Neuronal Atf4 contributes to RGC neurodegeneration after optic nerve crush.

(A) Neuronal knockout of Atf4, but not Ddit3, by intravitreal AAV2-hSyn1-mTagBFP2-ires-Cre in conditional knockout (cKO) mice improves survival of RGCs 14 days after optic nerve crush, as assessed by immunostaining for the pan-RGC marker RBPMS. (B) Neuronal knockout of Atf4 similarly improves survival of RGCs, as assessed by the robust nuclear marker of injured RGCs, phospho-c-Jun (p-c-Jun). (C-D) Comparable levels of neuroprotection are conferred by Atf4 cKO and (C) Atf4/Ddit3 double cKO (dcKO), or (D) Eif2ak3 cKO, which results in deletion of the ISR-activating kinase Perk. Error bars are SEM. Statistical analysis: one-way ANOVA with Tukey’s multiple comparison with a single pooled variance (*p<0.05; ***p<0.001; ****p<0.0001).

The Perk-activated ISR is a prominent contributor to the transcriptional response to optic nerve injury, primarily acting through Atf4.

(A) Venn diagrams of 282 transcripts exhibiting significant differences by RNA-seq between uninjured retinas and 3 days after optic nerve crush (FDR<0.05). Amongst those, transcripts that exhibit differences between wildtype and cKO retinas after injury (p<0.01 and FDR<0.2) were classified as Perk-, Atf4-, or Chop-regulated. (B-D) Linear regression analyses comparing the impact of neuronal Perk deletion (Eif2ak3 cKO), Atf4 deletion (Atf4 cKO), and Chop deletion (Ddit3 cKO) on the same 282 injury-responsive transcripts. Names of representative Atf4-dependent transcripts are burnt orange text and representative Chop-dependent transcripts are gold text. Significant changes are indicated by colored dots (FDR<0.05).

Perk regulates RGC-autonomous transcriptional changes through canonical Atf4 target genes and influence on c-Jun-regulated programs.

(A-B) Volcano plots of Ingenuity Pathway Analysis (IPA) implicate Atf4 and c-Jun as potential Upstream Transcriptional Regulators of injury-induced expression changes in (A) wildtype retina, and (B) those whose activity after injury is reduced by neuronal Perk deletion. Red orange, blue, and light blue dots indicate Z>2.5, p<10-13; Z>1.5, p<10-10; Z<-2.5, p<10-13; and Z<-1.5, p<10-10, respectively. (C) Heat map of RNA-seq data, showing known and putative Atf4 target genes after injury in wildtype or cKO retinas. (D-F) Linear regression analyses comparing current whole retina RNA-seq cKO data to published data from a similarly designed experiment (Syc-Mazurek et al, 2022). Data from that independent study is indicated on the y-axis by &. (D) Strong correlation among 282 injury-responsive transcripts in the respective wildtype injury conditions between these two independent studies. Names of representative Atf4-dependent transcripts are burnt orange text and representative Chop-dependent transcripts are gold text. (E-F) Significant correlations between the impacts of (H) neuronal Perk, or (I) Atf4 conditional knockout (this study) and deletion of c-Jun from the majority of neural retina& (p<0.001). (J) Cross-study comparison of Ddit3 cKO (this study) and Ddit3 KO& shows strong correlation between the few Chop-dependent injury-responsive transcripts (red dots). (G) Volcano plot of retinal expression data three days after optic nerve crush (this study) for 597 transcripts that are significantly altered in RGCs at day 2 or day 4 after crush, as determined by scRNA-seq (Tran et al., 2019). 287 of these transcripts are detected at FDR<0.05 (red, upregulated, or blue, downregulated) or |log2FC|>0.25 (light red, upregulated, or light blue, downregulated) in whole retina. Only three transcripts (yellow dots) were significantly regulated in whole retina in the discordant direction from scRNA-seq findings. (H) scRNA-seq data set (Tran et al., 2019) reveals RGC-autonomous activation of canonical ISR transcripts. Dot plot at 2 and 4 days after injury, showing both pan-RGC and type-specific upregulation of transcripts demonstrated by whole retina RNA-seq to be Atf4- and/or Chop-dependent. Notably, some transcripts exhibiting low, non-uniform, injury induction in RGCs – Avil, Gm29374, Cdsn, and Fibin – were amongst those that appear to exhibit potential Chop-dependence but failed to reach threshold for inclusion in either retinal injury-dependence, Chop-dependence, or scRNA-seq pseudo-bulk analyses. Conversely, Stk32a, the lone Chop-dependent transcript reported by Ddit3 knockout but not neuronal Ddit3 cKO is not regulated by injury within RGCs, suggesting non-autonomous modulation of this transcript by Chop in retina.

