Pten inhibition dedifferentiates long-distance axon-regenerating intrinsically photosensitive retinal ganglion cells and upregulates mitochondria-associated Dynlt1a and Lars2

Mammalian central nervous system (CNS) projection neurons fail to spontaneously regenerate damaged axons (Curcio and Bradke, 2018; Williams et al., 2020). Several approaches succeeded in promoting various extents of axonal regeneration, for example, after optic nerve crush (ONC) (Benowitz et al., 2017; Chun and Cestari, 2017; Ghaffarieh and Levin, 2012; Williams et al., 2020; Kim et al., 2018; Park et al., 2008; Trakhtenberg et al., 2018; Moore et al., 2009). Nevertheless, even in the approaches targeting potent pro-growth tumorigenic factors, only a rare subset of the axons regenerates the full-length (Gokoffski et al., 2020; de Lima et al., 2012; Lim et al., 2016). A number of developmentally regulated genes have been found to underlie the developmental decline (Chen et al., 1995; Goldberg et al., 2002) in intrinsic capacity of retinal ganglion cells (RGCs) (and other CNS projection neurons; Poplawski et al., 2020) to grow axons (Apara et al., 2017; Blackmore et al., 2012; Trakhtenberg et al., 2014, 2018; Moore et al., 2009). However, the tumor suppressor gene, Pten is one of the most potent gene regulators of axon regeneration discovered to date. Pten suppresses axon regeneration through inhibition of the mTOR pathway (Park et al., 2008), and Pten knockout (KO) was shown to promote various extents of axon regeneration from the RGCs that included a subset of αRGCs (Duan et al., 2015; Jacobi et al., 2022). Although experimental gene therapy knockdown (KD) of Pten expression in adult RGCs promotes long-distance (i.e. at least the full length of the optic nerve) axon regeneration in a small subset of RGCs (Park et al., 2008; Kim et al., 2018; Yungher et al., 2015), it is concerning for clinical use (Keniry and Parsons, 2008), and safer downstream effectors of Pten KD and mTOR pathway regulation are being investigated (Duan et al., 2015; Li et al., 2016; Bray et al., 2019; Lukomska et al., 2021; Park et al., 2010). Here, we used a new method for capturing specifically the rare long-distance axon-regenerating RGCs for single-cell RNA-sequencing (scRNA-seq) analysis, and also compared their transcriptomes with embryonic RGCs. Our approach enabled us to: (1) investigate why, despite Pten KO in all RGC types, some do not regenerate axons, others (such as αRGCs) regenerate short-distance axons (Jacobi et al., 2022), and only a rare subset regenerates long-distance axons (Duan et al., 2015), and (2) characterize the relationship between long-distance axon regeneration promoted by targeting Pten and the developmental decline in intrinsic axon growth capacity.

Multiple experimental approaches stimulate short-distance (i.e. up to 1.5 mm past the injury site) axon regeneration, but few stimulate long-distance axon regeneration (i.e. full-length of the optic nerve or longer). Moreover, even in the approaches which lead to long-distance axon regeneration, only a rare subset of RGCs regenerates axons 3 mm or longer, whereas the majority of the responding RGCs regenerate axons only a short-distance and then stall growth (Duan et al., 2015; Li et al., 2016; Bray et al., 2019; Kim et al., 2018). Because long-distance axon regeneration could provide insights into the mechanisms of full-length axon regeneration, we developed a novel surgical technique (that allows visual confirmation of appropriate targeting) for injecting CTB into the end of the optic nerve ∼3 mm distally from the injury site (as opposed to proximally, at 1.5 mm, as was done in another study; Jacobi et al., 2022), which enables retrograde labeling of the long-distance axon-regenerating RGCs, thereby prioritizing the identification of the rare subset of long-distance axon-regenerating RGCs over capturing many RGCs and mostly those that regenerate only short-distance axons.

Then, using scRNA-seq, we analyzed RGCs that responded to Pten KD by regenerating long-distance axons (i.e. full-length of the optic nerve, ∼3 mm from the injury site, or longer). Pten was knocked down using intravitreally injected adeno-associated virus serotype 2 (AAV2), which preferentially transduces RGCs and expresses anti-Pten short hairpin RNAs (shRNAs) known to promote axon regeneration after ONC (Yungher et al., 2015; Kim et al., 2018; Li et al., 2016; Ribeiro et al., 2020). Two weeks after ONC, long-distance axon-regenerating RGCs were isolated by fluorescence-activated cell sorting (FACS) from retinal cell suspension. At 12 h before sacrifice, Alexa Fluor-488-conjugated Cholera toxin subunit B (CTB) axonal tracer was injected into the optic nerve 3 mm distally from the ONC site. CTB was retrogradely transported to the RGC soma in the retina via long-distance regenerated axons. Only a small subset of the Pten KD-treated RGCs (identified by the expression of mCherry reporter) were CTB⁺ (Fig. 1A-D). No CTB⁺ RGCs were found in the retinas treated with the control vector expressing scrambled shRNAs and an mCherry reporter (Fig. 1E), as there were no regenerated axons in the control to uptake the CTB that was injected distally from the injury site. An mCherry reporter (identifying AAV2-transduced cells) and Alexa Fluor-488 (identifying long-distance regenerating RGCs) double-positive RGCs were isolated by FACS (Fig. 1A-E). The transcriptomes of 101 long-distance axon-regenerating RGCs (that passed quality control; see Materials and Methods) is a statistically appropriate representative sample of the target population comprised of only ∼90 RGCs per retina (representing 0.02% of the total retinal RGCs) that respond to Pten KD by regenerating long-distance axons at 2 weeks after ONC (Yungher et al., 2015; Kim et al., 2018) (101 is also comparable in number with 120 Pten KO short-distance axon-regenerating RGCs from a recent study; Jacobi et al., 2022).

