began by assessing the whole retina expression of transcripts that have been demonstrated to be differentially expressed by RGCs at 2 or 4 days after optic nerve crush (pseudo-bulk scRNA-seq |Diff>0.3| and FDR<10^-20). Of 615 such transcripts, 597 are represented in wildtype whole retina 3 days after injury. We find that 290 of these transcripts exhibit differential expression (|log₂FC|>0.25) in retina, with all but three modulated by injury in the same direction as reported by scRNA-seq (Figure 3G). Moreover, the majority of the 282 injury-responsive transcripts we have identified in whole retina exhibit concordance with changes found in scRNA-seq of RGCs (Supplemental Figure S2A-C). Importantly, many transcripts identified as Perk-, Atf4-, and/or Chop-dependent in this and other studies are among those that are upregulated neuron-autonomously within multiple subtypes of RGCs following optic nerve crush (Figure 3G). Consistent with this, the relationships we detected between Perk-dependent transcripts and Atf4- or c-Jun-dependent transcripts in whole retina are maintained when re-assessed using DEGs confirmed to be injury-regulated within RGCs by scRNA-seq (Supplemental Figure S2D-E). These results imply that the injury-induced Atf4 transcriptional program that we detect by whole retina transcriptomics of cKO mice primarily reflects RGC-autonomous expression changes.

This canonical Atf4 program, inclusive of the contribution of Chop, contrasts with a recent report of two non-canonical, largely non-overlapping, programs controlled by these two ISR transcription factors (Tian et al. 2022). We next investigated potential sources of this discordance. The proposed parallel neurodegenerative programs were deduced in part by RNA-seq of FACS-enriched injured RGCs expressing gRNAs targeting Atf4 or Ddit3. We therefore began with an assessment of how the injury responses detected in whole retina in the current study compare with those reported for RGCs collected by FACS. Using only transcripts independently identified as RGC-autonomous expression changes by scRNA-seq (Tran et al. 2019) for this cross-study comparison, we find a highly significant correlation (R=0.74) between injury-regulated transcripts under control conditions (i.e., without gene targeting) between these independent studies (Supplemental Figure S3A). This suggests that these distinct approaches similarly report prominent RGC-autonomous expression changes, inclusive of numerous known direct targets of Atf4, with a steep linear regression (slope = 1.57) that, unsurprisingly, suggests greater sensitivity for these changes using FACS-sorted RGCs. Despite that similarity, linear regression reveals little correlation between the impact of cKO- and gRNA-mediated knockout on 597 RGC-autonomous, injured-regulated transcripts. We find only very weak, shallow correlation between the effects of Atf4 gRNA and Atf4 cKO (R=0.13, slope = 0.33, p<0.005). cKO-sensitive signature Atf4 target genes Chac1, Phgdh, Sesn2, Slc7a5, and Slc6a8, amongst others, are not among the 125 of these 597 (20.9%) transcripts reported to be significantly affected by Atf4 gRNA (Supplemental Figure S3B) (Crawford et al. 2015; Garaeva et al. 2016; Han et al. 2013; Liu et al. 2018; Oh-Hashi et al. 2013; Park et al. 2017). We also find no correlation between the influence of Ddit3 cKO and gRNA targeting Ddit3, with no reported inhibition by this gRNA pool of the Ddit3 KO- and cKO-sensitive transcripts Gm13889 and Stbd1 among 121 (20.3%) significantly affected transcripts (FDR<0.05) (Supplemental Figure S3C). Notably, the reported effects of gRNA targeting Ddit3, like Atf3 gRNA, include no reduction of Ddit3 transcript but greater than 80% suppression of the pro-apoptotic and pro-regenerative Atf3 transcript (Tian et al. 2022), an effect not observed in Ddit3 germline (Syc-Mazurek et al. 2022) or conditional knockout mice (present study), despite ample sensitivity for Atf3 modulation using whole retina transcriptomics. These analyses argue that more specific and effective disruption of target genes by cKO primarily accounts for the striking discordance between the transcriptional programs of Atf4 and Chop suggested by these distinct approaches.

Neuronal Atf4 knockout limits axon regeneration by Pten-deficient RGCs

The commonalities that we uncovered between c-Jun- and Perk-regulated transcription suggested the hypothesis that, like c-Jun, Perk/Atf4 could be involved not only in promoting apoptosis but also in promoting regeneration. To investigate this possibility, we first examined the Perk- and Atf4-dependence of regeneration-associated genes (RAGs) and the injury-induced downregulation of mature and subtype-specific RGC markers, finding that many exhibit at least partial Perk- and Atf4-dependence (Figure 4A-B). We therefore crossed mice harboring the floxxed Pten alleles and those harboring floxxed Atf4 alleles or floxxed Perk alleles to generate homozygous dcKO mice. Following intravitreal injection of AAV2-hSyn1-Cre, we performed optic nerve crush, comparing axon regeneration in these dcKO mice to that of Pten cKO mice (Figure 4C-D). We find that neuronal Perk deficiency or Atf4 deficiency limits the efficacy of this regenerative intervention. These data support an integral contribution of the ISR to the axon regenerative program in these CNS neurons, and indicate that, as with Dlk and c-Jun, Atf4 is involved in both pro-regenerative and pro-apoptotic transcriptional changes.