Perk-Atf4 contributes to RGC axon regenerative potential after optic nerve crush.

(A-B) Heat maps of select injury-responsive transcripts revealed by retinal RNA-seq three days after optic nerve crush in wildtype and cKO mice, focusing on (A) genes implicated in axon regeneration, and (B) the maintenance of mature and subtype-specific RGC phenotypes. (C-D) Double conditional knockout (dcKO) of neuronal Perk (Eif2ak3; n=4) or Atf4 (Atf4; n=7) reduces regeneration enabled compared to conditional knockout (cKO) of the tumor suppressor Pten alone (n=8). Error bars are SEM. Statistical analysis: two-way ANOVA with Sidak’s multiple comparisons test (**p<0.01).

Neuronal targeting of Cre expression in retina.

Expression of tdTomato in whole mount retina of heterozygous Ai14 mice reports the expression of Cre recombinase in the neurons of the ganglion cell layer (GCL) following transduction by intravitreal AAV2-hSyn1-mTagBFP2-ires-Cre. Co-labeling using antibodies against βIII-tubulin (TuJ1) or NeuN labels RGCs and all GCL neuronal nuclei, respectively.

Neuronal targeting of Cre expression in retina.

(A-C) Scatterplots showing expression level (transcripts per cell x % expressing cells) of the 282 injury-responsive genes in 3 days post ONC bulk RNA-seq identified in this study in RGC scRNA-seq dataset from Tran et al. (2019) at 1, 2, and 4 days post crush. 271/282 injury-responsive genes were detected the scRNA-seq dataset and showed a general correspondence in trend at each time point. This was most clear at 4dpc in which 232/271 detected genes showed a >0.5 log2 fold change difference from control in the same direction as this study. Genes exhibiting >1 log2 change are labeled. This demonstrates that our expression differences from bulk-RNA seq identify RGC-specific expression changes and are in agreement with previous pseudo-bulk analysis of RGC scRNA-seq data from similar time points. (D-E) Similar correlations and slopes upon re-analysis of linear regression between (D) Perk- and Atf4-dependent transcripts and (E) Perk- and c-Jun-dependent transcripts using 287 genes detected as injury-responsive by both whole retina RNA-seq (|log2FC>0.25|) and RGC scRNA-seq (Tran et al., 2019) suggests that these relationships are driven primarily by RGC-autonomous expression changes.

Distinct knockout strategies, rather than RNA-seq approaches, account for discordance between Atf4- and Chop-dependent transcription reported in this and a previous study.

(A) Strong linear regression correlation among 479 injury-induced RGC-autonomous expression changes (as determined by scRNA-seq, Tran et al., 2019) that are detected in both whole retina (this study) and FACS-enriched RGCs (FDR<0.1, Tian et al., 2022) in control (i.e., without gene targeting) conditions. This suggests that both RNA-seq approaches readily detect RGC-autonomous changes in known Atf4 target genes (burnt orange text) and putative Chop target genes (gold text), with greater sensitivity afforded by enriching for RGCs demonstrated by Slope>1. (B-C) Linear regression comparing the impacts of targeting (B) Atf4 or (C) Ddit3 by cKO or gRNA on 597 transcripts that are significantly altered in RGCs at day 2 or day 4 after crush, as determined by scRNA-seq (Tran et al., 2019). X-axis data is from the current study (whole retina cKO), and Y-axis data is from Tian et al., 2022 (FACS-sorted RGCs expressing gRNAs)#. gRNAs significantly modulate transcripts unaffected by cKO (green dots), but do not significantly impact the expression of many transcripts affected by cKO (blue dots).