We compared the transcriptomes of RGCs that regenerated long-distance axons in response to Pten KD with adult uninjured and injured RGC scRNA-seq transcriptomes (Tran et al., 2019; Rheaume and Trakhtenberg, 2022 preprint). The long-distance axon-regenerating RGCs expressed canonical pan-RGC markers (Rbpms, Slc17a6, Tubb3 and Sncg) in a similar range as uninjured or injured (non-treated) RGCs (Fig. 1F). We then analyzed developmental regulation and KD efficiency of Pten gene expression, by comparing adult Pten KD-treated and untreated with embryonic RGCs (Lo Giudice et al., 2019). Expression of Pten was substantially upregulated during RGC maturation from embryonic to adult, and KD of Pten by AAV2 expressing anti-Pten shRNAs substantially reduced Pten expression in the Pten KD long-distance axon-regenerating RGCs to below embryonic level (Fig. 1G). The Pten KD long-distance axon-regenerating RGCs also segregated into two clusters (Fig. 2A).

The embryonic retina contains RGCs at various stages of development, as RGCs are born on different days in the embryonic retina (Prasov and Glaser, 2012; Bhansali et al., 2014; Lo Giudice et al., 2019). Together with the adult RGC scRNA-seq atlas (Tran et al., 2019; Rheaume and Trakhtenberg, 2022 preprint), the embryonic RGC scRNA-seq dataset (Lo Giudice et al., 2019) enabled generation of a developmental pseudo-timeline spanning the progression from embryonic into adult cell states (Fig. 2B,C). In analyzing the relative transcriptomic changes progressing from embryonic into adult RGC state (along the developmental pseudo-timeline), we unexpectedly found that some adult RGC subtypes remain more similar to the embryonic state, whereas others change more during maturation (Fig. 2C). We then identified marker genes, Gal, Fxyd6 and Gnb4, that are expressed in both embryonic RGCs and in adult RGC subtypes C33/C40 that are the most similar to embryonic RGCs, which further supports the existence of adult subtypes that retained features of embryonic cell state (Fig. 2D).

We then bioinformatically mapped (see Materials and Methods) the Pten KD long-distance regenerating RGCs to the UMAP of embryonic and adult RGC atlas, and found that almost all Pten KD long-distance regenerating RGCs were assigned to embryonic RGCs and to adult M1 intrinsically photosensitive (ip) RGC clusters C33/C40 that are the closest to embryonic RGCs (Fig. 2E). Specifically, Pten KD long-distance regenerating RGC cluster A mapped to cluster C40 and embryonic RGCs, whereas Pten KD long-distance regenerating RGC cluster B mapped to cluster C33 (Fig. 2F). When the Pten KD long-distance axon-regenerating RGCs are bioinformatically mapped only to the adult atlas RGC UMAP (without embryonic RGCs), almost all (96%) are assigned only to the non-α M1 ipRGC clusters C33/C40. By contrast, using the same method to bioinformatically map the Pten KO mostly short-distance axon-regenerating RGCs to the atlas RGC UMAP, the cells are assigned primarily to non-ip αRGCs, consistent with the original report from which the scRNA-seq dataset on the Pten KO short-distance axon-regenerating RGCs was obtained (Jacobi et al., 2022) (Fig. S1). Modest upregulation of only one αRGC marker Spp1 in the C33/C40 Pten KD long-distance axon-regenerating is consistent with Pten KO upregulating Spp1 (Kim et al., 2014; Sato et al., 2006; Lu et al., 2013) (Fig. S1D).

We also found that transcriptomes of untreated adult RGCs overall, and particularly transcriptomes of 19 clusters that survived 2 weeks after injury (Rheaume and Trakhtenberg, 2022 preprint) (Fig. 2G,H), partially reverted towards embryonic state, but not as close as the Pten KD long-distance regenerating RGC transcriptomes (which are the closest to the embryonic state; Fig. 2G). Moreover, C33/C40 subtypes to which the Pten KD long-distance axon-regenerating RGCs mapped, are the closest to embryonic state even in an uninjured state (Fig. 2H), and the proportion of Pten KD long-distance regenerating RGCs that mapped to these clusters is substantially enriched compared with the proportion of C33/C40 in the RGC atlas (Fig. 2I).