Figure 4.

Discussion

Cellular stress signaling pathways typically couple the promotion of growth and repair with priming for apoptosis, a feature of stress responses that may aid in the elimination of cells that are irretrievably damaged (Hotamisligil & Davis 2016). As prominent effectors of stress signaling, transcription factors of the bZIP family, including c-Jun and Atf3, therefore can contribute either to apoptosis or to recovery, depending on context, complicating efforts to harness transcriptional programs for axon regeneration independently of neurodegenerative programs (Jacobi et al. 2022; Kole et al. 2020; Raivich et al. 2004; Simon & Watkins 2018; Syc-Mazurek et al. 2017b; Tian et al. 2022; Watkins et al. 2013). Multiple lines of evidence indicating that the Perk-activated ISR contributes to RGC apoptosis after optic nerve crush raised the appealing possibility that this pathway may represent a distinct branch of the injury response that might be targeted to selectively reduce neurodegeneration without suppressing repair programs (Hu et al. 2012; Larhammar et al. 2017; Tian et al. 2022; Wang et al. 2020). The cKO studies reported here, however, reveal that the Perk-activated ISR is more integral to the entire RGC injury response than previously appreciated, contributing substantially, in conjunction with c-Jun, to both apoptotic and regenerative programs. Though MAP kinase stress signaling and the ISR are known to exhibit some degree of crosstalk in other settings (Brown et al. 2016; Danzi et al. 2018; Pakos-Zebrucka et al. 2016), the exceptionally close coupling of these largely independent pathways by Dlk signaling seems to be a critical feature of the response of RGCs to axon injury, with disruption of Atf4, like that of c-Jun, limiting both neurodegeneration and axon regenerative potential.

By examining in parallel the neuronal knockout of Perk and its effectors Atf4 and Chop, along with comparisons to related studies, the current work has provided insights into the mechanisms by which the ISR influences the fates of injured RGCs. These conditional knockout mouse experiments reveal that instead of non-canonical Atf4 and Chop transcriptional programs, each with distinct contributions to neurodegeneration, Atf4 is the principal effector of the ISR transcriptional response, with Chop playing a modest role within a canonical Atf4-mediated program. Though RGC neuroprotection in germline Ddit3 knockout mice has long been interpreted as evidence of its central role in the neuronal ISR, our findings suggest that further investigations of Chop may instead uncover its non-autonomous mechanisms regulating the survival of injured RGCs. The unexpectedly subtle role for neuronal Chop aligns well with expression profiling data from Chop-null mice (Syc-Mazurek et al. 2022) but stands in contrast to the extensive impact of CRISPR-based targeting of Ddit3 on the RGC transcriptome, which includes potent suppression of Atf3 and mimics with remarkable precision (R>0.9, slope=1) the impact of Atf3 gRNA (Tian et al. 2022). It is possible then that the differential impact between Ddit3 cKO and Ddit3-targeting gRNA on Atf3 mRNA underlies their distinct effects on transcription and neuroprotection, though this hypothesis is not sufficient to clarify why this Atf3-suppressing Ddit3 gRNA pool does not recapitulate the inhibition of RGC axon regeneration by Atf3 gRNA (Jacobi et al. 2022). With cKO experiments revealing that Atf3 upregulation is dependent on Atf4 and c-Jun rather than Chop, it will be of interest to elucidate how much of the commonality among the transcriptional programs of these bZIP factors is attributable to interdependence of their expression and how much to their combinatorial heterodimerization. Among other non-exclusive possibilities, the commonalities between Atf4- and c-Jun-dependent transcription could reflect: (1) dependence of Atf3 elevation on c-Jun-Atf4 heterodimers; (2) dependence on Atf4 for induction of Atf3, which then heterodimerizes with c-Jun; or (3) regulation of common target genes through distinct DNA binding sites. The current study provides a framework for deciphering these networks in part by leveraging cross-study comparisons of retina, RGC, and single-cell RNA-seq data to uncover robust RGC-autonomous expression signatures of Atf4 and other transcription factors.

Given the potential for the ISR to influence neuronal fate through multiple translational and transcriptional mechanisms, it is perhaps surprising that its primary contributions to the fates of injured RGCs are mediated by a single transcription factor, Atf4. Nevertheless, at least four lines of evidence contend that the cKO experiments reported here provide a reliable picture of its contribution. First, the extensive commonality in injury-regulated transcripts across studies, whether utilizing retinal expression data or filtering for RGC-autonomous expression changes, argues against AAV2-hSyn1-Cre or other technical aspects of our approach altering the injury response. Secondly, support for the modest impact of Chop disruption on RGC injury-regulated genes is provided by the highly similar findings in the independently generated and validated Ddit3-null and Ddit3 cKO mouse lines. Third, we find remarkable overlap between expression changes and phenotypes mediated by Perk and Atf4, which we previously demonstrated to be in the same pathway after optic nerve crush and other Dlk-activating insults (Larhammar et al. 2017). Finally, the interaction between Perk/Atf4 and c-Jun-regulated transcription uncovered by our cross-study analysis predicts an impact of Perk or Atf4 knockout on RGC axon regenerative potential that is validated by dcKO with Pten. This result is consistent with recent findings that early injury responses, including regenerative and degenerative responses that we have found in this study to be driven by Perk/Atf4, are similar in non-regenerating wildtype and regenerating Pten-deficient RGCs (Jacobi et al, 2022). Together, these conditional knockout studies highlight the critical roles for ISR-activated transcriptional programs in determining the fates of distressed neurons and the intimate link between injury-activated neurodegenerative and axon regenerative programs.

Acknowledgements

We thank Katie Steck, Shivani Kulkarni, and Talia Sisroe for technical assistance. This work was supported by grants from Mission Connect, a project of the TIRR Foundation, NIH grants R01NS112691 and R01NS076708 to T.A.W., the Glaucoma Research Foundation, and NIH grant R00EY029360 to N.M.T. Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number P50HD103555 for use of the Neuroconnectivity Core facilities. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Materials and methods

Animals

All animals used were in C57BL6/J background strain. Animal care and experimental procedures were approved by the institutional animal care and use committee (IACUC) at Baylor College of Medicine, according to NIH Guidelines.

Animal resource table

Antibodies resource table

Adeno-associated virus (AAV)

Adeno-associated viral vectors containing human synapsin-1 (hSyn1) promoter driving expression of mTagBFP2-IRES-Cre or mTagBFP2-IRES-NLS-smGFPmyc(dark) were packaged into AAV2 capsids at the Optogenetics and Viral Vector Core at Duncan Neurological Research Institute, Houston.

Intravitreal injections

Animals were anesthetized with isoflurane. Artificial tears ointment (Covetrus #11695-6832-1) was applied on the eyes. In case of bilateral injections, non-injected eye received artificial tears. The eyes were sterilized and prepared by 3 repeated applications of 5% ophthalmic betadine (Henry Schein #6900250), followed by Opti-clear ophthalmic eye wash (Akorn #NDC 17478-620-04) and wiped dry. Topical anesthetic 0.5% proparacaine HCl ophthalmic solution (Henry Schein #1365345) was applied. A 5-µl Hamilton syringe loaded with a custom 33-gauge needle (Hamilton #7803-5) was used to puncture the eyeball and relieve some of the intraocular pressure. The Hamilton needle was inserted back into the same puncture site and 2 µl of AAV in titers ranging from 10¹² −10¹³ vg/ml was delivered per eye.

Intra-orbital optic nerve crush (ONC)

Animals were dosed with 1mg/kg buprenorphine sustained release formulation 1 hour prior to surgery. At the time of surgery, they were anesthetized with isoflurane. Artificial tears ointment (Covetrus #11695-6832-1) was applied on the non-surgical eye. The surgical eye was sterilized and prepared by 3 repeated applications of 5% ophthalmic betadine (Henry Schein #6900250), followed by Opti-clear ophthalmic eye wash and wiped dry. Topical anesthetic 0.5% proparacaine HCl ophthalmic solution (Henry Schein #1365345) was applied on the surgical eye (left). Incisions were made in both the conjunctival layers using a pair of Vannas scissors (World precision instruments #501777). Two pairs of suture-tying forceps (Fine science tools #1106307) were used to gently clear the soft tissue in the intra-orbital space behind the eye until the optic nerve was visible. A pair of Dumont forceps (Fine science tools #1125325) were used to manually crush the optic nerve for 5s. The eyeball was gently pushed back into the orbit. Animals were then returned to their home cages and monitored until sternal recumbency was observed.