The existence of subtypes that retained features of embryonic cell state raises the possibility that they might be the RGCs which regenerated long-distance axons in response to Pten KD, particularly considering that the long-distance regenerating RGCs transcriptomically mapped almost exclusively to these embryonic-like adult RGC subtypes. C33/C40 are M1 ipRGCs within a subset of Opn4⁺ clusters – C7, C8, C43, C22, C31, C40 and C33 – the last four of which are ipRGCs (Tran et al., 2019). Although all Opn4⁺ clusters (Fig. 2J) are closer to embryonic state on the pseudo-timeline compared with all other clusters (Fig. 2H), the RGCs which regenerated long-distance axons in response to Pten KD mapped to the two closest to embryonic state ipRGCs C33/C40 (Fig. 2H,I). Pten KD long-distance axon-regenerating RGCs mapping to C33/C40 is partially consistent with the hypothesis that RGC types that are more resilient to injury (Pérez de Sevilla Müller et al., 2014; Tran et al., 2019; Rheaume and Trakhtenberg, 2022 preprint) are more responsive to Pten inhibition for regenerating axons (Li et al., 2016; Duan et al., 2015; Bray et al., 2019), but a recent study (Jacobi et al., 2022) identified αRGCs as the primary type responding to Pten KO by regenerating at least short-distance axons (Fig. S1). However, RGCs which regenerated long-distance axons in response to Pten KD appear to have originated from the non-α ipRGC subtypes C33/C40. It is also possible that other RGC subtypes responded to Pten KD but dedifferentiated towards an embryonic state (which facilitated long-distance axon regeneration) and thus transcriptomically mapped to embryonic-like C33/C40.

To resolve between these possibilities, we analyzed whether Pten KD long-distance axon-regenerating RGCs are enriched only for C33/C40 markers that are also enriched in embryonic RGCs (which could be a consequence of any subtype upregulating embryonic genes during dedifferentiation), or whether they are also enriched for C33/C40 markers that are unique to a mature cell state and not expressed in embryonic RGCs (which would suggest origination from C33/C40). We found that a substantial portion of the differentially expressed genes (DEGs) in the Pten KD long-distance regenerating RGCs partially reverted their levels of expression towards an embryonic RGC state (Table S1, and see Fig. 3C), while retaining the expression of embryonic RGC and embryonic-like adult RGC subtype markers, Gal, Fxyd6 and Gnb4 (Fig. 2D). We also found that C33/C40 markers not expressed in embryonic RGCs (Rheaume and Trakhtenberg, 2022 preprint) are enriched in the Pten KD long-distance axon-regenerating RGCs, namely Adra2a, Bmp7 and Rasgrp1, as well as Baiap3, Ucp2 and Xylt1, injury-induced downregulation of which was rescued by Pten KD (Fig. 3A,B). These data suggest that the long-distance axon-regenerating RGCs are more similar to embryonic-like C33/C40 because they originated from these subtypes, and that this is not a consequence of dedifferentiation from other subtypes.

Next, we identified groups of up- and downregulated genes, which partially reverted their expression towards an embryonic RGC state, as well as those that are not developmentally regulated (Fig. 3C and Table S1). Axon regeneration may require recapitulation of the molecular mechanisms involved in developmental embryonic axon growth (Filbin, 2006; Hilton and Bradke, 2017; Yaniv et al., 2012; Harel and Strittmatter, 2006; Goldberg et al., 2002; Chen et al., 1995; Poplawski et al., 2020; Moore et al., 2009). However, the adult CNS environment is different from embryonic and is further altered by lesion. For example, the glial scar and immune cells (that responded to lesion) were not present along the developmental axonal path, and therefore axon regeneration may also require targeting the molecular mechanisms that were not involved in developmental axon growth (Kim et al., 2018; Chen et al., 2000; Shen et al., 2009; Geoffroy and Zheng, 2014; Low et al., 2008; Yiu and He, 2006). DEG analysis of Pten KD long-distance axon-regenerating RGCs revealed patterns of gene expression that are consistent with both approaches (Fig. 3C). Therefore, we selected representative DEG candidates consistent with each approach for testing in the axon regeneration assay.

We selected Dynlt1a, which reverted its expression towards an embryonic RGC state (Fig. 3D), and the non-developmentally-regulated Lars2, which was upregulated in long-distance axon-regenerating RGCs but otherwise expressed at a relatively low basal level across conditions (Fig. 3E). We focused on these genes, because they could regulate mitochondrial dynamics involved in axonal regeneration (Luo et al., 2016; Han et al., 2016; Kreymerman et al., 2019; Zhou et al., 2016; Cartoni et al., 2016; Cheng et al., 2022). Dynlt1a is involved in bi-directional cargo (e.g. mitochondria; Fang et al., 2011) transport (Ligon et al., 2004) and regulates neurite growth in culture (Chuang et al., 2005). Lars2 is a nuclear DNA (nucDNA)-encoded aminoacyl-tRNA synthetase (aaRS), required for translation of mitochondrial DNA (mtDNA)-encoded proteins by aminoacylation with Leucine (L) to make the mtDNA-encoded tRNA-L (Sohm et al., 2004). MtDNA encodes the full set of tRNAs required for translation of mtDNA-encoded genes (which are not compensated for by the nucDNA-encoded tRNAs), and nucDNA encodes the full set of aaRSs that aminoacylate mtDNA-encoded tRNAs with their cognate amino acids (which are not compensated for by the aaRSs that aminoacylate the nucDNA-encoded tRNAs) (González-Serrano et al., 2019; Wang et al., 2021). We found that all aaRSs specific to the mt-tRNAs are differentially expressed in the Pten KD long-distance regenerating RGCs, with Lars2 being the most highly upregulated (Fig. 3F). Moreover, Lars2-aminoacylated L is significantly more enriched than any other amino acid in all 13 mtDNA-encoded proteins (Fig. 3G,H). Therefore, we hypothesized that involvement of Lars2 and Dynlt1a in regulation of axonal mitochondrial dynamics may render them plausible downstream effectors of Pten KD for promoting axon regeneration.

We then tested whether Dynlt1a and Lars2 are sufficient to promote axon regeneration, using an established assay (Park et al., 2008; Kim et al., 2018; Yungher et al., 2015; Lukomska et al., 2021; Trakhtenberg et al., 2018). AAV2 vectors expressing Dynlt1a, Lars2 or mCherry control, were injected intravitreally. ONC was performed 2 weeks later. To visualize the regenerating axons or their absence, CTB was injected 1 day before sacrifice at 2 weeks after ONC. The number of regenerating axons and RGC survival was quantified (see Materials and Methods for details; experimental timeline in Fig. 4A). We found that Dynlt1a and Lars2 promoted axon regeneration at least 2 mm and 3 mm past the injury site, respectively, compared with only minor axonal sprouting (as expected) in control (Fig. 4B,C). No spared axons were detected in either group. Dynlt1a (but not Lars2) also promoted RGC survival compared with the control group (Fig. 4D,E).

To gain further insight into the mechanisms through which Lars2 and Dynlt1a promote axon growth, we analyzed the gene network upregulated in the Pten KD long-distance axon-regenerating RGCs. We found that the biological processes most co-enriched in the gene network (involving Lars2 and Dynlt1a) upregulated in the Pten KD long-distance axon-regenerating RGCs were related to mitochondria, axonal growth and neurodevelopment (Fig. 5A,B). Furthermore, our finding that Lars2 promotes axon regeneration, which belongs to the gene ontology biological process (GO:BP) ‘positive regulation of neuronal axonal projection’, has linked it to the GO:BP ‘mitochondrial translation’ (under which Lars2 was previously annotated) (Fig. 5A,B). Thus, a cross-talk between mitochondrial translation and axonal growth processes, and involvement of a co-upregulated gene network, may underlie our finding that Lars2 and Dynlt1a promote axon regeneration independently from each other.

The failure of CNS projection neurons to spontaneously regenerate long-distance axons after injury or in neurodegenerative disease presents a major medical problem (Curcio and Bradke, 2018; Williams et al., 2020), as even in the approaches targeting clinically-concerning tumorigenic factors, only a rare subset of axons regenerate the full-length in animal models (Gokoffski et al., 2020; de Lima et al., 2012; Lim et al., 2016). To tackle this problem, studies have shown that the intrinsic axon growth capacity, which declines during maturation in the mammalian CNS (Goldberg et al., 2002), can be reactivated to some extent by targeting neuronal developmentally regulated genes (e.g. Klf4, Klf9) (Trakhtenberg et al., 2018; Moore et al., 2009), whereas axonal injury itself tilts the transcriptome of adult CNS neurons towards an embryonic state (Poplawski et al., 2020) (although failing to elicit axon regeneration). On the other hand, targeting the tumor suppressor Pten (Park et al., 2008) promotes various extents of axon regeneration, mostly short-distance axons from αRGCs and other RGC types (Jacobi et al., 2022), but also long-distance (i.e. full-length of the optic nerve or longer) regeneration from a rare subset of α and/or ip RGC subtypes (Duan et al., 2015).

Here, we demonstrate that, although bulk-RNA-seq has previously left unresolved whether injury itself tilts the transcriptome of all or only some adult CNS neuronal subtypes towards an embryonic state (Poplawski et al., 2020), scRNA-seq-enabled cluster-specific analysis revealed that only a subset of RGC subtypes reverts their transcriptomes towards an embryonic state. However, marginal dedifferentiation by injury alone fails to elicit spontaneous axon regeneration (as although neurons attempt, they fail to regenerate axons after injury; Sellés-Navarro et al., 2001). Unexpectedly, we also found the existence of adult RGC subtypes that retained features of an embryonic cell state, and identified Gal, Fxyd6 and Gnb4 as marker genes that are expressed in embryonic RGCs and then downregulated during maturation in all RGC subtypes, except a subset of Opn4⁺ RGCs, primarily subtypes C33 and C40 (that are transcriptomically more similar to embryonic RGCs than any other RGC subtype). We then showed that Pten inhibition-promoted long-distance (i.e. full-length of the optic nerve or longer) axon regeneration is associated with partial dedifferentiation towards an embryonic state of the responding M1 ipRGCs from subtypes C33 and C40. It is possible that these subtypes are able to respond to Pten KD by regenerating axons long-distance, because during maturation they retained features of an embryonic cell state (in contrast to other RGC subtypes), which enhanced their axon regeneration response to Pten KD. It would be important to investigate in future studies: (1) whether features of an embryonic cell state retained in the C33 and C40 subtypes are necessary for their enhanced axon regeneration response to Pten KD; (2) whether these features in the C33 and C40 subtypes also enhance axon regeneration promoted by other treatments (e.g. Klf9 KD); (3) whether experimentally reinstating features of an embryonic cell state in other RGC subtypes would also enhance their axon regeneration in response to Pten KD or other treatments.