Immunolabeling of retinae

Animals were euthanized by anesthesia overdose, followed by decapitation. The eyes were harvested and fixed in 4% paraformaldehyde for 1 hour. The retinae were dissected out in 1X phosphate-buffered saline (PBS) and then blocked for 30min in 5% goat serum, 0.5% Triton X-100 and 0.025% sodium azide in 1X PBS. The retinae were then incubated for 5 days in primary antibody diluted in blocking buffer solution in 4°C. They were then washed three times in 1X PBS with 0.5% TritonX-100 for 30 min each, and then moved to appropriate secondary antibody solution prepared in blocking buffer and incubated overnight. This was then followed by three washes again with 1X PBS with 0.5% TritonX-100 for 30 min each. Retinae were then mounted in Drop-n-Stain EverBrite^TM Mounting Medium (Biotium #23008) onto slides and imaged using Zeiss Axio Imager Z1 fluorescence microscope.

RGC survival assessment

Animals that underwent optic nerve crush surgeries were euthanized two weeks post crush by anesthesia overdose, followed by decapitation. Retinae were dissected out, fixed and immunolabeled as described above. Whole mounted retinas were imaged at 10x to capture 7-11 images per retina in all fluorescently labeled channels. Images were blind coded and quantified for surviving RBPMS positive and/or p-c-Jun positive cells through a combination of hand counting and RGC counter tool of the Simple RGC plug-in in ImageJ FIJI. The average count of all images per retina was calculated to obtain mean cell count per image and plotted to show RGC survival. For the plot showing percentage survival of RGCs, the counts were normalized to mean count of uninjured retinas for each experimental group. Plots and statistical analyses were generated using GraphPad Prism.

Axon regeneration assessment

Animals that underwent optic nerve crush surgeries were intravitreally injected on the surgical eye with 2 µl of Alexa 594-conjugated-cholera toxin β (Life technologies #C22842) at day 14. Animals were euthanized on day 15 by anesthesia overdose followed by decapitation. Surgical eye and optic nerve were dissected out together by first cutting the optic nerves at the chiasm through intracranial dissection and making a single cut along the length of the control uninjured optic nerve. Incisions were then made in the orbital bones to remove the ceiling of the orbital cavity and gently release the surgical eye and optic nerve out together. The retinae and optic nerves were then drop fixed together in 4% PFA. The connective tissue around the optic nerves were dissected out. The optic nerves were then processed for tissue clearing by incubating in 100% methanol for 4min and then moved to Visikol-1 overnight on a shaker at 4°C. They were then moved to Visikol-2 for 2 hours with shaking at room temperature and then mounted in Visikol-2 onto slides. The whole mounted and cleared optic nerves were imaged using Zeiss Imager Z2 fluorescence microscope and Apotome 2.0. The nerves were imaged by capturing 3-5 sequential Z-stacks at 10x magnification along the length of nerve with a range of 200um and Z-stacks of 100 images each. The Z-stacks were stitched together using FIJI stitching algorithm and max intensity projections were created in FIJI. The optic nerve crush site was identified by the point where the brightest CTB labeling ends. The line tool in FIJI was used to mark regions 0.75mm and 1.5mm from the crush site. A vertical line was drawn across the optic nerve at these respective distances and the number of axons that course through the drawn line at each distance was manually counted. In regions where there were too many axons that could be accurately counted, the Z-stacks were used for quantification. Every 10^th image in the z-stack was counted for number of axons coursing through the drawn lines at 0.75mm and 1.5mm from the crush site and the total number from each Z-stack calculated to provide an estimated number of axons. Plots and statistical analyses were generated using GraphPad Prism.

RNA preparation and Next-Generation Sequencing

Animals that underwent optic nerve crush surgeries were euthanized 3 days post crush by anesthesia overdose and decapitation. The retinae were quickly dissected out in 1X PBS and snap frozen in dry ice-ethanol bath and stored in −80°C until ready to be processed for RNA extraction. RNA was extracted using the Qiagen RNeasy micro kit (#74004). Extracted RNA was sent to Novogene Co. Ltd (Beijing, China) for Next Generation Sequencing using their Illumina NovaSeq platform for mouse mRNA sequencing. RNA libraries were prepared according to Novogene procedures by polyA capture (or rRNA removal) and reverse transcription of cDNA. Illumina PE150 technology was employed to sequence the samples. Sample reads were aligned to mouse reference genome using HISAT2 algorithm. Gene expression analysis was performed using Novogene pipeline. Venn diagrams were generated using Deep Venn (arXiv:2210.04597), heatmaps were generated using Morpheus (https://software.broadinstitute.org/morpheus), pathway analyses were performed using Ingenuity pathway analysis (Qiagen) and single cell mouse retinal ganglion cell atlas was obtained from Broad single cell portal.

Data availability

RNA-sequencing data is available from the Gene Expression Omnibus through series accession number GEO: GSE223321.