We also identified novel mitochondrial factors involved in axon regeneration (Luo et al., 2016; Han et al., 2016; Kreymerman et al., 2019; Zhou et al., 2016; Cartoni et al., 2016; Cheng et al., 2022). Dynlt1a is involved in bi-directional axonal transport of mitochondria along microtubules (Fang et al., 2011; Ligon et al., 2004; Chuang et al., 2005), which may be needed to supply mitochondria-generated energy (and ‘building materials’) for assembling the regenerating axonal segments. We found that Lars2-aminoacylated L (to mt-tRNA-L) is the most prevalent amino acid in (and therefore Lars2 is a limiting factor for production of) all mtDNA-encoded proteins, and Lars2 is the most highly upregulated mt-tRNA-specific aaRS in the Pten regenerating RGCs. Thus, Lars2 may enable synthesis of axonal mitochondria that are needed to supply energy for assembling the regenerating axonal segments. We also provided further insight into the mechanisms through which Lars2 and Dynlt1a promote axon growth, by showing that the biological processes most co-enriched in the gene network (which involved Lars2 and Dynlt1a) upregulated in the Pten KD long-distance axon-regenerating RGCs were related to mitochondria, axonal growth and neurodevelopment. Furthermore, our finding that Lars2 promotes axon regeneration, which belongs to the gene-ontology biological processes (GO:BP) ‘positive regulation of neuronal axonal projection’, has linked this biological process to the GO:BP ‘mitochondrial translation’ (under which Lars2 was previously annotated). Thus, a cross-talk between mitochondrial translation and axonal growth processes, and involvement of a subset of co-upregulated gene network, may underlie the ability of Lars2 or Dynlt1a to promote axon regeneration independently from each other. It will be important to investigate the roles of Lars2 and Dynlt1a in mitochondrial dynamics of regenerating axons, and whether co-targeting Lars2 and Dynlt1a will lead to a more robust axon regeneration than targeting each factor alone. Future studies also need to test whether over the long-term (i.e. longer than 2 weeks after ONC) axon regeneration achieved by Lars2 and Dynlt1a has the potential to regenerate beyond the optic chiasm, through the optic tract, and potentially to the postsynaptic target neurons in respective brain regions.

Our experimental approach prioritized identification of the rare subset of RGCs that regenerates long-distance (∼3 mm from the ONC site or longer) axons, rather than capturing many RGCs, and mostly those that regenerate axons only over a short distance (up to 1.5 mm from the ONC site). In order to reliably inject axonal tracer into the end of the optic nerve 3 mm beyond the injury site, we developed a surgical technique for appropriate targeting with visual confirmation (see Materials and Methods). Our approach yielded insights into long-distance axon regeneration that are more relevant for providing clues to developing axon regeneration treatments, and the non-tumorigenic Lars2 and Dynlt1a (identified through this approach) are viable candidates for the development of axon regeneration therapies. The experimental approach in another study injected retrograde tracer proximally to the injury site (1.5 mm away), which enabled the capture of many RGCs that regenerate short-distance axons (Jacobi et al., 2022), and did not use comparative analysis with embryonic RGCs as we did in our approach. Therefore, that study revealed different factors and insights than reported here. For example, it showed that primarily (82%) αRGCs (and 18% of multiple other RGC subtypes of which non-α ipRGCs represented only ∼1.8%; see figure 2G in that study) respond to Pten KO by regenerating axons short distance, whereas our study revealed that primarily non-α M1 ipRGC subtypes C33 and C40 respond to Pten KD by regenerating long-distance axons. Comparative analysis between short-distance axon-regenerating RGCs from that study (Jacobi et al., 2022) and long-distance axon-regenerating RGCs identified herein is shown in Fig. S1. However, this analysis did not include cells from another similar study (Li et al., 2022) that also found short-distance axon-regenerating genes, because retrospective analysis found those cells to be microglia/macrophages (Theune and Trakhtenberg, 2023 preprint). Also, Gal, which has been previously implicated in peripheral nervous system axon regeneration (Holmes et al., 2000), was one of the few identified RGC axon regeneration-promoting genes in that study (Jacobi et al., 2022), whereas we identified Gal as one of the markers of adult RGC subtypes C33/C40 that retained features of an embryonic cell state, which may have enabled enhanced long-distance axon regeneration in response to Pten KD. Gal is enriched in RGC subtypes C33/C40 nearly 3-fold relative to αRGCs and even more relative to other RGC subtypes in the uninjured RGC atlas. However, after ONC and Pten KO, Gal is also upregulated in αRGCs, although these cells did not dedifferentiate towards embryonic cell state.

Our study suggests that the degree of similarity of a neuron to the embryonic transcriptomic state is a predictor of its responsiveness to axon regeneration treatments, and that the adult neuronal subtypes that retain embryonic cell state features may have enhanced long-distance axon regeneration response to treatments. Considering that Pten KD downstream factors, Dynlt1a and more so Lars2, achieved axon regeneration ∼full-length of the optic nerve by 2 weeks after injury (despite fewer short-distance regenerating axons compared with Pten KD; Yungher et al., 2015; Kim et al., 2018), and considering that targeting other factors in the Pten pathway also promotes axon regeneration (even without complementary co-treatments) (Li et al., 2016; Bray et al., 2019; Lukomska et al., 2021), it is possible that several effectors may promote axon regeneration through tapping into subtype-specific and non-subtype-specific pathways, which could tilt the transcriptome towards an embryonic transcriptomic state. For example, we found that expression of the axon regeneration-facilitating Tet1 demethylase (which is upregulated by a dedifferentiation treatment co-expressing Oct4/Sox2/Klf4/Myc) (Lu et al., 2020) is substantially downregulated in the injured RGCs, but its expression is preserved (and even modestly upregulated) in the long-distance axon-regenerating RGCs that responded to Pten KD. Moreover, the Sox2 component of that treatment, which is not expressed in injured or uninjured RGCs, is upregulated in the long-distance axon-regenerating RGCs that responded to Pten KD. Furthermore, several other known axon regeneration-promoting genes were also co-upregulated in the long-distance axon-regenerating RGCs by Pten KD, including Braf (O'Donovan et al., 2014), Akt3 (Campion et al., 2022), Igf1R (Dupraz et al., 2013) and Sprr1a (Bonilla et al., 2002).

Taken together, our findings reveal the existence of adult neuronal subtypes that retained features of the embryonic cell state, demonstrate that the RGCs which regenerate long-distance axons in response to Pten KD originate from these embryonic-like adult subtypes and dedifferentiate towards an embryonic cell state, and identify novel axon regeneration-promoting genes, which suggest that mitochondrial protein synthesis may be rate-limiting in axon regeneration.

All animal studies were performed at the University of Connecticut Health Center with approval of the Institutional Animal Care and Use Committee and the Institutional Biosafety Committee, and performed in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Visual Research. Mice were housed in the animal facility with a 12 h light/12 h dark cycle (lights on from 07:00-19:00) and a maximum of five adult mice per cage. The study used wild-type 129S1/SvImJ mice (The Jackson Laboratory). Optic nerve surgeries and injections, and intravitreal injections, were carried out on mice of both sexes at 8-12 weeks of age (average body weight 20-26 g) under general anesthesia, as described previously (de Lima et al., 2012; Trakhtenberg et al., 2018; Kim et al., 2018; Lukomska et al., 2021). The viruses included AAV2 vectors expressing anti-Pten shRNA or scrambled shRNA control (both co-expressing mCherry reporter and using published sequences; Yungher et al., 2015; Kim et al., 2018; Lukomska et al., 2021), as well as Dynlt1a [open reading frame (ORF) of ENSMUST00000169415.2], Lars2 (ORF of ENSMUST00000038863.8), and mCherry (titers ∼1×10¹² GC/mL; VectorBuilder). Viruses (2 μl per eye) were injected intravitreally in 8-week-old mice, avoiding injury to the lens, 2 weeks prior to ONC surgery. Mice were randomly assigned to experimental or control conditions. This lead time allowed for sufficient transduction and expression of the shRNAs and transgenes in RGCs at the time of ONC. Transduction efficiency was ∼30%, which is similar to previous reports that also used AAV2 to target the RGCs (Yungher et al., 2015; Kim et al., 2018; Lukomska et al., 2021; Trakhtenberg et al., 2018, 2014).

To label the RGCs that regenerated axons in response to treatment with anti-Pten shRNA, axonal tracer CTB conjugated to Alexa Fluor 488 dye (C34775, Thermo Fisher Scientific) was injected (1% CTB in 1 μl PBS) into the end of optic nerve, ∼3 mm from the ONC injury site, 12 h before the enrichment of the RGCs from retinal single-cell suspension by immunopanning for Thy1 (as described previously; Trakhtenberg et al., 2014; Rheaume et al., 2018) for FACS (using BD Biosciences FACSAria cell sorter with FACSDiva v8.0 software) at 2 weeks following ONC. Enrichment of RGCs by immunopanning was necessary, because FACS of millions of retinal cells (from 10 retinas) would adversely affect primary neurons such as RGCs, which comprise <1% of all the retinal cell types (Jeon et al., 1998; Masland, 2012; Smith and Chauhan, 2015) (e.g. recovery of RGCs from whole retinas for scRNA-seq was limited to only 432 RGCs from multiple retinas; Macosko et al., 2015). Cells enriched for RGCs by immunopanning were then subject to FACS for mCherry⁺/CTB⁺ cells, and immediately processed by plate-based droplet scRNA-seq (see below). To inject CTB into the end of the optic nerve, the anesthetized mouse skull was exposed, drilled on the Bregma, and cut along the coronal suture line. Then, ∼5 mm width of the bone was removed laterally up to the cortical area, under which the optic nerve enters the optic chiasm. Parts of frontal cortex, striatum, thalamic nuclei and nucleus of the stria terminalis of amygdala were removed with a surgical spatula to expose the optic nerves along the anterior cranial fossa (taking care to minimize damage to the blood vessels) just enough to enable visualization of the optic nerve segment that enters the optic chiasm. To inject CTB, using stereotaxic apparatus, a 50 μm diameter glass needle (pulled from borosilicate glass capillaries BF150-86-10, World Precision Instruments, using Flaming/Brown P-97 micropipette puller) was guided and inserted into the optic nerve segment adjacent to the optic chiasm. After visually confirmed injection, coagulant was added to stop bleeding and the skin was sutured. Standard stereotaxic surgery analgesic regimen was administered. Mice were kept warm using a heating-pad, food and water placed in a Petri dish, and mice were checked regularly through to sacrifice 12 h later.

Standard histological procedures were used, as described previously (Kim et al., 2018; de Lima et al., 2012; Kurimoto et al., 2010; Trakhtenberg et al., 2018; Lukomska et al., 2021). Briefly, anesthetized mice were transcardially perfused with isotonic saline followed by 4% paraformaldehyde (PFA) at 2 weeks after ONC, the eyes and optic nerves were dissected, postfixed for 2 h, the retinas were dissected-out and optic nerves were transferred to 30% sucrose overnight at 4°C. The optic nerves were then embedded in OCT Tissue Tek Medium (Sakura Finetek), frozen, cryostat-sectioned longitudinally at 14 µm and then mounted for imaging on coated glass slides. Free-floating retinas were immunostained in 24-well plate wells and, after making four symmetrical slits, flat-mounted on coated glass slides for imaging. For immunostaining, free-floating retinas were blocked with the appropriate sera, incubated overnight at 4°C with primary anti-βIII-Tubulin (1:500, rabbit polyclonal; Abcam, Ab18207) antibody, then washed three times in PBS, incubated with fluorescent secondary antibody (1:500; Alexa Fluor, A21207, Thermo Fisher Scientific) for 4 h at room temperature, washed three times again in PBS and mounted for imaging. Images of the regenerating axons in the optic nerve and surviving RGCs in Fig. 4 were acquired using a fluorescent microscope (Zeiss, AxioObserver.Z1). Retinal flatmount images of RGCs positive for CTB and/or mCherry in Fig. 1 were acquired using a confocal microscope (Zeiss, Confocal LSM800).

To visualize the regenerating axons or their absence after treating with the viral vectors expressing Dynlt1a, Lars2 or mCherry control, axonal tracer [Alexa Fluor 488-conjugated CTB (C34775, Thermo Fisher Scientific) 1% in 3 μl PBS] was intravitreally injected 1 day before animals were euthanized 2 weeks following ONC. Longitudinal sections of the optic nerve were examined for possible axon sparing (Kim et al., 2018). No spared axons were found in control, and no evidence of axon sparing was found in experimental conditions (i.e. at 2 weeks after injury, no axons were found distal from the injury region of the optic nerve). Regenerated axons (defined as continuous fibers, which are absent in controls and are discernible from background puncta and artefactual structures) were counted manually using a fluorescent microscope (Zeiss, AxioObserver.Z1) in at least four longitudinal sections per optic nerve at 0.5 mm, 1 mm, 1.5 mm, 2 mm and 3 mm distances from the injury site (identified by the abrupt disruption of the densely packed axons near the optic nerve head, as marked by a rhombus in Fig. 4B), and these values were used to estimate the total number of regenerating axons per nerve, as previously described (de Lima et al., 2012; Kim et al., 2018; Trakhtenberg et al., 2018; Lukomska et al., 2021). RGC survival was quantified in retinal flatmounts as previously described (Kim et al., 2018; de Lima et al., 2012; Kurimoto et al., 2010; Trakhtenberg et al., 2018; Lukomska et al., 2021) by immunostaining with an antibody to βIII-Tubulin (neuronal marker Tuj1; encoded by Tubb3) and counterstained with DAPI to label the nuclei, taking advantage of the selective expression of βIII-tubulin in RGCs. ImageJ software was used to count βIII-Tubulin⁺ cells from images taken at 1-2 mm from the optic nerve head in four directions, then averaged to estimate overall RGC survival per mm² of the retina. Investigators performing the surgeries and quantifications were masked to the group identity by another researcher until the end of the experiment.

Custom designed Drop-seq barcodes from Integrated DNA Technologies (IDT) were delivered into wells of two 384-well plates. All primers in one well shared the same unique cell barcode and billions of different unique molecular identifiers (UMIs). An Echo 525 liquid handler was used to sequentially dispense lysis buffer, primers (custom designed Drop-seq barcodes from IDT) and reaction reagents, totaling 1 μl, into each well in the plate for the cell lysis and cDNA synthesis using a modified Drop-seq/SmartSeq2 protocol. Following cDNA synthesis, the contents of each well were collected and pooled into one tube using a Caliper SciClone Liquid Handler. After treatment with exonuclease to remove unextended primers, the cDNA was PCR amplified for 13 cycles and then fragmented and amplified for sequencing using a Nextera XT DNA sample prep kit (Illumina) using custom primers (Table S2) that enabled the specific amplification of only the 3′ ends. Paired-end FASTQs were generated using BCL2FASTQ v2.18.0.12 (Illumina). A digital expression matrix was constructed for each pair of FASTQs using Drop-seq tools v1.13 (http://mccarrolllab.com/dropseq) as follows: Bam creation with Picard (v2.9.3) FastqToSam; cell and UMI tagging, filtering, trimming with Drop-seq tools TagBamWithReadSequenceExtended, FilterBAM, TrimStartingSequence, PolyATrimmer; alignment with STAR (v2.5.4a) to the mm10-1.2.0_genome and transcriptome from CellRanger (for comparisons with 10x Genomics datasets); sorting with Picard (v2.9.3) SortSam; merging and tagging with Picard (v2.9.3) MergeBamAlignment and Drop-seq_tools (v1.13) TagReadWithGeneExon; and a gene-cell expression matrix of raw 3′ end counts (in CSV format) was produced with Drop-seq_tools (v1.13) DigitalExpression. Cells not expressing RGC genes such as Rbpms and Tubb3, as well as poor quality cells or doublets, were excluded.

BAM files, raw counts, normalized matrices and cell metadata (e.g. type assignment) were obtained for the mouse embryonic RGCs from Gene Expression Omnibus (GEO) deposit GSE122466 (Lo Giudice et al., 2019), and for the mouse adult RGC atlas and the injured RGCs from the GEO deposit GSE137400 (Tran et al., 2019). BAM files were converted to FASTQ files using CellRanger bamtofastq software. FASTQ files were aggregated where appropriate using CellRanger and then mapped to the CellRanger mm10-1.2.0 transcriptome. Batch correction was performed for separate batches using the FindIntegrationAnchors and IntegrateData functions from Seurat v.4.0.3 (Stuart et al., 2019; Hao et al., 2021). The same cells that passed the original quality checks and the same cell-to-type assignments from the original analyses (Lo Giudice et al., 2019; Tran et al., 2019) were used in the present study. Normalized count matrices were obtained for the mouse Pten KO RGCs (120 cells in total) from the GEO deposit GSE202155 (Jacobi et al., 2022). Comparative analysis of scRNA-seq datasets was performed using the R package Seurat v.4.3.0 (see below).

The plate-based-generated SmartSeq2 scRNA-seq dataset (Pten KD long-distance axon-regenerating RGCs) was merged with the 10x Genomics single-cell platform-generated scRNA-seq datasets (embryonic, adult atlas and adult injured RGCs) using SAVER (Huang et al., 2018), with default parameters for the plate-based-generated scRNA-seq, for the downstream comparative analyses between the datasets. All datasets were normalized using Seurat's NormalizeData function with default parameters (Stuart et al., 2019; Hao et al., 2021). The sex-specific genes Xist, Eif2s3y and Ddx3y were excluded for dimensionality reduction but retained for downstream analyses. Embryonic and adult RGCs were aligned using the mutual nearest neighbors algorithm from Batchelor (Haghverdi et al., 2018) as part of the Monocle v.3 (Cao et al., 2019), which was used to generate the merged embryonic and adult atlas RGC UMAP. Monocle was also used to determine the pseudo-timeline structure of the merged (embryonic/adult RGC) UMAP. The UMAP cell embeddings, generated by Monocle, were transferred to a Seurat object containing the same datasets, and the hyperparameters (umap.n.neighbors=10; umap.metric=‘euclidean’; umap.min.dist=0.1; n_epochs=200; learning_rate=1; repulsion_strength=1; negative_sample_rate=5; approx._pow=0; spread=1) were used to generate a Seurat UMAP model. The CellTools algorithm (Rheaume and Trakhtenberg, 2022 preprint) was used to map the Pten KD regenerating RGCs to their cell type origins, and the injured and Pten KD long-distance axon-regenerating RGCs were individually assigned the same pseudo-timeline score as their nearest reference neighbor. The counts-matrix for the plate-based-generated SmartSeq2 scRNA-seq dataset of the 120 short-distance axon-regenerating RGCs was obtained from the GEO deposit GSE202155 (Jacobi et al., 2022) and merged as above with the 10x Genomics single-cell platform-generated scRNA-seq datasets using SAVER (Huang et al., 2018), with default parameters for the plate-based-generated scRNA-seq for the downstream comparative analyses between the datasets. Normalization, mapping to the reference UMAP and assignment of the pseudo-timeline scores for the short-distance axon-regenerating RGCs was also performed as above. Comparative analysis of scRNA-seq datasets was performed using the R package Seurat v.4.3.0.

Heatmaps were generated using Superheat (Barter and Yu, 2018), with the average expression of each gene per group scaled using z-scores, as previously published (Rheaume et al., 2018). The genes were ordered by the log2 fold-change between average expression in the Pten KD long-distance axon-regenerating RGCs and average expression in the injured control RGCs from clusters C33 and C40. Violin plots were generated using ggplot2 and Seurat's VlnPlot function (Stuart et al., 2019; Hao et al., 2021). Expression data for specific genes was extracted using Seurat's VlnPlot function. Violin plots were generated using the ggplot2 geom_violin function, and overlayed categorical scatter (violin point) plots were generated using ggbeeswarm (https://cran.r-project.org/web/packages/ggbeeswarm/).

Functional enrichment analysis was performed using the R package gprofiler2 on genes upregulated (determined using Seurat's FindMarkers function) in Pten KD long-distance axon-regenerating RGCs relative to injured untreated RGCs (excluding C33 and C40), with all genes expressed in injured RGCs set as background (Kolberg et al., 2020). False discovery rate (FDR) was used for multiple testing correction. A subset of GO:BP terms containing the genes Dynlt1a and Lars2 were plotted in a Gene-Concept Network Plot using the clusterProfiler and enrichplot R packages (Wu et al., 2021).

All tissue processing, quantification and data analysis were carried out masked throughout the study. Sample sizes were based on accepted standards in the literature and our previous experiences. Sample size (n) represents total number of biological replicates in each condition. All experiments included appropriate controls. No cases were excluded in our data analysis, although a few animals that developed a cataract in the injured eye were excluded from the study and their tissues were not processed. The data are presented as mean±s.e.m. and was analyzed (as specified in the applicable figure legends) by ANOVA with or without Repeated Measures and a posthoc LSD test (SPSS). Significance of enrichment fold-change in Fig. 2I was determined using the EdgeR algorithm (McCarthy et al., 2012). Significance for DEGs in the heatmaps in Fig. 3 was determined by independent samples Mann–Whitney U-test, using R software as previously published (Rheaume et al., 2018). All differences were considered significant at P<0.05.

Portions of this research were conducted at the High Performance Computing Facility, University of Connecticut. We are grateful to Paul Robson (The Jackson Laboratory for Genomic Medicine, Farmington) for single-cell RNA-seq service, Sophan Iv and Vijender Singh (Research IT Services, University of Connecticut) and Stephen King (High Performance Computing Facility, University of Connecticut) for assistance with bioinformatics tools. We thank Juhwan Kim, Tyler Steidl and Mahit Gupta (University of Connecticut School of Medicine) for technical assistance.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201644.reviewer-comments.pdf.